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>
357 lines
11 KiB
Text
357 lines
11 KiB
Text
# pack.ludic — `ludic pack`, the writer for the .lpak asset pack.
|
|
#
|
|
# The reader is in the compiler (selfhost/backend/stdlib/emit_pak.ludic), which
|
|
# is also where the format is specified. This is the only thing that produces
|
|
# one, and the two have exactly one contract between them: the entry table is
|
|
# sorted by name, because the runtime binary-searches it.
|
|
#
|
|
# ludic pack pack the manifest's `pack` roots to build/<name>.lpak
|
|
# ludic pack --out x.lpak a b pack directories a and b to x.lpak
|
|
# ludic pack --list x.lpak print what is in a pack
|
|
# ludic pack --verify x.lpak re-hash every entry against the table
|
|
#
|
|
# Paths are stored exactly as the game asks for them — a file at ./assets/kit/x.png
|
|
# is stored as "assets/kit/x.png" — so packing changes nothing about how the game
|
|
# is written. That is the whole point: `gltf_load("assets/kit/hiker", ...)` is the
|
|
# same line of code before and after.
|
|
|
|
# ---- little-endian integers -------------------------------------------------
|
|
#
|
|
# The runtime reads these back with a plain i32 load, so the byte order here is
|
|
# the byte order of the machine that runs the game. Both are little-endian on
|
|
# every target Ludic supports today; a big-endian port would byte-swap in the
|
|
# reader, not here.
|
|
|
|
function put_u32(f: pointer, v: int) -> void {
|
|
let b = bytes(4)
|
|
b[0] = v & 255
|
|
b[1] = (v >> 8) & 255
|
|
b[2] = (v >> 16) & 255
|
|
b[3] = (v >> 24) & 255
|
|
file_write(f, b, 4)
|
|
}
|
|
|
|
# FNV-1a over `n` bytes of `buf` starting at `off`. Not a security hash - it is
|
|
# there so `ludic pack --verify` can tell a truncated or bit-rotted pack from a
|
|
# good one. The arithmetic wraps in 32 bits, which is exactly what FNV specifies
|
|
# mod 2^32; the writer and the verifier compare the same bit pattern, so the
|
|
# sign an int puts on it never matters.
|
|
function fnv1a(buf: pointer, off: int, n: int) -> int {
|
|
var h = -2128831035 # 0x811C9DC5 as a signed 32-bit int
|
|
var i = 0
|
|
while i < n {
|
|
h = h ^ (buf[off + i] & 255)
|
|
h = h * 16777619
|
|
i += 1
|
|
}
|
|
return h
|
|
}
|
|
|
|
# ---- reading one file whole -------------------------------------------------
|
|
#
|
|
# read_file in the prelude NUL-terminates and does not report a length, which is
|
|
# fine for source text and useless for a PNG. This keeps the length.
|
|
|
|
var pk_len: int = 0
|
|
|
|
function read_blob(path: pointer) -> pointer {
|
|
pk_len = 0
|
|
let f = file_open(path, "rb")
|
|
if f == null { return null }
|
|
file_seek(f, 0, 2)
|
|
let n = file_tell(f)
|
|
file_seek(f, 0, 0)
|
|
let buf = bytes(n + 1)
|
|
let got = file_read(f, buf, n)
|
|
file_close(f)
|
|
buf[got] = 0
|
|
pk_len = got
|
|
return buf
|
|
}
|
|
|
|
# ---- gathering the files ----------------------------------------------------
|
|
|
|
# Every file under `root`, in byte order. `find | sort` rather than a walk of
|
|
# Fs.list: the runtime binary-searches on strcmp order, and a recursive walk
|
|
# emits "assets/a.png" after "assets/ab/x.png" because it descends per directory,
|
|
# which is not that order. LC_ALL=C is what makes sort agree with strcmp.
|
|
function pack_gather(root: pointer) -> []pointer {
|
|
var out = new []pointer
|
|
if not file_exists(root) {
|
|
err(`ludic pack: no such directory: {root}\n`)
|
|
return out
|
|
}
|
|
let listing = capture(`find {root} -type f ! -name '.DS_Store' | LC_ALL=C sort`)
|
|
let n = slen(listing)
|
|
var i = 0
|
|
while i < n {
|
|
let line = line_at(listing, i)
|
|
i = i + slen(line) + 1
|
|
let t = s_trim(line)
|
|
if slen(t) > 0 { push(out, t) }
|
|
}
|
|
return out
|
|
}
|
|
|
|
# every root's files, concatenated then re-sorted as one list, since the runtime
|
|
# searches one table across all of them
|
|
function pack_gather_all(roots: []pointer) -> []pointer {
|
|
var all = new []pointer
|
|
var i = 0
|
|
while i < len(roots) {
|
|
let one = pack_gather(roots[i])
|
|
var k = 0
|
|
while k < len(one) { push(all, one[k]); k += 1 }
|
|
i += 1
|
|
}
|
|
return pack_sort(all)
|
|
}
|
|
|
|
# insertion sort by byte order. The lists are hundreds of entries, not millions,
|
|
# and this keeps the one ordering guarantee the format makes in one readable place.
|
|
function pack_sort(xs: []pointer) -> []pointer {
|
|
var i = 1
|
|
while i < len(xs) {
|
|
let v = xs[i]
|
|
var j = i - 1
|
|
while j >= 0 and pack_cmp(xs[j], v) > 0 {
|
|
xs[j + 1] = xs[j]
|
|
j -= 1
|
|
}
|
|
xs[j + 1] = v
|
|
i += 1
|
|
}
|
|
return xs
|
|
}
|
|
|
|
# strcmp order: negative, zero or positive, comparing unsigned bytes. This is the
|
|
# ordering the format promises and the runtime's binary search assumes.
|
|
function pack_cmp(a: pointer, b: pointer) -> int {
|
|
var i = 0
|
|
var r = 0
|
|
var done = false
|
|
while not done {
|
|
let ca = a[i] & 255
|
|
let cb = b[i] & 255
|
|
if ca != cb { r = ca - cb; done = true }
|
|
else if ca == 0 { r = 0; done = true }
|
|
else { i += 1 }
|
|
}
|
|
return r
|
|
}
|
|
|
|
# round `v` up to the next multiple of 16, so every blob starts aligned and a
|
|
# reader can hand a mapped pointer straight to something that wants alignment
|
|
function align16(v: int) -> int {
|
|
let r = v & 15
|
|
if r == 0 { return v }
|
|
return v + (16 - r)
|
|
}
|
|
|
|
# ---- writing ----------------------------------------------------------------
|
|
|
|
function pack_write(out_path: pointer, files: []pointer) -> bool {
|
|
let n = len(files)
|
|
if n == 0 { err("ludic pack: nothing to pack\n"); return false }
|
|
|
|
# Layout is decided before a byte is written: the entry table has to carry
|
|
# absolute offsets, and those are only knowable once every size is known.
|
|
let names_at = 32 + 16 * n
|
|
var name_off = new []int
|
|
var names_len = 0
|
|
var i = 0
|
|
while i < n {
|
|
push(name_off, names_at + names_len)
|
|
names_len = names_len + slen(files[i]) + 1
|
|
i += 1
|
|
}
|
|
let data_at = align16(names_at + names_len)
|
|
|
|
var data_off = new []int
|
|
var data_len = new []int
|
|
var hash = new []int
|
|
var total = data_at
|
|
i = 0
|
|
while i < n {
|
|
let blob = read_blob(files[i])
|
|
if blob == null { err(`ludic pack: cannot read {files[i]}\n`); return false }
|
|
push(data_off, total)
|
|
push(data_len, pk_len)
|
|
push(hash, fnv1a(blob, 0, pk_len))
|
|
total = align16(total + pk_len)
|
|
i += 1
|
|
}
|
|
|
|
run(`mkdir -p "$(dirname {out_path})"`)
|
|
let f = file_open(out_path, "wb")
|
|
if f == null { err(`ludic pack: cannot write {out_path}\n`); return false }
|
|
|
|
# header
|
|
let magic = bytes(5); magic[0] = 'L'; magic[1] = 'P'; magic[2] = 'A'; magic[3] = 'K'; magic[4] = 0
|
|
file_write(f, magic, 4)
|
|
put_u32(f, 1) # version
|
|
put_u32(f, n) # count
|
|
put_u32(f, names_at)
|
|
put_u32(f, data_at)
|
|
put_u32(f, 0); put_u32(f, 0); put_u32(f, 0)
|
|
|
|
# entry table, in the sorted order the runtime's binary search depends on
|
|
i = 0
|
|
while i < n {
|
|
put_u32(f, name_off[i])
|
|
put_u32(f, data_off[i])
|
|
put_u32(f, data_len[i])
|
|
put_u32(f, hash[i])
|
|
i += 1
|
|
}
|
|
|
|
# name heap
|
|
i = 0
|
|
while i < n {
|
|
file_write(f, files[i], slen(files[i]) + 1)
|
|
i += 1
|
|
}
|
|
|
|
# pad up to the first blob, then each blob followed by its alignment padding
|
|
pack_pad(f, data_at - (names_at + names_len))
|
|
var at = data_at
|
|
i = 0
|
|
while i < n {
|
|
let blob = read_blob(files[i])
|
|
if blob == null { err(`ludic pack: {files[i]} vanished mid-pack\n`); file_close(f); return false }
|
|
file_write(f, blob, pk_len)
|
|
at = at + pk_len
|
|
let want = align16(at)
|
|
pack_pad(f, want - at)
|
|
at = want
|
|
i += 1
|
|
}
|
|
file_close(f)
|
|
|
|
print(`packed {string(n)} files, {string(total / 1024)} KiB -> {out_path}`)
|
|
return true
|
|
}
|
|
|
|
function pack_pad(f: pointer, k: int) -> void {
|
|
if k <= 0 { return }
|
|
let z = bytes(k + 1)
|
|
var i = 0
|
|
while i < k { z[i] = 0; i += 1 }
|
|
file_write(f, z, k)
|
|
}
|
|
|
|
# ---- reading a pack back (list / verify) ------------------------------------
|
|
|
|
# the NUL-terminated name at `off`, copied out. Ludic's `+` on a pointer is
|
|
# string concatenation, not address arithmetic, so an offset into a buffer has to
|
|
# be sliced rather than added.
|
|
function pack_name(buf: pointer, off: int) -> pointer {
|
|
var n = 0
|
|
while buf[off + n] != 0 { n += 1 }
|
|
return str_sub(buf, off, off + n)
|
|
}
|
|
|
|
function pack_u32(buf: pointer, at: int) -> int {
|
|
return (buf[at] & 255) | ((buf[at + 1] & 255) << 8) | ((buf[at + 2] & 255) << 16) | ((buf[at + 3] & 255) << 24)
|
|
}
|
|
|
|
function pack_open_read(path: pointer) -> pointer {
|
|
let buf = read_blob(path)
|
|
if buf == null { err(`ludic pack: cannot read {path}\n`); return null }
|
|
if pk_len < 32 or buf[0] != 'L' or buf[1] != 'P' or buf[2] != 'A' or buf[3] != 'K' {
|
|
err(`ludic pack: {path} is not a pack\n`)
|
|
return null
|
|
}
|
|
return buf
|
|
}
|
|
|
|
function cmd_pack_list(path: pointer) -> int {
|
|
let buf = pack_open_read(path)
|
|
if buf == null { return 1 }
|
|
let n = pack_u32(buf, 8)
|
|
print(`{path}: {string(n)} entries`)
|
|
var i = 0
|
|
while i < n {
|
|
let e = 32 + 16 * i
|
|
let no = pack_u32(buf, e)
|
|
let dl = pack_u32(buf, e + 8)
|
|
print(` {string(dl)}\t{pack_name(buf, no)}`)
|
|
i += 1
|
|
}
|
|
return 0
|
|
}
|
|
|
|
function cmd_pack_verify(path: pointer) -> int {
|
|
let buf = pack_open_read(path)
|
|
if buf == null { return 1 }
|
|
let n = pack_u32(buf, 8)
|
|
var bad = 0
|
|
var i = 0
|
|
while i < n {
|
|
let e = 32 + 16 * i
|
|
let no = pack_u32(buf, e)
|
|
let dof = pack_u32(buf, e + 4)
|
|
let dl = pack_u32(buf, e + 8)
|
|
let want = pack_u32(buf, e + 12)
|
|
if dof + dl > pk_len {
|
|
err(` truncated: {pack_name(buf, no)}\n`); bad += 1
|
|
} else if fnv1a(buf, dof, dl) != want {
|
|
err(` corrupt: {pack_name(buf, no)}\n`); bad += 1
|
|
}
|
|
i += 1
|
|
}
|
|
if bad > 0 { err(`ludic pack: {string(bad)} of {string(n)} entries failed\n`); return 1 }
|
|
print(`OK {string(n)} entries verified`)
|
|
return 0
|
|
}
|
|
|
|
# ---- the command ------------------------------------------------------------
|
|
|
|
# The roots to pack: an explicit list of directories on the command line, else
|
|
# the manifest's `pack` lines, else assets/ when it exists. A game that keeps its
|
|
# assets where the convention puts them needs no configuration at all.
|
|
function pack_roots(m: Manifest, from: int) -> []pointer {
|
|
var roots = new []pointer
|
|
var i = from
|
|
while i < arg_count() {
|
|
let a = arg(i)
|
|
if a == "--out" { i += 1 } # its value is the pack, not a root
|
|
else if not s_starts(a, "--") { push(roots, a) }
|
|
i += 1
|
|
}
|
|
if len(roots) > 0 { return roots }
|
|
i = 0
|
|
while i < len(m.packs) { push(roots, m.packs[i]); i += 1 }
|
|
if len(roots) > 0 { return roots }
|
|
if file_exists("assets") { push(roots, "assets") }
|
|
return roots
|
|
}
|
|
|
|
# the pack a project builds by default: build/<name>.lpak
|
|
function pack_default_out(m: Manifest) -> pointer {
|
|
return `build/{project_name("")}.lpak`
|
|
}
|
|
|
|
function cmd_pack() -> int {
|
|
var out_path = ""
|
|
var i = 2
|
|
while i < arg_count() {
|
|
let a = arg(i)
|
|
if a == "--list" { return cmd_pack_list(argn(i + 1, "")) }
|
|
if a == "--verify" { return cmd_pack_verify(argn(i + 1, "")) }
|
|
if a == "--out" { out_path = argn(i + 1, ""); i += 1 }
|
|
i += 1
|
|
}
|
|
let m = read_root_manifest()
|
|
if out_path == "" { out_path = pack_default_out(m) }
|
|
let roots = pack_roots(m, 2)
|
|
if len(roots) == 0 {
|
|
err("ludic pack: nothing to pack.\n")
|
|
err(" put the assets under assets/, or name the roots in package.ludic:\n")
|
|
err(" pack \"assets/kit\"\n")
|
|
return 1
|
|
}
|
|
let files = pack_gather_all(roots)
|
|
if not pack_write(out_path, files) { return 1 }
|
|
return 0
|
|
}
|