Sixteen test programs under tests/*_pins_test.ludic, on today's code, each one tick from a set
state through the public step (wildlife_tick_ground / _bird / wildlife_tick, wildlife_release,
wildlife_new): the state it goes to, its timer, and how many of the package's rolls it took and
which (the next roll is read off a newborn's yaw, compared with a stream of the same seed). IDLE,
WANDER, ALERT, WARY, FLEE, CHARGE, GO_DRINK, GO_FEED, DRINK, APPROACH, EAT, TRAPPED, FLY, LAND,
PERCH; the droppings' roll before them; the newborn's seven (eight for a bird); what the step
leaves alone. tests/pin/ holds the shared step, the dice reader and two species of its own (a tame
horse, a crow that never lands on its own). Compiled, not run here: 27.2 holds to them unchanged.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
A registry marked `@Machine(Deer.mood)` is the transitions of a machine over that enum field of the
records a state's Table<Deer> holds. Its record has from and to (the enum's variants), on: string (an
action's name, "" for a transition the tick asks), guard: fn(Row<Deer>, reads...) -> bool and
enter: fn(Row<Deer>, reads...) -> void; the states are the enum's variants and the start is the
field's default. The rows are data (an .lres or defs), the names the studio already edits.
Written by the compiler (machines.ludic, machines_write.ludic): for each action an `on` names, a row
reducer in the registry's file (named ..__machine__DeerSteps, so it sits beside the program's own
row reducer on the same action, after it): the row's state, the first transition from it on that
action whose guard passes, the field set, enter run - guards and enters called by name. When a row
leaves a state on a guard alone, `state DeerStepsMachine` (the kept row view) and
deer_steps_tick(m: mut DeerStepsMachine, s: mut Herd, reads...), one transition a row a tick.
Nothing allocates.
The table is the whole machine: the field written anywhere else - an assignment, or a `machine`
block's become over it - is a type error (check_stmt.ludic, ck_machine_write). Guards and enters take
the row first, are the record's module's, keep a row reducer's rules (and may be handed the row);
a guard writes nothing through it. The graph is checked, each error at its row (in the .lres when
the rows are there): a state never reached from the start, a state with no way out, an `on` naming
no action or an action with no @Target, a self-transition with no guard, two ways out of a state on
one trigger behind an unguarded first. Also refused: @Machine off a registry, a field that is not a
plain enum with a default, a @Column field, no table (or two) of the record, a transitions record of
another shape, a machine outside its table's state's module.
ludic schema's code section gains `machines` (registry, record, field, enum, table, start, states,
actions, tick, module, at); ludic deps names a machine's reducer `reducer Deer in Herd.deer on Spook
(machine DeerSteps)`. vocab @Machine; docs annot-machine, kw-machine; LANGUAGE.md "A machine as
data"; examples actions/machine (+ deer_steps.lres) and ten rejects; test.ludic feat, reject and
schema cases (not run); changes/machines.md. Reseeded; bootstrap-cfree fixpoint holds (317642
lines); Maroon Lake's `ludic build --check` is clean against this tree.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
An action names one row of a ludic.base Table<T> by its handle, in a field marked @Target, and
`reducer Deer in Herd.deer on Spook(r: mut Row<Deer>, n: Noise, a: Spook)` runs once, for that row
alone (the table may sit down a path, S.w.tab). The drain resolves the handle (tb_row) and hands the
reducer a Row<T> - new in ludic.base: tb, row, h, rec - that the queue keeps, one per row reducer,
filled in place, so a targeted action allocates nothing; a stale handle runs nothing, and
LUDIC_ACTIONS_LOG=1 prints a line for it (@alloc_ok). Row reducers order among an action's by their
state's name, then the table's path.
Checked at compile time (actions_rows.ludic): the row reaches r.rec and r.h only - r.tb / r.row
refused, the view never assigned, stored, copied or handed on except to a @RowVerb (a function of
the record's own module taking Row<T> first; any other function taking a row is refused); a field
marked @Column (a table column mirrors it) is not written through r.rec; only the module owning the
state declares a row reducer; one @Target, an int, per action; the states between the row and the
action are read. `mut` is allowed on a Row<T> parameter.
ludic schema's code section gains row_reducers (record, table, state, action, target, predicted,
net, module, at) and row_verbs (name, record, module, at), and every action its target; row
reducers are left out of `reducers`. ludic deps and ludic-lsp name a row reducer
`reducer Deer in Herd.deer on Spook`. vocab: @Target, @Column, @RowVerb; docs/language pages;
LANGUAGE.md "A reducer on a row"; examples actions/rows and ten rejects; test.ludic feat, reject and
schema cases (not run); changes/row-reducers.md. Reseeded; bootstrap-cfree fixpoint holds (307497
lines); Maroon Lake's `ludic build --check` is clean against this tree.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
gg_cover_json and gg_solid_json ended in plain strings with "}}", which only a template literal reads as one
brace: the cover's closed the top-level object before "solid", and the solid's left a stray "}" at the end.
The committed golden is fixed by hand to what the generator now writes.
ground_fill_test also checks the golden as JSON: one value with nothing after it (Json.parse reads the
first value and ignores the rest, so it alone would not have caught this), an object, its eleven sections
in order. The generator notes what the studio asked: past the density patch an empty tile reads 0 for R and
G (only past the map's edge does a texel clamp), and several clearings multiply.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
tests/ground_fill_test.ludic compares gg_json (gf_hash over fixed inputs, Math.lerp cases whose order
matters, the yaw as model.vert turns by it, a hand-written density, one cover chunk with a clearing and
one solid chunk) line for line with tests/ground_fill_golden.json, and checks ground_fill draws exactly
the candidates and a solid layer's far bands a subset of band 0. The golden is written by
tests/gen/ground_fill_golden.ludic, run from the repository root; it is not committed here.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
ground_candidate answers (layer, chunk, band, cell) -> keep, x, z, scale, yaw, seed, wind from pure gf_*
steps and the density read between them; ground_fill draws its answers with the same operations in the
same order. A solid layer fills at step0 and band 0 whatever the camera, a far band drawing a stable
subset (draw 7 under (step0/step_b)^2), each thing an int id from (layer, chunk, cell), listed per chunk
into a caller's GroundSolids or answered by id. r3d_ground_clearing hands the trample over as discs.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Since the data's text became keys (26.4), a game's English patterns that take a table's name (Notify, txt_pat, a concatenation) drew the raw mark and key.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
remove is the inverse of add: the require line leaves package.ludic; the lock keeps exactly what the
remaining requires still reach, read from the store's copy of each locked package.ludic (no
network), so a package another still requires stays locked and what only the removed one brought in
leaves with it; each leaving package loses the ludic_modules/ symlink add made, never the shared
store entry. Refused (exit 1) when package.ludic does not require the module; source still
importing a removed package is a warning. With no store copy to read, only the named module leaves.
get --json diffs the lock before and after in memory and prints {added, removed, changed,
unchanged} on stdout (an entry as the lock records it: name, version, hash, kind, provides; a change
as name, from, to, from_hash, to_hash), the resolver's lines on stderr. Cases added to test-pkg,
not run.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
ludic-fmt --lint --json prints the violations --lint reports as one JSON array on stdout,
[{file, line, col, rule, message}] ordered by file, line and column (col where the rule knows it),
the summary on stderr, --lint's exit status, and never rewrites the baseline. ludic-fmt - refuses a
buffer that does not read as Ludic (a string/template/key literal left open, a bracket never closed
or closed by the wrong one) with exit 2 and name:line:col on stderr; - --lint judges a buffer as the
file --stdin-name names (--stdin-rel: that path relative to the project), against its baseline and
lint paths. ludic fmt --lint and ludic fmt - run it from the nearest package.ludic upwards (from
--stdin-name's directory when given). Hooks read nothing and write to stderr under --json or -.
Regression cases added to test-tools and ludic-dev test (fmt_editor_cases), not run.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
stdin is read once, whole, and read_file serves a copy of it wherever the program opens <path> -
the entry, an import through a barrel, a component's .xml / .lss, an .lres. Paths match after
normalising both ('/' separators, relative under $PWD, . / .. / // folded, case on Windows);
diagnostics keep the file's usual name, with the buffer's lines and columns. A <path> nothing
opens is one warning. On ludic build it implies --check. Reseeded; bootstrap-cfree fixpoint holds.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Keywords: module uses friend export internal numbers unsafe mut port bind action
reducer registry of as from def open alias component prop view (structure),
shows lasts then loads (scenes), system (ecs), dispatch (control), true false
null (operators). Attributes: @Ref @OneOf @Range @Unit @Asset @Color, @Node /
@Clip / @Material, @Tint @Derived, @Text / @Multiline, @Key, @AppendOnly /
@ByKey, @PerMap / @Chunked, @frame @max, @owns / @creates / @releases,
@deterministic @alloc_ok. Each is in tools/docgen/inventory.json; every fence
that is not marked skip parses (ludicc --fmt). annot-clearcolor's token loses its
quotes, which no reader strips.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
documentSymbol named `export property X` and `export registry` a function called
"property" / "registry": `export`, `unsafe` and leading @attributes are now a
lead-in to the declaration after them, which keeps its doc comment and is marked
exported. A record's field default was skipped with padv, which crosses lines, so
a property showed only its first field; a member now ends at a comma, the brace
or its line, past a fn type's parameters and a generic's arguments, and an
attribute's arguments are skipped rather than read as fields.
New symbols: state, event (and cancellable), action and port as records (struct,
event, event, interface) with their members; registry (with its record, for
go-to-type, and its line as detail); enum and its members; component and view
with props, state, fields, functions and events; reducer, named "S on A"; a
module-level let as a constant; def and bind bodies, module / friend / numbers
lines read past. A member of a component is not a name for the whole unit.
Hover and definition on an attribute's argument: @Node(model) / @Clip / @Material
to the field of the record it is written in, @Ref(Kits) to the registry (as any
top-level name), and hover on the attribute shows its docs/language page. A
fn-typed or generic field's type is shown whole. lsp_test (test-lsp) holds all of
it on a module file with each kind of declaration.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
`ludic syntax [--json] [-o FILE]` prints what `ludicc --emit-syntax` does (a line
per entry, or the JSON). `ludic-dev syntax` writes, between "ludic-dev syntax:
begin" / "end" lines, the keyword, type, phase and attribute tables of
ludic_syntax.h, LudicVocabulary's sets (JetBrains), ludic-mode.el's lists and the
language server's word tests (is_keyword_word and the rest; is_contextual_word is
every word the parser does not reserve, and every declaring or modifying one),
and every TextMate pattern marked "comment": "ludic-dev syntax: <group>" (shared
and the VS Code copy). The grammars gain module uses port bind action reducer
dispatch registry def open component prop view alias friend unsafe numbers of as
from mut system; import and extern colour as declarations; the phase clause
knows Overlay; the bitwise pattern matches | and ^ on their own again.
`ludic-dev syntax --check` - and check-vocabulary, whose old parser comparison it
replaces, and the regression suite (syntax_cases, one line per file) - fails when
a written list is behind, when a grammar lacks a keyword, type or phase, when
docs/language has no page for a keyword, type, phase or attribute, or when the
parser (a scan of selfhost/frontend: is_id / text == words, a == / ann ==
attributes) tests a word or reads an attribute vocab.ludic lacks, or the reverse.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
selfhost/frontend/vocab.ludic holds every keyword with its role (declaration,
modifier, statement, operator, constant), every declaration's form, the built-in
types and phases, every attribute with what it goes on, its arguments and a
one-line doc, the operators and the literal forms; `ludicc --emit-syntax` prints
it as JSON ("syntax_version": 1) and exits before reading any program. The
parser dispatches on words where it meets them, so the table is held to it from
the emitter's side: a keyword's "reserved" is is_reserved_word's answer, a phase
must pass is_phase_name and a field attribute listed as read must pass
at_field_known, or the emitter refuses to print. Reseeded (both seeds assemble;
bootstrap-cfree: out.ll == seed.ll).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The first cut (LGD1) stored every tile of every layer raw, scale channel and all: 288 MB a map, which
made Maroon's pack 803 MB. LGD2 keeps a tile index per layer (8 bytes a tile, read whole at the map's
load): 0 all zero, 1 one value in the word, else an offset and a length, the top bit set when the bake
deflated it (inflated into the tile cache with the runtime's z_inflate, a reused buffer; nothing per
frame beyond the cache). A layer keeps its scale only where it varies, the density 6 bits and the scale
4 (the same integer arithmetic as the bake). The bake is maroon-lake's tools/bake/ground_density.py;
the dev copy here writes the same format stored. Maroon 23.2 MB, Lamar 17.2 MB.
The deflated flag is tested as a negative length: `b & 0x80000000` compiled to broken IR (an i32
`and` with a bare `-`), a compiler fault to fix separately.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
GroundFill gains scale_var (an instance's scale x (1 +- var), from its own hash) and band_grow
(x (1 + grow * band)): a painted density carries the local mean, and without them every flower of a
kind was the same size and the far clumps lost the growth that kept the cover as the step widened.
Both 0: as before.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
- the test runner re-makes every state between tests again: since 12fdc07 a state is made by its getter on
first use, so re-running L_init_globals left the last test's state in place (state/tested failed its
second test); @L_reset_states forgets every lazy state, and the runner calls it before each test
- diag_json_case counts errors, and an absent @Ref / @Tint / @OneOf target is a warning since d82dc31:
ref_unknown is 1 error and node_bad 8
- schema_hash.ludic prints 1: a bool is 1 or 0 as text (random_plain's 1 1 1)
- permap_check_case looks for the unit warning without the quotes the JSON escapes
- baked_test's inputs-hash test makes its directory: each test has a temp directory of its own
- ludic deps prints phase 25's five counters too
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
examples/lang/permap (two maps, a chunked table with a negative, a positive and a
missing chunk, constants and fn by name, a nested list, a cross-row @Ref, a reload
shrinking in place, a bad file's file:line:col; built under arena strict with an @On
frame root), --check fixtures for a wrong field, a wrong type, an unknown constant and
a dangling cross-row @Ref, @Ref(PerMap) on an int, as PREFIX, and a @Unit warning.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
- @PerMap registry R of T from "file.lres" reads <maps root>/<map>/file.lres at run time
(package.ludic's new `maps` line, default assets/maps); @Chunked(n) one file per chunk,
{cx}/{cz} in its name. No as PREFIX, no constants, no def, never open.
- permap_gen: state R, r_load/_clear/_find/_path; chunked: RChunk, r_in/_out/_slot/_find/
_clear/_path; a typed reset and fill per record over lres.ludic, rows, lists and list
records pooled per table / chunk slot, @alloc_ok on what grows at high water.
- permap_consts: lres_consts__(), the program's int and float constants by name.
- permap_check: ludicc --check reads every map directory (fields, types, constants, fn,
@OneOf/@Range/@Ref, cross-row @Ref keys in the same map, a key twice); --no-maps skips.
- @Ref(PerMapTable) is refused on a non-string field; the schema says scope/chunk per
registry and scope map on such a Ref, and lists the canonical @Unit spellings; any other
@Unit spelling is a warning naming the canonical one.
- reseeded (mac + win seeds; bootstrap-cfree out.ll == seed.ll).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
A flat pooled tree (parallel lists reused from file to file), the file's bytes in a
buffer that grows only for a bigger file, strings unescaped into a kept scratch and
interned, the path built in a kept buffer, the program's constants by name (a sorted
table the compiler writes), and an open-addressing key hash rebuilt in place. Nothing
allocates past its high water; only an error's message is made, when a file is wrong.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
ground_density_bake(path, key, version, hash, pngs) cuts each layer's PNG (R the chance a cell keeps one,
G an optional scale) into the terrain's 64 m tiles, one layer decoded and let go at a time, as a bake's
file; ground_density_open(...) reads it - or, when it is missing or stale, a dev build's own copy cut from
the PNGs - a tile at a time into a 256-tile clock, never a layer whole. ground_fill(s, cx, cz, band, g)
places a GroundFill row's instances in a stream chunk: a candidate per cell of the band's step, jittered,
kept when a hash of (layer, chunk, band, cell) falls under the density (times the game's trample,
r3d_on_ground_trample, where the row asks); scale, yaw, seed and wind from the same hash. Nothing calls
it yet: behaviour unchanged.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The seat height, the horse's walk, gallop, acceleration and turn (VE_HORSE_*), its stamina (a gallop
over 6 m/s spends 6 a second, a walk gives back 3, 5 needed to gallop) and hay (0.02 a metre,
VEHICLE_HAY_MIN), the boat's stroke, turn and wind (VE_ROW, VE_ROW_TURN, VE_WIND) and where a rider
gets off (the horse's 1.2 m flank, the boat's 1.25 m wade) move into VehicleSpec (spec.ludic), held
per kind in VehiclesState.ve_specs and kept across a reset. vehicles_spec(kind, spec) sets one,
vehicle_spec(kind) asks it; a kind with no spec cannot be called. VEHICLE_HORSE and VEHICLE_BOAT stay
the kinds. The tests hand Maroon Lake's numbers in, and a new one holds a spec's walk and seat_h.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
A prebuilt tool, bin/ludic-ui-preview (built by ludic-dev build, shipped beside ludic-lsp, run as
`ludic ui-preview [--font DIR]`): ludic.ui with a backend that records every draw call as a line,
and mock component classes the studio describes (load / model / calls). frame t runs ui_show at
interface time t; text is measured on the CPU with the game's font.json metrics; locale goes
through ludic.i18n; tree / box / rules inspect the frame. No game code, no GPU, no network. The wire
protocol is frozen as tools/ui-preview/protocol-v1.md; smoke.txt is a transcript to run.
ludic.ui gains, all additive: UiClass.make_of, UiBackend.said / emitted, UiRule.text / at (with
comment line breaks kept so rule lines count true, and a class's styles read under its .lss path),
and ui_root, ui_find, ui_rules_of, ui_rule_value, ui_building, ui_class, ui_class_load,
ui_file_forget, ui_errors_clear; ui_nine_cuts_into exported.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Each UI component: module, place, doc, xml and lss paths, props and state
(type, default as written, place, doc), states_read (the header's states),
derived fields with their types, functions and events as the template calls
them (states and instance stripped), and the native tags its template uses.
natives: every ui_native / ui_native_input call with a literal tag - tag, via,
handler, place. schema_version stays 1; the lists are additions.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
impostor_from_baked / impostor_refill read a bundle of two DX10 .dds (albedo, normal: BC7_UNORM, every
level) through tex_load_dds_at and gpu_tex_compressed - no gpu_tex_mips - and check each is the
impostor's size; sampled as a painted atlas is (clamped, no anisotropy). No BC on the GPU, or a bake
that does not fit, and the caller paints RGBA8 and makes mips as before; the fog re-open reads the BC7
file again. The handles and every draw are unchanged. Compile-only: nothing run.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
GrassKind (grass_kind.ludic) carries what grass.ludic, grass_gpu.ludic, grass.vert, grass_cull.comp and
model.frag's BLADE path held as constants: the cell, s0 / d0 / radius, the row bands and rows, the blade
profile (neck, root, swell, tip, bend), the heights, clumps, widths and arch, where it grows (slope, snow,
the photograph's test), the root / tip / gone-to-seed colours, the far tufts, the root occlusion, the
roughness, the sheen and the wind. Its defaults are those constants exactly, so a game that sets nothing
draws as before. The shaders read them as u_gk_* (the cull as five more Params vec4s, 288 bytes);
grass_reset.comp writes each band's index count so a kind of other rows reaches the draw in order.
grass_quality holds the kind to a settings tier (never denser, never farther); R3D_GRASS_* still win.
grass_profile_w / _bend are the one profile, for the meshes and a game's preview. grass.mesh takes only
the cell: the rest of it is an older copy that already differs from grass.vert, left as it draws.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
ludic.lab: lab_png_write / lab_png_write_from (raw 8-bit, 1-4 channels, stored deflate) in png_write.ludic,
importable alone with its own LabPngState; png_convert.ludic's previews of float textures (R32F min..max,
RG16F x255, HDR x/(1+x) + sRGB); lab_ppm_to_png on the same encoder.
render3d: bake_load.ludic - impostor_from_baked / impostor_source / impostor_refill (a fog re-open reads
the bake), sky_baked_in and sky_precompute trying the bake at the start yaw (sky_compute is the
convolution, and sky_ibl_bytes always uses it), carpet_from_baked / carpet_bytes / carpet_finish,
bake_part_count / _len / _off. Compile-only: nothing run.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
A package on ludic.base alone. DevlinkNet is the transport (a game binds Udp, the test a fake);
DevlinkWorld's members are the verbs, each defaulting to "not offered" (err unbound): ping, hello (with
Build.schema_hash as 16 hex digits), cam_get / cam_set / cam_release, goto, map_load, time, weather, shot,
pause / resume / step. One request a frame, parsed in place from a fixed 8 KB buffer and answered into
another; map_load, shot and step answer later by id, one at a time, 5 s at most. Loopback senders only,
opened only when enabled() (dev_tools), co-op refuses everything but ping and hello. tests/devlink_test:
each answer byte for byte, bad arguments, a stranger dropped, co-op, the late answers and their timeout,
a closed build that never opens. Compiles; not run (the Master runs the package tests).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
gltf_factors reads pbrMetallicRoughness.baseColorFactor, metallicFactor, roughnessFactor and the
emissiveFactor into the primitive (pr.fac), and prim_factors hands them to every program that draws its
textures (actors, the scatter's meshes and levels, the impostor bake) as 1 - factor, so a program never
given them - or a material that gives none (pr.fac null) - multiplies by 1 and draws as before. An
untextured material with a colour (or a metal/roughness) samples a pure white for it, so the factor is
exactly what it draws: horse_cornea is its own colour, not the 200 grey every untextured part had.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Both constants were worked out by hand and a byte off each: the tiles' bake wrote "LAAK" and "LDT2", which
ludic bake --check refuses, and r3d_baked_read, holding the same wrong constant, would have refused
ludic.base's correct files (the sun's shadow among them) while accepting its own. Now 0x4B41424C and
0x3254544C, checked against the strings; the unused LTT1 constant is gone. A re-bake is needed.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
--emit-schema FILE writes the compiler's resolved view once the program type-checks: every
record (fields, types, defaults as written, docs, places, attributes), every registry with its
entries in their final order after the open-registry merge (key, constant, index, file:line:col
of the entry and of each field value, and which file contributed which keys), every const, and
the zero-argument functions a fn value can name. Deterministic, schema_version 1; the runtime is
left out. `ludic schema [file] [-o FILE]` wraps it.
--check --diagnostics=json prints every error as one JSON array on stdout: the checker's and the
module rules' all, a parse or lowering error as the last. Tokens and nodes now carry a column.
Fields take several @attributes; @Ref(Registry), @OneOf(PREFIX_), @Range(lo, hi), @Unit("..."),
@Asset("..."), @Color on a field and @AppendOnly / @ByKey on a registry change nothing but go into
the schema, and @Ref naming no registry is an error (every one reported). Fixtures:
examples/lang/attributes.ludic, examples/rejected/ref_unknown.ludic, cases in ludic-dev test.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
A world swap unmounts the old map pack before terrain_reload, and the tiles' file (a baked file in that
pack) stayed open until terrain_unload. terrain_tiles_close closes it and stops the page pool; the next
map's terrain_from_baked / terrain_reload opens its own and makes the pool again.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The game names the tiles' and the shadow's bake files (key, version) before the map is made; r3d_init's
terrain step and terrain_reload then open them with terrain_from_baked / terrain_shadow_from_bytes and
generate nothing, and fall back to the survey (saying so once) when a bake is missing or stale. Without
the call nothing changes.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The tiles' photograph side is 0 when the map has none: no photograph or coarse photograph sections, no
decode, ter_o answers 0, and no coarse photograph texture - as such a map has always drawn. The cut and
terrain_tiles_bake no longer require one.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
baked_open refuses a header whose key is not followed by a zero byte; a key a multiple of 8 long got
none. Now the payload starts at 32 + ((len(key) + 1 + 7) / 8) * 8, as bake_write writes it.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
stream_capture(s, cx, cz, band) points the stream at a chunk of its own (made on the first capture),
calls the registered r3d_stream_fill for that cell and band, and returns how many instances it
emitted; stream_captured(i, k) reads them from stream_scratch. Build-time only; nothing in a frame
calls it.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
baked_tex.ludic reads ludic.base's baked form by hand (render3d uses no package): the LBAK header, the
key ("png_sheets" / "cutouts", the Bakes rows) and the generator version, then the payload -
w, h, channels, depth and the samples for a PNG (the caller frees them as it would a decode), a whole
.dds for a cut-out (tex_load_dds_at: a .dds at an offset). Missing, or another key or version, and the
PNG path runs as before. Compiles with Maroon Lake (game, lab, lab/bake/textures).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
prec_put / prec_get / prec_size (Physics' spec): PREC_FULL, 12 bytes, for trees, boulders and stones
(u16 x, z in 1/65536 of the chunk; u16 y in cm above y_base; u8 scale over the kind's lo..hi, yaw, seed,
wind), and PREC_PACKED, 7 bytes, for the stream kinds (56 bits, least significant first: x 12, z 12,
y 12 in cm, scale 6, yaw 6, seed 4, wind 4). An encode takes the cell a value falls in, a decode its
centre, so a round trip is stable; a record is read from a []byte at an offset, never a pointer.
stream_emit_baked and layer_chunk_put take (recs, at, fmt) and decode through it; the earlier sb7_*
pair is gone.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Maps ship outside the game, each one a content-addressed pack (.lmap: the .lpak format) downloaded to
the save root. The runtime already served reads from packs mounted at boot (packs.index); now one can
come and go while the game runs:
- Fs.mount(path) -> bool maps a pack over the ones mounted before it (searched first, as a later
packs.index line is); Fs.unmount(path) -> bool gives the mapping back (munmap, UnmapViewOfFile on
Windows) and closes the gap in the search order. Each slot keeps its length and path for it. A
FILE* still open over one of its entries (a baked_open_range) is closed first. Still 8 packs at most.
- baked_path(map, file) is assets/baked/maps/<map>/<file> for a map's bake - what its .lmap carries
and the game's own pack never does - and assets/baked/<file> for the rest.
The IR assembles for macOS and Windows (llvm-as). Compile-only: nothing mounted or run here.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
sb7_pack / sb7_field are the one pack and unpack of the stream kinds' record (56 bits, least significant
first: x 12, z 12, y 12 over the chunk's y span, scale 6, yaw 6, seed 4, wind 4, read as a low and a high
word), which the bake and the game both import; stream_emit_baked decodes a prefix of it.
layer_chunks_begin / layer_chunk_put / layer_chunk_drop keep a fixed layer (trees, boulders, stones:
12-byte records) as a slab per resident chunk in its instance list, an n x n index made at load, the
list re-uploaded at most once a frame. Nothing calls them yet: behaviour unchanged.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
stream_set_grid(s, ox, oz) puts a stream's cells on a grid from a corner (0 keeps world zero, today's);
stream_emit_baked(s, recs, count, keep, x0, z0, y_base, lo, hi) emits the first keep 12-byte records
of a baked chunk (the layout agreed with Physics: u16 x, z in 1/65536 of the chunk, u16 y in cm above
y_base, u8 scale over the kind's lo..hi, u8 yaw, seed, wind) into the chunk being filled, and
stream_band_keep(count, share) is a band's prefix. Nothing calls them yet: behaviour unchanged.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
ZInflate (z_inflate_new) holds the bit reader, the RFC tables and every table and scratch list a block
builds, rebuilt in place: an inflate allocates nothing, and a worker thread inflates in a context of its
own (made on the program's thread). z_inflate_in / z_uncompress_in / z_gunzip_in take it; z_inflate /
z_uncompress / z_gunzip and every caller (image, atlas, tiled, render3d's png_decode) are unchanged and
work in RtInflateState's one context. The old per-call lit/dist tables were also leaked on an error
return; there are none now. Compiles: Maroon Lake headless, examples/library/tiled_p2 and tiled_p6.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Heights as u16 over each 64 m tile's own minimum and step (a tile spanning 200 m steps 3 mm), normals
octahedral 8 + 8, the photograph RGB8, and the coarse level (2048: heights f32, normals, photograph)
- 30 + 32 + 48 + 16 + 8 + 12 MB, about 146 MB a map where the float tiles were 192 plus nothing coarse.
The writer puts the whole copy back to the quantized values as it goes, so the build's own queries,
the physics, the placements' bake and every machine read the same numbers; a tile read decodes them
into the pools the queries and the page pool already use.
terrain_tiles_bake(path, key, version, inputs_hash) writes a bake's file (ludic.base's LBAK header,
the tiles as its payload) from a made map; terrain_from_baked(path, key, version, half, ox, oz) opens
one at boot in place of terrain_use_dem / terrain_use_ortho, and terrain_init then generates nothing
(and bakes the sun's shadow from the coarse level unless terrain_shadow_from_bytes gave it). Without a
bake the cut writes the same format to <dir>/<key>.tiles and reads it back. The GPU's coarse level is
made from the file's coarse sections (the blit from the whole maps is gone).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
What is deterministic is made at build time by the game's own code (`ludic bake`) and read at run
time. baked.ludic, baked_hash.ludic, baked_stream.ludic:
- The header, 32 bytes little-endian: "LBAK" | u32 format (BAKE_FORMAT 1) | u32 generator version |
u32 payload offset | u64 inputs hash | u64 payload length | then the key (UTF-8, zero-padded to 8),
then the payload at its offset, whose layout is the bake's own.
- bake_write(path, key, version, inputs_hash, payload, n) makes the directories and writes it;
bake_inputs_hash(inputs) is FNV-1a 64 over each space-separated input: its path, a 0, its bytes.
- baked_open(path, key, version) -> []byte: the payload as a view of the file (no copy), or null for
a missing file or another format, key or version. The runtime never hashes inputs; `ludic bake
--check` holds a baked file to them at build time.
- baked_open_range(path, key, version) -> BakedFile kept open (header checked once),
baked_read(bf, at, n, into) -> count read into the caller's buffer from payload offset `at`
(held to the buffer and the payload), baked_len, baked_close: a streamed bake (placements by chunk,
terrain by tile) with nothing made per read. Pack-aware through file_open: on macOS a packed entry
is an fmemopen over the mapped pack (a seek is free), on Windows a temporary copy of the entry made
once when it is opened.
- baked_path(map, file), bake_missing(bake_st, key) (a dev build's line, once a key).
tests/baked_test.ludic: a payload round-trips for its key and version only, a streamed read at an
offset and one past the end, the inputs hash follows path and bytes. Written, type-checked, not run
(compile-only rule).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Each pair runs the work as the renderer always has and reads the result back, or makes the same
textures and fills them from bytes instead: impostor_bytes / impostor_from_bytes (and impostor_fill,
for a fog that opens past a layer's cards), terrain_shadow_bytes / terrain_shadow_from_bytes,
sky_ibl_bytes / sky_ibl_from_bytes (sky_precompute split into sky_ibl_make and the convolutions).
A bundle is a word count, a word length per part and the parts (bake_pack / bake_unpack), checked
against the textures' shapes before any byte is used. gpu_vk_readback.ludic reads a texture's level 0,
every layer, as stored; r3d_baked_read reads a bake's file (ludic.base's LBAK header: key and version
must match) since render3d cannot import ludic.base. Nothing calls them yet: behaviour unchanged.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
runtime/native/inflate.ludic: ZInflate holds the bit reader, the RFC tables and every table and scratch
list a block builds, made once (z_inflate_new) and rebuilt in place - an inflate allocates nothing, and a
thread that inflates holds a context of its own. RtInflateState keeps one, so z_inflate / z_uncompress /
z_gunzip and their callers are unchanged; z_*_in take the context.
render3d png_jobs.ludic: png_parse (reads, walks the chunks, makes every buffer), png_work (inflates,
unfilters, packs; makes nothing, touches no state), png_take (frees, sets tex_*); tex_prefetch runs
png_work over a list on every core through Job.parallel_for and png_decode takes a waiting result, so
what uploads and in what order is unchanged. Nothing calls tex_prefetch yet. Compiles with the game.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The build's placements ask exact heights over the whole map before any ring exists, and through the
tiles that read the file ~25 000 times in one frame. The copies now stay after the cut and the game
calls terrain_tiles_build_done() when the world is built (the end of world_things, and after a swap's
terrain_reload); every answer is the same before and after, from the copy or the tile. The GPU half
merged here (r3d/tp-rt, r3d/tp-shd: the page table, the pool and the shaders) is unchanged by it.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
At the cut the whole 4096 height, normal and photograph textures go, replaced by a coarse
2048 level (the CPU's tt_coarse uploaded; the normal and photograph blitted down on the GPU,
the photograph mipmapped) and a pool of fine tiles in three array textures - heights, normals,
photograph, each layer a tile with a one-texel border - addressed through a tt_n^2 page table
(u_tp_page: slot + 1, 0 = the coarse level). The pool holds the tiles within the reach (900 m,
or the fog wall's when nearer) and a ring, and is made again when the fog wall changes its size
(terrain_pages_fog). Once a frame (tp_frame, beside tt_frame) tiles past the reach and two tiles
go and the wanted ones come in nearest first, 8 a frame, written into a staging buffer kept for
the process (two halves, one per frame in flight) and copied into their layers inside the frame's
own command buffer - no submit of their own - with the page table re-uploaded only when it
changed. tp_bind binds the pool (or 1-layer stand-in arrays while paging is off, u_tp_on = 0) for
every program that reads the ground: terrain_bind_height (models, scatter, shadow_bind, grass),
terrain_bind_prog, the sun pass and the shadow bake. grass_cull.comp reads through the same page
table. R3D_VKMEM prints the pool's line. Tiles off, nothing changes. Compile-only: not run.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Most of a build was one clang -O2 on one .ll (the game: 32 s of a 41 s headless build, one core).
Both paths that assemble - the CLI's (build.ludic: ludicc --emit-llvm, then clang) and ludicc's own
(-o, which ludic bundle and the examples use) - now cut the program's IR into N parts with llvm-split
(externalizing what the parts share), compile them with one clang each in parallel (-x ir -O<opt>
-mmacosx-version-min=11.0, the link's own clang taking the objects where it took the .ll), and remove
the parts and objects after. N is $LUDIC_JOBS, else min(cores, free GB / 1.5).
It needs an llvm-split and a clang of the same LLVM (Homebrew's LLVM 22 writes attributes Apple's
clang 17 cannot read): $LUDIC_LLVM, else /opt/homebrew/opt/llvm/bin. With either missing, on Windows
(its shell cannot run the parts at once yet), with LUDIC_SPLIT=0, or when a part fails, it compiles the
.ll whole as before.
$LUDIC_OPT=1 is a developer's faster build; ludic bundle sets LUDIC_OPT=2 for its compile whatever the
shell says.
Measured before the compile-only rule (this Mac, 12 cores, one build at a time):
- the game headless: 37-41 s -> 13-15.5 s (8 parts / by free memory), peak 2.1 GB -> 1.0-1.1 GB;
- the lab headless: 43.1 s -> 12.7 s, peak 2.4 GB -> 1.0 GB;
- the game at LUDIC_OPT=1, split: 11.8 s (fps cost not measured).
Both built and linked clean; the goldens and a headless shot of the result are not run here.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Within a pass a row writes only its own still-masked texels and reads only neighbours already let go,
which no row writes that pass, so the result is the single-threaded one. The worker takes a DilateJob
of plain buffers and allocates nothing. tex_dilate_bytes (safe_api) and examples/rendering/dilate.ludic,
which checks it against the old loop on RGB and RGBA atlases of sizes that do not divide (DILATE OK).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
terpage.glsl is the reference block (tpSlot, tpUV, terHeight, terHeightSmooth, terNormalXZ,
terOrtho, tpOrthoRes, tpOrthoLod), pasted by section into terrain.vert, terrain.frag,
tersun.frag, model.vert, grass.vert and grass.mesh. u_tp_on = 0 reads the old samplers with the
old coordinates and filtering; on, a resident tile is read at level 0 from u_tp_h / u_tp_nrm /
u_tp_ortho, anything else from the coarse map now bound under the old names. The B-splines use
the FULL map's texel and take every tap through the page, so a tile edge stays one surface;
blurred photograph reads (lod 1-2.5) stay on u_ortho with the level moved down by the coarse
map's ratio. The fragment stages drop their unused u_height. SPIR-V rebuilt: terrain programs
carry 23 samplers (21 in the fragment stage), up from 19.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
A switch a library reads only as a process starts - libmalloc's MallocLargeCache, which on a game
keeps ~200-300 MB of freed load buffers - has to be in the environment before main. For a .app started
from Finder, the Dock or `open` that is Info.plist's LSEnvironment; `app env` is repeatable and each
K=V becomes a string in it. bundle_case's probe app carries one and checks it with plutil.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The volumetric pass marched 800 m in a fixed number of steps whatever the wall: it now stops at the
wall with steps in proportion (at least 8) - 0.57 -> 0.33 ms at 30 m by the pass timers. SSGI skipped
only past 900 m; it now also skips past 0.85 of a wall, where the fog is solid (u_ao_far). applyFog
samples the fog's colour only where its weight is above zero. A card layer whose casters inside the
wall are 80% or more of it (a kilometre's wall) casts from its static list and is not refiltered and
re-uploaded every 4 m of camera motion. Fog off: the frame unchanged (0 pixels over 8).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
terrain_tiles_to(dir, key) before a map is made: once made, its heights, photograph and baked normals
are written tile by tile (64 m, TT_TEX) to <dir>/<key>.tiles - fresh every time, header last, so no
other generator's or a torn file is ever read - and the whole copies go. Every read goes through the
tile: resident, else read from the file there and then into a 2048-tile clock, so terrain_height,
terrain_height_smooth, terrain_ortho* and terrain_chunk_heights answer exactly what the whole copy
did (R3D_TT_CHECK: worst 0.0 m over 4000 points) whatever is resident - two machines and the boot's
placements agree to the bit. terrain_chunk_heights takes render3d_st mut for it.
terrain_height_near(read-only): the tile if it is in, else a coarse 2048^2 level (worst 0.91 m), never
the file - for line checks that must not take render3d_st mut. terrain_texel() is the texel size,
terrain_tiles_prefetch(x, z, r, budget) reads ahead, and a frame that reads more than 8 tiles says so
once. R3D_TERRAIN_TILES=<dir> turns it on for a run.
At the overlook with MallocLargeCache=0: 1731 MB off, 1658 MB on; 192 MB written in ~200 ms; the
frame unchanged (0 pixels over 8).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The DEM (R16 with mips, 44 MB) was read only by terrain_generate and kept for the map's life; a
world swap loads its own again. The height-field sun shadow was one RGBA32F for three values and an
unused fourth: the lowest lit height stays 32-bit (a receiver 3000 m up compares against it to a
quarter metre), the occluder distance and the cloud mask go beside it in RG16F (64 -> 32 MB), baked
in a pass of their own (TS_AUX): a pipeline takes one colour format for all its targets, and drawn
together into R32F + RG16F the writes went nowhere and nothing was lit.
-75 MB at every fog level, off included (2008 -> 1933 MB at the overlook); the fog-off frame is
unchanged (0 pixels over 8).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
A 1 km fog gave a 1040 m zone, more of the world than no fog's 900 m (2405 MB and 1029 nav tiles against
2294 MB and 753 in the fog profile). zones_radius is min(max(fog + 40, 150), 900); the test holds 1 km
and 860 m at 900.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The ring held 2 x 64 MB of host memory for a frame that uses 3 MB at most (2761 draws at the
overlook, 1.7 MB in town): it starts at 16 MB a half, and a frame that ever runs out grows it for the
frames after (gvk_ring_grow, making the kept sets again; R3D_RING_KB=<n> starts small to watch it).
-95 MB at every fog level, off included.
A streamed layer's arrays and its stream's arena are made for the reach a fog wall leaves (twice its
area's share, at least 4096 instances) and emptied to refill within the frame budget
(fog_streams.ludic). The near streams' reach is already inside most walls, so it is -9 MB at 30 m.
R3D_VKMEM adds the ring's high-water mark, the largest buffers and the streamed layers' share.
Footprint at the overlook: 2008 MB off (2108 before this phase), 1748 MB at 175 m, 1682 MB at 30 m.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
An empty world at the game's size held 51839 KB of Jolt's heap (200000 bodies, 65536 pairs, 20480
contacts, a 32 MB temp heap), the jolt= floor under every walk. Pairs and contacts come from what
moves - drops, boats, a few walkers - and a valley's still things make none. Now 11743 KB at 200000
bodies and 8461 KB at the 98304 the game opens for (jolt_floor_test holds it under 16 MB). The temp heap
still falls back to malloc for a step that wants more.
lib/macos-arm64 rebuilt; lib/windows-x64 needs native/build.sh on the PC (jph_world_new's new sizes).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
A cascade that begins past the wall was fitted and cleared every frame for nothing: the array is now
made with the ones that begin inside it (2 at 30 m, 3 at 175 m, 5 without a fog) and made again when
the wall changes that count; shadow maps 80 -> 32 MB at 30 m, 48 MB at 175 m. A layer the array lacks
clamps to its last, read only past the wall where everything is fogged.
The sky: at walls of 110 m and nearer the whole sky is the fog's colour, overhead too, with a soft
glow at the sun or moon and no stars or clouds through it, easing out by 175 m; from 175 m on the
horizon band alone, and off as before.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Inside the wall every instance is a mesh whose LOD2 casts its shadow, so a card is neither drawn nor
cast there: fog_impostors.ludic frees a layer's atlases while wall + margin < its near (trees 420-480
m, rocks 320 m) and paints them again from the model when it opens past it (impostor_paint, split out
of the bake). Impostor atlases get memory of their own, so a release goes back to the driver instead
of leaving a hole in a shared block. R3D_FOG_AT=<frame> with R3D_FOG_TO=<m> changes the wall mid-run.
At the overlook: 2102 MB off, 1824 MB at 30 m (249 -> 67 MB of atlases; the rest is the ground cover's
card atlases, drawn inside the wall); opening re-bakes 16 layers in 42-46 ms, back to 2102 MB.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
- The arrows, a press on the select and the pad step past a disabled option (ct_opt_step), wrapping
as before.
- A value naming a disabled option is shown as the nearest enabled one (the lower of two as near) and
set to it once the build is done: an event fired while the tree is built is cleared with the frame's,
so the clamp is queued (ct_clamp_n / _i) and fired after nt_fired_clear.
- The < or > whose next option is disabled is drawn disabled (its part's :disabled), so a cycler shows
where the options stop being on; it still steps past them.
- ct_build_select moves to controls_opts.ludic with the rest (controls_build.ludic 119 -> 104 lines).
Golden ui_select_disabled: options 3 and 4 disabled, a value of 4 clamps to 2, > is drawn disabled
there, > wraps to 0 and < from 0 steps back to 2. The 38 ui examples pass.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Every allocation carries the owner its maker was wrapped in (models, terrain, impostor atlases,
scatter, grass, shadow maps, frame targets, sky, water, game textures, made in a frame), and the
report prints each owner's images and buffers, the 25 largest images and the scatter layers',
streams' and terrain's CPU copies. Nothing drawn changes.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
zones_set / zones_clear / zones_merge, zones_contains / zones_nearest and the clusters' bounding circles,
over fixed arrays of ZONES_MAX (16). WildlifeWorld.active freezes an animal outside (no state change, no
dice), NpcWorld.active a walker (and no birth outside), VehicleWorld.active a moored boat, and
ThingsWorld.restocks keeps a morning's restock inside; every default is "everywhere", so an unbound game
is unchanged. Tests: the merge (near players one cluster, far two, a player leaving splits a chain),
contains, nearest, the radius; wildlife, npc, vehicles and things pass as before.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
For loading the world by each player's fog rather than a fixed 900 m. keep_near_test: a zone round
one corner brings in its tile, a bigger one its neighbours, an edge nudged back and forth twenty times
changes nothing, and a zone moved 2 km away lets them all go.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Whatever is cut at the wall is now cut inside full fog: at 0.3-1.0 the last metres before the cut
were 80-95% fogged against fully fogged ground behind, which drew a line of tufts at 70 m and a
dark band of blade tips at 30 m. The blades thin out while the fog is still coming in.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
cam_sphere_visible refuses a sphere wholly past the wall (terrain patches, scatter cells, stream
chunks, grass tiles, water), actors past it are skipped for draws, casters and outlines, the grass
reach and the streams' generate-and-gather reach end at it, and the card shadows cast only from
the wall's share of a layer (fog_casters.ludic, rebuilt every 4 m). r3d_beyond_fog is exported for
the game to skip animating what will not be drawn. Render-only: no query of the ground or the world
changes, and off is the old frame (0 pixels over 8 against 160ce96, twice per side).
Draws per frame at the overlook: 2757 off, 1104 at 175 m, 484 at 30 m.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Master's windowed fog profile failed the fence at walk t 35 s: +512 B from value_new / value_put /
value_slot / value_lists under bd_class and cmp_keycap_model - a KeyCap first shown mid-walk.
It is a first build, bounded, not a per-show leak: an unmounted instance goes to in_free and the next
show of its class reuses it (in_reuse, `renew`), its props and model objects kept and filled in place.
Only a NEW instance makes them - bounded by how many of that class are up at once, and the ones
unmounted less than two builds ago. in_reuse already declared the record; the props and model objects
and their first fill, made afterwards in bd_class, were not. They are now: in_reuse marks a new record
`fresh`, and bd_class's first fill of it goes through bd_first_fill (@alloc_ok) - a reused instance's
fill stays judged, so a real per-show leak would still fail.
Golden ui_first_show: a component first shown at frame 700, once the fence is judging, then hidden and
shown twice more: bad 0 (the toolchain before this fails frame 701: +464 B value_new from value_slot
under cmp_cell_model, value_put grow - the profile's failure). The 37 ui examples pass.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Every lit program blends toward the horizon colour (the prefiltered sky with the sun's inscatter,
so it carries the day's grade); the sky's horizon band is pulled to the same colour, wider the
closer the wall. The far plane, scatter and stream cull reach and the shadow cascades are clamped
to the wall; the reflection pass shares the shaders and the cull. 0 is off: every branch is gated
on u_fog_wall > 0 and r3d_reach hands back the far it was given. R3D_FOG_WALL=<m> for a shot.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The valley self-test died after the Lamar swap, loading Maroon's pine LODs:
gvk_fr_blk (the device suballocator's free ranges) grew past its @max(4096).
Nothing failed to coalesce: a range was carved at the buffer's bare size, so
what was left after it started unaligned and the next buffer, rounding its
start up, left a sliver before it that no later buffer could use. One host
block held 3,968 small buffers and 3,718 free ranges between them - 496 KB
free in all, 133 bytes a hole. Those merge away when a neighbour is freed, so
nothing leaked; the count simply rose with live buffers.
gvk_mem_new now carves `span`, the size rounded up to the alignment, and
records it as the allocation's length, so a free gives back exactly what was
carved and the rest of a range always starts aligned.
The same repro (the lab's `tests` scene, headless, a fresh test root): the list
never reached 256 slots (it passed 4096 before), the run completes (exit 0),
LAMAR OK and THINGS OK. r3d_small was 0 throughout, as ECS said.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
A small window (r3d_small: the launcher's UI process) draws its interface flat
and tests no depth, but gvk_screen_make gave it a 32-bit depth image the size
of the screen - 8 MB at 1920 x 1080. In small mode the screen has none: the
pass and a clear with no depth attachment already leave depth alone (clearing
DEPTH_BUFFER_BIT with no attachment is skipped), and pipelines follow the
pass's formats.
With the launcher's own 1920 x 1080 backdrop (maroon-lake game/launcher-small),
the launcher window measured 378 -> 295 MB of footprint at 25 s; its headless
shot (R3D_SMALL=1, so no depth) differs from the unchanged build by 0.279% of
pixels, the resized backdrop, and two runs of each are identical.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The atlas is white glyphs on alpha and the overlay reads only the coverage,
but it was uploaded RGBA8: Maroon Lake's 3072 x 2688 atlas was 43 MB of Metal
memory with its mips, the largest single thing in the launcher window.
overlay_font now keeps the alpha alone as R8 (ov_font_tex; any other PNG
shape is uploaded as before) and overlay.frag reads .r - the white fallback
texture reads 1 there as it did from .a.
Launcher window, graphics regions: the atlas 43.0 -> 11.5 MB; process
footprint at 25 s 402 -> 378 MB. The headless launcher shot is identical to
the unchanged build's (0 pixels differ; two unchanged runs differ from each
other by 1.69%, and the new one by the same against them).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
With one frame in flight, a buffer filled every frame (the interface's vertex
buffer, ov_vbo) was still being read by the frame on the GPU when the next
frame filled it, so gvk_buf_reserve released it and made a new VkBuffer and
its memory every frame - and every new buffer took a prime submit of its own.
Each handle now keeps a second buffer (gvk_bsp_*): a busy one swaps with it
when it is free, else the busy one becomes the second and one new buffer is
made; releasing a handle lets both go.
The launcher window (--launcher-ui, R3D_VK_PROF, per 120 frames): buffers
made/destroyed 120/120 -> 0/0, command buffers 240 -> 120 (one a frame). Its
footprint does not move with it (358 MB at 40 s, both): the churn was work,
not memory.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Beside Os.heap_bytes: malloc_zone_pressure_relief(NULL, 0) on macOS (weak, so a libc without it reads
0) and HeapCompact(GetProcessHeap(), 0) on Windows - kernel32 only, so the Windows build imports
nothing new; the bytes it says it released. Once after a load, never per frame.
Measured, for the record: on macOS it does NOT reach the large-block cache. A C program that frees six
15 MB blocks still holds 90 MB of MALLOC_LARGE (empty) after relief on every zone (it returns 0); only
MallocLargeCache=0 in the environment AT PROCESS START turns the cache off (read at malloc's init -
set later, it does nothing). Maroon Lake's watcher sets it for the processes it spawns.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Set before r3d_open: the frame's uniform ring is 2 x 4 MB (was 2 x 64), memory blocks are 8 MB with
anything over 4 MB on its own (was 64 and 16), and the descriptor pools are an eighth of a world's.
Maroon Lake's launcher window draws a picture and text and never becomes the game; its launcher home
and Settings page draw pixel-identical on it. The full-size path is unchanged (steady: STEADY OK).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
rt_init opened every windowed program's window before main, and render3d's gvk_open then only
retitled it. win_open makes the window when there is none (cocoa.ll and win32.ll alike), so for a
program that has gvk_open the runtime now leaves it (window_later(), an intrinsic: windowed and
render3d present): a process that never reaches the renderer - Maroon Lake's launcher watcher, which
only spawns the game and waits - never makes a window, an NSApplication or AppKit's heap.
Audited every window native reachable before the renderer opens (settings, telemetry, rescue, the
watcher reach App.* and Input.*): on macOS each that loads W_win / W_app / W_view / W_mtl / W_glctx
checks it for null; win_close, win_running, win_text, win_held, win_cursor_mode, win_gl_scale and the
pad and touch reads load none. On Windows each that loads W_hwnd / W_hdc checks it; the rest load none.
Measured, windowed, R3D_DEV=1 R3D_PLAYTEST=2, both killed after:
- the game straight to play: the window, the Vulkan swapchain (1920x1080) and the valley's models
come up, alive at 30 s;
- the launcher (R3D_GAME=launcher): the watcher 11 MB -> 3.7 MB, its launcher window alive at 10 s.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Maroon Lake's launcher watcher - a process that only spawns the game and waits - held 58 MB, 48 MB of
it MALLOC_SMALL. Measured with malloc stack logging it was not the state defaults but L_grow: every
program sized every property type's per-entity store to MAX_ENT (1024) slots at start, 1,583 stores,
entities or none. And every state record was made with its defaults in L_init_globals before Boot.
- emit_lazy.ludic: a program's (not the runtime's) state global is left out of L_init_globals, and
every read of it calls @S_<global>(), which makes it on first call from its own initializer - after
every registry and plain global, so a default may read them (the init-order crash cannot come back
through a state). What a getter makes is declared (@lp_fdecl): a state first touched in play is made
once and not judged as a frame's keep. A function value's trampoline calls the getter too.
- L_grow starts the stores at ECS_FIRST (8) and doubles as entities come, as it always did past MAX_ENT.
The watcher (with the game's watch step moved before the systems' defs): 57 MB -> 11 MB; the only
state it makes is UiState (14 KB). What is left: AppKit's window, opened by rt_init before main for any
windowed program (~4 MB), and the runtime's font, image and 2D inits (~1.2 MB).
Goldens: the arena, fence, value and json goldens; the 36 ui examples; 30 of the 32 ECS test cases
(sprite_render and sprite_atlas time out under a plain runner with the toolchain before this too).
Package tests: ludic.base, save, settings, i18n.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
gvk_swap_make freed none of what it made - the surface caps, the count, the
format and present-mode lists, the create info, the handle out - and replaced
gvk_swap_images without freeing it: a few hundred bytes every time the window
changed size, went to or from Retina, or the chain came back suboptimal. Each is
now freed on every way out (the minimised return and the failed create included),
and the old image list before the new one is made.
A windowed memory walk built against it (main 5ee600d2, 300 s, the autopilot
opening screens every 6 s): footprint 2816 MB at 120 s, 2818 MB at 300 s; the
fence judged 0 frames kept.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
A load freed only the parsed trip's nodes (Json.free) and kept every string the parser made, because
a loader might keep one; free_all freed every string and every key, so on a tree a migration had
added a literal to (`Value.put(v, "sver", Value.str("2"))`) it freed the literal and aborted.
- The parser marks the string values it makes (JP_OWNED, in the node's otherwise unused num) and
interns object keys (a few names, never freed); free_all frees only marked strings, never keys,
and a list's spares too. A setter that gives a marked node other text (value_set_str/_strs,
value_into_str/_strs, value_become) frees the parser's text first. value_as_int / _as_float read a
string as 0 as before.
- ludic.base sv_str returns intern(...): every package load that keeps a text read from a section
(minimap labels, a Thing's look, photo tags and files, an effect's label, ...) holds its own copy.
Golden json_free_edited: parse, put a literal key and string in, set a string, keep an interned copy,
free_all - 1100 loads: the copy reads on and nothing grows (the toolchain before this aborts, 134).
Tests: ludic.save, base, settings, minimap, things, photo, effects; json_saves, value_list_regrow; the
36 ui examples.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
gvk_sc_make's cache (program + bound textures -> a set in the kept pool) evicted
nothing until the pool's 8192 sets filled. A texture freed - a photograph's
thumbnail, a wall frame, any picture a screen loads - left its entries and their
sets behind for good, since its handle's generation moved on and the key could
never match again: 262 KB over 12 minutes of the user's play, and bound only by
the pool, hours away.
gvk_tex_release counts the releases (gvk_sc_dead); at GVK_SC_DEAD_MAX (256) the
next frame's start resets the kept pool and empties the cache (gvk_kpool_reset,
which waits for the frame in flight), and each draw still in use makes its set
again the first time it is drawn - the budget is the live combinations plus at
most 256 textures' dead ones. The declared reason on gvk_sc_make was the
swapchain's, copied; it now says what the cache is and what bounds it.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
- ludic.ui cv_join (units_pieces.ludic:93): a var() value's pieces were joined by +, a text left
behind per piece on every memo miss; it is written at its length at once (units_join.ludic), and
the miss is its own declared function (cs_vars_miss), bounded by the memo.
- ludic.hints hn_slots (rail.ludic:41): hints_reset made a new rail every time; the rail is
emptied in place, made again only for another count (rail_slots.ludic).
- value_spare_put (value.ludic:138): a list's spares grow only past the most it has ever cut off;
declared as such.
- ludic.save: save_read reads, parses and frees the file's text - the tree's strings are the
parser's own - and says whether the text was whole (SaveRead.intact) for the words of a refusal.
Maroon Lake's trip and home reads kept the whole file on every load and every menu card read.
Tests: ludic.save (10), ludic.hints (8); the 36 ui examples; value_list_regrow, json_saves and
alloc_fence_declared unchanged.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The user's walk showed the footprint rising outside the Ludic heap. Jolt's bytes were already counted;
the navmesh's were not, so nothing could see them. Now:
- ludic.nav nav_heap_test: fifty crossings of a four-tile map by the resident index hold the tiles in and
the native bytes to the first crossing's; a thousand crowd walkers in and out grow nothing; a reset
gives back every byte the mesh took.
- ludic.physics cross_heap_test: a 1 km map of 64 m chunks kept to a ring round a player crossing
corner to corner and back, each chunk a heightfield, twelve owned posts and six owned scaled hulls
as the game makes them: six more crossings hold the bodies, the shapes and Jolt's bytes and peak
exactly. (A post made as an offset of a cylinder, with only the offset owned, leaked the cylinder
every time: the game's solid_pillar owns its cylinder directly.)
lib/macos-arm64 rebuilt; lib/windows-x64 needs native/build.sh run on the PC for the new exports.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
From the user's play, reproduced on Maroon Lake's customisation screen with the pointer scripted over
each control (the lab's R3D_CLICKS): nothing on it was ever pressed or hovered.
- A screen's root <div> round a positioned panel (a component's root holding the kit Screen's
modal) lays out to 0x0, and the hit test (fr_pick) and :hover (fr_under) only walked into a child
whose box held the point - so nothing under it was reachable. A box with no size holds no point and
clips nothing: both look through it now (fr_empty), without hovering it.
- A control's parts (a select's < and >, a range's thumb) are made by ct_part, which never set
`hovered`: `select .ui-prev:hover` could not match anywhere. It is set as an element's is.
Golden ui_pointer_through: the panel's button pressed and hovered through the empty root, and a
select's > hovered (before this: pressed 0, hovered 0). The 36 ui examples pass.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The user's play of the bundle: the character shook while seated. chr_sit_tick eased the body onto the
seat point at the ground's height, then the same frame's chr_travel handed it to the walker, which
pushed it off a log, a bench or a boulder beside the seat; the next frame pulled it back. Unwalked, the
game's walker parks as it does under a rider. A hold's turn toward the work fought the seat's yaw the
same way. The new test sits the body against the boulder: 4.63 m from a seat at 5.7 before, still after.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Every tree on the map was drawn as an impostor card into all the cascades,
the near two included, whose receivers (within 60 m) are shaded by the near
trees' own LOD2 meshes cast beside them. The cards now cast from cascade 2 on.
Aspen view, same main and foundations, 240 frames, with the figure capsule:
shadow instances 323,587 -> 202,743; shadow GPU 4808-4942 -> 4579/4578 us;
median frame 17200-17693 -> 17178/17205 us. The look held: shots of the view
taken twice a side are identical within a side (0.00% of pixels past 8) and
differ by 0.04% across (every pairing).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
In cascades 2 and out (receivers from 60 m) a skinned figure of more than
eight pieces - a dressed person, up to 86 of them - casts one capsule its own
height (radius 13% of it) instead of every visible piece: past 60 m nobody
can tell a garment's shadow from the body's. Animals (a piece or two), rigid
props and the near two cascades are unchanged. The capsule is made once.
Aspen view, same main (30ec1722) and foundations (15f1020), 240 frames, twice
a side: shadow draws 1889 -> 1799; shadow GPU 4808/4819 -> 4710/4714 us;
median frame 17200/17221 -> 16910/17128 us. Most of the far cascades' actor
draws turned out to be rigid props (the census's "skinned" counts draws, not
actors), so this is the smaller of the two.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
From the user's play: clicking an item in the pack opened nothing, and the wardrobe's picks did
not respond - both are a component whose button says `emit click` (ItemCell, LookCell, ListRow),
answered by its user's on-click. An on-* attribute's name is kept as a browser would have it, so
on-click is on-press (tpl_event), and the emit looked for "click" and found nothing. An emitted
name now goes through the same tpl_event.
And a settings cycler's left arrow did the right one's job: any press let go on a select stepped it
forward. ct_activate_at steps back when it is let go over the select's .ui-prev; Enter and the rest
of the select still step forward.
Goldens ui_emit_click (the mouse and a press both reach on-click; with the toolchain before this,
neither does) and ui_select_arrows (1 -> 0 on <, then 2 on > >; before this, 1 -> 2 on <). The 33
ui examples pass.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
From the fence's kept frames in 22 minutes of play:
- A component field that is a record or a list of records was `value_put(o, k, view_val_T(x))`,
a whole new tree every frame (HudPrompt's notifications). view_fill.ludic generates view_set_T /
view_set_list_T / view_fill_T that fill the object and list under the key in place.
- value_list_fit dropped the items it cut off and value_item made new ones as the list grew back,
a Value per item per regrowth (value_item / value_set_strs); the cut-off items are now the
list's spares (Val.spare), and an item of another kind is turned rather than replaced.
- ludic.ui: a scroll box's "scroll" and a slider's "change" fired a fresh Value.float a frame
(sc_walk, scroll.ludic:36); ui_fire_float takes one from a ring kept with the state (fired.ludic).
ui_object_fit_into is exported, for a draw that keeps its list.
- Json.read_file(path): read, parsed, and the file's text given back - Json.parse(Fs.read_text())
kept the whole file on every read (Maroon Lake's settings peeks).
Golden value_list_regrow: a list alternating 6 and 2 items every frame keeps nothing (the toolchain
before this fails it: +128 B new Val from value_item). Game compiles; ludic.i18n/settings/hints/
base tests pass (ludic.ui has none); arena and fence goldens unchanged.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
settings_fill_json(v) writes every saved setting into an object the caller keeps, making a Value
only for a key seen the first time (value_set_str / value_set_int): Maroon Lake's Settings pages
rebuilt the whole tree on every change and dropped the old one. A text read from a file is interned,
and so is a muted card's key ludic.hints does not know, so the parsed tree can be let go whole.
Test: the store filled into a kept object changes its Values in place (6 tests pass).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
i18n_use re-read and re-parsed the .po on every change of language (ev_SettingsChanged, SetLang)
and dropped the old tables for good: the play of 22 minutes left lines_of 650 KB, fs_read_text
409 KB, PoEntry 113 KB and unquote 273 KB unreachable, and three of its @alloc_ok sites were
declared but unbounded. Each language's tables are now filed by index once built (tables.ludic,
I18nTables) and put back by reference; a new probe (i18n_init) or a different file for a code
(lang_add) reads it again.
Test: a language switched back to is its kept tables, not its file read again; a new probe reads
it (9 tests pass).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The user saw kit grass, ferns and flowers lit up at night. Nothing was
emissive and every layer draw binds the scene's lighting; what differed was
occlusion. At night the sky's light is nearly all the light, and it is
scaled by ambient occlusion: the meadow's blades are heavily occluded near
the ground, while a kit plant's own AO map knows nothing of the grass around
it, so it took the full sky. By day the sun hides the difference.
model.vert hands on each vertex's height above the model's base (v_lh), and
a non-blade FOLIAGE plant under 2 m scales its AO from 0.35 at the ground to
1 at 0.9 m. Crowns and the blades are untouched.
Measured in the lab (R3D_AT=tree, same binary, old and new SPIR-V): the fern
against the meadow beside it was 2.05x at 23:00 and 1.66x at 13:00; now 1.58x
and 1.60x - the same relation by night as by day. The flowers' brightest
tenth at night 78 -> 52. The noon frame moves 8.4% of pixels past 8 (the fern
and flower bases a little darker; run-to-run noise is part of that).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The user's play found the aspens' branches swinging like rope. Three terms:
- the whole-tree bend reached 22% of the tree's height at a gust's top (three
metres on a 14 m aspen); 4.5% now, and the side-to-side part at a quarter of
that and slower;
- a branch term: nothing within 0.35 m of the trunk, growing with the distance
out, phased by the direction the branch leaves the trunk (constant along a
branch, so it moves as one piece and its neighbours are out of step);
- the aspen's leaf flutter took its phase from each vertex at eleven radians a
metre, which bent every twig along its length; about three now, and half the
amplitude.
Measured on the lab's `aspen` view (10-frame bursts, crowns only, same binary
with the old and new SPIR-V): pixels moving more than 8 over ten frames 16.8%
-> 9.7%, frame to frame 4.4% -> 2.1%. SPIR-V rebuilt with `ludic-dev shaders`
(vertex stages only; the manifest is unchanged).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
A called boat and one got out of drifted across the lake on the wind (the user's play of the bundle).
The tests hold a boat at its berth in a full wind for eight seconds, and one dropped out on the lake
where it was left.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Every read of the screen made bytes(screen_w * screen_h * 4) and dropped it:
the user's manual play left 22,970,368 B unreachable (exactly 3024x1898x4)
at gpu_vk_draw.ludic:1673 - one per photograph (shots' ph_grab, through
gpu_read_screen_bytes). The read-back is now gvk_read_px, made once and made
again (the old one freed) only when the screen's size changes.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
A block took whatever @lp_site held when it was made. Only malloc, calloc, realloc, concat, the number
texts, the float text, a substring and Text.* set one, so every other runtime helper - intern,
Text.repeat and the string builders, the Fs, Os, unicode, uuid and crypto helpers - was charged to
whichever line had allocated last (walk 9 blamed gvk_tex_storage, m4_new, kept_push$int and
survey_op_reward for intern's 16 B copies).
- Any `call ptr @lp_*` now takes a site of its own, its kind the callee's name when no better one is
known (intern, str_repeat, fs_list, ...). What the helper makes, in however many blocks and
through whichever helpers it calls in turn, is that line's.
- When the call returns, @lp_site goes back to site 0, now named "(runtime) (no site) unsited": a
block made with no site of its own says so instead of borrowing the last one.
- The ECS stores' grows and a mod's registered stores - the only allocations emitted outside
emit_bind - take a site each.
Golden alloc_fence_sites: two lines take turns keeping memory, Text.repeat and a record, every frame
judged. Each report names its own line (25 x +32 B str_repeat at :11, 25 x +16 B new Box at :12);
the toolchain before this charges 23 of the 32 B texts to the new Box line. The other fence and arena
goldens are unchanged; ludic.base's tests pass; the game's frame ratchets are 0 on main 7cb2b164.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Past the 4 MB store lp_copystr falls back to the heap, and past the table
(49152 texts, or 64 probes) lp_intern copies on every call: either way a
program interning without bound would grow unseen. Both paths now call
lp_intern_over, which reports through lp_cap_over once ("intern (4 MB of
text, 49152 distinct texts) is full"), so warn says it and fail stops the
run (exit 87) like any capacity past its promise.
Checked with a compiler built from these sources (selfhost-build): 60000
distinct texts under R3D_ALLOC_FENCE=warn print the line once and every text
comes back right; under fail the run exits 87.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Walk 9's fence named 256 frames of +16 B at gvk_tex_storage:244 and
m4_new:92, always under HudDay's and HudGuide's models. Neither site
makes 16 B: the blocks were intern's copies (lp_copystr), which never set
lp_site and so were charged to whatever allocated last. The HUD's clock
and the guide's distance are a new text every few seconds, and each first
one was a malloc kept for good.
lp_copystr now copies into @lp_istore, 4 MB in the binary (untouched pages
cost nothing), and falls back to the heap only past it; lp_free ignores a
pointer into the store, so a text freed after it was interned is no fault.
value_num_text - the text a screen shows for a number, dropped whenever
the number changed - is interned the same way and its string() given back.
Checked: a program interning 100000 texts gets every one back right, the
same text as the same pointer, a freed one harmless; the headless valley
compiles, with frame_allocs, frame_keeps and birth_leaks at 0.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
water_reflection_pass made view(water_saved, 16, 16) and view(water_saved, 32, 16)
inside the block that makes the reflection target, so every time the target was
made again (a resize, a render scale, a settings change) two more 16-byte views
were made and the last two dropped: the windowed walk's fence caught one such
frame, +32 B. water_saved is the state's for good, so its views are made once.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
AVFAudio keeps a 64-byte AudioQueueOwner for good on every AVAudioPlayer play
after the clip has finished - measured one a play, whatever is called around
it (prepareToPlay, pause, no rewind) and still there after the player is
released; a stop before the play made one every time. The windowed walk showed
it as AudioQueueOwner 73 -> 87 in a minute.
A sound is now decoded once into a PCM buffer and played by a voice of its own
on one shared engine: a player node (the buffer scheduled again on each play,
looping for -1), a varispeed (the rate) and a small mixer (volume and pan).
snd_playing compares the uptime clock with the end worked out when the clip was
played (asking the node where it is made two AVAudioTime objects a call), and
snd_play drains an autorelease pool of its own. The C interface is unchanged.
A harness driving snd_* directly (400 plays past the end, the playing flag
checked during and after each, a loop and a stop, the setters) holds the heap
flat to a block and gets the flag right 400 of 400; the windowed valley
compiles and links against it.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Two blind spots from the windowed walk, where the heap grew about 1 MB a minute and the census read
`frames 0 bad 0 kept 0`:
- @alloc_ok memory was never held to its reason. Every R3D_ALLOC_DWIN judged frames (600) each site's
declared bytes are set against their high-water mark: a new high adds to a streak, a flat window
takes one off, a fall ends it. R3D_ALLOC_DRISE (6) is "declared but unbounded", said once per site
with its line; fail mode exits 86. The census adds `unbounded N` and a `dsite` row per declared
site by its growth since judging began. A list pushed forever grows by doubling, rising too seldom
to make a streak; a record or text made every time (a re-mount's defaults) is what it catches.
- Mem.play() (every screen opened) restarted the warm-up, so memory kept every frame was never flat,
the cap never came, and nothing was ever judged. A rewarm now keeps the first deadline, and past it
has R3D_ALLOC_REWARM frames (120) of grace.
Goldens: alloc_fence_unbounded (a record a frame under @alloc_ok: exit 86, named, census unbounded 1);
alloc_fence_rewarm (kept every frame, Mem.play() every 360: judged and failed at frame 3000 - the
toolchain before this runs all 6000 frames and exits 0). alloc_fence_leak, _declared, _auto,
alloc_ok_private and the four arena goldens unchanged; the game's frame ratchets 0.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
input_device_commit made words(6), words(IN_PADS * 6) and words(IN_TOUCH * 3)
every windowed frame and dropped them - the walk's exit scan found 7 MB of
them unreachable after ten minutes (headless never polls devices, so no
headless run saw it). They are made in in_init with the rest of the input
state and filled in place. input_text's UTF-8 buffer is the state's too,
sized for the most the window hands over (64 units, four bytes each): it
made one per keystroke.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
render3d's stream_new holds its pool through a local (`let live = s.chunks; push(live, new
Chunk)`): the flow edge from s.chunks to live carried ESC to nothing, the Chunk records were
LOCAL, and the arena reset them under the stream - the row and horse scenarios' crash at
0xdddd... in fn_stream_update. An ESC class is now HEAP too, so every alias of kept memory is,
and a value stored through it is kept. Bidirectional alias edges were tried first and over-kept
through returns (el_place, rim).
- examples/lang/arena_alias.ludic: the stream_new shape; poisoned it read 3 3000, now 3 1518
- examples/modules/alloc_ok_private.ludic: @alloc_ok on a module's private function and on a
statement in its private generic, declared at run time (it already passes: a guard)
- the game: frame_allocs, frame_keeps, owned_leaks 0; the lab builds under `arena strict`; the row
scenario poisoned (R3D_ARENA_CHECK=1, 2400 frames) runs clean, bad 0 kept 0
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
A state's field default reading a registry (jn_life_sp: []int = jn_life_species_new(), which reads
Species[sp].population) ran before the registry was filled, because globals were initialized in
declaration order: every gate scenario crashed in L_init_globals. Each global's initializer is now
followed - through the functions it calls and a record's field defaults - to the globals it reads,
and those are initialized first (a depth-first post-order; the source order kept between globals
that need nothing of each other, and in a cycle). Putting every state last is not enough: some
tables read a state's instance too. A test (a state whose default reads a registry declared after
it) crashes on d483c92 and prints '2 4' now; Maroon Lake's headless game loads and plays 180 frames.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
- the call graph is by name, and a local or a parameter named as a function (a float bd, part, bx)
linked to that function: a name the function binds itself is never an edge now.
- drain_actions, generated, calls every reducer; a reducer is reached from its action's dispatch,
so the drain's calls are not edges.
- a fresh value flowing into a local that also holds kept memory is still the frame's (storing it
anywhere kept would have made it ESC): HEAP now keeps only a push's growth off the arena.
- the arena starts at its first use rather than at the first frame mark, so boot's temporaries are
scratch too - dead once the Start handlers return - and a scratch site is never a birth.
Plus ludic.hints' rail and three of ludic.update's one-off lines declared. The arena goldens pass
poisoned. Maroon Lake (d79d189f): 39/11/66 -> 30/9/0.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
ludic.update: whether this copy can update itself is worked out once, when the updater is configured
(the panel asked every frame, building the path to the executable each time); the notes are a list
the state keeps, filled by update_notes_for when the version or the language changes, which the
game calls from its update tick. The feed's address is declared (once, when a check starts).
ludic.steps, ludic.effects, ludic.telemetry: what is left is made on an event and declared with its
bound - an arc's tables, a chapter's columns, a step's fact, an effect's start, end and clear, the
fact pool's growth, the player id, a props record's nesting stack - and two pushes into a caller's
kept list, at most a chapter's steps and the ring's size.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
render3d: shadow_fit, water_reflection_pass, layer_partition_lods and the
GPU cull's scratch are made with the state; v3_dist is scalar; the pushes
into lists sized at start-up, the caps probe, the table growth, the loads
and the constructors declared with their bounds (one statement a line);
the renderer's name made once with the device; the two error messages
given back; the dead lupine models removed.
runtime: a component's text is held interned in its value cell (one copy
per distinct text), so the getter's own text goes with its frame instead
of being kept by ludic.ui's model - 80 of the 83 keeps.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Two more holes of the arena's family, found from ECS's births:
- a function called through a function value (UiClass.make's cmp_x_new, a step list, a System's tick)
has its result flow nowhere the analysis can see, so what it returned looked LOCAL - and a caller
keeping it (ludic.ui's instance table) would keep scratch. Every function taken as a value (fn f)
now has its result kept.
- an entry block has no name, and the analysis only walked named declarations: what an entry stored
was never seen. It is walked now.
examples/lang/arena_fnval.ludic (a factory in a field, its records kept by a pool across frames)
crashed poisoned before and prints '5 1053' as the heap does now; in ludic-dev test. Maroon Lake:
component constructors are kept, not births (birth_leaks 115 -> 103); what fn values return is kept
(frame_allocs 194 -> 210, frame_keeps 151 -> 162); 5293 sites local, 6586 kept.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
With the arena on, a site the escape analysis proves LOCAL is the frame's scratch - made and gone
with the frame - so frame_allocs now counts only what frame code still takes from the heap. And a
scratch site is the arena's whenever the game's frames run (a frame, a click handler, a reducer), so
it is a leak at birth only when boot's code (a Start handler) reaches it, before the first frame.
Maroon Lake: frame_allocs 337 -> 194 with the game's own fixes, birth_leaks 189 -> 115.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The analysis made every component function a frame root, event handlers (cmp_x_on_delete) too, and
every reducer, whether its action is dispatched every frame or once a trip: a component's 'on'
handlers are no longer roots, and a dispatch is an edge to its action's reducers, so a reducer
counts only when frame code dispatches it. A push into a field declared @max(n) is bounded by the
fence's own check and no longer counted. Maroon Lake: frame_allocs 395 -> 337, frame_keeps 188 -> 169.
ludic.photo: the roll's order and a page of it are kept lists refilled in place (the pack's page
asked for both every frame), its kept lists say @max(256), and a shot's tags, a photograph's fact
and a new roll are declared (once per shot, sale or trip). 18 allocs and 9 keeps -> 0 and 0.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
gvk_screenshot kept a buffer the size of the screen per shot (the valley scan's largest unreachable
site), and its PPM header and row buffer; each goes on every way out now. Compiled (steady).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The last kept bytes of the leak gate's pack, shop and journal screens (memory_final, count mode).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
With the arena running every free and every realloc took the slow path (a call to check the range,
then the fence's own test). Now lp_free checks the arena's range inline and hands anything else to
libc unless the fence is tracking; lp_realloc goes slow only for a scratch request, a tracked run or
an arena block. Ready for when the final run's medians ask for it.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The exit scan crashed two of the gate's scenarios (world, swim: SIGBUS and SIGSEGV in lp_mem_scan at
0x0e00000c65800000 and 0x04000004e461c000): a word of data with its high bits set was handed to
malloc_size, and a zone faulted looking it up. A candidate must now be 16-aligned, at or above 4 GB
(macOS's page zero), below 2^47, and outside the frame arena before malloc_size sees it; a block's
own words are read only once malloc_size has said it is one.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
What @alloc_ok covers (a function and its callees, a statement, a statement in a generic's body on
every instance) was counted apart in the frame's verdict, but its sites still carried the bytes the
report and the census rank by, so a declared site was listed as if the frame failed for it. Declared
bytes now have their own per-site counter and never enter live, a site's row or the verdict:
examples/lang/alloc_fence_declared.ludic, all three forms after warm-up, passes the failing fence
('bad 0 kept 0', 1488 bytes declared), with and without the arena, and an undeclared site in the same
frame is still the one listed.
The reachability scan's sites are now the largest first (R3D_ALLOC_SCAN_TOP, 24 by default), and
R3D_ALLOC_SCAN_FILE=<file> appends every site that holds unreachable bytes: the whole table to triage.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The valley's reachability scan (R3D_ALLOC_SCAN) listed render3d start-up objects nothing held; all were
dropped after being handed to the driver or copied into a key: gvk_module's SPIR-V bytes, create info
and handle slot; gvk_program's key parts, .spv paths, bindings and layout create infos; the compute
program's the same; gvk_sampler's and gvk_view_of's create infos; gvk_zero_vbuf_get's 64 KB of zeros;
gvk_layout_key's result (always a copy now, freed by gvk_pipeline once its key holds it) and
r3d_program's defines. Each goes once the handle it made has been read. Compiled (steady).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
A default's node is emitted wherever its record is made, some of it in code the analysis never walks
(a scene's body, a test's); a mark from a walk elsewhere took effect there unchecked. The emitter now
asks for scratch only inside a function or @On body the analysis walked.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
A field's default is one expression, walked at every 'new' of its record: marked LOCAL by a frame's
temporary, it stayed marked when a record a pool keeps was made from it, and that record's list came
from the frame's scratch. The marks are now taken off any node one walk found kept.
examples/lang/arena_defaults.ludic is the case (a pool's record made in frame 3, a temporary of the
same type every frame): foundations 1315baf crashes on it poisoned; this prints '497 124747', as the
heap does. In ludic-dev test.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
memory_final's gate crashed in all 13 scenarios at the first frame of play, R3D_ARENA_CHECK=1 reading
0xDD in ludic.ui's nd_take: a node the pool keeps had come from the frame's scratch. Two holes:
- a call to a generic (ui_kept(list, n)) names the generic, and the analysis knows only its instances
(ui_kept$UiNode), so the callee looked unknown - and an unknown callee was taken to keep nothing.
A generic's call now reaches every instance, and an unknown callee keeps everything it is handed,
but for a short list of intrinsics known to keep nothing; view() shares its list's storage.
- a push's growth into a parameter's list was LOCAL whenever the list was not seen kept, though a
parameter may be a state's list. A site is LOCAL now only when its class is neither ESC nor HEAP.
examples/lang/arena_pool.ludic is the shape (a pool keeping records across frames through a generic
push): built with --arena it prints '7 3498' poisoned on every reset and with the arena off, in
ludic-dev test. On the valley every ludic.ui pool site is kept; 5480 sites local, 6431 kept.
Also, from ECS: a record's field defaults are stored into it when it is made, a global's initializer
is kept, and a component's own functions are frame roots (they run while its page is open).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Target's framebuffer and its colour and depth textures, a Mesh's index buffer (gvk_buf_new /
gvk_buf_delete now @creates / @releases GpuBuffer too), a Prim's mesh and an Actor's own skin clone.
ludic deps --resources over main: resource_drops 0, owned_leaks 0; a probe freeing a Target's fbo
alone was reported for its colour and depth (2), then removed. Compiled (steady).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The escape analysis now follows which state a kept value is stored into (a state parameter, a state
global - each its own class - through the flows to and from it), and every heap site in the fence's
table carries that owner. The census writes, per owning state, what its sites hold and how much that
grew since judging began: 'owner NotesState holds 6400 (+5600 since judging began)'. A keep() or
intern() is a site of its own for this, never scratch and never reported as a keep or a birth. It
needs the analysis, so the arena's (or --escape-report's) build; this is what a soak watches.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
A field marked @owns(PhysShape) holds a handle its record owns. A function that releases one owned
field of a record (body_free(w, s.body)) and neither releases nor hands on another owned field of
the same record type (s.shape) gives the first back and loses the second - the phys_remove bug, at
compile time. ludic deps --resources (or --owned) lists them; owned_leaks is a number --check
ratchets. A test: the function that frees a solid's body alone is the one found; the one that frees
both is not.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
A pipeline has no release: it is kept in the cache for the program's life (gvk_pipe_build stores it).
With every render3d handle annotated, ludic deps --resources over main e447acdd reads 0 drops; a
GpuBuffer made and bound to a local in a game function was reported (1), so the 0 is a real one.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
A function marked @creates(PhysShape) makes a handle one marked @releases(PhysShape) gives back.
ludic deps --resources lists every creating call whose handle is thrown away, or bound to a local
that is never released, passed on, stored or returned, and resource_drops is a number --check
ratchets. A test: a thrown-away create and one bound and never handed on are the two found; one
stored in a state and one released are not. A record's owned fields and a borrow form (a shape
used by several scaled ones) are the second half, with a resource type.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The analysis gives each local its declared or inferred type (a record's field, a list's element,
a call's result, words/floats) and takes a value whose type is a number or a bool out of every flow
and store: a float copied out of a frame's floats into a state's no longer makes the frame's list
kept (shadow_fit, water_reflection_pass, layer_partition_lods). frame_keeps 190 -> 181 on the game;
birth_leaks 564 -> 581, the lists that copy was hiding now seen as made and dropped outside a frame.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Mem.scan() and R3D_ALLOC_SCAN=<frame> (at that frame's mark, when no function is running) walk
every heap block reachable from the program's globals - the states among them - conservatively: each
word that is a heap block's start (malloc_size says so) is followed, blocks made before tracking too.
A tracked block nothing reaches is a leak whatever a frame's totals say; they are summed by site and
printed ('alloc-scan: frame 39 - 32 bytes in 2 blocks ... reachable from no global or state', then
the sites). R3D_ALLOC_EXIT=1 runs the same scan as the program quits. A test: two records dropped in
frame 20 are the two found, the one pushed into a state is not.
The escape analysis now takes free(x) as giving x back (ES_FREED, flowing to what reached x) and a
print's argument as used up, so ludic deps --births lists only what is never given back.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Mem.scan() and R3D_ALLOC_SCAN=<frame> (at that frame's mark, when no function is running) walk
every heap block reachable from the program's globals - the states among them - conservatively: each
word that is a heap block's start (malloc_size says so) is followed, blocks made before tracking too.
A tracked block nothing reaches is a leak whatever a frame's totals say; they are summed by site and
printed ('alloc-scan: frame 39 - 32 bytes in 2 blocks ... reachable from no global or state', then
the sites). R3D_ALLOC_EXIT=1 runs the same scan as the program quits. A test: two records dropped in
frame 20 are the two found, the one pushed into a state is not.
The escape analysis now takes free(x) as giving x back (ES_FREED, flowing to what reached x) and a
print's argument as used up, so ludic deps --births lists only what is never given back.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The retire, free-range and spare-id lists (4096), the per-buffer flags and the prime handles (16384),
the stage and the actor spares (2048), the per-program uniform offsets (4096) and a draw's set key
(130) are each @max'd at the room gvk_startup_state / actor_init made, so outgrowing one ends a dev
run with its name instead of quietly copying. Compiled (steady).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The texture and framebuffer tables grow to 4096/1024 and gvk_prime's queue is made in
gvk_startup_state; gpu_unit_2d's and the actor lists' lazy starts go. @alloc_ok on layer_room (a layer
outgrowing its cap), overlay_init, gvk_read_screen, gvk_hdr_metadata (a settings change), DLSS's
gsl_struct/gsl_fn, an actor's part tables (given back by actor_release) and the pool's fallback.
Birth leaks freed: gpu_caps_probe's create structs, gpu_caps_fake's text, gvk_note's line,
gvk_layout_key's table and each key it grew from, ov_text_wrap's slices, the default sky path.
(ludic deps --births still lists freed sites: its walk does not see free().) Compiled.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
'@max(64) boxes: []Box' on a property's or a state's field is a promise: the grow path of a push to
that field (only the grow path, so nothing is paid until it doubles) checks it, and growing past n is
reported by the fence - 'PoolState.boxes grew past its @max(16) (it holds 16)' - counted under
count, said under warn, and under fail (a headless or dev build's default) the run ends with exit 87.
Mem.over("what") is the same for a package's own table: ludic.base's StrTable past its most
(sb_intern) and ludic.ui's memo past three quarters of MM_CAP no longer quietly copy per call. The
census counts overflows.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
ludic deps --births lists every site the escape analysis finds kept by nothing and not the frame
arena's - boot and load code, a function spanning frames, frame code with the arena off - which is
made and dropped and never given back; birth_leaks is a number --check ratchets. A text used up by +
or == where it is made is freed at once and not counted; nor is what @alloc_ok covers.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Seventeen inline messages (allocation failures, missing conversions, the no-pipeline and compressed-
target warnings, resize and swapchain lines, the test-frame camera line, DLSS evaluate, shadow memory,
stream and terrain debug, the water test) are each a function of their own under @alloc_ok, taking
numbers, so the paths that say them hold no site. post_measure's two 1x1 exposure targets are made in
post_init and DLSS's evaluate structs in gsl_init; DLSS's camera up is a state field. The lists play
pushes into (retired, free ranges, spare ids, the per-buffer gpu flags, the stage and the actor spares)
get their room at start-up (gvk_room_*, actor_room), so a push fits. Compiled (steady).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
keep(x) copies a string, a slice (header and elements) or a record (shallow) onto the heap; intern(s)
hands back one heap string per distinct text from a fixed table in the runtime (FNV-1a, 65536 slots,
copied the first time; past 49152 only copied). Both are how frame code keeps what it made on purpose:
the escape analysis takes the copy as the heap's and leaves the argument LOCAL.
The analysis now records why a class escapes (the store, the event, the global it reached) and
ludic deps lists every allocation frame code makes and keeps - fkeep lines, 'ludic deps --keeps',
the frame_keeps number --check ratchets - leaving out what is under @alloc_ok and a push's growth
(25.5's capacities). --arena-strict (or 'arena strict') makes each an error naming the store, before
anything is emitted. A test: a template stored into a state is the one error; keep and intern of the
next two, an @alloc_ok push and a scratch temporary are not; 195 frames of arena resets under
R3D_ARENA_CHECK=1 later the kept and interned texts read as made, and intern gives the same string.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Render3dState's cam_planes, gpu_u_tmp, q_scratch, ov_nine_buf, the caster test's four points,
ter_bw, ter_scr and water_saved default to their buffers (as ludic.anim's clip state does), so the
lazy starts in cam_planes_update, gpu_tmp, terrain_height_smooth, ter_scratch, the water pass and
gvk_startup_state go. q_euler makes its views of q_scratch once (guarded on the view now). Compiled.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Behind ludicc --arena (or 'arena on' in the program's package.ludic): the escape analysis runs and a
LOCAL site's allocation raises @lp_want for that one call, so it comes from the frame's arena. Two
halves in one mmap reservation (R3D_ARENA_MB each, 256 by default), bump-allocated with a 16-byte
size header, flipped at each frame mark: a frame's scratch is good through the next frame, then its
half is started again (R3D_ARENA_CHECK=1 fills it with 0xDD first). The heap takes over when no frame
is running, off the main thread, or past the half's end; lp_free ignores an arena block and
lp_realloc copies one out. R3D_ARENA=0 turns it off at run time; the census reports each half's
high-water mark. A program that builds text, a list and a record per frame: 29998 heap blocks made
and 18002 freed without it, 8 and 8 with it and 544 bytes of scratch a frame, the same output.
A dispatch's 'new' fills the queue's kept record (E_NEW.b), so 25.2 no longer counts it and 25.3
treats its fields as kept. '@frame' is keyed by property and field: a 'run' field is a root only in
a property that marks it, and 'tick' stays a System's.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
shadow_pass's fbo status and probe block, shadow_bind's two location prints, layer_partition_lods'
LOD line and layer_update's dump are each a function of their own under @alloc_ok (debug switches
only), so the passes themselves hold no allocation site. Compiled (steady).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The renderer starts itself from the first frame (gpu_select), so the device, its tables, the
manifest and programs, grass, shadows, sky, terrain textures and the actor pool read as frame
allocations: each is @alloc_ok as start-up, with gvk_fail (a failure) and the actor census and
texture dump (debug switches). ov_nine's four corner/uv arrays are one floats(16) made in
gvk_startup_state; ac_in_light's four points are made there too. Compiled (steady, and ludic deps
over main 4316ff97).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Each morning's repopulation made a new WildAnimal (and the game a new drawing, actor and skeleton,
keyed by an ever-rising key) for every animal that had gone. A legend's record is never taken, nor
one still in a crowd. reuse_test: a hundred removes and makes are one record.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
emit_escape.ludic: every value is in a class, joined by flow edges (a let, an assignment, an argument
into its parameter, a result into the call) and store edges (a field, an element, a push). HEAP (a
parameter, a state, a global, what an unknown call hands back) flows forward; ESC (stored into
something HEAP, into a global, into an event's fields or named values, handed to an unknown callee)
flows backward, and from an ESC or HEAP target along a store. A load is its base's class. A site that
is neither ESC nor in a function reaching Mem.frame is LOCAL (Node.uns = ES_SCRATCH). ludicc
--escape-report prints each site and the totals; nothing is emitted differently yet - the arena that
allocates the LOCAL sites is next.
@alloc_ok on a generic now covers its instances (kept_push$NetFact is under kept_push's), and a
statement's @alloc_ok is carried on the node (Node.uns), so a generic's clone keeps it.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
A frame holding a new peak of facts made a record and grew the free list; now nothing is made until
64 are held at once. The two record-count tests hold the 64.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
ludic deps --allocs (25.2) found phys_float's five pushes per drop reaching the water and
phys_sink's filtered copies per removal.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
ludic deps --allocs (25.2) found npc_pick_name making a list of the body's names at every spawn and
np_left making a new list of the recent names at every retire.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
actor_own_skin(a) gives an actor its own clone of its model's skeleton and actor_release frees it
(skin_clone_free: the pose, matrices, views and scratch the clone made; the rest data stays its
source's). gvk_startup_state, at the end of gvk_init, makes the scratch and tables a draw used to
make on first need (gvk_tmp_buf, the pipeline fast cache, the set key and per-program uniform
offsets pre-sized to 4096 programs, skip/seen/size words, spare and retired lists, the grass cull's
words), and those frame paths no longer start them. Compiled (steady).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
@alloc_ok with its reason on what is made once per resource and kept (textures, programs, samplers,
views, layouts, memory blocks, the pipeline cache in gvk_pipe_build), on resize and swapchain
remakes, on screenshots and dumps, on a world being set up (streams, layers, water, post, bakes), on
loads (gltf_load, skin_load, tex_load*, png_decode, fonts) and on R3D_PROF / drawstats.
gltf_cached's hit path is its own and makes nothing; the miss (gltf_cached_load) is declared.
m4_look_at (now over m4_look_at_xyz) and m4_inverse work on scalars, and cam_update makes no scratch.
Left: lazy first-use starts, error and debug prints, and scratch made and freed each call (churn,
plan 25.3). Compiled (steady, and ludic deps over the game).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
A field declared '@frame run: fn(...)' makes every function stored in it a frame root, as a System's
tick is. @alloc_ok("why") before a statement takes that statement out of frame_allocs and makes what
it allocates declared at run time; a function holding one keeps the fence's scope depth and puts it
back at its return, so a return inside the statement cannot leave the scope open. @alloc_ok above
'export function' was lost - export parses the declaration one call down - and is now carried to it.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The census reads ludic.physics' jph_heap_bytes/_peak and the Vulkan runtime's lvk_ac_bytes/_peak by
name (dlsym, so nothing a package declares is declared twice) and prints jolt, vulkan and the rest
of the heap apart. An @alloc_ok function counts a scope in and out (@lp_fdecl): what it and its
callees make is marked declared in the side table, reported as 'declared' in the census and left
out of a frame's verdict and of Mem.kept().
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The census reads ludic.physics' jph_heap_bytes/_peak and the Vulkan runtime's lvk_ac_bytes/_peak by
name (dlsym, so nothing a package declares is declared twice) and prints jolt, vulkan and the rest
of the heap apart. An @alloc_ok function counts a scope in and out (@lp_fdecl): what it and its
callees make is marked declared in the side table, reported as 'declared' in the census and left
out of a frame's verdict and of Mem.kept().
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
gvk_pipeline_fast's miss (the build and the six cache pushes) is its own function, so the fence can
declare it: @alloc_ok("pipeline cache: one per variant the game draws") once lang/memory-fence's
compiler is merged (the annotation is not known on this base). Compiled.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
deps_reach's graph gains the handlers and each @On body (an emit reaches its event's listeners).
Roots: a handler in a frame phase, every reducer, an @On body, and a function stored as a System's
tick. Every allocating construct in what they reach - new, a list literal, push (grow), text built
by + or a template, words/floats/buffer/bytes - is a falloc line with the shortest chain from a
root (root>..>last six), and the program's count is frame_allocs. @alloc_ok("why") on a function or
a handler takes it and what only it reaches out; the reason is required. ludic deps --allocs lists
them, and frame_allocs is a number --check ratchets. Maroon Lake starts at 2661.
Not yet: @frame on a step list's field (only 'tick' is a root field so far), statement-level
@alloc_ok, the three lints.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Jolt's Allocate/Reallocate/Free/AlignedAllocate/AlignedFree go to libc's malloc with the size in a
16-byte header, and atomic counters keep live bytes, the peak and the blocks asked for (its job
threads allocate too). The fence reads the three exports extern_weak to judge Jolt by its plateau.
jolt_heap_test: a ground in and out 1100 times holds 46,482,514 bytes after the first 100 and after
all of them, the peak does not move, and a closed world gives back every byte it took. The macOS
library is rebuilt; the Windows DLL still has to be rebuilt on the PC.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Plan 25.1c: vk_mac.ll's @lvk_ac (posix_memalign under a 16-byte header of scope/offset/size, atomic
counters) passed at every render3d create/destroy (49 sites; the caps probe keeps its own null pair);
lvk_ac_bytes/_peak/_allocs/_scope_bytes for the fence, Vk.alloc_bytes. vk_win.ll: null and 0.
MoltenVK 1.4.2 counted 0 live bytes through them in steady. Fence findings: gvk_pipeline freed its 17
create infos (and reuses one bufs list); actor_init fills ac_spare with 512 records (actor_fresh,
m4_new, v3_new at first placement in play). Compiled, not run (the user's call).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The fence found np_fact_record making a record in five frames of every scenario: the pool grew to
each new peak of facts held at once, and q_unheld's free list and the queue's two lists grew with it.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The census's native line is malloc's live bytes over every zone since judging began less what
Ludic's tracked blocks kept - the libraries' and drivers' growth, read before the census file is
opened. A tracked block counts malloc_size(), not the size asked for, so kept is what the heap pays
and the residual carries no rounding. @malloc_zone_statistics is declared once, by the fence or by
Os.heap_bytes.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Every allocation the compiler emits goes through @lp_malloc/@lp_calloc/@lp_realloc/@lp_free, and a
Ludic-level one first stores its site (function, file, line, kind) in @lp_site. Off, that is one load
and a predictable branch (30 M allocations: 0.87-0.91 s against 0.87-0.90 s on leaks2).
On (the default in a headless build, and windowed under R3D_DEV), tracking starts at the first frame
on its own and judging once R3D_ALLOC_WARM frames in a row kept nothing (600) or R3D_ALLOC_WARM_MAX
after (re)start; Mem.play()/Mem.rewarm() sends a load back to its warm-up. A judged frame that ends
holding more than it began with is reported by site with its callers (the unwinder, taken only once
judging) and fails the run with exit 86 (R3D_ALLOC_FENCE=off|count|warn|fail). R3D_ALLOC_CENSUS
writes the totals and top sites at exit. The build's defaults are --fence=, --fence-warm=,
--fence-census= or a fence line in the program's package.ludic; the environment overrides them.
The runtime is IR (emit_fence_ir.ludic, generated from a template); tracking is a side table in one
calloc'd region, so no block carries a header and pointers crossing to natives stay safe. Examples
alloc_fence, alloc_fence_leak and alloc_fence_auto with cases in ludic-dev test; reseeded.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The encoder appends into RtJsonState's buffer (grown only past the biggest document yet): ints and
Q16.16 fixeds written as digits in place, floats through string() and freed. Json.encode copies the
answer out once; Json.write_file hands the buffer to Fs.write_text (.tmp + rename) and keeps nothing.
json_saves.ludic: exact text, the file equals encode, parse round-trips, 1000 saves grow 0; clean
under MallocScribble. json_quote (the + builder) is gone.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
A prompt that comes and goes as a player walks (co-op's netleak: in_get, cmp_*_new, bd_class, value_slot/put)
made a new record, props and model on every mount, and an action's call answered into a new Val (ev_call_with).
examples/library/ui_remount: two thousand comings and goings hold the heap at 0, and the counter starts at 0.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
An event was a tree of values encoded into a new string, a dropped line was never given back, and
every batch and save joined the queue into another. Now: telemetry_props / telemetry_str / _int /
_bool write the properties as JSON into a buffer the state keeps; the line is written beside it,
its time worked out from the clock's seconds (TelemetryWorld.clock_s, was stamp); its bytes go
into one ring (ring_bytes, at most queue_max lines), the oldest overwritten past either; a batch
and the file are written into a third kept buffer and go as bytes (TelemetryTransport.send takes
the bytes and their length; Http.body_bytes, Fs.write_bytes). The facts are pooled.
tests/ring_test: ten thousand events past the cap in lines and in bytes, 0 bytes of heap; the line
exact JSON, escaped, 2000-02-29 right; the batch puts the id in; the file round-trips.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
A co-op guest loading its models segfaulted in copyBufferToBuffer under vkQueueSubmit (0x68, AGX
LegacyBlitContext): a buffer primed into the frame's command buffer was released later in that
frame, and gvk_buf_release destroys a buffer no draw has marked, so the copy read a freed one. The
copies now run in their own command buffer before the frame's begins, checked against the slot's
handle as they are recorded.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Math.max, Math.sqrt and the rest (and the bare min/max/abs/clamp) are computed inline by the
emitter, and L4 gave each the unknown type, which agrees with everything: Maroon Lake's trail
put Math.max(5, n) into Notify's string field a1 and it failed in LLVM ("%t63 defined with type
i32 but expected ptr"). It was never about two dispatches on a line - one is enough. The checker
now mirrors the emitter: a float/double first argument gives that type (sign an int); otherwise
min/max/abs/clamp keep the first argument's type, sign/floor/ceil/round/posmod/wrap/ping_pong are
ints, the rest fixed. Named arguments or an argument it cannot type leave it unknown.
rejected/math_into_text is the case. Reseeded.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
A chunk's trunks and boulders put in and taken out made a new shape each time and freed none. The
table is by the body's index, made once at the world's size. tests/reuse_test: 1000 owned bodies put
and removed hold no more shapes and 32 bytes of heap.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Metal pools the command storage a command buffer grew to and grows it only when one encodes more than
any before (IOGPU under vkQueueSubmit < fn_gvk_once_end: +2-7 KB a window while a frame's draws
climbed, 192 KB in the trail scene). gvk_draw submits the frame's command buffer after
gvk_cb_cap_of draws (R3D_CB_DRAWS, 0 = none) and r3d_warm_commands draws that many through the
normal path once r3d is ready. steady: a frame drawing 20 to 200 actors grows 0 (was 2.7-7.5 KB).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Found by reading every builtin (Os.platform's 8 KB per call started it). Log builds its line only at
or above the threshold and frees it; DateTime.format folds through + so its pieces go; Input.text
encodes into one buffer; Path/Mime/Fs/Os free their temporaries on every path; string results of
Text/Path/Mime/DateTime/Os dirs are fresh and Text frees a fresh argument. Reseeded.
runtime_temps.ludic: 19.8 MB -> 0 over 20,000 rounds, 64 KB -> 0 over 200 of file work; clean under
MallocScribble. string_temps still 0.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
lp_os_platform and lp_os_arch malloc'd 8 KB per call (the uname buffer) and kept none of it:
string_temps now asks both every round, 327 MB over 20,000 before, 0 after. Reseeded. render3d:
MoltenVK made a mapped buffer's MTLBuffer at its first bind (fn_gvk_draw +4 blocks in the boat
window); gvk_buf_reserve queues it and the next frame's command buffer copies 4 bytes out of it.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
stream_update made a Chunk and a words copy per new chunk (fn_stream_update +39 blocks / 13.2 KB a
window with the hiker in the drifting boat). The pool holds STREAM_MAX_CHUNKS records; the arena
is twice the layer's cap; eviction compacts it; a chunk that cannot be kept is gathered from the
scratch. stream_clear_all frees records, lists and streams. steady: 300 new cells, cap 256: 156 KB
before, 0 / -4.9 KB after, and every kept chunk's data checked against its cell.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
np_queue made a NetMsg and a buffer per message, every flush a new keep list, and every message
that came in a NetMessage and a buffer; nr_text a new string per text read. A party plays at
dozens a second, so all of it is kept now: a pool of MTU-sized records for the queue (a peer's two
lists swapped by the flush), the inbox's records in two halves swapped when a message finds the
inbox empty, net_written's one record, a relay hello's and a STUN request's bytes in one scratch
buffer, and the texts read out interned. Tests: 6000 messages with the heap flat (Os.heap_bytes),
and an inbox record that holds still across a drain and comes back two drains on.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
tests/reuse_test: 200 walkers made and removed hold one place and 0 bytes of heap (Os.heap_bytes);
twenty closes and opens keep the same shape and walker lists.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The pools kept every command object ever recorded, so the heap grew each time a frame drew more than
any before (fn_gvk_draw under vkCmdBindVertexBuffers/BindIndexBuffer/Draw in the leakcheck; ~650 B a
draw). Off: 3.7 ms a frame either way over 520 actors. steady: 20 to 200 actors grows 5-7 KB with it
off and 120 KB with it on (bound 16 KB).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
th_fact made a ThingFact per placement, removal and use, and the valley places and removes Things
all day. The records come back two drains after they were handed out, so a reader placing a Thing
while it reads the last drain's list never sees one change (the test holds exactly that).
thing_use takes ThingsState mut, since it pushes a fact.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
No bake and no new file: the skeleton comes from the skin's parents and rest pose (its own joints
and their ancestors - a kit holds several rigs), a clip from anim_read_doc's channels. Built by
native/build.sh from the pinned release; the Windows DLL imports KERNEL32 alone and passes there.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Vk.heap_bytes pulled the Vk module - and on lang/uifree the GL window path - into a plain program,
which then failed to link (_cgl_offscreen, lgl_GetError). Os.heap_bytes is malloc_zone_statistics
through a weak reference (0 where there is none, and on Windows). Reseeded. Docs for it and for
Json.free / Json.free_all.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
gvk_tex_storage freed none of ici, out, req and vci (8 blocks a two-texture model); tex_load_ex kept
dds_path_of's string. Found with malloc_history over 400 load/release rounds (712 bytes a round, all
of it these). steady: the model's bound is 4 KB over 200 (was 1.6 MB), and the frame is the least of
three settled windows - a valley self-test beside it read 87 KB once.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
s[a .. b] is always a copy, so it is freed once a +, a comparison or print has read it. Reseeded.
string_temps.ludic adds a slice compared and a slice concatenated each round (960 KB over 20,000
before, 0 after) and a kept slice read after its +.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The left half of a + chain, a template's pieces and holes, a number's text and a side made only to be
compared are marked fresh and freed after the +, ==, != or print that reads them. lp_int_str and
lp_long_str move their digits to the start of the buffer, so the pointer they return is the one
malloc gave. Reseeded. examples/lang/string_temps.ludic: kept intermediates stay good, and 20,000
rounds grow the heap 0 bytes (2.9 MB before).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Under the suite's parallel load a frame read ~80 KB high: MoltenVK's completion handlers release a
finished command buffer on their own thread and lagged. Settled, 12 of 12 runs four at a time pass;
the frame's bound drops from 64 KB to 4 KB.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
A freed texture or buffer id goes on a spare list the next one takes; tex_note_size keeps sizes by id.
model_release frees prims, meshes, material names, the skin and leaves gltf_cached. actor_release
takes an actor off the stage and actor_new reuses its record (ECS's Things come and go all day).
steady.ludic: an actor round at 0 bytes over 2000.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
thing_put and thing_spawn made a new record for every Thing placed, and the world places and removes
them all day (an animal's sign and bed, a drop, a fish), so play grew by a record per placement.
Removed records wait in a queue compacted in place; once 32 wait, the oldest is set back as new
with a new uid. ludic.base's grid_reserve makes the buckets for n rows now, since a rehash in play
leaves the old bucket array behind. Tests: a record comes back only past the lag, as new, with a
new uid, the indexes agree, and the spare queue stays bounded.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
A game placing props while it plays (a sign an animal leaves, a fish, a bobber) loaded each through
gltf_load, which reads the file and parses its document every time. gltf_cached finds the same
(dir, file, node) by comparing the names in place and hands back the model the first load made;
a model whose document is read after the load (ludic.anim's clips) still goes through gltf_load.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
- Json.free_all (value_free_all): a parsed tree's nodes, lists and strings. render3d frees each
glTF document that way at the next load; the names kept out of it are copies (a primitive's
material, a skin's joints, an animation clip's name in ludic.anim) - a model's strings were
~640 KB left behind per load
- jp_number made a digits list per decimal in a document and never freed it
- gvk_layout_id matches a mesh's layout as numbers in a scratch made once, against the layouts
known end to end; a new mesh no longer builds a key string
- gvk_mem_new puts a new block into the record of one given back rather than appending, so a
buffer made again every few frames no longer grows the block lists
steady.ludic adds a glTF parsed and freed whole 200 times: 0 bytes (38,400 before the digits fix),
beside the buffer path and the frame, still 0. A model loaded and let go still keeps ~2 KB a round
(texture and buffer handles are not reused yet); it is bounded at 8 KB a round.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
n x n samples of chunk (i, j) of a size_m grid from the terrain's corner, row-major into the
caller's buffer with nothing allocated, each the (1 4 1) / 6 B-spline filter of the texels under it
(ter_spline_at): what ludic.physics' jph_shape_heightfield_bspline makes of the whole map, so a
chunk's physics ground matches the drawn one and its neighbours' to the bit. The read-back lives
for the whole map (terrain_generate to terrain_unload). examples/rendering/chunks.ludic, in the
suite, checks the shared edges, a sample against the filter by hand, and a short buffer refused.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
gvk_mem_new made a one-slot []pointer per allocation, and turned the requirement's size and
alignment into strings to read them as ints; gvk_list_drop_last rebuilt the spare-record list to
drop its last entry; gvk_mem_id did the string round trip on every free. One slot is kept
(gvk_map_slot), int() truncates a long, the spare list pops. Every Text.to_int(string(x)) in
render3d is int(x) now.
Vk.heap_bytes() (vk_mac.ll: malloc_zone_statistics' size_in_use; 0 on Windows) and
examples/rendering/steady.ludic, in the suite: a buffer released and made again 5000 times and
600 whole frames gain 0 bytes each - the allocator before this, 1,120,000 over the 5000.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
nav_load_index indexes a saved mesh's tiles (cell, offset, size, ref) from their headers and keeps the file open, with no tile in. nav_tiles_keep reads in the tiles within rin of any player and removes those past rout of all; the file is read through Ludic's pack-aware file_open. The file format is unchanged. tiles_test: tiles come and go with the point, a path works across the seams once they are in, and a reset closes all. Measured on the baked maps (the 900 / 1100 m ring round the start): Maroon 617 of 1807 tiles in, 8.9 MB instead of about 27 MB for a person and 5.4 MB of about 14 for a large walker; Lamar 391 of 1147. Both libraries are rebuilt; nav 17/17 on the Mac and the PC, DLL KERNEL32 only.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Metal aborted a net.sh guest ("MTLPixelFormatBC7_RGBAUnorm is not color renderable"): the crash
report's main thread was in vkQueueSubmit from gvk_once_end, MoltenVK encoding a vkCmdBlitImage
through a render pipeline. water.ludic bound its reflection by sampler name and then asked the
BOUND texture for mips - on Vulkan the bound texture was not the reflection but the last one
bound, since 23.3 a BC7 kit texture. gpu_bind_sampler now makes its texture the bound one, as
OpenGL did (and water binds it explicitly); gvk_tex_mips and gvk_mips_now skip compressed images;
gvk_pass_begin leaves out a compressed colour attachment and says so.
tests/net.sh passes (main e515bd87 built against it; host and guest agree on 196 animals, the
people, the board and the lost hiker).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The first generation pass's R32F height is kept as ter_height_tex (not copied into an RGBA32F),
so every reader of the height - placement, the read-back, physics, selftest16's 9 mm - sees the
same 32 bits. ternormal.frag writes the baked normal's x and z into ter_normal_tex (RG16F), and
the six places that read it (terrain.frag twice, tersun.frag, grass.vert, grass.mesh,
grass_cull.comp, which takes a third texture at binding 6) rebuild y. SPIR-V regenerated.
Maroon Lake's play, headless Vulkan: 2793 -> 2647 MB. The camp's frame: 0.066% of pixels differ by
more than 8 (mean 0.024/255), isolated grass blades at the slope gate. ludic-dev test 307/307.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The per-frame sweep over physics, character, vehicles and nav found three fact makers that made a record per fact: contacts and splashes, the character's and the vehicles'. Each keeps a pool now, as wildlife's and npc's do (q_unheld). phys_sink, asked on every removal, built two new lists each time; it filters into a kept spare pair and swaps. ludic.nav's paths and crowds already allocate nothing, and none of the four builds a string per call. What remains at load, reset or a world swap is not per frame. vehicle_feed takes its state mut. vehicles_test: 1000 fed facts, drained turn by turn, use at most 4 records. All packages 421.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
dispatch lowered to ludic_act_push(k, new A { ... }): a fresh record every dispatch, and an input
system dispatches Move and FrameTime every frame. The queue now keeps a list per action
(kept<k>, used<k>); ludic_act_new__A hands out the next (made only when all are queued), new's
emitter fills it field by field as it fills a fresh one (every field, default or given), and
drain_actions sets every used<k> back to 0 once the queue is empty. A reducer only ever reads
its action during the drain, so nothing sees a record after it is reused. Actions are visible
to the program's file, where the queue lives, as reducers already were.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
wl_fact made a new WildFact per fact, about one a frame, and Ludic never gives one back. ludic.base's q_unheld(q, pool, out) lists the records a queue no longer holds: neither waiting nor handed out by its last drain, which is good until the next drain. ludic.wildlife and ludic.npc now keep a pool, hand out a free record, and make a new one only when every record is held. Taking a record writes the state, so wl_fact / np_fact and the few callers holding their state read-only take it mut. Tests: q_unheld's rules (an empty drain holds on, the safe side); 3000 prints drained frame by frame use at most 4 records; 500 arrivals drained turn by turn use at most 4. base 37, wildlife 27, npc 17; all packages 420.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The device's textureCompressionBC is asked for and remembered (gvk_has_bc). tex_load_ex prefers a
DX10 .dds with the full chain beside the .png (not for an edge-padded cut-out atlas):
texture_dds.ludic reads BC7 / BC5 / BC4, gpu_tex_compressed makes the image with every level and
no colour-attachment use (a compressed image is only sampled and copied into), and
gvk_tex_upload_blocks copies each level's blocks from one staging buffer. A colour map is BC7
sampled as sRGB, a data map BC7 read as it is. examples/rendering/bc.ludic holds it, in the suite;
ludic.lab's plate carries its .dds (its three shots render at 56-60 dB against the .png's).
ludic-dev test 307/307, no Vulkan SDK in the environment.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
native/build.sh builds MoltenVK v1.4.2 from its pinned, checksummed tag (its dependencies at the
commits its ExternalRevisions pins), thinned to arm64, id @rpath/libMoltenVK.dylib, signed ad hoc;
its licence goes in native/LICENSE-MoltenVK. Every render3d program links it through its rpath -
the package's lib/ while developing, Contents/Frameworks in a bundle, where ludic bundle puts and
signs it - and vk_mac.ll also looks for @rpath/libMoltenVK.dylib (after an SDK loader, so the
validation layer still stacks in development). The suite's SDK stand-in (vk_env) is gone: the
render checks draw on the MoltenVK the package carries. gpu_is_gl() removed; nothing calls it.
ludic-dev test 306/306 with no Vulkan SDK in the environment.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
nav_crowd_add_fixed adds a walker with no steering, and nav_crowd_place puts it (snapped to the mesh) where the player is, with the velocity they have, each frame. The others plan round it by that velocity, and whatever it was pushed by is undone at the next placement. crowd_test: six walkers aimed through a standing player come no closer than 1.71 m and all get past, and the player stays where it was put. nav 16/16 on the Mac and the PC; DLL KERNEL32 only.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
nav_crowd_add takes how hard a walker keeps its distance, and WildSpecies.spacing (default 2) is a species' room from others walking by, which the valley hands the crowd. Both libraries are rebuilt; nav 15/15 on the Mac and the PC, DLL KERNEL32 only. wildlife 26/26.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
- gvk_sampler built a key string on every call it was reached (a texture read two ways in turn
misses its per-texture cache each time); samplers are found by their seven numbers instead
- gvk_tex_mips_into allocated a VkImageBlit per level, every frame (the exposure measure's
chain); it comes from the scratch ring, as does gvk_tex_grow_mips's copy
- gvk_program_new made a new program - modules, pipelines - on every call, and every actor asks
for one; a variant is now made once and shared (a program is immutable), and one that cannot
be made stays 0 for every later asker
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
npc_set_crowd hands a person to a crowd. Its walk no longer steps or asks the way: its want is goal_x / goal_z / goal_speed (0 when it is not walking). npc_moved hands back the crowd's place: position, ground, heading from velocity, gait. Arrival, the act and the give-up after twenty seconds without progress are unchanged. Nothing allocates. npc 16/16.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
wildlife_set_crowd(a, true) hands a walker to a crowd. Its step no longer moves it, and its want is two fields on it: goal_x / goal_z (the point it heads for or faces) and goal_speed (0 when it stands). A route reads those and hands back where the crowd put it through wildlife_moved(a, x, z, vx, vz). That sets its place, its ground, its heading from its velocity and its prints, and updates its row's columns so the spatial index finds it there. Nothing allocates per animal per frame: the want is the animal's own fields, and moved writes in place. The prints and the turn counters moved to walked.ludic (steer.ludic was at the 100-line limit). crowd_test holds it; with crowd off every old test is unchanged. wildlife 26/26.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The shim builds DetourCrowd from the same pinned Recast & Detour tag. nav_crowd_start puts a crowd on a kind's mesh, with the mesh's tastes as its filters. Walkers are added (snapped to the mesh), sent and sped through verbs, stepped together, and read back into nav_agent_*. nav_crowd_us times the stepping with the OS clock. Dropping a mesh drops its crowd first. On the test meadow, twenty walkers crossing head-on never come closer than their two radii (0.7003 m), all arrive, and a step costs about 12 us. nav 15/15 on the Mac and the PC; the DLL still imports KERNEL32 alone.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
e634177 started a layer at 256 instances and grew it in layer_add and the stream gather, but a
game writes l.inst directly (Maroon Lake's track prints, trees and rocks), past what had been
grown: "index out of range: 2048, len 2048" in play. layer_new takes its cap up front as before;
layer_reserve(l, n) is exported for a caller that writes l.inst itself once layers start small
again, opt-in.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
17.10 found almost every animal turn-back was for a climb, where the navmesh (40-45 degrees) is stricter than the step's 1.2 per metre. So the frame-sized test was refusing a few centimetres of steep ground: a pebble a walker steps across. It also refused sooner the faster the frame rate. wl_climb takes the ground half a metre ahead along the way the step actually goes. climb_test holds it at 60 and 240 Hz: a pebble is crossed and a wall is not. Both tests fail under the old rule. wildlife 24/24.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Found with malloc_history over 1200 frames of play without the interface (the first Ludic function
on each allocating stack, live bytes at two marks): sh_loc built `name + "[0]"` per program per pass
per frame - 24.6 MB of 33 in the window; its callers pass both names as literals now. tick_set fills
the frame loop's one Tick instead of tick_new making one a frame. A ridden boat or horse said
VEHICLE_MOVED as a new fact every frame; the metres are added up in VehiclesState and taken once
(vehicle_moved_take / _kind / _x / _z). Play's growth without the interface: 92 -> 14.7 MB a minute
(maroon-lake tests/leakcheck.sh); what is left is phys_push (Physics' fix on lang/nav) and the HUD's
strings. Packages 407, ludic-dev test 305 pass.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
- model_release(model) frees each primitive's mesh, and each texture once no other model uses it:
the glTF texture cache counts the primitives using each texture (a path loaded again, and a LOD
chain borrowing its LOD0's material, each take a reference); the last one frees the texture and
forgets it along with any remembered material naming it, so a later load is a fresh one
- a scatter layer's instance and sort arrays start at 256 and double as layer_add or a stream
fills them, to the layer's cap, instead of the whole cap up front (0.4 GB in Maroon Lake)
- gvk_tex_read's copy struct comes from the scratch ring
- examples/rendering/release.ludic holds it (RELEASE OK on Vulkan and OpenGL), in the suite
smooth renders pixel-identical before and after (max |d| 0). ludic-dev test 306/306.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
gvk_pipeline_fast cached only pipelines it made, so one the driver refused was attempted again on
every draw, each attempt building its key string and create structs anew; the refusal is cached
too (a Vulkan program's variants are fixed when it is made, so it would fail the same way).
stream_evict made a new chunk list per eviction and never freed the evicted chunks' records; it
compacts the list in place and frees each evicted chunk with its data.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
terrain_generate, terrain_use_dem and terrain_use_ortho each overwrote the previous height
texture (256 MB of RGBA32F at 4096^2), the heights read back (64 MB), the DEM texture and the
orthophoto's texture and pixels. terrain_reload unloads first, so the world swap was already
clean; any other caller building a map over a live one now lets the old go, and the read-back
array is reused, being the same size for every map.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
phys_pose, phys_ray, phys_push and phys_walker_pose each made a new record per call, some several times a frame, and Ludic never gives one back. They now fill the caller's PhysPose / PhysHit / PhysPush / PhysWalk and return whether they found anything. phys_overlap returns a count; phys_overlap_id(i) reads each id back. The package's own uses (phys_resolve, phys_row, phys_walker_move) read the values buffer directly. Tests take small allocating helpers. physics 20, character 15, vehicles 8.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
- every texture upload malloc'd a CPU copy of its pixels and never freed it (236 MB over the
valley's boot, and again for every model loaded later): the pixels are converted straight
into the mapped staging buffer
- the per-level and per-layer views gvk_view_of made outlived their texture, and on MoltenVK a
view keeps its Metal texture alive: a released texture takes its views with it, and the cache
is keyed by numbers instead of a string built on every call
- the Vulkan structs filled for a draw, pass, barrier, descriptor set, buffer, allocation or
upload (about sixty call sites) come from a reused 1 MB scratch ring (gvk_tmp)
- the descriptor-set cache and the retired buffers are emptied in place, not replaced; the grass
cull's dispatch arguments are made once
- macOS drains an autorelease pool each frame (Vk.frame_pool, lvk_frame_pool in vk_mac.ll)
- R3D_VK_PROF reports Vulkan objects made and destroyed by kind, and every cache's length
- examples/rendering/smooth presents through render3d, so it runs on Vulkan too
The full valley on headless Vulkan loads to 2.18 GB and holds (it passed 8 GB while loading
before). ludic-dev test 305/305, selfhost-test 33/33.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The rest of the leak Physics found: a drain that carried facts gave its list away and made a new
one, every queue every frame something happened. The queue keeps two lists and hands one out while
the other fills; a drained list is good until the next drain of that queue (nothing in the game or
the packages keeps one past it). With ab34f82's empty drain and in-place clear, a queue allocates
nothing in steady state. queue_test holds the reuse; every package (403) and lab/unit (90) pass.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Every per-entity store (@S_ components, @H_ flags, alive, kind, freelist, owners) is a heap block
L_grow doubles from 1024 as L_alloc hands out a slot past it, the new slots zeroed; each site loads
the store's base where it indexes it (ecs_base, its registers %ecsb* so a raw function's t0 labels
cannot collide). Prop.has bounds against @L_cap, Pool.capacity answers it, a mod's registered
stores grow with the rest, every main grows the stores once before anything reads them. A snapshot
records its slot count first and a load grows to it before reading back. The overflow stop of
1c7ce84 is gone with the wall. ludic-dev test 305 passed, selfhost-test 33 passed; 1000 / 5000 /
100000 entities spawn and count.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
nav_us() gives the microseconds spent in nav_path across the loaded meshes. The shim times each path with the OS's monotonic clock (clock_gettime, or QueryPerformanceCounter from KERNEL32), so the DLL still needs no C++ runtime. Both libraries are rebuilt, and 13 tests pass on the Mac and the PC.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The count is wildlife_turned(), the 75-degree fallback an animal takes when the way was not enough. It is a measure only and is not saved.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
A step round what is in the way, a target planned again after twenty seconds without progress, and a target given up are each pushed as NPC_F_STUCK (how, x, z), so the game can see where the way failed a person. The lake test expects one when a walker turns back from the water; the way test expects none when the world gives the way.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
It sets a target straight away from the threat (40 m when it flees, 20 m when it is wary) and heads for it with wl_head, so the way port takes it round water and cliffs. Where the world gives no way, it still runs straight away as before. way_test now has a flee round the wall's end.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
L_alloc handed out a fresh slot without a bound, so the 1025th spawn wrote past the end of every
component array. It stops with a located message naming the store's size and where many things
belong (ludic.base's Table). Growable stores are plan 24.8: the save, rollback snapshot and mod
table write the stores whole at a compile-time size, so that is a file-format change. Reseeded;
ludic-dev test 305 passed, selfhost-test 33 passed.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
300 parses of a 446 KB glTF: 2118 MB before, 528 MB with the one-allocation string, 94 MB freed.
A headless Maroon Lake at play: 2334 -> 2195 MB by footprint.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
A streamed world loads and lets go of a chunk's entities together, so a chunk is a first-class
group: chunk_of(x, z, size) packs a chunk's cell into an int, tb_remove_all(tb, ix, v) removes every
row an index files under one value (the observers told of each). IntIndex kept one list per value
up to the largest, which a packed chunk number (hundreds of millions) turned into hundreds of
millions of empty lists and a 12 GB test run; a value past 1024 now gets its slot through an IntMap,
so a sparse value costs one list.
ludic.things files every Thing under its 256 m chunk (TH_CHUNK), kept by every move;
things_chunk_count and things_drop_chunk (the port and the facts told of each removal, as
thing_remove tells them) are what the memory budget's streamed chunks ask. The moves are their own
file (moves.ludic). Tests: ludic.base 35, ludic.things 8, all under a 2 GB cap (24 MB peak).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Every pool thread runs the worker at once with the same states, so a mut state in a worker was a
race nothing reported. check_worker_ref refuses a worker whose leading states include a mut one;
threads.ludic's total moves into the words the worker is handed, under the mutex. The seeds are
regenerated (ludic-dev reseed). ludic-dev test 305 passed, selfhost-test 33 passed.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The living animals around a point from the grid (a rare species from its own list), into a list
the caller keeps in its own state, so a frame's proximity question walks the cells it covers
rather than every animal ever made. wildlife_alive_of is the species index's count.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The vehicles are rows of a ludic.base Table with kind and owner as columns, every player's own by
an owner index and a nid map: vehicle_of walks only that player's rows, vehicle_by_nid is a map
lookup, and a player who leaves has their rows removed instead of the whole list rebuilt. ve_add
takes the owner, so the owner is filed once and never assigned around the table. Positions stay
on the record: a boat moves every frame and there are at most two a player. vehicles_test holds
the table (9 tests); the game builds against it and lab/unit passes (88 tests).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
tb_on_add / tb_on_remove take a fn(int) (its states supplied, like a system's) told each handle as
its row is made and before it goes (ecs_hooks.ludic), so what follows a table - a drawing, a
message - does so without a scan. Every row's tick now also stamps its 64-row block, and
tb_changed_since / tb_added_since skip the blocks nobody wrote since: a delta over a large table
costs the blocks that moved. A row moved into a removal's gap keeps its own tick (it was not
written); a removal is told by the hook. ecs_test holds both. The ludic.wildlife table
(5ac3d48, written while builds were held) now builds and passes its 19 tests.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Written while builds are held (agents share 16 GB; no build or run until the user lifts it), so
it has not been compiled or tested. The animals are rows of a ludic.base Table: x, z, species and
alive as columns, a 32 m grid over the living, the living by species, and a nid map.
wildlife_by_nid is a map lookup instead of a scan of every animal ever made, wildlife_count walks
only its species, wildlife_nearest asks the grid, wildlife_set_alive writes the flag and the row
together, and wildlife_refile files the tick's moves once after it (wildlife_verify holds the
columns to the records). A guest puts its whole table away and brings it back. A test case covers
them; to run once builds are allowed: ludic test packages/ludic.wildlife.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Ludic frees nothing a safe program allocates, and every package's fact queue is drained once a
frame: q_drain handed back its list and made a new one each time, most often of an empty queue.
An empty drain now returns the queue's one shared empty list (never to be pushed to), and q_clear
clears in place - the live list is never one a drain handed out. A queue that held facts still
gives its list away, so what leaks is in proportion to what happened, not to the frame rate.
queue_test holds it (an empty drain is the same list twice; a drained list is untouched by later
pushes and clears).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
A mechanic that keeps many of something keeps them as rows of a Table<T> rather than a list it
scans. Removal swaps the last row in; a handle (22-bit slot, 9-bit generation) goes stale when its
entity is removed. tb_set_f / tb_set_xz / tb_set_i stamp a change tick and refile the row in every
index over that column in O(1): tb_grid (a doubly linked spatial hash, rings outward for nearest,
rehashing as it grows), tb_index (a cached query: the rows of each value of a kind column, gated by
an active column), tb_nearest_of / tb_within_of (a rare kind from its own list), tb_nearest_where
(a predicate on the record), tb_within_recs (into the caller's list), tb_changed_since /
tb_added_since, IntMap. No question allocates or writes: Ludic frees nothing, and the old lists'
per-call copies leaked every frame.
ludic.things keeps its Things as a Table<Thing> with x, z, kind and active as columns; every verb
writes the record and the row together (a Thing carries its handle and table, so thing_hide(t)
still needs no state), thing_set_on / thing_set_xz / thing_place_at are the silent forms the game
used to do by assignment, and things_verify holds the columns against the records.
things_near(_of) fill a caller's list, thing_of_kind walks a kind, things_count_of counts one, and
things_tick visits only the kinds that tick. thing_find answers the first-placed by uid.
Against a []Record scanned (M4 Pro): nearest 0.37 / 1.5 / 7.2 us at 10k / 100k / 1M (list 30 /
307 / 3075), by id 0.09 us at 100k (list 17), a move refiled in 11-44 ns. ecs_fuzz_test holds the
grid, the kind index and the record queries against a scan through 9000 random changes.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
- grass_cull.comp decides each candidate blade once a frame (place, ground, density, water,
slope, frustum, colour field) and writes survivors into three bands by distance, one indirect
draw each; grass_inst.vert only bends and places the vertices. OpenGL keeps grass.vert's
per-vertex path; R3D_GRASS_GPU=0 compares
- compute programs take sampled textures after their buffers (gpu_compute_tex / gpu_dispatch_tex),
and every dispatch now records a compute-to-draw memory barrier
- the blades as a sward: 4 m cells, bands with five, three and one-quad blades, spacing doubling
every 18 m to 70 m, never narrower than a pixel; lit facing the sun and leaning to the sky,
shadow looked up above the ground (it read the terrain as its caster), a colour ramp that
leaves only the sheath dark, clumps, dry patches and a tussock shade worked out per blade
- scatter layers flagged grass are skipped while the blades draw (and under R3D_NOGRASS);
R3D_BLADES, R3D_NOBLADES win over the game's setting; R3D_GRASS_S0/D0 for measuring
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Checked on both machines with a bundled Jolt probe: the licence and the library in place, the app
signed (Mac) and the probe's ball landing at the same height on both.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
tools/native/lib.sh passes -implib rather than /implib: Git Bash rewrites an argument starting
with / into a Windows path, and lld-link was handed 'C:\Program Files\Git\implib;...'. Both DLLs
import KERNEL32 alone, so no C++ runtime ships beside them.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The water floats the hull, the wind pushes every boat as a force, the lake bed stops it at the shore;
the swell takes the outward share of a stroke. ludic.physics: phys_hull_add (a buoyant box),
phys_row, phys_bow_speed, and a test that floats one a quarter under, rows, turns and grounds it.
The vehicles tests row a real Jolt hull on the fake shore; the horse moved to horse.ludic.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Conflicts with 0.R5's state defaults in ludic.wildlife resolved; wl_step writes the push's answer, so
its callers take WildlifeState mut again.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
- ludic.wildlife: WildlifeWorld.push / pushed_x / pushed_z (defaults change nothing); a ground
walker's step is slid clear of a still thing with a body from its species' size and scale, and
a step that gets nowhere turns away as the water does. Birds are not pushed; no dice added.
- ludic.npc: NpcWorld.push / pushed_x / pushed_z (a person 0.35 m round) in npc_walk, and
NpcWorld.clear (default true), which npc_line_ok asks at 1 m so a detour does not cut a trunk.
- ludic.aim: AimView.clear replaces AimView.ground and the ground march; the pick asks the line
a body's width short of where the ray enters the thing (aim_clear_at(t, back)), so a solid
thing does not hide itself. probe.ludic (the ground-agreement probe) and AimView.dry are gone.
- ludic.photo: PhotoLens.clear replaces PhotoLens.ground, PhotoLens.trunk and the march.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
- a CAMetalLayer on the view (cocoa.ll win_metal_layer), VK_EXT_metal_surface, QuartzCore linked
with Vk.*; the drawable measured after the layer sets the backing scale
- vk_mac.ll opens MoltenVK directly after any loader: a bundle ships only libMoltenVK.dylib
- a covered window is not presented to (win_visible); one frame in flight on macOS
(R3D_VK_INFLIGHT), with images, buffers and descriptor pools held until it is done
- win_held / win_mouse / win_pad / win_touch / win_text / win_present pass slice elements (arg_buf):
every windowed program died on its first input poll
- variants.list and SPIR-V for the three NEAR_FADE foliage programs
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The package keeps the rules and hands them over each step: speed and heading, whether a jump
leaves, CHAR_STEP and the boots' slope limit. The walker (Jolt CharacterVirtual) does gravity,
landing, steps, slopes and following the ground down. Too steep is ground too steep facing the way
the body wanted to go; stepping off an edge is not. ludic.physics: phys_walker_move (follow the body
if something else moved it, then step), the slope limit, the jump always honoured. The character's
tests run on a real Jolt walker over the same fake world: 15 pass, a knee-high boulder now a step.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Five new tests: stand and walk at the asked speed, a low stone is a step and a tall one a wall, a
jump up and back, a crate shoved aside, a hull turned by a torque and carried by a force.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Maroon Lake: 149 parameters became read-only. What stays mut is a real write - in the packages mostly a
lazy start inside a question (things_all, gear__ensure), which is what to take out next.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
collide.ludic (circles in a cell grid that could not be removed) and its Render3dState fields are
deleted, and the reload example stops asserting colliders. A game still calling col_* must move to
ludic.physics first (Maroon Lake's src/solid).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Our own shim (native/shim/jph_shim.cpp) over Jolt built from the pinned tag with cross-platform
determinism and no fp contraction: shapes as handles (box, sphere, capsule, cylinder, dome,
offset, heightfield, mesh), still/kinematic/dynamic bodies by id with real removal, rays,
top_at and push (what CharacterGround asks), overlaps, buoyancy against the PhysWater port,
contacts recorded in C and drained as PHYS_HIT / PHYS_SPLASH facts. Fixed 1/60 steps, at most four
a frame. Nine tests against the real library, including a pile run twice to the bit.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
native "<target>" "<path>" in a package's package.ludic; the compiler records the libraries of
every package a program imports and writes them into the IR (; ludic-native:), so ludicc -o,
ludic build, ludic test and ludic bundle all link one list. macOS: an rpath to the package and to
Contents/Frameworks, where ludic bundle copies and signs each library and drops the build
machine's rpath. Windows: the import library, the .dll copied beside the exe (--natives-out for
the bundle). tools/native/lib.sh builds from a pinned, checksummed source with clang on both
machines; ludic.nativeecho is the worked example; the shim rules are in packages/README.md.
Linked at build time rather than dlopen (docs/PACKAGES.md says why). Reseeded.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The old kit read Input.mouse_x() * gl_scale; ludic.ui read it raw against a layout sized in
drawable pixels. A backend says how many screen pixels one pointer unit is (pointer_scale);
render3d's is gl_pixel_scale(), 1 on Windows and headless.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
A reach through Port.member() follows that member's binding (or its default), and Registry[i].field -
or a local holding Registry[i] - follows that field in each entry, so a question asked of a port or a
table that also holds verbs no longer reaches the verbs. In Maroon Lake that took the valley's
'what is this Thing called' from 47 states it could change to 1. examples/state/write_reach.ludic.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
What only becomes known inside the drain - a value another reducer just set, the map in play - no longer
has to be faked into the action, so a verb that reads several systems while it changes one is a reducer
instead of an act handed its states. A second state to write is still refused, and the message says the
way out. examples/actions/reads.ludic; reseeded.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
ludic.ui runs a frame's presses after drawing it, and what they dispatched waited for the end of the
phase - after the host had presented - so a button's change showed a frame late. Decided: drain, not a
phase per action. ui_show and ui_press drain the queue once the presses have run, so the code after
them and the frame presented next see the change; a host that runs presses some other way calls
drain_actions() before it presents.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
A game's exported UiAct collided with ludic.ui's, which nothing outside ludic.ui uses: it is private
to ludic.ui now, and a private record of one spelling in two modules never clashed. A real clash - a
type named like one a package exports - is still refused, and the message names the package and the
way out (`ludic_ui exports it, and exported names are one namespace - rename this one, or declare it
without export inside a module of your own`).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
A step list or a registry of fn values takes no state and still reaches every state its steps take.
The compiler now writes `reach <n> <function>` - every state a function can come to by a call, a
`fn f` it writes or a global holding fn values it reads, to a fixed point - and ludic deps reports
widest_reach beside widest_function, with how many of those states the function does not take
(Maroon Lake: app_boot, 72, all 72 through fn values). --widest N lists the N functions that take the
most states with what each reaches; --reach N orders them by reach. A baseline without widest_reach
does not hold it until rewritten.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
`function kind_of(f: CharFact)` for a CharacterFact was taken on trust and failed in the code writer
as "member access on non-aggregate". A capitalised type - plain, in a slice, or a generic's argument -
must name a declared property, record, state, event, enum, action or packed value type, or a type
parameter of its declaration: `kind_of's parameter f: there is no type CharFact`.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
On Maroon Lake the prune gave home_keep_records(base_app_st, home_st, r: RunRecords, save_app_st) a
second save_app_st at the front, and every call a second argument: a state declared after a plain
parameter was not counted as declared. It is now, and a call to such a function is never given the
state again. `ludic build --check` let the duplicate through and clang refused it; the checker now
refuses a function that names two parameters alike (`add names two parameters n`).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
ludic.base's Queue<T> carries a QueueTag (its name and pending count): queue_new(name), q_push(q, v),
q_drain(q), q_clear(q) take no BaseState, and core_undrained(tags) names the given queues still holding
facts. A reducer on a package's state can now call that package's verbs (wallet_earn(wallet_st, n)).
ludic migrate state --prune (ludicc --migrate-prune) takes out each state parameter a function no
longer uses, nor anything it calls, and the argument that fills it - including an argument for a
parameter the callee has dropped, which is taken out before the call is checked, so a generic's T is
told by the argument that says it. A reducer keeps its state. --dry-run now counts the edits it would
make. Every package was moved with it: 608 base_st parameters and their arguments, 1889 edits.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
- the function that calls every reducer (and the action queue) is written in the program's own
file: in the first action's file it belonged to that module, depended on every module with a
reducer, and joined a game's modules into one 69-module cycle (examples/actions/modules and a
ludic deps case hold it); the state instances, the queue and the reducers make no deps edges
- ludicc --check / ludic build --check lower the program too and write nothing, so the code
writer's refusals are in it: a bind to a function that is gone, an unknown name (and the checker
now refuses fn <missing> itself); rejects bind_missing_fn, unknown_name, registry_count_key
- def R count is refused: its constant would be PREFIX_COUNT, the registry's size
- a file's module, package, trust and numbers-float are tables, and from the check on the lookups
of functions, enums, records, globals and externs are too (tagged enums kept as a list): Maroon
Lake's check-only build went from about 20 s to 7 s including lowering, its IR from about 2
minutes to under 10 s; duplicate declarations are found by table, not a pair of loops
- threads.ludic's pool check gives each call a little work, so a busy machine cannot run them all
on the caller before a worker wakes (it failed one run in three under load)
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
It lowered to an i32 compare of two pointers, which the IR refused. Two strings with the same text
are equal now, a null only to a null, and a failure says expect_eq failed (got "camp", want
"lake"). examples/library/testing_strings.ludic (two tests fail on purpose, and the output is
checked); ludic.base's actions_test uses it again.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
widest_function: the most states any function or entry point of the program's own takes, with
which one; --check holds it like the other numbers and --baseline writes it.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
action Name { fields } is a typed record; reducer State on Action(s: mut State, a: Action) { ... }
in the module that owns the state takes exactly that state and the action (a second state is
refused); dispatch Action { fields } queues one from anywhere, the queue supplied by the runtime.
The queue is drained at the end of every phase of the frame loop, after every phase of ludic.base's
core_tick_all, and by drain_actions(): in dispatch order, each action's reducers in the order of
their states' names, an action a reducer dispatches queued behind, a queue still growing after 64
rounds stopped with the action named. Examples actions/pack, phases, runaway; rejects for a second
state, a reducer on a non-action and an unknown dispatch; ludic.base's actions_test; LANGUAGE.md.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
- component Name (a: mut A, b: B) { ... }: every getter, default, function and event takes the
header's states before the instance; a member's call to another passes them on; the glue is
supplied them; the template never sees them; a read-only one is read-only in every member
- ludic migrate state: a component's members' needs go into its header (added to an existing one,
mut added where now changed); a field read or a member call inside a component is the compiler's,
so nothing is written inside a name and no ', )' is left; an entry point that declares states
already gets the rest after them
- a program's module named like a package gets <Name>AppState; a program's own file its own state;
a friend module's files go by directory; a package's settable var stays state
- it writes only under the programs and directories given (and runtime/ with --runtime), and
refuses the whole run naming any other file that would have to change
- a name a package already moved into its state is rewritten through it; a read of the runtime's
var through the runtime function that answers it (gl_w: gl_width())
- a state's instance supplied by the runtime is not a uses reference
- tests: state/component, rejected/state_component_ro, rendering/ui_render3d, migrate component
and foreign cases
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
ludic migrate state packages <every example program> packages/ludic.lab/example/plate.ludic
1804 vars into 126 states, 64 into lets; 23498 edits in 460 files
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
A local bound straight from another module's global (let t =
thing_cur), or from such a local, is followed within its function, and
a write through its field or element is listed as a warning after the
counted writes. A reference from a function's result is not followed.
Reseed.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The parse, the types and the module rules (export, uses, layers, ports,
registries) run and nothing is emitted or linked: about three seconds
on Maroon Lake. In this mode the checker asks vis_check at each
reference it resolves, with a global's initializer and a registry's
entries seen from the files the emitter would use. Reseed.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
bind Purse { money: g_money } writes the getter bind_Purse_money in the
bind's own file, so a one-line wrapper per member is not needed; a
member that takes something is refused a variable. Reseed.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The modules of one layer use each other freely and may go round;
anything outside the layer is held to the module's uses, and a layered
module with no uses reaches nothing outside it. The cycle check walks
the graph with each layer as one node, so a cycle leaving a layer is
refused. ludic deps shows the layers and counts the largest cycle
without their own edges. Reseed.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Every file is compiled, linked and every test block run as up to N jobs
at once (xargs -P; default the CPU count), each test with a TMPDIR of
its own, and the report is read back in file order. Windows keeps the
sequential path.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
With LUDIC_DEPS=<file> the compiler records every reference the
visibility pass resolves (from module, to module) and every assignment
to another module's global. ludic deps prints the modules,
dependencies, largest cycle, cross writes and globals written from
outside, with --graph, --dot, --writes, --uses MOD, --check FILE and
--baseline FILE. Reseed.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
A property or event a module does not export no longer collides with
another module's of the same spelling: each private one is renamed for
its module, with the types, new, emit and @On written in that module.
Exported ones stay one namespace; two events of one spelling are now
refused. Generic records, entity components and component or view
records stay global. Reseed.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
A template literal nested in a {...} hole crashed the parser: the outer
literal ended at the inner backtick. The lexer (and ludic-fmt's) now
reads a hole as code, taking strings, chars and templates in it whole.
A decimal literal past 2147483647, or a hex one of more than eight
digits, was wrapped into a negative int; it is a long with its value
now, and giving one to an int is refused. Reseed.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
A scene is an entry in the open registry LabScenes, shown on a plate: a
flat lit disc, the package's own small sky with the sun at one hour, and
a frame clock. Headless each camera (lab_shot, lab_shot_at) settles and
is written as build/lab/<scene>/<shot>.png (a stored PNG, written here);
in a window N and P step through them. tools/lab/run.sh and sheet.py
make the contact sheet; tools/lab/plate_build.py makes plate/.
ludic.render3d gains r3d_plate_mode (no terrain, grass or water is made)
and r3d_sky_path. quit() in a program with no frame loop links. The
example takes ~120 MB resident, 417 MiB peak footprint. Reseed.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The entries are read and checked against the registry's record as a
registry ... from file is, errors at the resource file's line, and they
are defs of the module that wrote the line: the registry must be open to
it, and they take that module's place in the stable order. Reseed.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
A private function, var or const no longer collides with the same
spelling in another module, in no module or in the runtime: where two
meet, each private one is renamed for its module (seed$shop), with every
reference its module writes that no local shadows. Exported names stay
one namespace. Nothing is renamed without a clash. Reseed.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
A line is cast at a point the FishingWorld port says is water (depth, or afloat), bait (asked,
then said as FISHING_BAIT_USED for the game to take) halves the wait, the bite picks a species -
a legend only on a lure, likelier in deep water and with skill, else an everyday one by weight -
and the fight is a marker, a green band that moves above Relaxed and closes on Hard or for a
fish whose data says so, a Gentle ring of one press, and a legend's runs and the line it takes
back. Length and weight come from FishSpecies data. The package never draws, speaks or touches
a pack; every step is a FishingFact (CAUGHT, BAIT_USED, LINE_SNAPPED, SLACK, MISSED, ...). Its
dice are its own Rng. A legend on Gentle fights as a legend (the old ring lost him at once).
Uses ludic.base only; thirteen tests over a fake lake.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Items (base price, sell fraction) and vendors (stock, buys, refuses) are the game's data handed
over at boot; standing, the day, a shortage and "buys beyond its list" are the ShopWorld port;
money is ShopPurse and the pack ShopPack. The week's wanted and glutted items come from an Rng
seeded by the week, never the world's. The balance invariants - a shop is not a fountain,
friction falls and never inverts, raw keeps its order - are the package's tests (nine).
ludic.crafting: its private names carry the package's prefix too (a private name is still one
global namespace with the game's - ludic.shop's `sp_seed` met the game's).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
A recipe is `def CraftRecipes key { out, n, ins: [...], ns: [...], station, minutes, hold,
group, learned }`. The pack is the CraftPack port (count, take, add, room, and `leaves`: what
an item leaves once used, so water gives back its bottle); stations are CraftStations (ready,
start), unbound meaning hands only. craft_make hands a click recipe over or starts a timed one
at its station, refunding when the station refuses; the hold runs from the screen that shows it.
Uses ludic.base only; ten tests with a fake pack and fake stations.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Minutes of fuel that burn down in game time, twice as fast in the rain with
nothing over the ring, and the hours it held under cover. Lighting and feeding
are decided (fire_decide), costed in wood (fire_cost) and applied
(fire_apply) apart, so a machine that does not own the world can ask the one
that does. Warmth and cooking by distance are queries. FIRE_LIT,
FIRE_WENT_OUT and FIRE_LOW_FUEL are facts; the rain, the tarp, the ban and
who owns the world come through the FireWorld port. Its own save section.
Uses ludic_base only; 8 tests.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Warmth, food, water and energy as bars of 0..100; what an hour of doing something
costs them (the package's rates: a full bar is a day at a trip's median); the
countdown when one runs out, reported as NEEDS_CRITICAL / NEEDS_RESTORED /
NEEDS_COLLAPSED facts. The body's work, the air, clothing, the sun, the
difficulty and the effects come through the NeedsWorld port. Its own save
section. Uses ludic_base only; 13 tests.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
A step is (kind, param, need) and a question about state: a kind has a
shape (yes, a count, a mask counted, every bit of a mask) and a pool
(world, any, sum, or, max, each), handed in by the game as StepsKind
records. An arc is chapters of up to STEPS_PER_CHAPTER steps; steps_check
sticks each met step of the current chapter (STEPS_MET) and opens the
next when all are (STEPS_CHAPTER), so a step met before its chapter
ticks at once. With company each seat's shares are held per chapter,
pooled as the kind says (steps_met, steps_party_got), counted for an
EACH step (steps_each, steps_lacking), and a leaver takes theirs along.
Port: StepsWorld { value(kind, param, player) (required), party,
present }. The package knows no kind's meaning and no chapter's words.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
latest.json asked once and polled (the newest version and its notes, in
a language or English); then Velopack's releases.<os>.json, the plan (the
delta chain when the installed full package is on disk, no delta is
missing and the chain weighs less than the full one; the newest full
package otherwise, named as the feed names it), each package streamed to
the packages folder with its bytes on a bar and an eased pace, its
SHA-256 by Get-FileHash or shasum in a child, Update patch per delta and
Update apply --waitPid. macOS keeps its packages outside the bundle and
names --rootDir and --packageDir; a translocated app is not installed.
Ports with the runtime as every default: UpdateWorld (platform, exe, pid,
now_ms), UpdateNet (get, download, poll, text, received, free),
UpdateProc (spawn, poll, free). Config: feed, app_id, version, root,
packages, mac_packages, scratch. It words nothing: states, UPDATE_ERR_*
codes and numbers, and UPDATE_STARTED / _STOPPED / _APPLYING facts - the
game exits on the last.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
ludic.clock, .effects, .inventory, .wallet, .weather and .tracks say
'uses ludic_base' (ludic.base uses nothing). ClockWorld, PackRules,
WeatherWorld and TracksWorld are export ports whose defaults are the old
fallbacks; clock_bind, inv_bind, weather_bind and tracks_bind are gone,
and the tests bind their fakes. The base README's toy does the same.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
A module that says uses must name every package module it reaches; a
package with no module line is named for its directory (ludic_render3d)
for this rule, so a mechanic reaching into the renderer is caught. Only
the engine's runtime needs no naming. 'module x uses' with nothing after
it reaches no other module. A port whose every member has a default
answers with its defaults when unbound, and 'new' of a port says to bind
it. ludic-dev test runs 'ludic test packages'. Reseed.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Events are encoded once into lines held in memory; a batch is the first
lines joined, with the player id put in where a mark stood, POSTed on
Http's own thread every flush_ms or at flush_at events; a failure keeps
its lines and doubles the wait up to backoff_max; the queue is on disk
every save_ms and read back at the next start; off (the enabled port)
drops the request in flight, empties the queue and deletes the file; and
a run that may not send (can_send) keeps nothing. The player id is the
machine's own id (MachineGuid, IOPlatformUUID, /etc/machine-id) hashed
with the game's salt, else random; it lives in the game's id file beside
whatever else the game keeps there.
Ports: TelemetryWorld (can_send, enabled, now_ms, stamp) and
TelemetryTransport (send, poll, drop; unbound: Http). Config is a record
(host, key, path, lib, queue_file, id_file, id_salt, and the timings).
Facts: TELEMETRY_SENT, _FAILED, _ID. Not a System: it runs from a
launcher's first frame, before any world exists.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The runner's list starts from the declared systems, in the order the
compiler gives an open registry, and core_add appends after them.
registry_test covers it; the README says ludic test runs the package.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Random.range is rng_range, so a package's own rng_range(a, b, c) took
every Random.range call silently. Where the program calls such a method
and the target resolves to a function of its own, the function is
refused, naming the namespace method and the call. Reseed.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
On a float or double they emitted an i32 compare and clang refused the
IR; they compare as the wider float kind now and a failure prints the
numbers with %g. Reseed.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The seeded random numbers lived in the ECS runtime, so a tool or a
program of test blocks got "unknown function rng_range". They are
runtime/native/rng.ludic now, spliced on Random.* or a bare rng_*/seed
call nobody defines, and imported by core.ludic for a game. A bare
value_*/json_* call nothing in the program defines splices the value
tree the way Value.* does. Reseed.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
A ring of prints per kind; a port for the trip's time, the reader's tier, a
bearing in words and a hook the game draws each print through. A read is a
TrackRead fact on tracks_reads() every time and a count the first time; other
sign is told with tracks_note. The count and last bearing are its save section.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The kinds and their odds are data the game hands in; a port asks whether this
machine owns the world, the hour, the night, the odds from a kind, what the
story wants of the sky (a front, a hold, a clear night), the front's kind and
length, and a seed. Its own dice; a new spell, a clearing, lightning and a
wanted front are facts on weather_facts(); its own save section.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Test blocks were never type-checked, and the checker is what makes a
generic call real; they are checked like entry now. Reseed.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
A directory stands for its *_test.ludic files and the files straight
inside any tests/ under it. The runner answers --list and runs one test
by name, and ludic test runs every block in a child process, so tests
no longer share globals. A failed expect names the file it is written
in (the path the compiler was given) and its line. Reseed.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
ludic.inventory: counts per item kind, a PackRules port (stack_limit), add
what fits / take all or none / set outright, an ItemChanged queue, and a save
section keyed by item NAME so a game can reorder its items with no migration.
ludic.wallet: earn, spend only what is there, lose down to zero, set, a
MoneyChanged queue and its own save section. Six and five test blocks.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
A def from another module into a registry that is not open is refused,
and defs go through visibility (the registry exported, its module in the
definer's uses). Index order: the declaring module's entries, then the
other modules' by module name, each in reading order. A registry entry
is emitted as its own file's code, so what it names is seen from its
own module. Reseed.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
A ClockWorld port (owns_world, rest_scale, seed), verbs to set and advance it,
clock_new_day, a DayTurned queue (stayed up past midnight, or slept), a
calendar whose month, year and season are pure functions of the day, and a
System (PH_INPUT) with its own save section. Nine test blocks in
tests/clock_test.ludic with fakes for the port.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
A port is a record of function values a module calls through
(Clock.now()) without naming the module that answers. A port used and
never bound, a bind missing a required member or naming one the port
lacks, and a second bind are refused; the binder must see the port, and
what it binds is checked from its own file. Reseed.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
ludic.effects: effects_add / effects_add_after / effects_clear / effects_run,
effects_sum / effects_has / effects_at / effects_live, an EffectEnded queue
(ran out, pushed out by the bonus cap, crowded out of a full ring) and a
System with its own save section. Nine test blocks in tests/effects_test.ludic.
ludic.base: q_new collided with render3d's quaternion q_new the moment a game
imported both, so the queue constructor is queue_new.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
A scoped friend sees those modules' private names and only the exports
of every other; a plain friend module still sees everything. Reseed.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
A reference from a module that declares uses into a module it does not
name is refused, exported or not, naming the use and the fix. A module
with no uses clause keeps the old rule; a package's module is always
usable; a friend is not held to it; a cycle in the declared graph is
refused; LUDIC_VIS_REPORT=1 lists the violations as uses: lines. Reseed.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
A mechanic package depends on ludic.base and nothing else: ports (records of
function values for now) for questions, queues for facts, verbs for changes,
phases for order and its own versioned save section. The runner inits, resets,
saves and loads systems in the order added and ticks them phase by phase; a
missing save section is a reset. Tests for each piece and a worked route
between two toy mechanics live under tests/. The old ludic.core (engine ECS
components) is used by examples/library and stays.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
A program's own `run` was never called: every call to it lowered to the
built-in System.run (C's system()), and clang failed on the IR.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
- the scrollbar is .ui-scrollbar with a .ui-thumb, styled by the default sheet or a theme; a press
on the thumb holds it at the point it was taken and the pointer moves it in proportion, a press on
the track jumps the thumb's middle there and holds it, and neither presses what is under the bar
(it used to centre the thumb on the pointer and press the row beneath as well);
- while a popover is up only scroll boxes inside it take the pointer, and a press outside it that
closes it forgets any press in progress;
- a transition ends exactly on its value (it stopped a rounding error short, at every frame rate).
Tests: ui_popover_mouse, ui_scrollbar, ui_ease (60/120/180/240 Hz).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
- <input type="number" min max step>: typed digits replace its value (a minus when min allows one,
a point when step has a fraction), Enter or leaving the field commits them held between min and
max, the left and right arrows step it.
- <input type="key">: Enter or a click starts it listening and the next key is its value, Tab, the
arrows and Enter included; Esc stops it listening, Backspace clears it, `shown` names the value,
and ui_capturing() tells the host to leave its keys alone (through the frame that ended it).
- note="..." under any control's label (.ui-labels > .ui-label + .ui-note); a range's (or number's)
value with decimals, format="percent" and a unit.
- A popover with `anchor="id"` (the nearest element of that id) or a bare `anchor` (the element
before it) sits beside it by `placement` (right, left, bottom, top), its margin the gap, opens to
the other side where it would leave the screen, and is kept on it.
- A tooltip's title splits into lines at a newline or a written \n.
- text-shadow, inherited, drawn sharp under the text.
- ui_opacity(): the group opacity a native's draw is at.
- border-image is drawn as painted with no background colour, tinted by one, and not at all when
that colour is transparent (it drew a white nine-slice).
- A picture file is drawn untinted, as a browser draws one; an atlas cell (prefix:name) still takes
the color around it, and an <img> with its own color is tinted by it.
- The render3d backend loads a picture again when its file changes or appears (a failed load was
kept for ever), and at once after ui_image_reload(path); takes a path with a drive letter (C:/...)
as a path rather than an atlas; and slices a nine-slice by its texture's own width and height.
- ui_inputs.ludic and ui_look.ludic; LANGUAGE.md and the changeset say all of it.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
- flex-shrink: a line whose children want more room than it has takes it back from those that
shrink, in proportion to shrink times size and never below a min size, a fixed size or the
content. A scroll box shrinks (and scrolls), and a box in a column shrinks as far as the scroll
boxes inside it let it, so a list in a column takes the room its siblings leave with no height.
`flex: grow shrink`, and `flex-shrink` by name.
- A row wider than its room measures its texts (and boxes without a width) again in the room the
rest leave them, so a line wraps beside an icon.
- calc() with + - * / and brackets over px, %, em, rem, vw, vh, plain numbers and var(); a width
or height keeps its percentage, taken of the room it is given. Lengths lists (padding, margin)
keep a calc()'s spaces.
- `order` places a box's children without touching the tree.
- width: 0 and height: 0 are 0: an unset size is UI_AUTO now, not 0.
- A component root that is itself a component takes each user's class, style and id in turn (the
outermost's id wins), and the rules of all their sheets are weighed by specificity together; a
user's rule wins a tie. The parent's sheet used to override the child's whatever its specificity,
and a middle component's sheet was lost.
- The default sheet lays .ui-track out as a row, so a range's fill is as tall as its track and a
checked box's knob goes right; a range's thumb is centred on the fill's end.
- ui_boxes.ludic.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
- An action whose first word is `set` or `emit` followed by `(` is a call, so a component's
`on set(v: int)` is pressed as `set(4)`; `set x = ...` is still the action.
- A `string` prop (or state, or parameter) given a number in a template reads it as text through
ludic.ui's new `ui_val_text`; `label="{3}"` used to arrive as "".
- Two components of one name (or of names that differ only in case, which share their functions'
names) are an error at the second declaration that names the first's file and line, instead of a
"function cmp_x_def_y is defined twice".
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
- `popover`: a top layer that keeps the pointer and the keyboard, closed by a press outside or Esc.
- `title` tooltips after half a second's rest, styled by .ui-tooltip.
- `<progress>` and `<meter>`.
- Atlases take rows and number cells, and importing ludic.ui/render3d.ludic is enough to draw with
render3d.
- The compiler reports two declarations of one name before it type-checks, so a package global
clashing with a program's reads as that, not as 29 type errors.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
A game no longer writes a host:
- The runtime reads Input itself: focus and keyboard navigation (Tab, arrows, Enter/Space,
autofocus), the pointer (hover, :active, click on release, drag), and scroll boxes with the
wheel, a draggable scrollbar, clipping and scroll-into-view.
- HTML's controls are built in (button, checkbox, radio, range, select, text, key capture), made
of plain parts a stylesheet styles, each reporting with on-change and event.value.
- ludic.ui/render3d.ludic draws with render3d's overlay: textures, named atlases (icon:NAME),
nine-slice border-image, rounded rects and rings, clipping, and a scale.
- Hooks for the program's language, sounds and clock (ui_translator, ui_sounds, ui_clock).
- ui_dev: hot reload, errors on screen, LUDIC_UI_DUMP.
- CSS:
- colours as #rgb / #rrggbbaa / rgb() / rgba() / names;
- border-radius and outline (following the radius), box-shadow, background-image and a tinted
border-image;
- group opacity, @keyframes / animation, transition;
- :focus, :focus-visible and :focus-within.
- HTML mixed content, and boolean attributes.
- render3d gains tex_width / tex_height, and the XML reader keeps text runs in order.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The compiler turns a component declaration into:
- a record of its props and state;
- a constructor, a props setter, a model and a call;
- a class, registered with ludic.ui at start.
Its template and styles are read from beside it and compiled in, with @import inlined, and a
missing template fails the build.
In the runtime:
- each mounted instance keeps its own props and state, and `set` in a template writes the state;
- styles are scoped to the component;
- class, style and id on a component's tag land on its root, styled by the parent's sheet too;
- ui_reload re-reads a component's files from disk and keeps instances.
The package's own module is now ludic_ui, so a program may call a directory of its own ui.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
- Natives: ui_native(tag, measure, draw) makes an element the program draws itself; ui_fire
runs its on-<event> with event.value. input, select and textarea have simple defaults.
- React:
- keyed <each>, <let>, <provide> context through components, <fragment>;
- named slots, default props on <component>;
- on-mount / on-unmount;
- inline text inside text elements.
- CSS:
- custom properties and var();
- position relative/absolute/fixed with insets and z-index;
- em/rem/vw/vh and @media;
- wrapping text and ellipsis, overflow;
- + and ~ combinators, :nth-child(an+b), :checked, :active.
- Developing: errors with file:line, ui_errors(), ui_reload() keeping state, and an
inspector-style ui_dump.
- view fields infer the type of a literal or a named function's result.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
A `view Name { field = x; function q(..); on e(..) }` declaration is the one bridge between a
program and its UI: it writes view_<name>() -> UiView, whose model is a Value of every field and
whose call runs a query or an event by name.
ludic.ui is a template runtime:
- HTML-shaped XML screens and components, loaded at run time;
- {expression} bindings, if/else/each, props, slots, per-instance state;
- on-press / onclick actions (event, set, emit);
- component libraries (export="true", <import src as>).
Styling:
- stylesheets in <style> or importable .lss files (@import);
- CSS selectors (#id, .class, [attr=v], descendant and > combinators, :hover, :disabled,
:first-child, :last-child, :nth-child, :not) weighed by specificity;
- the box model and flex under CSS's property names.
Also:
- default parameters, and positional-then-named calls;
- Value gains a float kind;
- a function shadowing a runtime one is refused;
- an index is evaluated before the slice is read;
- runtime errors name the right file.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
lint lines in package.ludic - one_statement, max_file_lines, max_function_lines,
max_comment_lines, max_header_lines, paths, baseline - checked by ludic-fmt --check
<files> and ludic-fmt --lint (the project), at the line; a baseline ratchet lets a
rule arrive in a codebase that breaks it (--init-baseline), lowered as it is fixed.
check-impl reads the alias declarations too (it had been blind to every namespace
L6 moved out of the compiler), and eight methods get their pages; a test holds the
formatter to keeping type arguments together (L5).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
registry NAME of RECORD [as PREFIX] from "file.lres" fills a registry from a data
file of key { field: value } entries, read at compile time with the record as its
schema: every entry is checked like a def, errors name the resource file's line,
nested records and lists of them are bare { } / [{ }] typed by their fields, and
values are expressions in the registry's module. The entries are compiled in, so
nothing parses at start-up and a build that succeeds has validated its resources.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
registry NAME of RECORD [as PREFIX] is a global table; def NAME key { ... } in
any file is one entry, collected in source order and filled before any code
runs. Each entry gets an index constant (PREFIX_KEY), the table PREFIX_COUNT,
and a record with a key field gets it filled and a NAME_find(key). A record
literal naming a field its record lacks is now an error everywhere; a global's
initializer is lowered as its own file's code (its errors, and what its module
may see, were whichever statement came last).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The typed buffers are slices: words/floats/fixeds/doubles/pointers(n) make
zeroed, bounds-checked []int/[]float/... and the type names mean them. buffer(n)
is a []byte, with text_of, Fs.read_bytes/write_bytes and view(xs, start, n).
bytes(), indexing a raw pointer or bytes, free, resize, Memory.*, raw file calls,
data_of and C externs are refused outside unsafe { } / unsafe function, and a
project's own files may write unsafe only with --unsafe; the runtime and packages
are the platform. A slice passed to an extern goes as its data.
What the change found: Sync's atomics on a slice header, words(n) uninitialised,
input's fixed axes in ints, truetype's fixed outlines as ints, skin matrices
typed int, gl_shader's source table made from raw bytes. render3d gets safe
entry points (safe_api.ludic). Rendering is byte-identical; a frame costs the same.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
`alias meth(labels) = target` in a namespace block makes Ns.meth a call to
target with those labels (the target's own parameter names without a list). The
engine's 41 table-driven namespaces - 438 methods: Http, Udp, Process, Json,
Value, Screen, Input, Audio, World, Tiled, ... - leave emit_ns_call for
runtime/native/namespaces.ludic, spliced into every program; 532 lines of
compiler go and the seed shrinks by 23k lines of IR. The game's IR is byte
identical. The checker checks an alias call's arguments against its target.
Still built in: the inline namespaces (Math, Text, List, Vector, Color, Time,
Date, ...) and the methods that pick a target by argument type.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
property Pool<T> { ... }, function first<T>(xs: []T) -> T, map<T, U> over fn
types; a type writes an instance as Pool<Thing>, nested as deep as needed. The
parser names an instance Pool$Thing and remembers its generic and arguments; the
checker takes the generic declarations out, infers a call's type arguments from
its arguments or its result's declared slot, and makes each instance once as an
ordinary record or function, checked like any other. Errors print Pool<Thing>.
An instance keeps its generic's module and export (L3). ludic-fmt keeps type
arguments together while spacing comparisons and shifts.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
selfhost/check/ walks every function, the entry, tests, globals' initializers and
@On listeners with real scopes, and refuses mixed number kinds, text and numbers,
two record types, mismatched slices and fn types, wrong argument counts, wrong
returns and wrong push elements - every mix-up at once, each at its line.
LUDIC_CHECK_REPORT=1 lists them by category. pointer stays untyped (L7's).
What it found is fixed: render3d's HDR scan calling the float-bits extern f_lt
with floats; ludic.shooter's right-stick aim overflowing past half a push;
prof.ludic storing longs in []int; extern arguments now coerced to their
parameters. Text-returning runtime functions say string; Assets.ready says bool.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
A barrel's 'module NAME' makes its directory a module; a declaration other modules
use says 'export'. Private use from another module is an error naming the module
and where to mark it; 'friend module' sees everything (a test harness); a file in
no module is public and a package keeps its own module. LUDIC_VIS_REPORT=1 lists
every violation instead of stopping, so a codebase can be given its exports first.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
fn(int, float) -> bool is a type, fn name is any top-level function's value, and a call through
a local, a global, a record field, a slice element, a parameter or a result of a function type
is an indirect call; two function types mix only when equal, a call checks its argument count,
and a value may be null (examples/functions/values.ludic). Job.parallel_for keeps its worker
check.
render3d's scene is registered rather than required by name: r3d_on_draw, r3d_on_casters and
r3d_on_stream_fill (hooks.ludic). The two rendering examples register theirs - and had defined
scene_draw_casters with no parameter while the renderer passed one, which nothing checked.
render3d declares numbers float itself; smooth.ludic is converted to floats and returns when
r3d_init fails instead of running on into a segfault. Noise.* check their argument count (a call
one short crashed the compiler). selfhost-build says why it failed. The migration tool reads a
declared float as evidence. Seed regenerated.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
L1. Two vars, consts, enums, properties or events of one name in a program are an error naming
both places (functions already were); the first used to win silently. A let / var of a name its
own block already declared is an error (it used to shadow). A function with a result type whose
body can reach its end without a return is an error, asked structurally of the body - a return,
an if/else or a match with a default arm whose every branch ends - rather than handing back
whatever the result slot held. The game, the lab and every package example pass all three;
the lab had one harmless shadow, and the game's split screens had two real bugs of the kind.
examples/rejected/ holds the four refusals, checked by the test runner's new reject_case.
Seed regenerated.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Text.to_float reads a leading decimal number (strtod), the twin of Text.to_int, for data files.
A return with a value in a function that returns nothing now says so where it is, instead of
reaching clang as 'store void'. Seed regenerated.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Listeners are compiled into one @ev_<E> function and return branched to its exit, so every
listener declared after one that returned early - and every foreign listener - never heard
the event. The per-kind listener shape (return unless it is my kind) answered only the
first-declared kind. examples/events/answer.ludic holds it; seed regenerated.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
A numbers float file adapts decimal literals to a fixed operand or slot, and refuses to
promote a computed int to a float implicitly: there it is almost always float bits. Explicit
float(x) is always allowed.
render3d's numbers are float, converted by tools/migrate/floatbits.py - a whole-program
inference of which ints carried IEEE bits (union-find over flows, calls, returns, buffers,
nested buffers and lexical scopes) and a rewriter to operators, Math.* and float literals,
with float_bits / float_from_bits left only where bits really cross (runtime scratch
buffers, mixed buffers). Seed regenerated.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Input.text, App.monitor_count / window_to_monitor / window_fixed,
gl_sleep_us, and the grass and collide changes, uncommitted on main and
depended on by the game; carried here so the language work starts from
what the game actually uses. main's working tree is untouched.
`reseed` writes both, and the committed Windows seed carried none of the bounds
check (0 sites against 859 in the regenerated one), so a bootstrap there would
have built a compiler without it - the same half-a-change as the host seed, on
the other platform.
Not verified by running: a Windows bootstrap cannot be done from this Mac. What
is checked is that it is generated by the same `reseed` from the same source as
the host seed, and that it carries the emission the host seed does. The host's
own bootstrap is a proven fixpoint and selfhost-test's C-free bootstrap passes.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
0998cb5 committed the backend source and not the IR seed it is built from, so
`ludic-dev build` on a clean checkout rebuilt the compiler from a seed that
predates the change and produced one with no bounds checking at all. The check
only existed in whichever binary happened to be sitting in bin/.
That is the same half-a-change this project has just been bitten by twice - a
caller committed without the definition it needs, a table's declaration moved
without its parameters. A compiler change is the source AND the seed.
seedcheck.ludic:5: index out of range: 5, len 1
from a compiler bootstrapped out of the checked-in seed, which is the thing
that was not true before.
bootstrap: FIXPOINT (gen2.ll == gen3.ll). selfhost-test: 33 passed, 0 failed,
including the C-free bootstrap from the seed.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
A slice has carried { ptr, len, cap } since it existed and nothing ever read
the len: every index emitted a bare getelementptr. Running off the end of one
wrote into whatever the allocator had put next, and the program died somewhere
else entirely - a maroon-lake crash took a day to find because the stack named
a texture upload and the write was in a telemetry buffer five frames earlier.
Now each index loads the length and compares unsigned, which rejects a negative
index in the same instruction, and a failure aborts with the location the other
located errors use:
oob.ludic:9: index out of range: 7, len 3
`words` and the other raw buffers are unchanged - they are a bare malloc with no
length to check, which is the argument for moving off them.
The SPIR-V variants are regenerated in the same commit: shaders --check was
failing against the edited GLSL.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
scrollingDeltaY is a double and the Cocoa event pump truncated it to an int per EVENT
before accumulating. A trackpad or a Magic Mouse sends a stream of fractions of a line,
every one of which truncated to zero, so the wheel was dead; a notched mouse sends three
to ten lines at once, so it jumped. The fraction is accumulated now, a precise delta is
scaled from points to notches, and the remainder carries to the next frame.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
layer_flutter's mask was the vertex's height alone, so on a ten metre aspen every part of
the trunk above 1.2 m trembled with the leaves and the bole read as cloth. What separates
a leaf from a bole is distance from the model's centre line, not height; the mask is
radial now, and small plants keep the old one because a flower is all leaf.
NEAR_FADE dithers tree foliage out inside arm's length of the camera, in the depth
prepass, so the lit pass never sees those pixels and the equal-depth optimisation is
untouched. gl_Position.w is the view depth, so it costs one varying and no uniform.
Blades, cards and flowers are excluded - they live at the player's feet - and the shadow
pass keeps every leaf.
daylight_hand takes a reach in metres. The falloff was an inverse square windowed between
26 and 6 with both numbers hard-coded: two per cent of its own near field at ten metres,
so a torch lit your boots. It is a gentle power out to the reach now.
Measured in Maroon Lake, twice per side: the crown mask moves 1.7% of an aspen frame and
nothing at all in the meadow under it; the near fade moves 7.1% of a first-person frame
at a conifer and 0% where there is no foliage in the lens; the torch's reach lifts the lit
ground 20 to 45% and leaves the sky bit-identical.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Actor.cast_hidden separates drawn from casting. ac_visible rejected a hidden actor from
the shadow pass too, so hiding the player's own body for first person took the player's
shadow with it - and in a sunlit basin your own shadow is what tells you where you are.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
layer_flutter(l, v) gives a scatter layer a per-leaf tremble. The vertex stage offsets
each leaf by a phase taken from its own place on the card, so neighbouring leaves are
never in step, and writes the result out as a varying the fragment stage uses to flash
the leaf's pale underside as it turns - which is the part that reads, since a still
frame of a tremble is a still frame of nothing. One uniform, one varying, no extra pass.
And the sway itself was measured in METRES: hgt * hgt * 0.35 is right for a 40 cm
flower and puts ten metres of sideways into a 14 m trunk, so every tall tree in the
valley stood bent over like a fishing rod. It is a fraction of the model's own height
now - the tip moves a few per cent of the tree whatever the tree is, and the base does
not move at all.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
water.frag takes u_wade - a point and a strength - and puts spreading rings and a
patch of churn into the surface normals there, so a wader marks the water and a
swimmer works the whole of it. It is one uniform in a fragment stage that was
already running, applied after the distance flattening so a disturbance close to
the camera survives it, and it measures no frame cost at all.
Also records the two renderer features that shipped without a changeset.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
The existing occlusion is tuned as AMBIENT occlusion: a wide radius answering "how open is
the sky here". That is the right question, and it leaves every object in the frame
hovering - because what says a log is ON the ground rather than in front of it is a hard,
narrow darkening in the last few centimetres where the two meet, and at a metre and a half
of radius that darkening is spread so thin it is not there.
A second, tight pass: eight taps inside 40 cm, which at any normal distance is a handful of
pixels and stays in cache, with a much tighter range check than the ambient one so a wall
across the room does not darken the floor in front of it.
Small by frame area, because contact occlusion is - 1.8% of the frame, 7% of the region
around the things it grounds. GL 7.0 -> 7.1 s, VK 7.3 s over 400 frames.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Rain fell and the ground did not change: the weather was a curtain of particles in front
of a dry valley.
u_wet, 0 dry to 1 soaked, does the three things a wet surface does. Darkens the albedo.
Drops the roughness hard - which is what makes a soaked meadow read as soaked rather than
merely dark, because the sky glances off it. And pools: water finds the low places and
flat ground holds what a slope sheds, so the puddle mask is the macro noise the materials
already use gated on slope, and a puddle takes the world's up for its normal rather than
the ground's relief.
It joins the shore's existing wet band rather than fighting it - the same term from a
different cause.
34.1% of the frame changes between dry and soaked. GL 7.1 s either way over 400 frames.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
You walked through a meadow and every blade ignored you, which is the most noticeable
thing missing from every step the game asks you to take.
u_push (x, z, radius) bends blades away from a body and DOWN - a trodden stem is shorter
as well as leaning, and leaving the height alone made them splay outward like a fan
instead of being walked through. Applied after the blade's own yaw has put it into world
axes and before it is tipped onto the ground normal, so the push is a world direction
rather than something in the blade's private frame.
Radius 0 means nobody is there, so nothing is paid for when it does not apply. Measured:
GL 6.9 s, VK 7.3 s over 400 frames, unchanged.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
The grass had a gust built from two sines; the trees had no gust term AT ALL, only a
per-instance wobble. So the meadow rippled and the canopy above it swayed to an unrelated
rhythm, and nothing ever crossed the valley.
wind.glsl is prepended to every stage (programs.ludic, and shaders.ludic for the SPIR-V
build) so grass, crowns and water read the same field at the same world position. It
declares no uniforms on purpose: u_time and u_wind already exist in several of those
files and redeclaring them is a compile error in whichever stage includes both.
The wavelength is what made it work. At 0.030 the front was 209 m crest to crest -
longer than the meadow you can see - so the whole frame sat in one phase and the gust
read as everything breathing together. At 0.085 it is 74 m and a front crosses a view in
a couple of seconds.
R3D_DEBUG_WIND=1 paints the field on the ground. It is the only honest way to show a gust
in a still image, and two shots 0.3 s apart move by 39/255.
400 frames: GL 6.9 s, VK 7.2 s. Backends agree to 0.68/255.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Two changes that have been sitting uncommitted in this checkout, gathered as a recovery
point before the next run of work.
grass.vert: the blade existence gate tested green DOMINANCE - g - max(r, b) - which is a
test for lush green and nothing else. A dry alpine meadow is yellow-green, its red as high
as its green, so the meadow scored zero and was thinned to the floor - a QUARTER of the
blades - on exactly the ground that should be thickest. Measured on Maroon's own ortho,
g - max(r, b) reads +0.026 at the camp and -0.002 six hundred metres away, flipping between
full density and a quarter over continuous meadow; g - b reads +0.076 and +0.014 and
separates plant from rock and snow just as well, because rock and snow are neutral and
vegetation is not. Bare ground in the near band went 82% -> 57% looking down.
Blade height is biased short (h3 * h3) rather than spread evenly, so a meadow is a dense
mat with taller stems out of it; an even spread read as a lawn that had been cut.
water.frag: the planar reflection is returned at 0.72 of the scene's exposure rather than
all of it, the fresnel mix caps at 0.70 rather than 0.86, absorption falls so the bed is
visible through the surface at the angles a player stands at, and the ripple field is
roughly halved - so a sheltered lake is one sheet of water with a mountain in it instead
of open chop to the horizon.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
The shipping default had NO anti-aliasing at all: the temporal resolve was removed, the
setting's first option went on saying "Temporal", and MSAA defaults to one sample, so
every machine without DLSS drew grass and needle cards with nothing smoothing an edge.
FXAA in the sharpen pass, which already reads the neighbourhood and runs last on the LDR
image. No history, so it cannot drag or smear a reflection. 25.5% less single-pixel
staircase on edge pixels.
The trap, recorded because it inverted the result: the unsharp delta must be computed
from the RAW image and only then applied to the anti-aliased colour. Centre from FXAA and
neighbours from the raw texture measures half smoothing and half signal, so the mask
sharpens precisely what FXAA softened - 29% WORSE than no anti-aliasing at all.
Depth of field for the photo mode: a disc whose radius is the circle of confusion,
normalised by focus distance so a landscape shot is not a macro, with taps rejected
unless they are at least as out of focus as the pixel they blur into - which is what
stops a sharp foreground haloing into a blurred background. Between the scene and the
bloom, so a blurred highlight still blooms. Skipped whole when the aperture is shut.
400 frames in ordinary play: GL 7.1 s, VK 7.2 s - the lens does not draw there.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Foliage roughness was pinned to 1.0 and grazing Fresnel switched off, so nothing green
in the game had a highlight anywhere. The glint off waxy leaves and wet needles is most
of what makes a stand look alive rather than painted.
The reason it was off is real: a crown is card quads, and at a grazing angle the card's
normal is a lie, so a specular lobe frosted whole crowns white against the sky. The sheen
returns as its own term gated on exactly that - it fades as the card turns edge-on, which
is where its normal stops meaning anything. A tight lobe for the glint, a weak wide one
for the waxy rim, nothing at the angles that frosted.
Translucency reaches 260 m rather than 140, and a dense crown passes 0.45 of it rather
than 0.3, so a backlit stand glows for as far as you can see it.
400 frames: GL 7.1 s, VK 7.4 s. Backends agree to 0.14/255.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Correcting the previous commit, which said this was unfixed. It is fixed, and I had the
wrong knob: raising the mudstone tint's magnitude from 0.14 to 0.38 made the face LIGHTER
rather than REDDER and it came out pale tan. The shader's own DEBUG_ALB view measured
red:blue at 1.56 on the peak where maroon mudstone wants nearer 4, which says plainly that
what needed moving was green and blue DOWN, not red up.
The built-in DEBUG_MAT view is what settled it, and I should have reached for it an hour
earlier instead of inventing my own: it shows the Bells as pure red, rockW = 1, so the
face is rock and always was. That also explains why painting the snow albedo bright red
changed nothing and I wrongly read that as "this surface is not drawn by this shader" -
there is no snow on those faces to paint.
The far tier's fallback moves with it, so the cheap tier keeps the near tier's mean.
Backends agree to 0.60/255.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
55% of every ground pixel was the orthophotograph, at every distance and on every
material. A survey image records WHERE the forest and the scree are well and what a
steep rock face is coloured badly - it is shot through kilometres of air at an angle no
player stands at. The photograph keeps the meadow and the forest and gives the rock back
to the rock, by material and by elevation and never by distance: a distance fade used to
live here and was removed because the photograph has the day's own shadows baked in, so
grass grew dark patches as you backed away and they slid as you walked.
The far tier's rock fallback is the maroon mean, not a neutral grey - TFAST_3 skips the
rock sample and leaves that constant standing in for a whole mountain.
Ortho-classified snow needs altitude now. Bright and unsaturated is what snow looks like
from a satellite and also what a sunlit rock face looks like. Gully snow is gated on the
snow line rather than two absolute heights.
Distance desaturation eased 1.9 -> 1.15: tuned on a valley whose rock was grey anyway.
NOT FIXED, and I want it recorded rather than implied: the Bells themselves are still
not maroon. The near and middle rock is warmer and measurably so, but the distant peak
does not respond to ANY of this - not the rock tint, not the ortho weight, not the snow
line, not the gully gate. Painting sA bright red left it unchanged, so whatever surface
that is, it is not this shader's snow and not this shader's material blend. Somebody
should find out what draws it before tuning any of these numbers further.
400 frames: GL 7.0 s, VK 7.2 s. Backends agree to 0.62/255.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Everything before this made the air a COLOUR APPLIED TO A SURFACE. Nothing put light
in the space between surfaces, so the basin had no shafts, no pooled mist and no rays
off a ridge at any hour, whatever was done to the fog.
A half-resolution march from the camera to the depth buffer, asking the same shadow
the rest of the frame asks - the cascades, the baked height-field shadow and the cloud
mask - so a shaft is cast by the actual trees and the actual ridge and a passing cloud
dims its own rays. Henyey-Greenstein scattering, because real air throws light forward.
Density and a separate ground-hugging mist layer ride the sun's elevation, so mist
forms in the cold at either end of the day and burns off by mid-morning.
Composited with the bloom pyramid's own tent upsample under ONE/ONE - what was wanted
and already there - and before bloom, so a shaft blooms. Into post_hdr, not post_scene:
post_scene is what the water refracts and shafts added there would sit under the lake.
The tuning that mattered was the sky term, which is added at every step: at 0.06 it
accumulated into a flat grey wash lifting lit and shadowed air equally, which is the
contrast a shaft is made of, and the valley came out one pale sheet. At 0.012 the sun
dominates and there is light rather than fog. I cut the density and mist three times
before the frame looked like air instead of paint.
R3D_NOVOL=1 for an A/B; Off in Settings skips the pass whole.
400 frames at 07:00: GL 7.1 -> 7.3 s, VK 7.2 -> 7.4 s. Backends agree to 0.08/255.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Aerial perspective is a curve now. A low sun shines through far more air than a
high one and shines ALONG the ground rather than down onto it, so density, height
falloff and forward scatter all ride the sun's elevation; overcast thickens the air
and flattens the scatter, because a grey sky has no disc to scatter from. `lowsun`
falls away BELOW the horizon as well as above it, or the middle of the night gets a
dawn's haze with no dawn to justify it.
The term that was missing entirely is distance DESATURATION. Blending a saturated
green ridge toward a saturated blue noon sky leaves a saturated ridge - which is why
the same valley read as a photograph at dusk, where the fog colour happened to be a
warm grey, and as a toy at one o'clock. A surface is now pulled toward its own
luminance faster than the fog itself arrives. Measured far/near saturation at the
camp: 07:00 1.11 -> 0.89, 09:00 1.04 -> 0.93, 13:00 0.98 -> 0.89.
The grade is the hour's too - nine literals bound at the draw, written by
daylight_set now. Noon is the case worth naming: direct sun is warm-white and the
only thing filling a midday shadow is a blue sky, so noon gets a cool balance over a
blue-lifted shadow with hard contrast, and dawn and dusk the reverse. Ground R-B,
lit vs shadowed: 07:00 +42.8/+14.2 -> +48.9/+15.1, 13:00 +32.2/+14.8 -> +25.2/+2.9.
Gain is left alone deliberately: the grade is `c * gain + lift * (1 - c)`, so warming
it warms the whole frame, and warming it at noon made one o'clock yellower than seven
in the morning - the opposite of the point.
The visible sky is relit. Turning a photograph on its axis does not change what
colour it was taken at, so every sunset had a mid-morning blue overhead. An analytic
sky supplies the chroma and the photograph keeps the luminance: the cloud stays where
it is and goes orange at dusk, the zenith goes deep blue at noon, and no second sky
is shipped. It fades out under the horizon and eases off under cloud.
The ground bounce follows the ground, crossing meadow to rock at the map's treeline
instead of being one green constant everywhere including above the scree.
R3D_NOAIR=1 restores all of it, so a before-and-after comes from one binary at one
hour; it joins R3D_NOCLOUD / R3D_NOSHADOW / R3D_NOGI.
Verified on macOS OpenGL, macOS Vulkan (MoltenVK) and Windows Vulkan (RTX 3070 Ti).
Backends agree: mean difference 0.15-0.88/255 within a machine. Across machines the
ORIGINAL renderer already differed by 5.02/255 at 19:12 and this build differs by
2.80, so cross-platform variance is pre-existing and did not grow. 400 frames: GL
7.4 s before and after, VK 7.0 s before and after.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
The platform gave these keys no code at all, so a game's rebinding screen could not take
one and nothing said why. Windows asked the active layout what they type and got nothing
(w_vk_char answers 0 for a key with no character); macOS let them fall through to
charactersIgnoringModifiers, which reports NSF1FunctionKey and its neighbours at 0xF704
and up - outside the 256-bit held set either way.
w_keyval and ev_keyval now name them, with the same codes on both: 132-143 F1-F12,
144-149 Home / End / PageUp / PageDown / Insert / Delete, 150 Caps Lock, 152-161 the
numpad digits, 162-166 its * + - . and /. The numpad's Enter is Enter.
Key.F1, Key.Home, Key.Numpad0 and the rest fold at compile time, and Input.key_label
names them without asking the layout - a key that types nothing is called the same thing
on every layout.
examples/library/input_typeless_keys.ludic covers the codes, the names, the held set and
the press edge; 150 passed in ludic-dev test, 33 in selfhost-test.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Crypto.random_* and Uuid.* read /dev/urandom, which Windows lacks, so every
byte was zero; the Windows target now calls RtlGenRandom (advapi32).
ludicc takes --title, which ludic build/run/bundle pass from package.ludic's
app name, so rt_init opens the window under that name instead of the
program name.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
The runtime opens a game's window before main, titled with the program's name, and
gl_open attached to it without passing its title on (render3d's gvk_open called win_open,
which returns early on Windows and opened a second window on macOS). win_set_title in
cocoa.ll / win32.ll (setTitle: / SetWindowTextW, UTF-8 -> UTF-16; a no-op without a window,
stubbed headless) retitles it from gl_open, and win_open on an open window retitles it on
both platforms.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Process.spawn/poll/kill/free - non-blocking child processes with no shell: posix_spawn on
macOS (process.ll), CreateProcessW with MSVC-quoted arguments and no console window on
Windows (process_win.ll), linked only when a program uses Process.*.
Http.save_to streams a response body into a file (NSURLSession with a run-time delegate
class on macOS, the WinHTTP read loop on Windows); Http.received / Http.expected report
progress while it is pending. Freeing a pending request cancels it and parks the slot
until the worker has finished.
App.window_hide / App.window_show take the game's window off the screen and back without
closing it; the run goes on while hidden. Docs, examples, tests and a changeset.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
The package overview, its commands, scripts and hooks, dependencies against
the lock with fetch/update/verify/vendor/add/remove, the app preview and the
asset roots. The bar over package.ludic now only prompts when something
needs doing.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Udp.open/port/send/recv/from_ip/from_port/close/resolve/local_ip/ip/ip_text, native
BSD sockets (udp.ll) and Winsock (udp_win.ll, -lws2_32), linked only when a program
uses Udp.*; example, docs and tests.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
glibc keeps sinf/sqrtf/floorf and the rest in libm, which macOS links implicitly;
examples/lang/floats.ludic failed to link on the Linux build-and-test job.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
emit_bin_vals lowered every pointer-typed ==/!= to @lp_str_eq, so two
distinct records compared their bytes up to the first zero byte. Content
compare now needs both sides to be text (string, or the untyped
pointer/ptr runtime code carries text in); other references use
icmp eq ptr, and string vs a non-text reference is a compile error.
Adds selfhost/tests/identity.ludic; both seeds regenerated.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
`float` (32-bit) and `double` (64-bit) with + - * / %, comparisons and unary minus.
Decimal literals take their type from context and stay `fixed` elsewhere; int and long
promote implicitly (LUDIC_WARN_FLOAT_PROMOTE=1 lists every promotion). float(), double(),
int(), long() and fixed() convert; floats(n)/doubles(n) buffers; float fields, globals,
constants and parameters; Math.* computes in float for float arguments; string/print
write the shortest round-tripping decimal; float_bits/float_from_bits expose the bits.
@deterministic code may not use floats. The f_* runtime helpers stay as they are.
Editors know the new type words; the JetBrains plugin is 1.5.0.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
- JetBrains plugin 1.4.0: semantic colours (builtin / vendor / own), template strings,
brace handling, run configurations and a test console, package.ludic and
package.lock.ludic editing (completion, docs, app preview, colour previews, asset
navigation), External Libraries for the runtime and packages, doc pages for built-ins
- ludic-lsp: go to definition for imports, document links, hover with inferred types,
type definition, signature help with parameters, docs from docs/language
- `import "dir"` resolves a barrel `dir/index.ludic`
- package.ludic `entry`, `script` and `hook before|after <command>`; `ludic <script>`,
`ludic script`, `ludic scripts`
- `ludic test --verbose` and `--test NAME`; the test runner filters by name
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
A reflecting body is clipped to its ellipse like every other unless it is an
unbounded sea, so a map whose reflection belongs to its lake (Maroon Lake, once
its sea sits below it) does not draw that lake's level over every hollow in the
survey. The terrain's wet shore and its forest and scree gates read the carved
lake's line inside its outline (u_lake), the rule grass already used. SPIR-V
regenerated.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Every R3D_* read goes through r3d_env_has / r3d_env (env.ludic): a headless
build honours them as before, a windowed build only when R3D_DEV is set to
anything but "0", so a shipped game never reaches a debug view, feature kill,
file writer or hardware fake. Documented in docs/SHIPPING.md.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
The Windows runtime keyed the held set by the character MapVirtualKeyA gave a
virtual key, so a game's WASD belonged to whatever the active layout put there,
and with an input method on every letter arrived as VK_PROCESSKEY. The typing
block is now read from the scancode (the arrows, numpad and F-keys still by
virtual key); macOS reads keyCode the same way. Input.key_label(key) names a key
in the player's own layout (Windows) or as its US character elsewhere.
Verified on Windows with SendInput into a live window: VK_Z carrying W's scancode
holds 'w', VK_PROCESSKEY carrying A's holds 'a', Caps Lock changes nothing.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
r3d_hdr_calibrate(peak, paper, black) feeds the tonemap's and the overlay's HDR10 variants and
the display's HDR metadata; ov_hdr_nits draws a calibration patch at a number of nits.
gpu_caps_probe asks the running Vulkan renderer's instance instead of making and destroying a
second one under Streamline's interposer, which left the next swapchain rebuild calling address 0.
The overlay gets an HDR10 variant, and the tonemap brightens HDR highlights by one factor rather
than per channel.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
A chunk's dispatch was sized for its largest tile, so the far tiles beside a
near one ran thousands of empty invocations: 9.8 ms of grass at 4K on an RTX
3070 Ti. One dispatch per tile, sized to that tile, brings it to 5.0 ms - still
three times the chunked path's 1.7 (36 fps against 41), so GF_MESH_GRASS is not
implemented yet and the Advanced row says a coming update.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
NVIDIA Streamline's interposer exports none of VK_KHR_acceleration_structure's
commands, so ray tracing looks each up per device with vkGetDeviceProcAddr and
calls it through a pointer: create (four pointers -> result), destroy (device,
handle, allocator), build sizes (device, type, info, counts, sizes), the
command-buffer build (buffer, count, infos, ranges) and the device address
(device, info -> u64). Both runtimes assemble; nothing uses them yet.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
The chunked grass path draws each chunk as one mesh-shader dispatch when the
setting asks and the card has VK_EXT_mesh_shader: work group y is a tile, x a
batch of 16 of its blades, and a blade the placement, density, frustum, water
or slope tests reject emits nothing - the instanced path still runs its eight
vertices to a degenerate position. grass.mesh generates the same blades as
grass.vert (grass_blade_mesh(4)'s rows, the same hashes, sway and lighting
normal), capped at the instanced path's 65535 a tile.
Vulkan: VK_EXT_mesh_shader with meshShader, and maintenance4 (glslang's mesh
stages declare LocalSizeId); vkCmdDrawMeshTasksEXT looked up per device, as the
Streamline interposer exports none; a *.mesh program's pipeline takes the mesh
stage and no vertex input, its bindings the mesh stage bit. gpu_has_mesh,
gpu_draw_mesh_tasks; r3d_mesh_grass and R3D_MESH_GRASS / R3D_NO_MESH.
bin/ludic-dev rebuilt: the committed binary predated the shader tool's mesh
support and compiled grass.mesh as a vertex stage.
PC (RTX 3070 Ti): the camp matches the chunked path; validation only the
no-window present-id message.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
HDR output: an HDR10 swapchain (A2B10G10R10, ST 2084 over BT.2020) when the
setting asks and the display offers it, with HDR metadata. The tonemap's HDR10
variant keeps the SDR picture up to a 200-nit paper white and rolls highlights
on to 1000 nits; the overlay's converts the interface to the same white. The
screen and LDR images go 10-bit with it; screenshots refuse while it is on.
OpenGL and the Vulkan SDR frame are unchanged. The instance asks for
VK_EXT_swapchain_colorspace. HDR metadata only where the loader has
vkSetHdrMetadataEXT: Streamline's interposer does not, and calling the thunk
crashed the game the moment the swapchain came up HDR10. PC 4K monitor: HDR10,
validation 0. R3D_HDR overrides the setting.
DLSS:
- the vertical jitter offset flips with Streamline's image (rows from the top):
unflipped, Quality resolved the ground into concentric rings;
- preset K in every mode: the default M put Performance at 18 ms a frame at 4K
on an RTX 3070 Ti (33 fps against 41 with DLSS off; with K, 60);
- the camera is jittered only while this frame holds a token and the last
evaluate worked.
R3D_DLSS_PRESET, R3D_CAM_LOG (the camera and DLSS state a frame) and
R3D_NOGRAIN for measuring.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
ludic-dev shaders: a variant whose first file is *.mesh compiles that stage with
glslang -S mesh for Vulkan 1.3, without the vertex stage's depth-remap wrapper
or invariant gl_Position (a mesh shader writes an array of positions and
remaps depth itself). Its SPIR-V keeps the .vert name, so the manifest and the
loader are unchanged. The 51 existing programs build identical SPIR-V.
runtime: lsl_call_piii(fn, ptr, i32, i32, i32) calls a command-buffer command
through a pointer. NVIDIA Streamline's interposer exports no
vkCmdDrawMeshTasksEXT, so it is to be looked up per device with
vkGetDeviceProcAddr. Both runtimes assemble.
Nothing uses either yet.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
grass_flush uploaded u_tiles (vec4[256]) with u_fv(4n floats). On Vulkan an
array element is copied at the size given and placed at the array's stride,
so each tile got one float: nonsense corners and zero blades a cell. The
meadow had no grass on the Vulkan renderer since cdfffa6 while the draw
counts looked right. u_f4v uploads n vec4s; OpenGL was never affected.
PC camp: chunked path matches R3D_GRASS_TILES=1, validation 0.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
streamline.ludic: slInit before the Vulkan instance, feature support per
adapter, a frame token per frame, Reflex sleep and PCL latency markers, and
DLSS super resolution on the lit HDR frame (Halton jitter, depth + zero
motion vectors with camera motion from clipToPrevClip, matrices carrying
the Vulkan path's y flip and depth remap). Bloom, tonemap and sharpen read
the upscaled size. The device asks for privateData and present_id, which
Streamline's hooks need. R3D_DLSS / R3D_REFLEX / R3D_SL_LOG for tests.
gpu_feature_implemented: DLSS and Reflex.
ludic bundle (Windows): app native "<dir>" copies native libraries beside
the executable.
Verified on an RTX 3070 Ti: DLSS Quality evaluates 1280x720 -> 1920x1080.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
vk_sl_prefer(1) before Vk.open() makes the Windows loader try sl.interposer.dll
beside the executable and fall back to vulkan-1.dll. Thunks for the sl* exports
and for calling feature functions from slGetFeatureFunction; macOS stubs.
Verified on an RTX 3070 Ti: slInit eOk, DLSS, DLSS-RR, Reflex and PCL supported,
DLSS-G reports no supported adapter (needs RTX 40).
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
The water reflection keeps the main camera's frustum planes, and ac_visible tested the actor itself
against them: the town's people, far above the lake with their images far below the frame, all drew
again. The image (2L - y) is what is tested now - town, 69 skinned reflection draws -> 38, the frame
byte-identical; PC camp reflection scene CPU 793 -> about 760 us. R3D_PROF splits the reflection's
CPU into setup, terrain, scene and sky.
Tried and not kept, with the numbers (PC, camp, Vulkan GPU timestamps):
- the ground's cost is its scanned material taps: without them the terrain pass is 555 us of 1568,
without the photograph 1029; the sun, the noise fields and the grain are 20-100 us each. Moving the
cheap tier in from 200 m to 60 m changes nothing, so it is the far tier's single taps over the
mountains. Skipping the normal-map taps past 900 m (where the detail normal is fully faded) saved
22 us and moved a few pixels - reverted.
- grass: cutting its radius to 300 m barely moves it, so the cost is the near blades' pixels; moving
their three noise fields to the vertices changed nothing (747 us) - reverted.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Where the frame goes, before optimising it further. R3D_PROF now prints the average frame: CPU before
the swap (the game's share apart), GPU and swap, and the renderer's CPU phases in frame order - the
shadow pass split into cascade fit, scatter casters and actor casters. On Vulkan the per-pass GPU
table comes from timestamp queries (host query reset asked for where the device has it); MoltenVK's
attribution is tile-based and not to be trusted per pass, the PC's is.
What it showed on the PC (camp): 4.3 ms CPU and 5.3 ms GPU a frame; the shadow pass was the largest
CPU phase (1.8 ms) and actors half of that. Every actor within 300 m was drawn into all five cascades,
though the outer two only shade receivers from 212 and 935 m out: 160 actors and 300 draws into each.
cast_band_reaches - the flowers' reach test, now shared - skips an actor for a cascade it cannot shade
(receivers counted from 0.85 of the previous split, where sunShadow's cross-fade begins).
PC camp, two runs each: mean frame 9553/9601 -> 8664/8656 us; CPU 4.3 -> 3.7 ms; shadow GPU 1.14 ->
0.79 ms; actor-shadow CPU 1.02 -> 0.60 ms; 2039 -> 1411 draws; self-tests 59/59, validation 0.
Mac: OpenGL shot viewpoints and the camp byte-identical; town 19 px at <= 2/255 on two flower stems a
few metres from the camera - the accepted leftover-binding difference, no shadow; self-tests 59/59 on
OpenGL and Vulkan. R3D_CAST_ALL=1 draws every caster into every cascade.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
The reflection pass is a second copy of the terrain, the vegetation and the actors, and it ran on
every frame of a map with water - looking straight down at a meadow included. The mirrored body is cut
into rectangles where the ground lies below it (32 m over the height map in 8 x 8 blocks, the sea
beyond it coarse), and the pass runs when one meets the view frustum and its water is not all dry
where it shows, outside the frame or behind the ground. Three wrong turns on the way, each kept in a
comment: bounding spheres (a 156 m cell reached into the view from behind the camera), sight lines
that never asked whether the point was in the frame, and a walk of every rectangle every frame (0.1 s
per 400 frames on the PC until the blocks). Built once in 6.6 ms.
Mac, the five shot viewpoints: view c (the meadow) 225 reflection draws -> none; a, b, d, e unchanged;
OpenGL frames byte-identical; self-tests 59/59 on OpenGL and Vulkan; MoltenVK validation adds nothing.
PC camp (lake in view): 1882 draws either way, 3.8 s for 400 frames either way, three runs each,
self-tests 59/59, validation 0. R3D_REFL_ALWAYS=1 runs the pass every frame; R3D_REFL_DBG=1 prints the
test once.
R3D_ACTOR_CENSUS=<frame> prints the lit pass's actors grouped by model: town is 17 models and 69 draws,
and only four rigid models repeat (17 actors) - too little for instancing to be worth a shader variant.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
tersun.frag rasterised every terrain patch a second time into a screen buffer, because on OpenGL the
cascade read inside terrain.frag fell off a driver cliff (about 6 ms a frame). Vulkan has no such
cliff: its terrain programs are the SUN_INLINE variant, which evaluates the same two tiers with the
same normal and cross-fade in the ground's own shader, and terrain_draw skips the pass - in the frame
and in the water reflection. OpenGL keeps the pass. R3D_SUN_PASS=1 keeps it on Vulkan, for comparing.
Mac Vulkan town 1759 -> 1692 draws; frames within 2/255 of the pass (15331 px, float rounding). PC camp
1984 -> 1882 draws, 3.9 s for 400 frames either way, self-tests 59/59, validation 0. OpenGL frames
byte-identical.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
A panel and its label alternated the white and font textures, and every switch closed the overlay's
draw range: 77-91 ranges a frame in town. The font atlas is now bound beside the image texture for the
whole flush, and v carries 4 x the vertex's mode (0 image, 1 font, 2 flat colour) on top of its real
coordinate, so only a different image texture or a clip rectangle closes a range. Town: 24 ranges.
OpenGL frames byte-identical at all five viewpoints; MoltenVK with validation adds no message.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Grass (Vulkan with multi-draw indirect): every visible tile is a record in one buffer, uploaded once a
frame, and each band draws its records 256 at a time. A record's firstInstance is its place in the
chunk times 65536; grass.vert's TILES variant reads that place's corner and indices per cell from
u_tiles. R3D_GRASS_TILES=1 keeps a draw per tile. OpenGL is unchanged.
Casters: a LOD level with no impostor is drawn into a shadow cascade only when its distance band,
widened by six times its height, the camera's height over the ground and the frustum's corner reach,
can touch that cascade's receivers. The flowers' mesh levels (6 - 30 m) leave the three outer
cascades. R3D_CAST_ALL=1 draws every level everywhere. OpenGL frames byte-identical at all five
viewpoints; alpha-tested shadow draws at a 460 -> 244.
gpu_has_mdi() guards both this and the GPU-culled trees' multi-record draws.
Camp: Mac Vulkan 2645 -> 2191 (grass) -> 2034 draws; PC 2657 -> 2046, self-tests 59/59, validation 0.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Every level of a kit tree or rock carries the same materials, so a GPU-culled layer merges each
material's levels into one mesh (layer_arena_build) and scatter_cull.comp writes a record per
(material, level) naming that level's index and vertex range; one indirect draw covers them all.
A conifer's lit pass and prepass go from 8 draws to 2, its shadow LOD from 6 to 2. Layers that do
not fit (card levels, other materials or attributes, 32-bit indices) keep the CPU path.
GPU culling is now the Vulkan default (R3D_GPU_CULL=0 turns it off). Camp benchmark, 400 frames:
PC 2791 -> 2657 draws, 4.3 -> 4.1 s (both runs); Mac 2791 -> 2657, 8.0/7.4 -> 7.7/7.2 s.
Self-tests 59/59 on both machines, PC validation 0 errors; frames within run-to-run noise; OpenGL
frames unchanged.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
gpu_rb_storage makes the image with the samples asked for; a pass takes its attachments' count and
every pipeline in it matches; a blit from a multisampled source resolves on the end of an empty
dynamic-rendering pass (colour averaged, depth from sample zero - vkCmdResolveImage cannot do depth).
gpu_msaa_max reads the device's colour-and-depth sample limits, and post_set_msaa clamps to it, so
the Anti-aliasing setting is live on Vulkan. The pipeline cache's pass key no longer packs samples
into three bits.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
gpu_feature_implemented(GF_VULKAN) is true: the Vulkan renderer draws the whole game (validation
clean, 105 fps at the camp on the RTX 3070 Ti). Ray tracing, DLSS, Reflex, HDR output and mesh-shader
grass still report false, so their rows keep saying they take effect later.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
- A block sub-allocator: 64 MB blocks per memory type, images and buffers kept apart, first fit with
alignment, freed ranges merged and empty blocks given back; anything over 16 MB still gets its own
allocation. The self-tests' second world holds 94 allocations instead of 8735 (the driver's limit
refused a shadow map before). Camp bench unchanged, 105.3 fps.
- shadow_set_res keeps the size that worked when the card has no memory for the new one, and records
it in shadow_refused, instead of ending with no shadow map; gpu_tex_ok says whether a texture has an
image behind it.
- Image barriers skip an image that was never made (a failed allocation used to crash there), and
R3D_VK_ERRLOG=<file> appends every Vulkan failure line by line, so a crash no longer takes the
message with it.
- R3D_VK_PROF reports draws asked for and not made, so a layer missing from a frame is never silent.
Validation on (VK_INSTANCE_LAYERS): the game's self-tests 61 OK, 0 errors, on the RTX 3070 Ti.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
- A draw's set carries its textures and lives in a pool of its own, keyed by each texture's handle,
generation and sampler; the uniform blocks are dynamic uniform buffers into the frame's ring,
bound with the draw's offsets, so a draw that changes only uniforms allocates and writes no set.
The sampler for an unbound slot is made once instead of looked up by string every draw.
Camp bench, RTX 3070 Ti, 400 frames, twice: 102.6 fps (53.3 before; OpenGL 114.3), sets 0.9 ms
against 8.5. MoltenVK (M4 Pro): sets 0.2 ms.
- Integer vertex attributes read by float inputs use USCALED formats (a_joints was UINT against a
vec4), and every shader input a mesh does not feed reads a shared zero buffer. NVIDIA drew
anyway; MoltenVK refused every skinned pipeline and the whole actor layer was missing from the
Mac's Vulkan frame. A draw skipped for want of a pipeline now says so, once per program.
- A failed image allocation is reported instead of bound as a null allocation.
Validation proven on (VK_INSTANCE_LAYERS): 0 errors over the game's self-tests (61 OK) and a frame.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
- sky_start_yaw: a game that turns the sky at boot sets it before r3d_init and the image-based light
is baked once at that yaw; sky_set_yaw to a yaw already baked bakes nothing.
- terrain_init_step: the height field and normals, the sun shadow, the materials, the patches and
programs as separate steps; r3d_load_count is 8, so a loading bar moves through the terrain (half
the start-up) instead of jumping over it. terrain_init runs the four in a row as before.
- The per-cascade actor cull centres its sphere at half the model's height and sizes it by the
larger of height and radius: centred at the radius, it dropped a head at a cascade's edge (11 px
in town, found by the drawstats comparison). Lake was byte-identical before and after.
OpenGL frames at the five viewpoints unchanged; the game's 59 self-tests pass.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
actor_draw_casters drew every visible actor within 300 m into all five cascades. The draw-call
count (R3D_DRAWSTATS) found about 1500 skinned shadow draws a frame in town against 69 in the lit
pass. The cascade is an orthographic box, so a bounding sphere (1.5x the actor's radius) outside
its clip x/y casts nothing into that layer and is skipped. OpenGL frames at the five viewpoints
are unchanged; the game's 59 self-tests pass.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
- r3d_open opens the window with nothing baked; r3d_load_count / r3d_load_step run the rest (sky
and light, terrain, shadow and screen targets, cover and actors, grass) so a game can present a
loading frame between them. r3d_init is the same calls in a row.
- sky_set_quality(width): the prefiltered sky light at 256 / 512 / 1024, baked again.
- post_set_msaa(samples): multisampling on OpenGL, remade in place (post_msaa_live is false on
Vulkan); post_free frees the multisampled framebuffer too.
- STREAM_BUDGET_US is a variable; r3d_fog_scale multiplies the fog the day sets.
- Vulkan buffers carry UNIFORM_BUFFER usage: the frame's ring is bound as uniform buffers and was
created without it (VUID-VkWriteDescriptorSet-descriptorType-00330, found on MoltenVK).
OpenGL frames at the five viewpoints unchanged; the game's 59 self-tests pass.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Measurement only: a tally at each of gpu.ludic's five draw doors, program
changes and texture binds, keyed by the open profiler pass. Off unless
R3D_DRAWSTATS is set; frames byte-identical with it off and on (Mac, OpenGL).
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
- r3d_set_anisotropy(level) updates every mipmapped texture already loaded (OpenGL parameter,
Vulkan sampler record), not only later uploads; the game's setting never reached the scanned
materials, which load before the settings are read.
- shadow_set_res(size) remakes the cascades at 1024 / 2048 / 4096 (shadow_res replaces the
SHADOW_RES constant); lighting.glsl reads the texel size from the map, so OpenGL frames at
2048 are unchanged.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
- Device: multiDrawIndirect, drawIndirectFirstInstance and drawIndirectCount where present.
- Buffers carry storage and indirect usage; a GPU-owned buffer is never swapped under a draw.
- Compute programs from shaders/compute.list (binding 0 parameters, 1.. storage buffers),
built by `ludic-dev shaders`; gpu_compute / gpu_dispatch / gpu_draw_mesh_indirect in gpu.ludic.
- R3D_VK_PROBE=1: a dispatch read back (OK on the RTX 3070 Ti).
- scatter_cull.comp: a tree layer's frustum test and LOD split on the GPU, with the lit, prepass,
impostor and shadow-LOD draws reading its records. Behind R3D_GPU_CULL=1 and off by default:
at the camp it is slower (43.0 fps against 53.3), because the frame's cost is per-draw
descriptor sets and it adds empty-level draws. Validation-clean; OpenGL frames unchanged.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
`fn name` read `while p < fn and ...` as a reference to a function called `and`, which broke
tools/ludic-cli/glgen.ludic and so the dev tool's own build. A reference now needs the name on
the same line and not one of and / or / not / in / is. The threads example checks `fn` as a
local before `and`. Reseeded; bootstrap-cfree reproduces the seed.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
- `fn name` names a top-level function as a value (E_FNREF, lowers to @fn_<name>); the worker
entry point for Job.parallel_for, which checks it takes (int, pointer-like) and returns void.
- runtime/native/threads.ll (pthreads) and threads_win.ll (Win32 SRWLOCK/CONDITION_VARIABLE): a
pool of one worker per core but one, parked between batches; every thread claims chunks by
compare-and-swap. Linked only into programs that use Job/Promise/Sync, by `ludicc -o`,
`ludic build` and the test suite's build helper.
- Sync.* is real: native mutexes, atomics as cmpxchg retry loops (neither clang takes atomicrw,
the PC's rejects seq_consistent), mutex-guarded channels, Sync.cpu_count from the OS.
- spawn/despawn on a pool thread stop the program with a located panic.
- examples/library/threads.ludic and its test; docs for fn, Job.parallel_for, Job.is_worker.
- Reseeded (bootstrap-cfree: out.ll == seed.ll). 141/141 on macOS; jobs, threads and the guard
pass on Windows from the reseeded Windows seed.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
R3D_VK_PROF at the camp view (about 2950 draws a frame) showed the frame spent in bookkeeping:
19.5 ms a frame finding pipelines by string key and 16 ms building descriptor sets.
- Pipelines: a mesh carries an interned vertex-layout id (dropped only when an attribute's shape
changes, not when an instance buffer is swapped), render state packs into an int and the pass
formats into another; the program's last hit is tried first. The string path only builds.
- Samplers: each texture keeps the sampler for its parameters until they change.
- Descriptor sets: a program's last set is reused within the frame while its blocks and resolved
textures are unchanged; a uniform write that repeats the value it already holds changes nothing.
Headless on the RTX 3070 Ti at 1920x1080: 21.7 -> 53.3 fps (pipelines 0.2 ms, sets 8.5 ms, inside
draws 10 ms a frame; OpenGL 114 fps). The camp frame is unchanged and validation-clean. OpenGL frames
byte-identical at the five viewpoints; 59 self-tests pass; VKRES OK.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
- A buffer re-uploaded after a draw this frame read it gets fresh storage, as OpenGL orphans
one; storage moved off or freed while the frame still reads it is destroyed after the frame's
submit. Draws mark the buffers they bind. No flush for buffer work.
- Mipmaps asked for mid-frame (the exposure measure, every frame) are recorded into the open frame
after its pass; growing a chain the first time keeps its one-shot path.
- R3D_VK_PROF prints, every 120 frames, draws and flushes per frame and the milliseconds spent
finding pipelines, filling descriptor sets and inside draws.
The gain was small - the camp view headless at 1920x1080 on the RTX 3070 Ti went from 21.3 to
21.7 fps (OpenGL: 114 fps) - so these flushes were not what holds the frame; the profile is how
the rest is found. The frame is unchanged and validation-clean; OpenGL frames byte-identical at the
five viewpoints with 59 self-tests passing.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
- gpu_select takes Vulkan for a window on Windows; gvk_init asks for VK_KHR_surface,
VK_KHR_win32_surface and VK_KHR_swapchain when the renderer will present. A window elsewhere
stays on OpenGL with the reason.
- gvk_open opens the window without a GL context and makes the screen images at its client area;
gvk_swap_make builds the swapchain on the Win32 surface (FIFO with vsync, mailbox or immediate
without) and rebuilds it when it is out of date, suboptimal or the window changes size.
- gvk_present blits the screen image into the acquired swapchain image, flipped (the screen keeps
OpenGL's bottom-up rows), and presents it.
- Samplers pointed at a texture unit with u_i and then fed by binding units - the actors do this -
read that unit's texture; on Vulkan they drew with the white stand-in (the tent, the log, the
chair, the workbench).
- gl.ll carries a weak win_gl_drawable for headless macOS builds.
On the RTX 3070 Ti the windowed build runs the valley through Vulkan at 3840x2160 with the HUD and
the frame-rate readout (16 fps: every upload still waits on the frame, and buffers are
host-visible). The headless Vulkan frame is unchanged; OpenGL frames byte-identical at the five
viewpoints with 59 self-tests passing. The actor fix is compiled and OpenGL-verified, not yet
seen on the PC; blades at the grass's middle distance still draw black on Vulkan.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
OpenGL links a vertex output to a fragment input by name; SPIR-V links them by location, and
glslang's --auto-map-locations numbered each stage in its own declaration order. The foliage
prepass's depth.frag declares v_wpos then v_uv where model.vert writes v_wpos, v_nrm, v_uv, so
the prepass read a normal as its texture coordinate, its alpha test cut every flower head and
leaf, and the lit pass (depth EQUAL) drew nothing over them. `ludic-dev shaders` now gives both
stages explicit locations: the vertex stage's out order numbers them and the fragment stage looks
each in up by name. All 45 variants checked: every fragment input sits on its vertex output.
Also:
- A clear still waiting for its pass when the framebuffer changes now runs on that framebuffer,
instead of becoming the load op of whichever pass began next.
- R3D_DUMP_ATLAS writes every impostor and card atlas a run bakes (build/atlas_<n>_*.ppm); the
40 baked on Vulkan match OpenGL's.
The PC's Vulkan frame now shows the flowers as OpenGL does, validation-clean. ludic-dev test 140
passed; OpenGL frames byte-identical at the five viewpoints; 59 self-tests pass.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Three things the first Vulkan frames on the RTX 3070 Ti showed against OpenGL on the same PC:
- Exposure: the adaptation pass reads the HDR scene's smallest mip, and a render target made
without pixels had one level, so exposure came from a single texel. A target asked for mipmaps
now grows a full chain (level 0 kept), and passes draw through level-0 views keyed by the
image's generation.
- Foliage: the depth prepass and the lit pass (depth EQUAL) are different variants. Vulkan vertex
stages now declare an invariant gl_Position so both land on the same depth.
- Alpha to coverage is enabled only on a multisampled pass. OpenGL ignores it without MSAA; Vulkan
with one sample dropped every fragment under half alpha.
vk_resources also checks a big-endian 16-bit RGB upload (a normal map). VKRES OK; ludic-dev test
140 passed; the PC's Vulkan frame is validation-clean; OpenGL frames byte-identical at the five
viewpoints with 59 self-tests passing.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
gpu.ludic's calls branch to the Vulkan backend when it was chosen and came up; every OpenGL
statement is unchanged, only guarded. R3D_GFX=vk selects it in a headless run (the window's
swapchain is the next milestone) and falls back to OpenGL, with the reason, when the device or
the SPIR-V manifest is missing.
- Render state is cached as before and turned into pipelines at the draw; u_* and sampler binds
go into the variant's uniform blocks; meshes, buffers and textures are gvk_* objects;
framebuffer binds are dynamic-rendering passes; same-size blits are image copies; the screen,
the photograph read-back, the present and the screenshot go through the frame.
- Work that submits on its own (uploads, read-backs, new or freed images and buffers) flushes the
frame first, so it runs in OpenGL's order. A read may take fewer channels than the image has
(the height field's R from its RGBA32F bake). Pipeline keys name vertex bindings by order, not
buffer handle, so re-pointed instance buffers keep their pipeline.
The valley renders at frame 90 validation-clean on the RTX 3070 Ti and on MoltenVK. OpenGL frames
byte-identical at the five viewpoints; 59 self-tests pass with no GL error.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
The drawing half of the Vulkan backend, compiled into render3d and not yet reached from it:
- gvk_program: a program handle as its manifest variant - two SPIR-V modules, a descriptor set
layout (the vertex block at 0, the fragment block at 1, the samplers at their manifest bindings)
and a pipeline layout. Nothing is compiled at run time.
- gvk_pipeline: one pipeline per program, recorded vertex layout, render state and pass formats,
built the first time that combination draws. Attributes the shader does not read are left out;
the front face is clockwise, since neither API flips y between clip space and its target rows.
- gvk_uniform / gvk_u_set / gvk_bind_texture: loose uniforms written into each stage's block at
the manifest's offsets and array strides; gvk_draw_set copies the blocks into a per-frame ring
at the device's alignment and fills a descriptor set from a per-frame pool, with a white 1x1
texture for a sampler nothing was bound to.
- The frame: one command buffer; a framebuffer bind ends the pass and the next begins at its first
clear or draw (a clear that comes first is the load op); attachments move to attachment layouts
for the pass and back to SHADER_READ_ONLY after it, a cascade drawn through a view of its layer.
gvk_present submits and waits; gvk_screenshot reads the screen image back bottom row first.
r3d.ludic now imports gpu_manifest.ludic too. Textures remember their size.
OpenGL frames byte-identical at the five viewpoints; 59 self-tests pass; VKRES OK and VKDEVICE OK.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
The renderer's projections are OpenGL's, whose clip-space depth runs from -w to w; Vulkan clips
everything below 0. `ludic-dev shaders` now wraps each vertex stage - its own main runs, then
gl_Position.z = (z + w) / 2 - so every variant's depth lands in [0, w]. The GLSL the OpenGL
renderer compiles is untouched. No y flip is needed: a Vulkan target's row 0 is where OpenGL's is
(NDC y = -1), so render to texture, sampling and gl_FragCoord agree between the two, and only the
present and the screenshot flip.
45 vertex modules regenerated (spirv-val clean, manifest unchanged); ludic-dev test 140 passed.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
gpu_vk_res.ludic gains what gpu.ludic's texture and buffer handles stand for on Vulkan:
- gvk_tex_storage / _upload / _mips / _read / _release: an image and view per handle, a full mip
chain when pixels come with it (the renderer asks for mipmaps after the upload) and one level
for a target, every level in SHADER_READ_ONLY between uses. Uploads are converted to what the
image stores: a missing alpha filled opaque (three-channel formats are stored with four), 16-bit
PNG samples byte-swapped when the unpack state says so, 32-bit float HDR halved into half
floats. Mips are blitted down level by level; a read-back brings level 0 home.
- gvk_sampler: one VkSampler per filter / wrap / compare / anisotropy combination, made when first
asked for, with GL's defaults where the renderer set nothing.
- gvk_buf_*: vertex, index and instance buffers behind one handle, kept when an upload fits.
Host-visible while the backend comes up.
gpu_vk.ludic switches on anisotropic sampling where the device has it and reads its limit.
examples/rendering/vk_resources.ludic checks it all: VKRES OK, validation-clean, every allocation
freed, on the RTX 3070 Ti (16x anisotropy) and on MoltenVK. One run on the Mac crashed while a
headless game run was using the GPU and did not come back in two reruns. OpenGL frames
byte-identical at the five viewpoints; 59 self-tests pass.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
The Vulkan backend is two files. gpu_vk.ludic is the device, memory and one-shot commands,
pure Vulkan, and still runs on its own (vk_device.ludic). gpu_vk_res.ludic is the resources in
the renderer's vocabulary - OpenGL's names for formats, filters and blend factors, which only
exist where Gl.* is named - starting with the format table. r3d.ludic imports both after
gpu.ludic; nothing calls them yet.
OpenGL frames byte-identical at the five viewpoints; 59 self-tests pass; VKDEVICE OK.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
The first part of the Vulkan side of gpu.ludic, not yet reached from the renderer: gvk_init
brings up the instance (portability enumeration where offered), the first discrete GPU, a
graphics queue and a device with the Tier 1 floor switched on (dynamic rendering,
synchronization2, descriptor indexing, timeline semaphores), and says why when it cannot so
the caller stays on OpenGL. gvk_mem_type / gvk_alloc pick and allocate memory (one allocation
per resource while the backend comes up), and gvk_once_begin / gvk_once_end carry uploads,
bakes and read-backs through submit and wait.
examples/rendering/vk_device.ludic runs it alone: VKDEVICE OK and validation-clean on the RTX
3070 Ti and on MoltenVK.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
A game ended its frame with Gl.swap() and shot it with Gl.screenshot, which tied it to OpenGL
and made it name Gl.* only so the parser would splice the GL runtime. gpu_present and
gpu_screenshot carry both (and name Gl.* themselves, so importing render3d is enough), and
r3d_present / r3d_screenshot are what a game calls. The Vulkan backend takes both over.
OpenGL frames byte-identical at the five viewpoints; 59 self-tests pass with no GL error.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
A Slang vertex and fragment shader draw a triangle into an image with dynamic rendering (no
render pass object: the pipeline names its colour format), image barriers move it to the
attachment and transfer layouts, and vkCmdCopyImageToBuffer reads it back into a PPM. It is
the path the Vulkan renderer's passes and screenshots take.
The corner comes from SV_VulkanVertexID: Slang compiles SV_VertexID to gl_VertexIndex -
gl_BaseVertex, which declares the draw-parameters capability.
Validation-clean and VKTRIANGLE OK on the RTX 3070 Ti (Windows) and the M4 Pro (MoltenVK),
13824 pixels covered on both.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
The last of milestone 1: no OpenGL call is left in render3d outside gpu.ludic but the clock
and the CPU-side buffer helpers.
- gpu_program builds a program and remembers the variant it came from (vertex, fragment and
defines as one line - the SPIR-V manifest's key), so a backend that cannot compile at run
time finds the pipeline for the same handle. gpu_use_program and gpu_program_free replace
32 uses and 2 frees; the overlay's program is recorded under overlay.vert|overlay.frag.
- gpu_query_new / _begin / _end / _result carry R3D_PROF's timers.
- gpu_open, gpu_vsync, gpu_renderer_name and gpu_resize_check carry the context.
- r3d_program_tess is gone: nothing called it and no tessellation shader exists.
- R3D_GLCHECK also checks after framebuffer and renderbuffer changes, viewport, draw buffers,
program use, texture parameters, the resize check and between frames.
OpenGL frames are byte-identical at the five viewpoints; 59 self-tests pass with and without
R3D_GLCHECK, with no GL error; ludic-dev test 140 passed.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
The overlay flushed into whatever framebuffer was bound last, which raised GL error 1286 on
every run: ov_flush now binds the screen and its viewport before it draws.
R3D_GLCHECK=1 checks each draw, clear, blit, upload and attachment for a pending error or an
incomplete framebuffer and names the target, and each sampler bind for a texture with no image
or a mipmap filter without mipmaps. Off, it costs one flag test.
Still open: an intermittent 1286 reported at "terrain shadow bake" after the self-tests (about
half the runs), and one macOS "unloadable" texture warning at shutdown while the post targets are
freed. No frame samples a bad texture.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
win_native_window / win_native_instance hand a Vulkan swapchain what it is created on: the
HWND and module instance on Windows (VkWin32SurfaceCreateInfoKHR), the game's view on macOS.
Headless builds define both as null (weak on macOS, so a windowed link keeps cocoa.ll's), so
code that asks for them links everywhere and simply finds no window.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
gpu_manifest.ludic loads shaders/spv/manifest.txt and answers what a backend asks while drawing:
which variant a program built by r3d_program is, where a uniform lives in its stage's block,
which binding a sampler has, what the vertex inputs are. examples/rendering/vk_manifest.ludic
resolves every program in variants.list against it (45 of 45) and checks a few offsets and
bindings. Nothing imports it yet.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Framebuffers and their attachments, renderbuffers (multisampled too), draw and read buffers,
completeness checks, blits, viewports, clears, the screen framebuffer, the multisample enable,
the wireframe switch and GL error checks now go through gpu_fb_* / gpu_rb_* / gpu_viewport /
gpu_clear / gpu_blit / gpu_check, and no other file names them. Each call is the one GL call it
replaces, in the same order: the fixed viewpoints render bit-identically and the game's
self-tests report exactly what they did.
What is attached to each framebuffer - colour slots, a depth texture or one layer of an array,
renderbuffers and their samples - is recorded as it is attached, for a backend that builds
render passes and image views. gpu_read_screen is the frame read-back a photograph takes.
R3D_GLCHECK=1 checks, around every draw, clear and blit, that the bound framebuffer is complete
and that no error is left behind, naming the framebuffer. It has already narrowed the old
"gl error 1286 at terrain shadow bake": the error is pending before a draw after the map self-
test, so it comes from a call that is not a draw, clear or blit.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Texture creation, 2D and array uploads, pixel packing, filters, wraps, comparison, border,
anisotropy, mipmaps, texture units, read-backs and freeing now go through gpu_tex_* and
gpu_bind_sampler, and no other file names them. Each call is the one GL call it replaces,
in the same order, so OpenGL renders bit-identically at the fixed viewpoints and the game's
self-tests report what they did before.
What those calls say about a texture - size, format, layers, min and mag filter, wraps,
comparison, mipmaps, anisotropy - is recorded per handle as it is set: a backend with
immutable images and separate sampler objects creates both from exactly that.
r3d_bind_tex takes a texture kind (GPU_TEX2D / GPU_TEX2D_ARRAY) instead of a GL target.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Every vertex array, vertex and index buffer, attribute pointer, instance divisor, stream
upload and draw call now goes through gpu_mesh_* / gpu_buffer_* / gpu_draw_*, and no other
file in the package names them. A Mesh records its layout as it is built - which buffer
feeds which attribute at what stride and offset, per vertex or per instance - so a backend
that bakes vertex input into a pipeline can read it back. On OpenGL each call is the GL it
replaces, in the same order: the five fixed viewpoints render bit-identically and the game's
self-tests report exactly what they did before.
scatter_attach takes the mesh rather than its vertex array; a mesh now frees every vertex
buffer it owns (glTF meshes used to keep all but the first); the two helpers nothing called,
mesh_grid_patches and mesh_instance_buffer, are gone.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Vulkan cannot compile GLSL when the game starts, so the programs render3d builds are listed
(shaders/variants.list, 45 of them, collected with R3D_PROGRAMS_LOG across the self-tests, the
screens, the viewpoints and the debug switches) and `ludic-dev shaders` compiles each into
shaders/spv/<id>.vert.spv and .frag.spv with a manifest of what the backend needs: each
stage's uniform block and member offsets, the samplers' bindings, the vertex inputs.
The GLSL is the renderer's own, assembled as programs.ludic assembles it, through glslang's
relaxed Vulkan mode, so gpu_uniform / u_* can write the same uniforms into a block on Vulkan.
Bindings are assigned by the tool (glslang's own numbering put several samplers of one stage
on binding 0): the vertex block is 0, the fragment block 1, samplers from 2 in name order,
shared across both stages. Every stage passes spirv-val; `ludic-dev test` rebuilds and compares
wherever the Vulkan SDK is installed.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
`ls a b` fails when `a` is missing even though `b` exists, so with VULKAN_SDK unset the
guard skipped the check on a Mac that has the SDK under ~/VulkanSDK.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Render state (depth, blending, culling, colour writes, alpha-to-coverage, depth bias,
scissor) and uniforms go through gpu_* / u_* and nowhere else; OpenGL state is cached and
only changes reach the driver. Frames are bit-identical to before at the fixed viewpoints.
R3D_GFX=gl|vk or gpu_request chooses a backend, falling back to OpenGL with a reason.
gpu_caps_probe() asks Vulkan, on Windows, for the 1.3 floor, ray tracing, mesh shaders, the
NVIDIA RTX generation, Reflex and HDR colour spaces, for a game's settings to grey out what
a machine cannot use; R3D_CAPS pretends to be a given card for tests.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
ludic-dev vkgen reads vk.xml into runtime/native/vk_api.ludic (constants, every struct's
<Struct>_sizeof and <Struct>_<field> offsets, one extern per command) and vk_thunks.ll.
Every size and offset was compiled against the SDK's C headers; `ludic-dev test` checks the
tracked files against the registry wherever the Vulkan SDK is installed.
vk_win.ll (vulkan-1.dll) and vk_mac.ll (libvulkan.1.dylib, MoltenVK) open the loader at run
time, so a program built with Vk.* starts on a machine without Vulkan. ludicc and ludic
build link both for any program that uses Vk.*. The seeds are regenerated for the new
compiler.
vk_probe reports what a machine's Vulkan can do; vk_compute dispatches a Slang compute
shader and reads the picture back, clean under the validation layer on an RTX 3070 Ti and
on an M4 Pro through MoltenVK.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
The delta was this frame's position minus the last, and the last started at 0,0, so the
first frame reported the cursor's whole distance from the corner as motion. A cursor-mode
change did the same, switching between a locked cursor's virtual reticle and the real
cursor. Maroon Lake's camera adds mouse_dy to its pitch and came up pointing at the ground.
examples/library/input_mouse_rebase.ludic covers both cases, plus ordinary motion and
re-setting the mode already in force.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
The queue was only emptied by the next outline_model call, so the last highlighted tree kept
its rim after the player looked away. r3d_frame now calls outline_frame, which drops a batch
the previous frame closed and nobody reopened.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
ov_text, ov_text_w and ov_text_wrap decode UTF-8 instead of drawing bytes 32-126. font.json
may list the atlas's code points in "codes"; an atlas without it reads as before (ASCII
from "first"). A missing code point draws as '?', a cut-off sequence as one '?'.
overlay_font(dir) loads or swaps the atlas at run time, for a language with its own script.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Needle-card crowns are many cut-out quads deep and a discarding shader turns early depth
rejection off, so every card behind the front one ran the full lighting shader. The near
tree LODs now write depth first (depth.frag, the same alpha coverage test), terrain under
them is rejected before shading, and the lit pass draws them with no discard and an equal
depth test (invariant gl_Position). R3D_NOPREPASS=1 restores the old path.
Dense forest on a Windows PC: 61 -> 78 fps at 3840x2160, 116 -> 164 fps at 1920x1080.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
http_link_flags grepped the IR for "@hs_", which every program's header
declares, so `ludic build` linked the HTTP transport and Foundation into
everything - invisible on macOS, a failed link on Linux, where it broke four CI
cases (ludic run, the installed layout, ludic new, the seed build through
`ludic build`). It now looks for a call to hs_send: examples/library/http.ludic
has one, snake.ludic (which still declares ten @hs_ names) has none.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
The asset-pack boot calls _NSGetExecutablePath, which glibc does not have, so
the Linux link of the compiler seed failed in every CI job - build-and-test,
cfree-fixpoint and every publish run - since packs landed. The last release CI
created was v0.7.0. The Linux shim now defines it over /proc/self/exe.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
http_win.ll implements http.ll's hs_* contract over WinHTTP: the request is a
record built on the game thread, the whole exchange runs on a worker thread,
TLS uses the system's certificate checks, and headers are answered from the
kept request handle. ludicc links it with winhttp instead of refusing Http.* on
Windows.
win32.ll decodes the splash and App.set_icon images with WIC (ole32): the
splash is a topmost borderless window at the artwork's size in points times the
display scale, composited over its background colour; the icon becomes an
HICON set on the window now or when win_open makes one.
Verified on the PC: examples/library/http.ludic prints its expected output, a
real GET to https://git.workshopsoft.io/ returns 200 with its body and
Content-Type, and the bundled Maroon Lake shows its splash centred at launch and
its icon in the title bar.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
The CLI's shell commands go through shell(), which is run() on POSIX and a
scratch script handed to Git for Windows' bash on Windows (exit codes read
directly there). compile_app links through `ludicc -o` on Windows, ludic run
starts the .exe, and ludic-dev build and ensure_ludicc assemble
selfhost/ludicc.win.seed.ll, which ludic-dev reseed now writes beside the
macOS seed. ludic bundle makes build/<name>/ with a GUI-subsystem exe carrying
the .ico beside `app icon` as an llvm-rc resource, game.lpak and packs.index;
ludicc gains --gui and --link, and quotes its whole link line for cmd.exe.
Verified: ludic-dev test 135/135, selfhost-test 32/32; on the PC a checkout
bootstraps from the Windows seed, ludic-dev build makes the toolchain, and
`ludic bundle` makes build/Maroon Lake/, which runs from its own folder.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
AppKit gives kVK_Delete (51) the character NSDeleteCharacter, 127, which is
what both ev_keyval (the held-key set) and win_poll (the per-frame key)
received, so Key.Backspace, which folds to 8, never matched. Both now map key
code 51 to 8.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
compile_app linked gl.ll when the IR named @lgl_* but never http.ll and
Foundation for @hs_*, so examples/library/http.ludic failed to link under
`ludic build` on every hs_* symbol while `ludicc -o` built it. http_link_flags
greps for @hs_ and links them in both modes; ludic-dev test now builds the
example through `ludic build` as well.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
The arrow Key constants folded to the codes of w/s/a/d, which is what the
per-frame key reports for an arrow, but cocoa.ll has always stored an arrow in
the held-key set under 128-131. So Input.key_down(Key.Up) read the W bit and
Input.move_i's arrow half never moved anything. Input.key keeps its WASD alias,
so games comparing it with 'w' (snake, chronorift) still take the arrows.
New example input_arrows.ludic, checked by ludic-dev test.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
A game asking for 960 x 540 got 960 x 540 pixels - a quarter of a 4K panel.
w_fit_scale turns the display's DPI into a whole-number scale (96 -> 1, 168 and
192 -> 2), brought down until the window fits; win_open and win_gl_resize size
the client area with it, win_gl_scale reports it the way cocoa.ll reports the
backing scale, and the mouse still arrives in the game's own units.
Verified on a 3840x2160 panel at 175%: windowed gl_triangle's 640x360 window
draws a 1280x720 frame.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
audio_win.ll implements audio.ll's snd_* contract over XAudio2 from IR: a
source voice per clip, restart on play, LoopCount for loops, SetVolume,
SetFrequencyRatio and a balance pan through SetOutputMatrix, with the COM
slots taken from the SDK's xaudio2.h. Clips are RIFF WAVE read through
lp_pak_open (weakly referenced), so a packed sound loads directly. ludicc links
it with xaudio2 and ole32, and silences xinput.lib's importeddllmain warning.
Verified: ludic-dev test 135/135, selfhost-test 32/32; on the PC a harness
loads, plays, pans, loops and stops clips, and Maroon Lake windowed reports
"sound: 31 of 31 clips loaded".
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
win32.ll implements cocoa.ll's contracts over user32 and xinput: the message
pump, the held-key set and frame key in Ludic codes, the mouse with raw input
for cursor mode 2, clip-based cursor modes released on focus loss, XInput pads,
and the software framebuffer present. win32_gl.ll puts the WGL 4.1 core context
on that window (vsync, borderless full screen); gl_win.ll's context code is now
lgl_wgl_core, shared with the headless hidden window, whose win_gl_* stubs move
to gl_win_nowin.ll. audio_win.ll links Audio.* silently for now.
Verified: macOS fixpoint, ludic-dev test 135/135, selfhost-test 32/32; on the
PC gl_triangle renders identically headless and in a real window, and Maroon
Lake builds windowed and runs a playtest to frame 240 in the desktop session.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
The UCRT's remove() deletes files only, so a tree removed bottom-up left every
directory behind. emit_win defines remove over DeleteFileA, falling back to
RemoveDirectoryA. Found by Maroon Lake's selftest26 ("1 left behind"), which now
passes on Windows with the rest of that game's self-tests.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
gl_win.ll creates a hidden-window WGL 4.1 core context and carries gl.ll's
float/memory helpers, with ldexp and QueryPerformanceCounter in place of the
libSystem calls. glgen now also writes gl_thunks_win.ll: the same 478 thunks,
calling through pointers that @lgl_win_load fills from wglGetProcAddress (and
opengl32.dll for GL 1.1). ludicc links the pair against opengl32/gdi32/user32
on a Windows target.
Verified: gl_api.ludic and gl_thunks.ll regenerate byte-identically, ludic-dev
test 135/135, selfhost-test 32/32, and headless gl_triangle on an RTX 3070 Ti
matches the macOS frame to within one level per channel.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
--target <triple> (default: the host, from OS=Windows_NT) selects the Windows
runtime. emit_win.ludic defines the POSIX names the backend already calls over
the UCRT and Win32 in IR, so rename replaces an existing file, ftell is 64-bit,
and fopen is binary. Known folders follow %APPDATA% / %LOCALAPPDATA% / %TEMP%,
and the driver speaks cmd.exe, writes .exe outputs and finds LLVM's clang.
Verified: macOS three-stage fixpoint, ludic-dev test 135/135, and on Windows
the Mac-emitted Windows IR and the Windows-built compiler's IR are identical
through two generations.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
A water body other than the reflecting one drew its whole bounding rectangle,
so the corners outside the lake's carved ellipse showed water over dry ground;
those bodies now discard outside the ellipse. r3d_fog_base (default 0) is the
height the height fog is measured from, so maps sharing one datum at different
heights get the same air.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
terrain_reload/terrain_unload release one map's height field, survey, photograph
and patch bounds and generate another at any TERRAIN_HALF; scatter_clear_all
(with streams), actor_clear_all, col_reset and mesh_free empty the scene;
water_body_add/water_bodies_clear draw several still-water planes with one
reflecting; terrain_sea separates the coast's sea level from the carved lake's,
and grass keeps off both. Fixes CDLOD patch bounds for TERRAIN_HALF != 4096 and
water_init leaking a mesh and program per call. examples/rendering/reload.ludic.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
The degenerate case - a point exactly on a collider's axis, where there is no
direction to push it - set the normal to (1, 0), which is already a unit vector,
and then divided it by the clamped d = 0.001 along with the genuine normals. The
push came out a thousand times too big: dead centre on a 0.5 m trunk moved a body
about 800 m rather than the 0.85 m that clears it.
Off-centre the arithmetic was right, and off-centre is how anything arrives at a
trunk on foot, so nothing in play ever hit it. A teleport, a spawn, or a world
generator dropping something onto an existing collider would have.
(ex, ez) / d is a unit vector for every d > 0, because d is its own length -
there was never anything to clamp and nothing that could grow. The normal is now
built once and explicitly, and the clamp is gone.
Verified through a game, which is the only harness this package has: render3d's
own float helpers need the Gl runtime spliced, so collide.ludic cannot be
compiled standalone for a unit test. Maroon Lake's selftest12 stands a body dead
centre on a trunk and asserts the push never exceeds the two radii added together
- which is the definition of being pushed clear, and so the tightest honest bound
available. It reports 798.88 m before this change and 0.80 m after, with the
off-centre case unchanged at 0.66 m.
`ludic bundle` builds an AppIcon.icns and macOS reads it out of the .app, so a
shipped game has an icon. `ludic build` produces a bare executable: no bundle, no
CFBundleIconFile, and so no icon at all - macOS draws the generic green "exec"
tile. That is the build a developer runs every day, which is why "the game has no
icon" can be true for months while the bundle is perfect. It was here: the .app's
icns validated against iconutil, the plist was right, the signature was right,
and NSWorkspace rendered the artwork - and the binary beside it still had the
exec tile.
App.set_icon(path) takes the bytes through lp_pak_open, so a packed path and a
loose one both work and this does not repeat Audio.load's trick of taking a
filesystem path only. NSData copies them, so the buffer goes straight back. A
missing or undecodable image leaves the existing icon alone rather than clearing
it; a bundled app is unaffected; headless links no AppKit and compiles it away.
Verified the Cocoa sequence against an ObjC twin doing the same message sends:
before, a bare binary's applicationIconImage is the generic 128x128 tile; after,
it is the 1024x1024 artwork.
Backend change, so selfhost/ludicc.seed.ll is reseeded: the bootstrap fixpoint
and the C-free rebuild from the seed both pass.
They were the macOS layout on every platform - the comment above them even said
"macOS/BSD layout" - so a game built on Linux wrote its saves to
~/Library/Application Support, a directory that means nothing there. Naming the
right directory is the entire reason a program calls these instead of joining a
path itself.
macOS save/config ~/Library/Application Support/<app>
cache ~/Library/Caches/<app>
Linux save $XDG_DATA_HOME or ~/.local/share/<app>
config $XDG_CONFIG_HOME or ~/.config/<app>
cache $XDG_CACHE_HOME or ~/.cache/<app>
macOS is unchanged to the byte, deliberately. save_dir and config_dir stay the
same directory there: Apple's home for a config file that is not an
NSUserDefaults plist is Application Support too, and a shipped game's settings
must not move out from under it. On Linux XDG separates the two and so does this.
An XDG variable that is set but EMPTY falls back to the default, which is what
the spec says and what an exported-but-unset variable looks like from a shell.
uname is read once and remembered rather than per call, because save_dir is
called on every save.
Verified on both branches. macOS by running it; the Linux branch by building the
probe's IR with the cached platform flag pinned, which exercises the emitted XDG
code exactly - unset, set, and set-but-empty all resolve as the spec says. The
suite's own case asserts the HOST's convention, so a macOS box covers the Apple
branch and CI covers XDG.
This is a backend change, so selfhost/ludicc.seed.ll is reseeded with it: the
bootstrap fixpoint (gen2 == gen3) and the C-free rebuild from the seed both pass.
A pack root is packed wholesale, and that is the right default - a game writes
`pack "assets"` and everything it opens is in the pack. What also goes in is
everything the game does NOT open: the preview renders a model pipeline leaves
beside its meshes, the intermediate a texture bake writes and never reads again,
the .blend the .gltf came out of. Nothing errors, nothing looks wrong, and the
app is simply bigger than the game. The only way out the manifest offered was
naming every file by hand, which is worse - a list that goes stale the day
someone adds a texture.
So `.packignore`, with gitignore's rules, because that is the file everyone
already knows. Anchored and floating patterns, `preview/` for directories only,
`*` and `?` stopping at a separator where `**` crosses one, `[a-z]` classes, `!`
re-includes with the last line winning, a deeper file beating a shallower one,
and no re-including out of an ignored directory.
The semantics are not claimed, they are checked: the implementation was diffed
against git itself over two fixtures - 35 paths, 19 patterns, nested ignore
files, directory negation, `[!0-9]` and `\#` escaping - and `git check-ignore`
and `ludic pack` agree on every path.
Two rules of its own, because a pack is not a working tree. `.packignore` is
never packed (nothing reads one at run time, and --no-ignore does not bring it
back). And it governs the project's own roots only: a package's resources - the
renderer's shaders above all - are added after the gather, so a stray `*.frag`
in a game's ignore file cannot quietly un-ship what it needs to draw anything.
`ludic pack` reports what it left out; `--no-ignore` packs everything so you can
see what a rule is costing. `ludic bundle` gathers through the same path, so the
two agree by construction.
A frame is drawn by more than one pass - a shadow map, a water reflection, the
scene - and actor_draw runs in each. An Actor's rim survives that because it is
a field on the actor; the queue did not, because the first pass to run emptied
it. A queued outline was drawn into whichever target came first and was gone by
the time the scene was drawn, so nothing appeared with every uniform, matrix
and mesh correct - which took a while to find.
A flush now closes the batch rather than clearing it, and the next
outline_model opens a new one.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Three things a game could not say, each of which had been worked around.
Audio.play_at(id, gain:, pitch:, pan:) fires a one-shot with its own gain,
pitch and stereo position. Audio.volume and Audio.pitch are global - they are
the options screen - so a game placing a sound in the world was fighting them,
and distance attenuation was simply not expressible. The backend already took
volume and rate per call; this adds setPan: alongside them and stops routing
through the master state.
ludic.render3d gains outline_model(model, mat, width, r, g, b): a rim around
something that is not an Actor. The outline pass walked the actor list and
stopped, so instanced scatter - a forest - could not be highlighted at all. It
is a queue flushed by the same pass, which is what gets the depth test right
when the caller does not control pass order.
And terrain_coast(cx, cz, margin, fall), the other way to make an island:
the sea around the survey's own edge rather than cut out of the middle of it.
terrain_island measures a radius from a centre, which drowns two thirds of a
real survey to make an island of the rest; this measures inward from the
boundary, so everything the data covers stays land and the coast is where the
data runs out. Both modes gained a strand - the last few metres of height
either side of the water line compressed, which stretches a cliff plunge out
into beach and shallows.
132 regression tests and the self-host fixpoints pass.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Maroon Lake keeps assets/polyhaven as a symlink into a shared checkout - which
is an ordinary thing to do, and what `ludic assets` encourages - and `find`
does not follow symlinks. The pack was written, reported success, and silently
omitted all 71 files behind the link, including the sky HDRI.
What that looked like from the outside is worth recording, because it is the
failure mode this whole feature has to avoid: the bundled game started, printed
one line about an HDRI it could not read, carried on, and then died in
terrain_height reading offset 0x1cd681c off a null pointer - the CPU height
field, never allocated, because r3d_init had given up several steps earlier. A
missing asset surfaced as a segfault a long way from the cause.
So: `find -L`, and a warning when a declared root contributes no files at all.
A root that packs nothing is nearly always a typo or a link into a tree that was
never fetched, and the warning costs one line where the alternative costs an
afternoon.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
`ludic build` produces a program. Double-clicked it opens a Terminal window, it
wears the generic executable icon, it calls itself whatever the file is called,
and it carries none of its assets. `ludic bundle` produces an application.
Everything it needs is in package.ludic, so the command takes no arguments: an
Info.plist and PkgInfo from `app` lines, an .icns built by sips and iconutil at
all ten sizes macOS asks for from a single source PNG, the asset pack in
Contents/Resources, and an ad-hoc signature - which is not optional on Apple
silicon, where an unsigned binary is killed rather than warned about. The bundle
identifier falls back to the package path reversed, so a project that never
thinks about it still gets a defensible one instead of two apps sharing a key
Launch Services hangs the Dock, saved state and permissions off.
A bundled game is moved to ~/Library/Application Support/<name> before main,
because Finder starts a .app with its working directory at "/" where no save
could ever be written. Reads come out of the pack, writes land somewhere real
and per-user, and the game's save code needs no change and no platform
knowledge.
The splash is the other half of looking like an application. A game that loads
165 MB spends a visible moment doing it with nothing on screen, which from the
outside is indistinguishable from a launch that failed. splash_show puts a
borderless window up from the same constructor that mounts the pack - before
main, so it appears while the process is still starting rather than after the
slow part it exists to cover - and reads the artwork out of the pack like any
other asset. It turns the run loop enough times to be mapped and composited
there and then; once composited the backing store survives a busy main thread,
so it stays up for the whole load.
Nothing hides it automatically. Only the game knows when its first real frame is
ready, and a splash that vanishes before that leaves the same black gap it was
covering, so the game calls App.splash_hide(). Headless there is no splash and
the call lowers to nothing.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
A Ludic game opened its assets by a path relative to the working directory, so
`build/mygame` ran only from the project root and there was nothing to give
anyone but "the binary, and also this whole tree". A game is not one file, but
shipping it has to be.
`ludic pack` writes a .lpak: a header, a name-sorted entry table, a name heap
and the blobs, 16-byte aligned. Entries are stored rather than compressed - PNG,
JPEG and glTF binary arrive compressed already, and a decompressor on the load
path would spend CPU to make the file no smaller.
The reader is spliced into the compiler at the one place every asset in a Ludic
program comes through: file_open. gltf_load, tex_load, Audio.load, Fs.read_text
and the renderer's own shader loads all bottom out there, so routing it through
@lp_pak_open reaches every one of them without any of them knowing. A read of a
packed path becomes an fmemopen over the mapped bytes; everything else is the
fopen it always was. Fs.exists and Fs.size consult the packs too, so a game that
guards a load with Fs.exists keeps finding its assets once they are packed.
The pack is mmap'd rather than read: 165 MB of textures costs one syscall at
boot and pages in only what is touched. @lp_pak_boot runs from
@llvm.global_ctors, before main, so a pack is mounted before the game's first
line - and it reads packs.index, a plain list, so the mount order is explicit
and a later pack shadows an earlier one.
Without a packs.index nothing mounts and every open goes to the filesystem
exactly as before, which is every `ludic run` during development. Packing is a
shipping step and is invisible until you ship.
The compiler reproduces itself byte-exactly and the C-free bootstrap from the
seed still holds.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
The water shader's `u_screen`, and the reflection target's size, were both taken
from gl_w/gl_h - the drawable. But gl_FragCoord in that shader runs over the
scene target, which is post_w x post_h. They match only at a render scale of 1;
at anything less the refraction and depth reads landed in the wrong corner of
the frame and the lake showed a squashed copy of it instead of its own bed.
Both now come from the scene target. The reflection is sized from it too, which
also stops it paying for pixels the water never samples. R3D_DUMP_REFL reads the
target's own w/h rather than recomputing them.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
`terrain_island(cx, cz, r, fall)` keeps land out to `r` and then scales the
terrain down into the water over `fall` metres and on to a shelf. Scaling rather
than blending to a fixed bed is what makes the coastline come out of the terrain
that is already there: low ground turns into beach and shallows, high ground
into cliff. `r = 0` leaves the survey alone, so nothing that does not ask for an
island is touched.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
The moon was fixed opposite the sun and worth a constant trickle of light. It
now carries a phase, 0 new .. 0.5 full .. 1 new again, and that one number
decides three things at once: the disc's terminator, the fraction of its light
that reaches the ground, and where in the sky it rides.
The moon is where the sun was `lag` of a day ago, so a full moon rises as the
sun sets and a new moon travels with the sun and is never seen. A new moon is
now a properly dark night, which is what makes a carried light worth having.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
An actor gains `outline` (metres of rim) and `ocol` (its colour). The pass draws
the model a second time with its vertices pushed out along their normals and its
front faces culled, so only the far side of the swollen shell survives - a
silhouette exactly `outline` metres wide. It is depth-tested against the scene,
so anything standing in front of the actor hides the rim too.
skin.vert grows an OUTLINE path for the push; outline.frag is the flat-colour
fragment shader it pairs with.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
`Gl.*` binds the whole OpenGL 4.1 core API — every entry point of the
platform gl3.h with every GL_* constant, generated by `ludic-dev glgen`
with per-call ABI thunks. Windowed builds get an NSOpenGLContext on the
existing window at Retina resolution; headless builds render into an
offscreen CGL context, so a program that uses Gl.* renders and
screenshots identically under the test harness. It links gl.ll, the
thunks and OpenGL.framework only when used; every other build stays
byte-identical.
packages/ludic.render3d is a physically based renderer written on that
surface: HDRI image-based lighting, GPU-generated terrain with scanned
PBR materials, CDLOD, cascaded shadows, glTF with skinning, instanced
vegetation with impostors, procedural grass, water, SSAO, and an HDR
pipeline with bloom, auto-exposure and ACES.
It also carries this session's work on it: the terrain at half its cost
(10.3 -> 5.4 ms of frame), the streaming hitch that got worse the longer
you played, a resize that emptied the world, and the packaging that lets
a game use the renderer from its own repository — `ludic assets`, the
material manifest shipping with the package, and shader lookup falling
back to the install root. See changes/ for each, with its numbers.
The camping game that drove all of it has moved out to its own
repository, Maroon Lake; examples/rendering/smooth.ludic stays as the
renderer's example here.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
The install root is not only the artifact: `ludic add` caches fetched packages
in <root>/store, keyed by content hash. install_staged moved the whole root
aside and replaced it, so re-running the installer — which is exactly what
`ludic upgrade` does — deleted the cache and forced every project to refetch.
The store is moved into the staged tree before the swap. Nothing else in the
root is preserved, because everything else is the toolchain and should be
replaced.
Verified by staging an install, planting a store entry, upgrading, and reading
the entry back.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
`ludic new my-game` wrote `program My-Game`, so the first thing anyone did with
a new project — run it — failed with `expected '{', got '-'`. The project name
is a directory name and the identifier is Ludic source, and they do not accept
the same characters: names are now folded into a valid identifier (my-game ->
MyGame, 2048 -> Game2048, a.b.c -> ABC) and a name that cannot be a directory or
a package is refused with the rule instead of being mangled into one.
An audit of every command's flags turned up more of the same shape, all fixed:
- build/run ignored unknown options, so `--headles` silently produced a windowed
binary, and `-o` with no path was silently dropped.
- `ludic fmt` printed the formatted text to stdout while its help said "in
place", so it appeared to do nothing. It writes now, with --check for the
report-only case a hook wants.
- `ludic test nosuch.ludic` deferred the error to the compiler.
- build-lib's failure messages ran `{tmp_dir()}` through the shell literally —
an interpolation written inside a non-interpolating string — and its argument
guess matched package.lock.ludic, then tried to compile the lockfile.
- Several messages still identified the tool as `x`.
`ludic-dev test` now scaffolds under four awkward names, builds and tests each,
and asserts a space-bearing name is refused — the case that shipped broken.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
`ludic help` ended with a section titled "contributing to the toolchain itself",
listing bootstrap, reseed, docs-gen and release tasks. None of that is available
to someone who installed the language — those tasks need the repository — so the
shipped tool was advertising work its user cannot do, in a namespace they have to
read past to find `new` and `run`.
The tasks move to a second program, dev.ludic -> bin/ludic-dev, built from a
checkout and excluded from every release artifact. `ludic` keeps the project and
package commands and nothing else; `ludic dev …` now explains where the tasks
went instead of failing as an unknown command.
What this shook out: the two programs share prelude/build/project/pkg, so the
helpers each had accreted in whichever file first needed them — cc(),
ensure_ludicc, the string functions, title_case, cmd_version — moved to where
both can see them. The argument-shift indirection added for the `dev` namespace
is gone with the namespace, so commands read argv directly again.
`ludic-dev test` asserts the split rather than trusting it: the staged install
must build a project, and `ludic dev build` there must fail while naming
ludic-dev. install.sh keeps building older tags, whose bootstrap goes through
main.ludic.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
docs-gen publishes install.sh with the site, but the workflow only ran for
docs/**, tools/docgen/** and tools/ludic-cli/**. v0.5.2 changed install.sh,
CHANGELOG.md and VERSION — so nothing matched, no deploy ran, and the fix that
release existed for was never served to anyone running the one-liner.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
~/.bashrc was only appended to when it already existed, so on an account without
one — a fresh container, a minimal image — `bash -i`, which is what most Linux
terminal emulators start, did not see the toolchain even though every other
shell did.
.bashrc is now created when missing. Unlike .bash_profile, whose existence stops
bash reading ~/.profile, a .bashrc that only sources the env file changes nothing
else about how bash starts.
Verified across zsh -i, zsh -c, bash -l, bash -i, sh -l and dash -l on a staged
HOME: all six resolve ludic, a second run writes nothing, and .bash_profile is
still never created.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
2026-09-05 22:33:35 +03:00
2198 changed files with 790474 additions and 87956 deletions
low latency, portability for MoltenVK), with every constant and every struct's size
(`<Struct>_sizeof`) and field offsets (`<Struct>_<field>`). Generated from the Vulkan
registry by **`ludic-dev vkgen`** into `runtime/native/vk_api.ludic` and `vk_thunks.ll`;
structs are plain memory filled by name with `Vk.put_i32` / `put_i64` / `put_ptr`. Every
size and offset was checked against the SDK's C headers (3128 facts). A C float is float
bits in an int; 64-bit values and non-dispatchable handles are `long`.
- **The loader is opened at run time**, never linked: `vk_win.ll` (`vulkan-1.dll`) and
`vk_mac.ll` (`libvulkan.1.dylib`, MoltenVK). `Vk.open()` returning 0 means no Vulkan, and
the program carries on. `ludicc` and `ludic build` link both files for any program that
uses `Vk.*`.
- `examples/rendering/vk_probe.ludic` reports what a machine's Vulkan can do;
`vk_compute.ludic` runs a Slang compute shader (`vk_compute.slang`) and reads the picture
back - on an RTX 3070 Ti and on an M4 Pro through MoltenVK, clean under the validation layer.
- **render3d `gpu.ludic`**: the seam between the renderer and a graphics API. Render state
and uniforms go through `gpu_*` / `u_*` and nowhere else (OpenGL frames unchanged);
`R3D_GFX=gl|vk` or `gpu_request` chooses a backend, falling back to OpenGL with a reason;
`gpu_caps_probe()` detects, on Windows, the Vulkan 1.3 floor, ray tracing, mesh shaders,
the NVIDIA RTX generation, Reflex and HDR, for a settings screen to grey out what a machine
cannot use (`R3D_CAPS=rtx50|rtx40|rtx30|amd|intel|none` pretends, for tests).
## v0.14.2 — 2026-09-14
### Fixes
- **`Input.mouse_dx` / `mouse_dy` no longer jump on the first frame or when the cursor mode changes.**
The delta was the position minus the previous frame's, and the previous position started at 0,0,
so the first frame reported the cursor's whole distance from the corner as motion, and switching
between a locked cursor's virtual reticle and the real cursor did the same. A camera that adds
`mouse_dy` to its pitch came up pointing at the ground with nobody touching the mouse. Both frames
now report no motion; every other frame is unchanged, windowed or headless, on macOS and Windows.
## v0.14.1 — 2026-09-13
### Fixes
- **A rim from `outline_model` no longer stays on after the caller stops asking** — the queue
was only emptied by the next `outline_model` call, so a game that highlights what the
crosshair is on left the last highlighted tree outlined for good once the player looked
away. `r3d_frame` now drops a batch nobody reopened this frame (`outline_frame`).
## v0.14.0 — 2026-09-13
### Features
- **Overlay text is UTF-8, and a font atlas can carry any script** — `ludic.render3d`'s `ov_text`,
`ov_text_w` and `ov_text_wrap` decode UTF-8 instead of drawing bytes 32–126, so a game can show
Turkish, German, Polish, Greek, Cyrillic or whatever its atlas holds.
- `font.json` may list the atlas's code points in `"codes"` (atlas order). An atlas without it is
read as before: consecutive code points from `"first"`, so existing ASCII fonts keep working.
- A code point the atlas lacks draws as `?`; a control character as a space; a malformed or
cut-off sequence as one `?`, so a string sliced mid-character still draws.
- `overlay_font(dir)` loads or swaps the atlas at run time, for a language that brings its own font.
## v0.13.3 — 2026-09-13
### Performance
- **Dense forest renders faster** — a depth prepass for the near tree foliage in `ludic.render3d`.
A crown of needle cards is many cut-out quads deep, and a shader that can `discard` turns
early depth rejection off, so every card behind the front one ran the full lighting shader.
In a dense stand at 3840x2160 that made the vegetation pass the largest in the frame. The near
tree LODs now write depth first with `depth.frag` (the same alpha coverage test), the terrain
beneath them is rejected before shading, and the lit pass draws them with no `discard` and an equal depth test. `R3D_NOPREPASS=1` restores the old path for comparison.
## v0.13.2 — 2026-09-13
### Fixes
- **`ludic build` links `http.ll` only into a program that uses `Http.*`.**
The check that added it grepped the IR for `@hs_`, which every program's header declares,
so the HTTP transport and Foundation were linked into everything. That cost nothing visible
on macOS and failed the link on Linux, where `ludic build`, `ludic run` and a new project all
broke in CI. It now looks for a call to `hs_send`.
## v0.13.1 — 2026-09-13
### Fixes
- **The toolchain links on Linux again** — CI builds, tests and publishes releases.
The asset-pack boot finds the directory beside the executable with `_NSGetExecutablePath`,
which Darwin's libc has and glibc does not, so every Linux link of the compiler seed failed
from the release that added packs onward: CI's build, suite and C-free bootstrap jobs, and
every `publish` run, so no release after v0.7.0 carried a Linux toolchain or was created by
CI at all. `tools/ci/linux_stdio_shim.ll` now supplies it, over `/proc/self/exe`.
## v0.13.0 — 2026-09-13
### Features
- **Windows target** — `ludicc` builds and runs headless programs on Windows, and builds itself there.
- **`--target <triple>`** — a triple naming `windows` selects the Windows runtime; without
the flag the target is the host, read at run time from `OS=Windows_NT`. The IR still
carries no triple, so clang assembles it for the machine it runs on.
- **The libc surface, in IR** — on Windows the header emits `emit_win.ludic`, which defines
the POSIX names the backend calls (`fopen`, `ftell`, `rename`, `opendir`, `mmap`,
`fmemopen`, `uname`, …) over the UCRT and Win32. Three of them fixed silent breakage
rather than link errors: `rename` onto an existing file (every `Fs.write_text` after the
first), a 32-bit `ftell`, and text-mode `fopen` rewriting `\n` as `\r\n`.
- **Known folders** — `Os.save_dir` / `config_dir` are `%APPDATA%\<app>`, `cache_dir` is
`%LOCALAPPDATA%\<app>`, `temp_dir` is `%TEMP%`, all with forward slashes; a bundle's
`home` line points at `%APPDATA%\<name>`.
- **The driver** — finds `C:\Program Files\LLVM\bin\clang.exe` when clang is not on
`%PATH%`, writes `.exe` outputs, speaks cmd.exe for its directory and cleanup commands,
and reads `argv[0]`, `%PATH%` and `$LUDIC_HOME` with either separator.
- **A window** — `win32.ll` is `cocoa.ll`'s contract over user32: the held-key set and frame
key in Ludic codes, the mouse in framebuffer pixels with raw input for cursor mode 2,
cursor hide/lock/confine that lets go when the window loses the foreground, XInput pads
in SDL order with rescaled deadzones, and the software `win_present`. `win32_gl.ll` puts
the WGL context on that window with vsync and borderless full screen; a headless build
links `gl_win_nowin.ll` instead. The process is per-monitor DPI aware.
- **Sound** — `audio_win.ll` is `audio.ll`'s contract over XAudio2: one source voice per
clip, loops, volume, rate and a balance pan through the output matrix, RIFF WAVE in PCM or
float. It reads a clip through the asset pack, so a bundled game's first `Audio.load`
succeeds - AVAudioPlayer takes a filesystem path, which is why macOS cannot.
- **The CLI on Windows** — `ludic build`, `ludic run` and `ludic-dev build` work from a Windows
checkout. Every shell command the CLI issues runs through Git for Windows' bash
(`$LUDIC_BASH` names another), `compile_app` links through `ludicc -o`, and a checkout
bootstraps from the new `selfhost/ludicc.win.seed.ll`, which `ludic-dev reseed` now
writes beside the macOS seed.
- **`ludic bundle` on Windows** — `build/<name>/` holding a GUI-subsystem `<name>.exe`, the
`game.lpak` and `packs.index`; an `.ico` beside `app icon` is compiled with `llvm-rc` and
linked in. `ludicc` gains `--gui` (no console behind the window) and `--link <file>`.
- **`Http.*` on Windows** — `http_win.ll` is `http.ll`'s contract over WinHTTP: a request
built on the game thread, the exchange on a worker thread, TLS with the system's
certificate checks, and response headers answered from the kept request handle.
- **The splash and the window icon** — decoded by WIC from the bytes in the pack. The splash
is a topmost borderless window at the artwork's own size in points times the display's
scale; `App.set_icon` sets the title bar and taskbar icon of a build with no icon resource.
- **`Gl.*` on Windows** — `gl_win.ll` makes a WGL 4.1 core context on a hidden window's DC,
and `gl_thunks_win.ll` (generated by `ludic-dev glgen` beside `gl_thunks.ll`) calls
every entry point through a table `@lgl_win_load` fills from `wglGetProcAddress`, falling
back to `opengl32.dll` for the 1.1 functions it will not return. A headless GL program
renders on the GPU there: `gl_triangle` matches the macOS frame to within one level per
channel.
### Fixes
- **Backspace fires on macOS** — the Delete key reports `Key.Backspace` (8).
AppKit gives the key marked delete (kVK_Delete, 51) the character `NSDeleteCharacter`, 127,
which is what both the held-key set and the per-frame key received, so `Key.Backspace` never
matched. `cocoa.ll` maps key code 51 to 8 in both.
- **Held arrow keys register** — `Key.Up`, `Key.Down`, `Key.Left` and `Key.Right` are 128-131.
They folded to the codes of w, s, a and d, which is what the single per-frame key
(`Input.key`) reports for an arrow. The held-key set has always stored an arrow under
128-131, so `Input.key_down(Key.Up)` tested the W bit and `Input.move_i`'s arrow half never
moved anything. `Input.key` keeps its WASD alias: a game comparing it with `'w'` still
takes the arrows.
- **`ludic build` links `Http.*`** — a program that uses the HTTP client builds through the CLI.
`compile_app` linked the OpenGL backend when a program named `@lgl_*` but never
`runtime/native/http.ll` and Foundation for `@hs_*`, so a `Http.*` program failed to link
under `ludic build` / `ludic run` while `ludicc -o` built it. It now links them the same way,
windowed and headless.
## v0.12.1 — 2026-09-13
### Features
- **ludic.render3d: `r3d_fog_base`** — the height the height fog's density is measured from (default 0, the world's y = 0). Two maps sharing one height datum, one far lower than the other, no longer give the lower one many times thicker air.
### Fixes
- **ludic.render3d: lakes are ellipses** — a water body other than the reflecting one was drawn as the whole rectangle of its bounds, so the corners between that rectangle and the lake's carved ellipse showed sheets of water over dry ground. Those bodies are now clipped to the ellipse; the reflecting body (the sea) still fills its rectangle.
## v0.12.0 — 2026-09-13
### Features
- **ludic.render3d: replace the world at run time** — a game can swap its terrain, scatter, colliders, actors and water for another map's without restarting.
- `terrain_reload(dem, emin, emax, base, ox, oz, ortho, half)` releases the current map's height field, survey, photograph and patch bounds and generates another at any `TERRAIN_HALF`; `terrain_unload()` is the release on its own.
- `scatter_clear_all()` (with `stream_clear_all()`), `actor_clear_all()` and `col_reset()` empty the scene for the next map; `mesh_free()` releases a mesh's GL objects.
- **Water bodies** — `water_body_add(level, cx, cz, ex, ez, reflect)` and `water_bodies_clear()` draw several still-water planes at their own levels; the first that reflects gets the planar reflection. `water_init` still means one reflecting plane.
- **Fix:** the terrain's patch quadtree placed patches at a fixed 32 m leaf, so any `TERRAIN_HALF` other than 4096 put patch bounds in the wrong place; leaves now scale with the map.
- **Fix:**`water_init` built a new mesh and compiled a new program on every call.
- `terrain_sea(level)` separates the sea's level (the coast, the strand, the shoreline, forest and scree shading) from the carved lake's, so a lake can sit above the sea. Unset, the sea is the lake's level as before.
## v0.11.1 — 2026-09-12
### Fixes
- **`col_resolve` no longer throws a body across the map when it starts dead centre on a
collider.** The degenerate branch — a point exactly on a circle's axis, where there is
no direction to push it — chose a unit normal `(1, 0)` and then divided it by the
clamped `d = 0.001` anyway, along with the real normals. The push came out a thousand
times too large: a body standing exactly on a 0.5 m trunk was moved about 800 m instead
of the 0.85 m that clears it.
Off-centre the arithmetic was correct, and off-centre is how anything arrives at a
trunk while walking, so it never showed up in play — only a teleport, a spawn or a
world generator placing something on an existing collider could land on the axis.
`(ex, ez) / d` is always a unit vector for `d > 0`, because `d` is its own length: there
was never anything to clamp. The normal is now built once, explicitly, and the clamp is
gone.
## v0.11.0 — 2026-09-12
### Features
- **`App.set_icon(path)`** — an icon for a binary that has no bundle to take one from.
`ludic bundle` builds an `AppIcon.icns` and macOS reads it from the `.app`, so a
shipped game has an icon. `ludic build` produces a bare executable, which has no
bundle, no `CFBundleIconFile` and therefore no icon at all — macOS draws the generic
green "exec" tile in the Dock. That is the build a developer runs all day and the one
they see every session, so "my game has no icon" is true long before it ships.
```ludic
App.set_icon("assets/app/icon.png")
```
The path resolves through the pack first and the filesystem second, like every other
asset, so the same call works packed and unpacked — it does not repeat `Audio.load`'s
trick of taking a filesystem path only. An image that cannot be found or decoded
leaves the current icon alone rather than clearing it, calling it in a bundled app is
a harmless no-op, and a headless build compiles it away to nothing.
## v0.10.1 — 2026-09-11
### Fixes
- **`Os.save_dir` / `Os.config_dir` / `Os.cache_dir` follow the platform.** They were the
macOS layout everywhere, so a game built on Linux wrote its saves to
`~/Library/Application Support` — a directory that means nothing there. Naming the
right directory is the entire reason a program calls these instead of building a path.
| | macOS | Linux |
| --- | --- | --- |
| `save_dir` | `~/Library/Application Support/<app>` | `$XDG_DATA_HOME` or `~/.local/share/<app>` |
| `config_dir` | `~/Library/Application Support/<app>` | `$XDG_CONFIG_HOME` or `~/.config/<app>` |
| `cache_dir` | `~/Library/Caches/<app>` | `$XDG_CACHE_HOME` or `~/.cache/<app>` |
An XDG variable that is set but empty falls back to the default, as the spec requires.
macOS is unchanged to the byte — `save_dir` and `config_dir` stay the same directory
there, because Apple's home for a config file that is not an `NSUserDefaults` plist is
Application Support too, and a shipped game's settings must not move out from under it.
On Linux XDG separates the two and so does this.
One consequence worth stating plainly: a game already shipped on Linux was writing to
the old macOS-shaped path, and this moves it. Those files are not migrated for you —
they are wherever `~/Library/Application Support/<app>` ended up on that machine, and a
game that has Linux players should move them once on startup.
## v0.10.0 — 2026-09-11
### Features
- **`.packignore`** — keep build artefacts out of the shipped asset pack.
A pack root is packed wholesale, so everything a model or texture pipeline leaves
beside its output ships too: preview renders, bake intermediates, the `.blend` a
`.gltf` came from. Nothing errors and nothing looks wrong — the app is just bigger
than the game. The only way out was naming every file in `package.ludic`, a list
that goes stale the day someone adds a texture.
Put a `.packignore` beside the assets instead. The rules are **gitignore's**, down
to the parts people rely on without thinking about them: patterns anchored by a
slash or floating without one, `preview/` for directories only, `*` and `?` stopping
at a separator where `**` crosses it, `[a-z]` classes, `!` re-includes with the last
line winning, a deeper file beating a shallower one, and no re-including out of an
ignored directory.
`ludic pack` reports what it left out, and `--no-ignore` packs everything so you can
see what a rule costs. `ludic bundle` gathers the same way. `.packignore` itself is
never packed, and the file only governs your own roots — a package's resources, such
as the renderer's shaders, are added afterwards and cannot be excluded by accident.
## v0.9.1 — 2026-09-11
### Fixes
- **`outline_model`'s batch survives the whole frame.** A frame is drawn by more than one
pass — a shadow map, a water reflection, the scene — and `actor_draw` runs in each of
them. An Actor's rim survives that because it is a field on the actor; the queue did not,
because the first pass to run emptied it, so a queued outline was drawn into whichever
target happened to come first and was gone by the time the scene was drawn. Nothing
appeared, with every uniform, matrix and mesh correct.
A flush now *closes* the batch rather than clearing it, and the next `outline_model`
opens a new one: every pass in a frame sees the same requests, and a caller still needs
no frame hook.
## v0.9.0 — 2026-09-11
### Features
- **`Audio.play_at(id, gain:, pitch:, pan:)`** — fire a one-shot with its own gain, pitch
and stereo position, leaving the master settings alone.
`Audio.volume` and `Audio.pitch` are global: they are there so a player can turn the game
down, and a game that used them to place a sound in the world would be fighting its own
options screen. This is the per-voice version, and it is what distance attenuation is made
of — a 3D game works out how far away a sound is and which side it is on, and says so.
`gain` rides on top of the master volume, so the options screen still wins; `pan` runs
-1 (hard left) to 1 (hard right).
A loaded sound is still one player, so firing the same handle again restarts it rather
than layering a second copy. Load a handle per variant when several need to overlap.
- **`ludic.render3d`: `outline_model(model, mat, width, r, g, b)`** — a rim around something
that is not an Actor.
An Actor has had an `outline` since the outline pass landed, and everything else in a
scene had nothing: the pass walked the actor list and stopped. Instanced scatter was the
gap that mattered, because a game that highlights whatever the crosshair is on could
highlight every object in the world *except* the twenty thousand most common ones — the
trees.
`outline_model` queues a model at a transform from anywhere in the frame and the outline
pass flushes it alongside the actors', which is what gets the depth test right: the rim has
to be drawn after the scene it is tested against, and a caller does not control pass order.
A layer whose vertex shader moves its instances — wind, or standing them on the drawn
terrain — should be handed the transform that shader arrives at.
- **`ludic.render3d`: `terrain_coast(cx, cz, margin, fall)`** — the other way to make an
island, and the one a real survey wants: the sea goes around the survey's **own edge**
rather than being cut out of the middle of it.
`terrain_island` measures a radius out from a centre, which suits a made-up map and
drowns most of a real one — an 8 km mountain survey loses two thirds of itself to make an
island of the rest. `terrain_coast` measures inward from the boundary instead: everything
the data covers stays land, the outer `margin` metres go under water, and the `fall` metres
inside that are scaled down into it. The band is wobbled by low-frequency noise, so what
comes out is headlands and bays rather than the square the data arrived in.
Both modes also gained a **strand**. Scaling alone hands a 500 m mountainside a 40-degree
plunge into the sea, which is a cliff coast and nothing else; the last few metres of height
either side of the water line are now compressed, which stretches them out horizontally
into beach and shallows. Inland lakes are untouched — they have their own bed.
## v0.8.0 — 2026-09-10
### Features
- **A boot splash**, customised by the game. `app splash` in `package.ludic` names
the artwork and `app splash_bg` the colour behind it; the runtime raises a
borderless window from a constructor that runs before `main`, reading the image
out of the asset pack, so it is on screen while the process is still starting
rather than after the slow part it exists to cover. Nothing hides it
automatically - only the game knows when its first real frame is ready, so the
game calls `App.splash_hide()`.
- **Asset packs.**`ludic pack` writes every asset a game opens into one `.lpak`
beside the binary, and the runtime mounts it before `main`. Nothing about how a
game is written changes: the pack is spliced in at `file_open`, the one place
every asset comes through, so `gltf_load`, `tex_load`, `Audio.load`,
`Fs.read_text` and the renderer's own shader loads all find it without knowing
it exists. `Fs.exists` and `Fs.size` consult the packs too. Entries are stored
rather than compressed and the pack is `mmap`'d, so 165 MB of terrain costs one
syscall at startup and pages in only what is touched. Several packs can be
mounted in order, and a later one shadows an earlier one - which is how a patch
replaces individual files without rewriting the base pack. See `docs/SHIPPING.md`.
- **`ludic bundle`** turns a built game into a real macOS `.app`: `Info.plist` and
`PkgInfo` from the manifest, an `.icns` built from one source PNG at every size
macOS asks for, the asset pack in `Contents/Resources`, and an ad-hoc signature
so it launches on Apple silicon. Everything comes from `app` lines in
`package.ludic`, so the command takes no arguments. A bundled game is also moved
to `~/Library/Application Support/<name>` at startup, because Finder starts a
`.app` with its working directory at `/` where no save could ever be written.
- **`ludic.render3d`: `terrain_island(cx, cz, r, fall)`** keeps land out to `r` and then
scales the terrain down into the water over `fall` metres. Scaling rather than blending
to a fixed bed is what makes the coastline come out of the terrain already there: low
ground becomes beach and shallows, high ground becomes cliff. `r = 0` leaves the survey
untouched.
- **`ludic.render3d`: a rim outline on an actor.** An actor gains `outline` (metres of
rim) and `ocol` (its colour). The pass draws the model again with its vertices pushed
along their normals and its front faces culled, so only the far side of the swollen
shell survives - a silhouette exactly `outline` wide, depth-tested against the scene, so
anything standing in front of the actor hides its rim too.
- **`ludic.render3d`: the moon has a phase.** One number, 0 new .. 0.5 full .. 1 new
again, decides the disc's terminator, how much of its light reaches the ground, and
where in the sky it rides - a full moon rises as the sun sets, a new moon travels with
the sun and is never seen. A new moon is now a properly dark night, which is what makes
a carried light worth having.
### Fixes
- **`ludic.render3d`: water read the window's size, not the frame it drew into.** The
water shader's `u_screen` and the reflection target were sized from the drawable, but
`gl_FragCoord` there runs over the scene target. They match only at a render scale of 1;
below that the refraction and depth reads landed in the wrong corner of the frame and the
lake showed a squashed copy of it instead of its own bed.
## v0.7.0 — 2026-09-10
### Features
- **A game can use the renderer from outside this repository.**`ludic.render3d` reads two
things from disk at run time — its GLSL, and the scanned CC0 materials — and both were
found only by a path relative to the working directory, so the renderer worked in a
Ludic checkout and nowhere else. A game living in its own repository now needs to copy
neither.
**The shaders come from the package**, wherever the package is. They belong to
`ludic.render3d` and ship with it, so the renderer looks for them beside the project
first (a Ludic checkout, where they are under `packages/`) and then under the install
root, `$LUDIC_HOME/packages/ludic.render3d` — the same place the compiler already
resolves `import "ludic.render3d/r3d.ludic"` from. Nothing to vendor, and no version of
the shaders that can drift from the version of the code that compiles them.
**`ludic assets [--force]`** fetches the scanned materials and the HDRI sky into
`assets/polyhaven/` of whatever project you run it in. The list of what to fetch is the
renderer's own — the renderer decides which materials it wants — so it moved out of the
repository's `assets/` and into the package as
`packages/ludic.render3d/assets.manifest`, where it ships with the toolchain. A game
does not keep its own copy of that list and so cannot fall out of step with the
renderer's material set. `ludic dev fetch-assets` is the same command from a checkout.
**A URL-shaped module built its binary into directories.**`project_name` took
everything after the last dot of the manifest's module path, which for a package
identified the way the package manager identifies them —
`git.host.io/user/name` — is inside the *host*: the build wrote
`build/io/user/name` instead of `build/name`. The last path segment comes first now, and
a dot inside that segment still separates namespace from package, so `ludic.render3d`
still builds as `render3d`.
**The camping game has moved out** to its own repository —
| the IR, to read | `ludicc src.ludic --emit-llvm -o src.ll` |
| the IR, to read | `ludicc src.ludic --emit-llvm -o src.ll` |
| the schema an editor reads (records, registries and their entries, consts) | `ludicc src.ludic --emit-schema schema.json` |
| every error, as a JSON array on stdout | `ludicc src.ludic --check --diagnostics=json` |
| the same, with an unsaved buffer on stdin standing for one of its files | `ludicc src.ludic --check --diagnostics=json --stdin-file lib/a.ludic < buf` |
bin/ludicdev test-tools # the editor-toolchain suite (ludic-fmt, ludic-lsp)
bin/ludic-dev test-tools # the editor-toolchain suite (ludic-fmt, ludic-lsp)
bin/ludic clean # remove build/, out.ppm and stray artifacts
bin/ludic clean # remove build/, out.ppm and stray artifacts
```
```
@ -67,7 +68,7 @@ The stdlib lives in the runtime (`runtime/`) and is surfaced as namespaces
and register its id in `tools/docgen/inventory.json`. Each documented
and register its id in `tools/docgen/inventory.json`. Each documented
namespace gets exactly **one** directory (the docs check enforces this).
namespace gets exactly **one** directory (the docs check enforces this).
3. Add or extend an example under `examples/` and a case in the test suite.
3. Add or extend an example under `examples/` and a case in the test suite.
4. Run `bin/ludic dev docs-gen --out build/pages && bin/ludic dev docs-check build/pages` — the
4. Run `bin/ludic-dev docs-gen --out build/pages && bin/ludic-dev docs-check build/pages` — the
check fails if any inventory symbol lacks a page or is still seed text.
check fails if any inventory symbol lacks a page or is still seed text.
5. Add a **changeset** for the user-facing change: a small file under
5. Add a **changeset** for the user-facing change: a small file under
[`changes/`](changes/README.md) with a `bump:` level and a one-line summary.
[`changes/`](changes/README.md) with a `bump:` level and a one-line summary.
@ -82,22 +83,22 @@ Releases are changeset-driven. Every user-facing change ships with a changeset
(step 5 above). Read the next release before cutting it:
(step 5 above). Read the next release before cutting it:
```bash
```bash
ludicdev release --dry-run # render the CHANGELOG section, write nothing
ludic-dev release --dry-run # render the CHANGELOG section, write nothing
```
```
Then cut it:
Then cut it:
```bash
```bash
ludicdev release [major|minor|patch] # omit the level to derive it from the changesets
ludic-dev release [major|minor|patch] # omit the level to derive it from the changesets
git push origin main --follow-tags
git push origin main --follow-tags
```
```
`ludicdev release` aggregates the pending changesets into a new `CHANGELOG.md` section
`ludic-dev release` aggregates the pending changesets into a new `CHANGELOG.md` section
— grouped by change type, with each changeset's markdown kept intact — bumps
— grouped by change type, with each changeset's markdown kept intact — bumps
`VERSION`, commits `chore(release): vX.Y.Z`, and tags it.
`VERSION`, commits `chore(release): vX.Y.Z`, and tags it.
**Pushing the tag is what publishes.** The `release` workflow builds the
**Pushing the tag is what publishes.** The `release` workflow builds the
toolchain from the IR seed, runs `ludic dev test`, `ludic dev test-tools` and `ludic dev bootstrap-cfree`
toolchain from the IR seed, runs `ludic-dev test`, `ludic-dev test-tools` and `ludic-dev bootstrap-cfree`
against the tagged tree, and only then creates the Forgejo release — with the
against the tagged tree, and only then creates the Forgejo release — with the
source tarball, a Linux toolchain build, a `.sha256` beside each, and that version's
source tarball, a Linux toolchain build, a `.sha256` beside each, and that version's
`CHANGELOG.md` section as the notes. It refuses to publish if the tag and
`CHANGELOG.md` section as the notes. It refuses to publish if the tag and
@ -113,10 +114,10 @@ macOS artifacts cannot be produced on the Linux runner — a `darwin-arm64` buil
needs a macOS host, and there is no cross-compile path (it would need the Xcode
needs a macOS host, and there is no cross-compile path (it would need the Xcode
SDK and a Mach-O linker). Attaching one therefore means either registering a
SDK and a Mach-O linker). Attaching one therefore means either registering a
macOS runner and giving it a job, or running the same command CI runs from a
macOS runner and giving it a job, or running the same command CI runs from a
Mac. Either way it is `ludicdev publish`, which only adds assets the release is missing:
Mac. Either way it is `ludic-dev publish`, which only adds assets the release is missing:
```bash
```bash
FORGEJO_TOKEN=… ludicdev publish v0.4.0
FORGEJO_TOKEN=… ludic-dev publish v0.4.0
```
```
Checksums are one `.sha256` file per artifact rather than a single `SHA256SUMS`,
Checksums are one `.sha256` file per artifact rather than a single `SHA256SUMS`,
@ -138,14 +139,14 @@ broken link at a time. Everything host-shaped has an environment override, so a
move can be rehearsed before it is committed.
move can be rehearsed before it is committed.
**Hosts and URLs.** The install one-liner is served from the documentation site,
**Hosts and URLs.** The install one-liner is served from the documentation site,
which publishes `install.sh` beside the pages that quote it (`ludicdev docs-gen`
which publishes `install.sh` beside the pages that quote it (`ludic-dev docs-gen`
copies it in; `docs-check` fails without it). Change the host in:
copies it in; `docs-check` fails without it). Change the host in:
| Where | What |
| Where | What |
|---|---|
|---|---|
| `install.sh` | `REPO_API`, `REPO_URL`, `INSTALL_URL` — each `${LUDIC_…:-default}`, so `LUDIC_REPO_URL=… sh install.sh` tests a move without editing anything |
| `install.sh` | `REPO_API`, `REPO_URL`, `INSTALL_URL` — each `${LUDIC_…:-default}`, so `LUDIC_REPO_URL=… sh install.sh` tests a move without editing anything |
| `tools/ludic-cli/project.ludic` | `install_url()` (`$LUDIC_INSTALL_URL`), used by `ludic upgrade` and `ludic doctor` |
| `tools/ludic-cli/project.ludic` | `install_url()` (`$LUDIC_INSTALL_URL`), used by `ludic upgrade` and `ludic doctor` |
| `tools/ludic-cli/forgejo.ludic` | `FORGEJO_API_DEFAULT` (`$LUDIC_FORGEJO_API`), used by `ludicdev publish` |
| `tools/ludic-cli/forgejo.ludic` | `FORGEJO_API_DEFAULT` (`$LUDIC_FORGEJO_API`), used by `ludic-dev publish` |
| `docs/site/site.json` | `repo_url`, the `start.terminal` one-liner, and the doc links in `nav_links` |
| `docs/site/site.json` | `repo_url`, the `start.terminal` one-liner, and the doc links in `nav_links` |
| Prose | `README.md`, `COMPILING.md`, `tools/editors/README.md`, and the two editor plugins' "server not found" messages |
| Prose | `README.md`, `COMPILING.md`, `tools/editors/README.md`, and the two editor plugins' "server not found" messages |
@ -155,9 +156,10 @@ copies it in; `docs-check` fails without it). Change the host in:
`is_ludic_file`), every editor asset (`tools/editors/shared/*.json`,
`is_ludic_file`), every editor asset (`tools/editors/shared/*.json`,
`vscode/package.json`, the JetBrains `LudicFileType`), and every source file
`vscode/package.json`, the JetBrains `LudicFileType`), and every source file
**Bakes you can look at, and three more of the renderer's textures read from one.** `ludic.lab` writes
raw 8-bit pixels (1 to 4 channels) as a PNG (`lab_png_write`, `png_write.ludic`, importable alone with its
own `LabPngState`) and turns float textures into honest previews (`png_convert.ludic`: R32F min..max as
grey, RG16F as red and green x255, HDR RGBA16F as x/(1+x) then sRGB). `ludic.render3d` takes three bakes
the game names (`bake_load.ludic`) and makes each as before when one is missing or stale: an impostor's
atlases as BC7 with their baked mips, through the compressed upload (`impostor_from_baked`,
`impostor_fill_bc7`; a fog opening past its cards reads the bake again instead of painting), the sky's image-based light at the start yaw per prefilter width (`sky_baked_in`; any other
turn of the sky is convolved), and the grass carpet (`carpet_from_baked`, `carpet_bytes`).
`impostor_from_bytes` returns null, keeping nothing, when the bytes are not the impostor's shape.
`@Color`, `@Node(field)`, `@Clip(field)`, `@Material(field)`, `@Tint(SLOT)`, `@Derived`, `@Text`, `@Multiline`, `@Key`, and `@AppendOnly` / `@ByKey` on
a registry - which change nothing but go into the schema; `@Ref` naming no registry is an error, and
so is `@Node` / `@Clip` naming a field that is not a glTF (`@Asset("gltf")`, or an `@Ref` to one), and
a listing `@OneOf` (`@OneOf(A, B)`, not a prefix `@OneOf(P_)`) naming a constant that does not exist, and `@Tint` naming no constant. A field may now carry several attributes. `ludicc --check
--diagnostics=json` (`ludic build --check --diagnostics=json`) prints every error as one JSON array
of `{file, line, col, severity, message}` on stdout; tokens and nodes now know their column.
`Build.schema_hash()` answers FNV-1a 64 of the program's own schema (the bytes `ludic schema` prints),
computed only when a program names it, and 0 under `ludicc --release`, which `ludic bundle` now passes.
A target no part of the program declares (an `@Ref` registry, an `@Tint` or listed `@OneOf` constant) is
a warning and `"unresolved": true` in the schema, so a package can name the game's registry; a name of
another kind is an error. `@OneOf` on a string field takes words, and every registry row's value is
checked against them.
The schema has a `components` list: each UI component's module, place, doc, template and stylesheet
paths, its `props` and `state` (type, default as written, place, doc), `states_read` (the states its
header names, apart from its model), `derived` fields with their types, and the `functions` and
`events` its template calls with their parameters (states and instance stripped), and the
registered native tags its template uses. A `natives` list gives every `ui_native` /
`ui_native_input` call with a literal tag: the tag, the function called, its handler and its place.
**ludic.lab: 16-bit PNGs** — `lab_png_write16_from(st, path, w, h, channels, px, at)` writes 16-bit samples (two bytes each, most significant first) the way `lab_png_write_from` writes 8-bit ones, so a test can write a height map render3d's decoder reads back as R16; `lp_png_bytes(stride, h)` is the size of one.
**Map-scoped tables: `@PerMap` and `@Chunked(n)` registries** — a registry whose rows live in each map's directory and are read when the map loads, or a chunk at a time.
`@PerMap @ByKey registry Props of PropRow from "props.lres"` reads `<maps root>/<map>/props.lres` (the root is `package.ludic`'s new `maps "assets/maps"` line), and `@PerMap @Chunked(64) registry Instances of InstRow from "instances/{cx}_{cz}.lres"` one file per chunk. The compiler writes the table's `state` and its verbs in the declaring module (`props_load`, `props_clear`, `props_find`, `props_path`; `instances_in`, `_out`, `_slot`, `_find`, `_clear`, `_path`) and a typed fill over a small runtime `.lres` reader (`runtime/native/lres.ludic`, spliced on demand): a row names any int or float constant by name (resolved at load), a `fn` value by name, nested records and lists. Rows, their lists and the records in them are pooled and refilled in place, and `_find` is an allocation-free hash, so a load allocates nothing past the table's high water and a frame may bring a chunk in. `@Ref(T)` into a map table is a string key, checked against the same map. `ludicc --check` type-checks every map directory's tables against their records with file:line:col diagnostics (`--no-maps` to skip). The schema gives each registry a `"scope"` (`"map"` / `"game"`) and `"chunk"`, and a map `@Ref``"scope": "map"`; `@Unit` now has canonical ASCII spellings (`deg`, `m/s2`, `N.m`, ...), listed as the schema's `"units"`, and any other spelling is a warning naming the right one.
