From 7d8882d28d48166a9c5b19232775cb6ba9791ae9 Mon Sep 17 00:00:00 2001
From: Orkuncakilkaya <orkuncakilkaya@gmail.com>
Date: Tue, 29 Sep 2026 17:17:20 +0300
Subject: [PATCH] inflate in a caller-owned context, and a PNG decoded in three
 steps (parked: the boot's PNGs move to build time)

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>
---
 packages/ludic.render3d/env.ludic      |   2 +
 packages/ludic.render3d/png_jobs.ludic | 181 +++++++++++++++++
 packages/ludic.render3d/r3d.ludic      |   1 +
 packages/ludic.render3d/texture.ludic  |  79 +-------
 runtime/native/inflate.ludic           | 270 +++++++++++++------------
 5 files changed, 337 insertions(+), 196 deletions(-)
 create mode 100644 packages/ludic.render3d/png_jobs.ludic

diff --git a/packages/ludic.render3d/env.ludic b/packages/ludic.render3d/env.ludic
index 4479bcf9..2a618bef 100644
--- a/packages/ludic.render3d/env.ludic
+++ b/packages/ludic.render3d/env.ludic
@@ -880,6 +880,8 @@ export state Render3dState {
   tex_channels: int = 0
   tex_depth: int = 0        # bits per sample (8 or 16)
   tex_file_len: int = 0
+  tex_pf: []PngJob = null        # what tex_prefetch decoded, for png_decode to take (png_jobs.ludic)
+  tex_zs: []ZInflate = null      # an inflate context per concurrent decode, made on this thread
   tex_anisotropy: fixed = 16.0
   tex_size_ws: []int = new []int
   tex_size_hs: []int = new []int
diff --git a/packages/ludic.render3d/png_jobs.ludic b/packages/ludic.render3d/png_jobs.ludic
new file mode 100644
index 00000000..2eefe0b1
--- /dev/null
+++ b/packages/ludic.render3d/png_jobs.ludic
@@ -0,0 +1,181 @@
+# png_jobs.ludic - a PNG decoded in three steps, so the middle one can run on a worker: png_parse (the
+# program's thread) reads the file, walks its chunks and makes every buffer; png_work inflates, unfilters
+# and packs into them with a context of its own and makes nothing; png_take hands the samples back and
+# tells the renderer their shape. tex_prefetch runs png_work for a list of files on every core, and
+# png_decode takes a prefetched image when one is waiting - so what a load uploads, and in what order,
+# is what it always was.
+property PngJob {
+  path: string = ""
+  z: ZInflate = null
+  idat: pointer = null
+  idlen: int = 0
+  raw: pointer = null
+  plte: pointer = null
+  stride: int = 0
+  rawlen: int = 0
+  w: int = 0
+  h: int = 0
+  bd: int = 0
+  ct: int = 0
+  channels: int = 0
+  fbpp: int = 0
+  oc: int = 0
+  out: pointer = null
+  ok: bool = false
+  taken: bool = false
+}
+property PngBatch { jobs: []PngJob = null }
+
+@alloc_ok("loading a texture or a font: a load, not a frame")
+function png_parse(render3d_st: mut Render3dState, path: pointer) -> PngJob {
+  let d = r3d_read_file(render3d_st, path)
+  if d == null { print(`png: cannot read {path}`);  return null }
+  let size = render3d_st.tex_file_len
+  if size < 8 or d[0] != 137 or d[1] != 80 { free(d);  print(`png: not a png: {path}`);  return null }
+  var w = 0;  var h = 0;  var bd = 0;  var ct = 0
+  let plte = bytes(768)
+  let idat = bytes(size)
+  var idlen = 0
+  var i = 8
+  var done = false
+  while not done {
+    if i + 8 > size { done = true;  continue }
+    let ln = be32(d, i)
+    let typ = i + 4
+    let body = i + 8
+    if ln < 0 or body + ln > size { done = true;  continue }
+    if tag4(d, typ, 73, 72, 68, 82) { w = be32(d, body);  h = be32(d, body + 4);  bd = d[body + 8];  ct = d[body + 9] }
+    if tag4(d, typ, 80, 76, 84, 69) { let m = min(ln, 768);  for k in 0 .. m { plte[k] = d[body + k] } }
+    if tag4(d, typ, 73, 68, 65, 84) { mem_copy(mem_off(idat, idlen), mem_off(d, body), ln);  idlen += ln }
+    if tag4(d, typ, 73, 69, 78, 68) { done = true }
+    i = i + 12 + ln
+  }
+  free(d)
+  if w <= 0 or h <= 0 { free(idat);  free(plte);  return null }
+  let j = new PngJob
+  j.w = w;  j.h = h;  j.bd = bd;  j.ct = ct;  j.idat = idat;  j.idlen = idlen;  j.plte = plte
+  j.channels = 1
+  if ct == 2 { j.channels = 3 }
+  if ct == 4 { j.channels = 2 }
+  if ct == 6 { j.channels = 4 }
+  let bppbits = bd * j.channels
+  j.fbpp = max(1, (bppbits + 7) / 8)
+  j.stride = (w * bppbits + 7) / 8
+  j.rawlen = h * (j.stride + 1)
+  # one zeroed scanline in front of the data, so row 0's "row above" is real
+  j.raw = bytes(j.stride + j.rawlen + 8)
+  for z in 0 .. j.stride { j.raw[z] = 0 }
+  if (ct == 3) or (bd < 8) {
+    j.oc = 1
+    if ct == 3 { j.oc = 3 }
+    j.out = bytes(w * h * j.oc)
+  } else { j.out = bytes(h * j.stride + 8) }
+  return j
+}
+
+# the part a worker may run: into what png_parse made, in the context it was handed
+function png_work(j: PngJob) -> void {
+  j.ok = false
+  let stride = j.stride
+  let raw = j.raw
+  if z_uncompress_in(j.z, j.idat, j.idlen, mem_off(raw, stride), j.rawlen) < 0 { return }
+  # reverse the per-scanline filters in place, then pack rows without the filter byte
+  for y in 0 .. j.h {
+    let line = stride + y * (stride + 1)
+    png_unfilter(raw, line + 1, line + 1 - (stride + 1), stride, j.fbpp, raw[line])
+  }
+  let w = j.w
+  let out = j.out
+  if (j.ct == 3) or (j.bd < 8) {
+    # expand palette / sub-byte grey to 8-bit RGB (palette) or 8-bit grey
+    let bd = j.bd
+    let maxv = (1 << bd) - 1
+    let plte = j.plte
+    for yy in 0 .. j.h {
+      let row = stride + yy * (stride + 1) + 1
+      for x in 0 .. w {
+        let bp = x * bd
+        let idx = ((raw[row + bp / 8] >> (8 - bd - bp % 8)) & maxv)
+        if j.ct == 3 { out[(yy * w + x) * 3] = plte[idx * 3];  out[(yy * w + x) * 3 + 1] = plte[idx * 3 + 1];  out[(yy * w + x) * 3 + 2] = plte[idx * 3 + 2] }
+        else { out[yy * w + x] = idx * 255 / maxv }
+      }
+    }
+  } else {
+    for yy in 0 .. j.h { mem_copy(mem_off(out, yy * stride), mem_off(raw, stride + yy * (stride + 1) + 1), stride) }
+  }
+  j.ok = true
+}
+
+# back on the program's thread: the samples (null if the stream was bad), and their shape told
+function png_take(render3d_st: mut Render3dState, j: PngJob) -> pointer {
+  free(j.idat);  free(j.raw);  free(j.plte)
+  j.idat = null;  j.raw = null;  j.plte = null
+  if not j.ok {
+    free(j.out)
+    print(`png: inflate failed: {j.path}`)
+    return null
+  }
+  var ch = j.channels
+  var bd = j.bd
+  if (j.ct == 3) or (j.bd < 8) {
+    ch = j.oc
+    bd = 8
+  }
+  render3d_st.tex_w = j.w;  render3d_st.tex_h = j.h;  render3d_st.tex_channels = ch;  render3d_st.tex_depth = bd
+  let out = j.out
+  j.out = null
+  return out
+}
+
+# the inflate context a decode on this thread (or the i-th of a batch) works in: made here, never by
+# a worker, and kept
+@alloc_ok("once per concurrent decode: an inflate's tables, kept and reused")
+function tex_z(render3d_st: mut Render3dState, i: int) -> ZInflate {
+  if render3d_st.tex_zs == null { render3d_st.tex_zs = new []ZInflate }
+  while len(render3d_st.tex_zs) <= i { push(render3d_st.tex_zs, z_inflate_new()) }
+  return render3d_st.tex_zs[i]
+}
+
+# Decode these PNGs on every core now, for png_decode (tex_load, the font) to take in its own time and
+# order. A file nobody takes is let go by tex_prefetch_drop.
+@alloc_ok("a load: the batch's list and each file's job")
+function tex_prefetch(render3d_st: mut Render3dState, paths: []string) -> void {
+  if render3d_st.tex_pf == null { render3d_st.tex_pf = new []PngJob }
+  let b = new PngBatch
+  b.jobs = new []PngJob
+  for i in 0 .. len(paths) {
+    let j = png_parse(render3d_st, paths[i])
+    if j == null { continue }
+    j.path = paths[i]
+    j.z = tex_z(render3d_st, len(b.jobs))
+    push(b.jobs, j)
+  }
+  Job.parallel_for(len(b.jobs), fn png_batch_one, b)
+  for i in 0 .. len(b.jobs) { push(render3d_st.tex_pf, b.jobs[i]) }
+}
+function png_batch_one(i: int, b: PngBatch) -> void { png_work(b.jobs[i]) }
+
+# a prefetched image of this path, not taken yet, or null
+function tex_pf_take(render3d_st: mut Render3dState, path: string) -> PngJob {
+  if render3d_st.tex_pf == null { return null }
+  for i in 0 .. len(render3d_st.tex_pf) {
+    let j = render3d_st.tex_pf[i]
+    if not j.taken and j.path == path {
+      j.taken = true
+      return j
+    }
+  }
+  return null
+}
+
+# what a prefetch decoded and nobody took, let go
+function tex_prefetch_drop(render3d_st: mut Render3dState) -> void {
+  if render3d_st.tex_pf == null { return }
+  for i in 0 .. len(render3d_st.tex_pf) {
+    let j = render3d_st.tex_pf[i]
+    if not j.taken { png_take(render3d_st, j) }
+    if j.out != null { free(j.out) }
+    j.out = null
+  }
+  List.clear(render3d_st.tex_pf)
+}
diff --git a/packages/ludic.render3d/r3d.ludic b/packages/ludic.render3d/r3d.ludic
index 69eb70e3..52e8561b 100644
--- a/packages/ludic.render3d/r3d.ludic
+++ b/packages/ludic.render3d/r3d.ludic
@@ -18,6 +18,7 @@ import "prof.ludic"
 import "drawstats.ludic"
 import "programs.ludic"
 import "texture.ludic"
+import "png_jobs.ludic"
 import "mesh.ludic"
 import "gpu_vk_draw.ludic"
 import "camera.ludic"
diff --git a/packages/ludic.render3d/texture.ludic b/packages/ludic.render3d/texture.ludic
index 3ec299fd..4e6003f5 100644
--- a/packages/ludic.render3d/texture.ludic
+++ b/packages/ludic.render3d/texture.ludic
@@ -115,77 +115,14 @@ function png_unfilter(raw: pointer, cur: int, prev: int, stride: int, fbpp: int,
 
 @alloc_ok("loading a model, a texture or a font: a load, not a frame (a guest loading a teammate's look is one)")
 function png_decode(render3d_st: mut Render3dState, path: pointer) -> pointer {
-  let d = r3d_read_file(render3d_st, path)
-  if d == null { print(`png: cannot read {path}`);  return null }
-  let size = render3d_st.tex_file_len
-  if size < 8 or d[0] != 137 or d[1] != 80 { free(d);  print(`png: not a png: {path}`);  return null }
-  var w = 0;  var h = 0;  var bd = 0;  var ct = 0
-  let plte = bytes(768)
-  let idat = bytes(size)
-  var idlen = 0
-  var i = 8
-  var done = false
-  while not done {
-    if i + 8 > size { done = true;  continue }
-    let ln = be32(d, i)
-    let typ = i + 4
-    let body = i + 8
-    if ln < 0 or body + ln > size { done = true;  continue }
-    if tag4(d, typ, 73, 72, 68, 82) { w = be32(d, body);  h = be32(d, body + 4);  bd = d[body + 8];  ct = d[body + 9] }
-    if tag4(d, typ, 80, 76, 84, 69) { let m = min(ln, 768);  for k in 0 .. m { plte[k] = d[body + k] } }
-    if tag4(d, typ, 73, 68, 65, 84) { mem_copy(mem_off(idat, idlen), mem_off(d, body), ln);  idlen += ln }
-    if tag4(d, typ, 73, 69, 78, 68) { done = true }
-    i = i + 12 + ln
-  }
-  if w <= 0 or h <= 0 { free(d);  free(idat);  free(plte);  return null }
-  var channels = 1
-  if ct == 2 { channels = 3 }
-  if ct == 4 { channels = 2 }
-  if ct == 6 { channels = 4 }
-  let bppbits = bd * channels
-  var fbpp = (bppbits + 7) / 8
-  if fbpp < 1 { fbpp = 1 }
-  let stride = (w * bppbits + 7) / 8
-  let rawlen = h * (stride + 1)
-  # one zeroed scanline in front of the data, so row 0's "row above" is real
-  let raw = bytes(stride + rawlen + 8)
-  for z in 0 .. stride { raw[z] = 0 }
-  if z_uncompress(idat, idlen, mem_off(raw, stride), rawlen) < 0 { free(d);  free(idat);  free(raw);  free(plte);  print(`png: inflate failed: {path}`);  return null }
-  free(d);  free(idat)
-  # reverse the per-scanline filters in place, then pack rows without the filter byte
-  var y = 0
-  while y < h {
-    let line = stride + y * (stride + 1)
-    png_unfilter(raw, line + 1, line + 1 - (stride + 1), stride, fbpp, raw[line])
-    y += 1
-  }
-  var out: pointer = null
-  if (ct == 3) or (bd < 8) {
-    # expand palette / sub-byte grey to 8-bit RGB (palette) or 8-bit grey
-    let maxv = (1 << bd) - 1
-    var oc = 1
-    if ct == 3 { oc = 3 }
-    out = bytes(w * h * oc)
-    for yy in 0 .. h {
-      let row = stride + yy * (stride + 1) + 1
-      for x in 0 .. w {
-        let bp = x * bd
-        let idx = ((raw[row + bp / 8] >> (8 - bd - bp % 8)) & maxv)
-        if ct == 3 { out[(yy * w + x) * 3] = plte[idx * 3];  out[(yy * w + x) * 3 + 1] = plte[idx * 3 + 1];  out[(yy * w + x) * 3 + 2] = plte[idx * 3 + 2] }
-        else { out[yy * w + x] = idx * 255 / maxv }
-      }
-    }
-    channels = oc
-    bd = 8
-    free(raw)
-  } else {
-    out = bytes(h * stride + 8)
-    for yy in 0 .. h { mem_copy(mem_off(out, yy * stride), mem_off(raw, stride + yy * (stride + 1) + 1), stride) }
-    free(raw)
-  }
-  free(plte)
-  render3d_st.tex_w = w;  render3d_st.tex_h = h;  render3d_st.tex_channels = channels;  render3d_st.tex_depth = bd
-  return out
+  let pre = tex_pf_take(render3d_st, path)
+  if pre != null { return png_take(render3d_st, pre) }
+  let j = png_parse(render3d_st, path)
+  if j == null { return null }
+  j.path = path
+  j.z = tex_z(render3d_st, 0)
+  png_work(j)
+  return png_take(render3d_st, j)
 }
 
 # Edge padding for cut-out atlases: pixels darker than `thresh` (the unused
diff --git a/runtime/native/inflate.ludic b/runtime/native/inflate.ludic
index 73fc2cb0..ecdc4aa5 100644
--- a/runtime/native/inflate.ludic
+++ b/runtime/native/inflate.ludic
@@ -15,54 +15,93 @@
 # megabytes of texture at load, and there the table is worth its 2 KB.
 # ============================================================================
 
-# ---- bit reader (DEFLATE packs bits least-significant-first) ---------------
-export state RtInflateState {
-  z_src:    pointer = null
-  z_len:    int = 0
-  z_pos:    int = 0
-  z_bitbuf: int = 0
-  z_bitcnt: int = 0
-  z_err:    int = 0
-  z_lbase: words = null
-  z_lext:  words = null
-  z_dbase: words = null
-  z_dext:  words = null
+# ---- the context ------------------------------------------------------------
+# Everything one inflate works in: the bit reader, the RFC's length and distance tables, and every
+# table and scratch list a block builds, made once with the context (z_inflate_new) and rebuilt in
+# place - so an inflate allocates nothing, and each thread that inflates holds a context of its own.
+# RtInflateState keeps one for the callers that inflate on the program's own thread.
+export property ZInflate {
+  src:    pointer = null
+  len:    int = 0
+  pos:    int = 0
+  bitbuf: int = 0
+  bitcnt: int = 0
+  err:    int = 0
+  lbase:  words = null
+  lext:   words = null
+  dbase:  words = null
+  dext:   words = null
+  lit:    words = null
+  dist:   words = null
+  clen:   words = null
+  lengths: words = null
+  dl:     words = null
+  offs:   words = null
+  firstc: words = null
 }
 
-function z_start(rt_inflate_st: mut RtInflateState, src: pointer, len: int) -> void {
-  z_tables_once(rt_inflate_st)
-  rt_inflate_st.z_src = src
-  rt_inflate_st.z_len = len
-  rt_inflate_st.z_pos = 0
-  rt_inflate_st.z_bitbuf = 0
-  rt_inflate_st.z_bitcnt = 0
-  rt_inflate_st.z_err = 0
+export state RtInflateState {
+  z_ctx: ZInflate = z_inflate_new()
+}
+
+@alloc_ok("once per context: every table an inflate works in, rebuilt in place ever after")
+export function z_inflate_new() -> ZInflate {
+  let z = new ZInflate
+  z.lbase = words(29)
+  z.lext = words(29)
+  for s in 0 .. 29 {
+    z.lbase[s] = z_len_base(s)
+    z.lext[s] = z_len_extra(s)
+  }
+  z.dbase = words(30)
+  z.dext = words(30)
+  for s in 0 .. 30 {
+    z.dbase[s] = z_dist_base(s)
+    z.dext[s] = z_dist_extra(s)
+  }
+  z.lit = z_table_new(288)
+  z.dist = z_table_new(30)
+  z.clen = z_table_new(19)
+  z.lengths = words(320)
+  z.dl = words(32)
+  z.offs = words(16)
+  z.firstc = words(17)
+  return z
+}
+
+function z_start(z: ZInflate, src: pointer, len: int) -> void {
+  z.src = src
+  z.len = len
+  z.pos = 0
+  z.bitbuf = 0
+  z.bitcnt = 0
+  z.err = 0
 }
 
 # Fill the bit buffer to at least `n` bits without consuming any (n <= 16, so the
 # buffer never shifts a byte past bit 15 and cannot reach the sign bit).
-function z_need(rt_inflate_st: mut RtInflateState, n: int) -> void {
-  while rt_inflate_st.z_bitcnt < n {
-    if rt_inflate_st.z_pos >= rt_inflate_st.z_len { return }
-    rt_inflate_st.z_bitbuf = (rt_inflate_st.z_bitbuf | (rt_inflate_st.z_src[rt_inflate_st.z_pos] << rt_inflate_st.z_bitcnt))
-    rt_inflate_st.z_pos += 1
-    rt_inflate_st.z_bitcnt += 8
+function z_need(z: ZInflate, n: int) -> void {
+  while z.bitcnt < n {
+    if z.pos >= z.len { return }
+    z.bitbuf = (z.bitbuf | (z.src[z.pos] << z.bitcnt))
+    z.pos += 1
+    z.bitcnt += 8
   }
 }
 
-function z_bits(rt_inflate_st: mut RtInflateState, need: int) -> int {
-  var val = rt_inflate_st.z_bitbuf
-  while rt_inflate_st.z_bitcnt < need {
-    if rt_inflate_st.z_pos >= rt_inflate_st.z_len {
-      rt_inflate_st.z_err = 1
+function z_bits(z: ZInflate, need: int) -> int {
+  var val = z.bitbuf
+  while z.bitcnt < need {
+    if z.pos >= z.len {
+      z.err = 1
       return 0
     }
-    val = (val | (rt_inflate_st.z_src[rt_inflate_st.z_pos] << rt_inflate_st.z_bitcnt))
-    rt_inflate_st.z_pos += 1
-    rt_inflate_st.z_bitcnt += 8
+    val = (val | (z.src[z.pos] << z.bitcnt))
+    z.pos += 1
+    z.bitcnt += 8
   }
-  rt_inflate_st.z_bitbuf = (val >> need)
-  rt_inflate_st.z_bitcnt -= need
+  z.bitbuf = (val >> need)
+  z.bitcnt -= need
   return (val & ((1 << need) - 1))
 }
 
@@ -81,7 +120,7 @@ function z_table_new(nsym: int) -> words {
 }
 
 # lengths[i] = code length of symbol i (0 = symbol unused)
-function z_table_build(table: words, lengths: words, n: int) -> void {
+function z_table_build(z: ZInflate, table: words, lengths: words, n: int) -> void {
   for i in 0 .. 16 {
     table[i] = 0
   }
@@ -91,7 +130,7 @@ function z_table_build(table: words, lengths: words, n: int) -> void {
   }
   table[0] = 0                 # length 0 means "not present"
   # offset of each length's first symbol
-  let offs: words = words(16)
+  let offs = z.offs
   offs[1] = 0
   for l in 1 .. 15 {
     offs[l + 1] = offs[l] + table[l]
@@ -103,14 +142,13 @@ function z_table_build(table: words, lengths: words, n: int) -> void {
       offs[l] += 1
     }
   }
-  free(offs)
 
   # ---- the fast lookup ----
   for i in 0 .. Z_FASTSZ {
     table[16 + i] = 0
   }
   # first canonical code of each length
-  let firstc: words = words(17)
+  let firstc = z.firstc
   var code = 0
   for l in 1 .. 16 {
     code = ((code + table[l - 1]) << 1)
@@ -143,17 +181,16 @@ function z_table_build(table: words, lengths: words, n: int) -> void {
       k += 1
     }
   }
-  free(firstc)
 }
 
-function z_decode(rt_inflate_st: mut RtInflateState, table: words) -> int {
-  z_need(rt_inflate_st, Z_FAST)
-  if rt_inflate_st.z_bitcnt >= Z_FAST {
-    let e = table[16 + (rt_inflate_st.z_bitbuf & (Z_FASTSZ - 1))]
+function z_decode(z: ZInflate, table: words) -> int {
+  z_need(z, Z_FAST)
+  if z.bitcnt >= Z_FAST {
+    let e = table[16 + (z.bitbuf & (Z_FASTSZ - 1))]
     if e != 0 {
       let l = (e >> 16)
-      rt_inflate_st.z_bitbuf = (rt_inflate_st.z_bitbuf >> l)
-      rt_inflate_st.z_bitcnt -= l
+      z.bitbuf = (z.bitbuf >> l)
+      z.bitcnt -= l
       return (e & 65535)
     }
   }
@@ -162,7 +199,7 @@ function z_decode(rt_inflate_st: mut RtInflateState, table: words) -> int {
   var first = 0
   var index = 0
   for len in 1 .. 16 {
-    code = (code | z_bits(rt_inflate_st, 1))
+    code = (code | z_bits(z, 1))
     let count = table[len]
     if code - first < count {
       return table[Z_SYMS + index + (code - first)]
@@ -171,7 +208,7 @@ function z_decode(rt_inflate_st: mut RtInflateState, table: words) -> int {
     first = ((first + count) << 1)
     code = (code << 1)
   }
-  rt_inflate_st.z_err = 1
+  z.err = 1
   return -1
 }
 
@@ -202,49 +239,33 @@ function z_dist_extra(sym: int) -> int {
   return (sym - 2) / 2
 }
 
-# The RFC tables above are pure functions of the symbol; compute them once rather
-# than dividing per match.
-
-function z_tables_once(rt_inflate_st: mut RtInflateState) -> void {
-  if rt_inflate_st.z_lbase != null { return }
-  rt_inflate_st.z_lbase = words(29)
-  rt_inflate_st.z_lext = words(29)
-  for s in 0 .. 29 {
-    rt_inflate_st.z_lbase[s] = z_len_base(s)
-    rt_inflate_st.z_lext[s] = z_len_extra(s)
-  }
-  rt_inflate_st.z_dbase = words(30)
-  rt_inflate_st.z_dext = words(30)
-  for s in 0 .. 30 {
-    rt_inflate_st.z_dbase[s] = z_dist_base(s)
-    rt_inflate_st.z_dext[s] = z_dist_extra(s)
-  }
-}
+# The RFC tables above are pure functions of the symbol, computed once per context (z_inflate_new)
+# rather than divided per match.
 
 # ---- block decoders -------------------------------------------------------
 # `out` is the destination window; returns the new write position, or -1.
-function z_stored(rt_inflate_st: mut RtInflateState, out: pointer, at: int, cap: int) -> int {
-  rt_inflate_st.z_bitbuf = 0
-  rt_inflate_st.z_bitcnt = 0                        # stored blocks are byte-aligned
-  if rt_inflate_st.z_pos + 4 > rt_inflate_st.z_len { return -1 }
-  let n = rt_inflate_st.z_src[rt_inflate_st.z_pos] + (rt_inflate_st.z_src[rt_inflate_st.z_pos + 1] << 8)
-  rt_inflate_st.z_pos += 4                   # LEN then its one's complement
+function z_stored(z: ZInflate, out: pointer, at: int, cap: int) -> int {
+  z.bitbuf = 0
+  z.bitcnt = 0                        # stored blocks are byte-aligned
+  if z.pos + 4 > z.len { return -1 }
+  let n = z.src[z.pos] + (z.src[z.pos + 1] << 8)
+  z.pos += 4                   # LEN then its one's complement
   var w = at
   for i in 0 .. n {
-    if rt_inflate_st.z_pos >= rt_inflate_st.z_len { return -1 }
+    if z.pos >= z.len { return -1 }
     if w >= cap { return -1 }
-    out[w] = rt_inflate_st.z_src[rt_inflate_st.z_pos]
+    out[w] = z.src[z.pos]
     w += 1
-    rt_inflate_st.z_pos += 1
+    z.pos += 1
   }
   return w
 }
 
-function z_codes(rt_inflate_st: mut RtInflateState, out: pointer, at: int, cap: int, lit: words, dist: words) -> int {
+function z_codes(z: ZInflate, out: pointer, at: int, cap: int, lit: words, dist: words) -> int {
   var w = at
-  var sym = z_decode(rt_inflate_st, lit)
+  var sym = z_decode(z, lit)
   while sym != 256 {
-    if rt_inflate_st.z_err != 0 { return -1 }
+    if z.err != 0 { return -1 }
     if sym < 0 { return -1 }
     if sym < 256 {
       if w >= cap { return -1 }
@@ -254,11 +275,11 @@ function z_codes(rt_inflate_st: mut RtInflateState, out: pointer, at: int, cap:
     if sym > 256 {
       let s = sym - 257
       if s >= 29 { return -1 }
-      let length = rt_inflate_st.z_lbase[s] + z_bits(rt_inflate_st, rt_inflate_st.z_lext[s])
-      let d = z_decode(rt_inflate_st, dist)
+      let length = z.lbase[s] + z_bits(z, z.lext[s])
+      let d = z_decode(z, dist)
       if d < 0 { return -1 }
       if d >= 30 { return -1 }
-      let distance = rt_inflate_st.z_dbase[d] + z_bits(rt_inflate_st, rt_inflate_st.z_dext[d])
+      let distance = z.dbase[d] + z_bits(z, z.dext[d])
       if distance > w { return -1 }
       if w + length > cap { return -1 }      # bounds once, not per byte
       var sp = w - distance
@@ -270,44 +291,43 @@ function z_codes(rt_inflate_st: mut RtInflateState, out: pointer, at: int, cap:
         k += 1
       }
     }
-    sym = z_decode(rt_inflate_st, lit)
+    sym = z_decode(z, lit)
   }
   return w
 }
 
-function z_fixed_tables(lit: words, dist: words) -> void {
-  let lengths: words = words(288)
+function z_fixed_tables(z: ZInflate, lit: words, dist: words) -> void {
+  let lengths = z.lengths
   for i in 0 .. 144 { lengths[i] = 8 }
   for i in 144 .. 256 { lengths[i] = 9 }
   for i in 256 .. 280 { lengths[i] = 7 }
   for i in 280 .. 288 { lengths[i] = 8 }
-  z_table_build(lit, lengths, 288)
+  z_table_build(z, lit, lengths, 288)
   for i in 0 .. 30 { lengths[i] = 5 }
-  z_table_build(dist, lengths, 30)
-  free(lengths)
+  z_table_build(z, dist, lengths, 30)
 }
 
-function z_dynamic_tables(rt_inflate_st: mut RtInflateState, lit: words, dist: words) -> int {
-  let nlen = z_bits(rt_inflate_st, 5) + 257
-  let ndist = z_bits(rt_inflate_st, 5) + 1
-  let ncode = z_bits(rt_inflate_st, 4) + 4
+function z_dynamic_tables(z: ZInflate, lit: words, dist: words) -> int {
+  let nlen = z_bits(z, 5) + 257
+  let ndist = z_bits(z, 5) + 1
+  let ncode = z_bits(z, 4) + 4
   if nlen > 286 { return 0 }
   if ndist > 30 { return 0 }
 
-  let lengths: words = words(320)
+  let lengths = z.lengths
   for i in 0 .. 19 { lengths[i] = 0 }
   # the code-length alphabet is transmitted in this fixed permutation
   # 16,17,18,0,8,7,9,6,10,5,11,4,12,3,13,2,14,1,15 — biased by '0' so it is one literal
   let order = "@AB08796:5;4<3=2>1?"
   for i in 0 .. ncode {
-    lengths[order[i] - 48] = z_bits(rt_inflate_st, 3)
+    lengths[order[i] - 48] = z_bits(z, 3)
   }
-  let clen = z_table_new(19)
-  z_table_build(clen, lengths, 19)
+  let clen = z.clen
+  z_table_build(z, clen, lengths, 19)
 
   var n = 0
   while n < nlen + ndist {
-    let sym = z_decode(rt_inflate_st, clen)
+    let sym = z_decode(z, clen)
     if sym < 0 { return 0 }
     if sym < 16 {
       lengths[n] = sym
@@ -319,10 +339,10 @@ function z_dynamic_tables(rt_inflate_st: mut RtInflateState, lit: words, dist: w
       if sym == 16 {
         if n == 0 { return 0 }
         prev = lengths[n - 1]
-        rep = 3 + z_bits(rt_inflate_st, 2)
+        rep = 3 + z_bits(z, 2)
       }
-      if sym == 17 { rep = 3 + z_bits(rt_inflate_st, 3) }
-      if sym == 18 { rep = 11 + z_bits(rt_inflate_st, 7) }
+      if sym == 17 { rep = 3 + z_bits(z, 3) }
+      if sym == 18 { rep = 11 + z_bits(z, 7) }
       for k in 0 .. rep {
         if n < 320 {
           lengths[n] = prev
@@ -331,51 +351,51 @@ function z_dynamic_tables(rt_inflate_st: mut RtInflateState, lit: words, dist: w
       }
     }
   }
-  z_table_build(lit, lengths, nlen)
+  z_table_build(z, lit, lengths, nlen)
   # the distance lengths follow the literal ones in the same buffer
-  let dl: words = words(32)
+  let dl = z.dl
   for i in 0 .. ndist { dl[i] = lengths[nlen + i] }
-  z_table_build(dist, dl, ndist)
-  free(dl)
-  free(lengths)
-  free(clen)
+  z_table_build(z, dist, dl, ndist)
   return 1
 }
 
-# Inflate a raw DEFLATE stream. Returns bytes written, or -1.
-function z_inflate(rt_inflate_st: mut RtInflateState, src: pointer, len: int, out: pointer, cap: int) -> int {
-  z_start(rt_inflate_st, src, len)
-  let lit = z_table_new(288)
-  let dist = z_table_new(30)
+# Inflate a raw DEFLATE stream in context z. Returns bytes written, or -1.
+export function z_inflate_in(z: ZInflate, src: pointer, len: int, out: pointer, cap: int) -> int {
+  z_start(z, src, len)
+  let lit = z.lit
+  let dist = z.dist
   var w = 0
   var final = 0
   while final == 0 {
-    final = z_bits(rt_inflate_st, 1)
-    let btype = z_bits(rt_inflate_st, 2)
-    if rt_inflate_st.z_err != 0 { return -1 }
-    if btype == 0 { w = z_stored(rt_inflate_st, out, w, cap) }
+    final = z_bits(z, 1)
+    let btype = z_bits(z, 2)
+    if z.err != 0 { return -1 }
+    if btype == 0 { w = z_stored(z, out, w, cap) }
     if btype == 1 {
-      z_fixed_tables(lit, dist)
-      w = z_codes(rt_inflate_st, out, w, cap, lit, dist)
+      z_fixed_tables(z, lit, dist)
+      w = z_codes(z, out, w, cap, lit, dist)
     }
     if btype == 2 {
-      if z_dynamic_tables(rt_inflate_st, lit, dist) == 0 { return -1 }
-      w = z_codes(rt_inflate_st, out, w, cap, lit, dist)
+      if z_dynamic_tables(z, lit, dist) == 0 { return -1 }
+      w = z_codes(z, out, w, cap, lit, dist)
     }
     if btype == 3 { return -1 }
     if w < 0 { return -1 }
   }
-  free(lit)
-  free(dist)
   return w
 }
 
+# the program thread's inflate, in the context its state holds
+function z_inflate(rt_inflate_st: RtInflateState, src: pointer, len: int, out: pointer, cap: int) -> int { return z_inflate_in(rt_inflate_st.z_ctx, src, len, out, cap) }
+function z_uncompress(rt_inflate_st: RtInflateState, src: pointer, len: int, out: pointer, cap: int) -> int { return z_uncompress_in(rt_inflate_st.z_ctx, src, len, out, cap) }
+function z_gunzip(rt_inflate_st: RtInflateState, src: pointer, len: int, out: pointer, cap: int) -> int { return z_gunzip_in(rt_inflate_st.z_ctx, src, len, out, cap) }
+
 # zlib wrapper (RFC 1950): two header bytes, then DEFLATE, then Adler-32.
-function z_uncompress(rt_inflate_st: mut RtInflateState, src: pointer, len: int, out: pointer, cap: int) -> int {
+export function z_uncompress_in(z: ZInflate, src: pointer, len: int, out: pointer, cap: int) -> int {
   if len < 2 { return -1 }
   let cmf = src[0]
   if (cmf & 15) != 8 { return -1 }
-  return z_inflate(rt_inflate_st, offset(src, 2), len - 2, out, cap)
+  return z_inflate_in(z, offset(src, 2), len - 2, out, cap)
 }
 
 # gzip framing (RFC 1952): a 10-byte header (magic 1f 8b, CM=8, FLG, 4-byte MTIME,
@@ -384,7 +404,7 @@ function z_uncompress(rt_inflate_st: mut RtInflateState, src: pointer, len: int,
 # the header + optional fields, inflate the body, and ignore the trailer — the
 # CRC is a redundancy check, not needed to decode (PNG likewise ignores ancillary
 # CRCs). Returns bytes written, or -1.
-function z_gunzip(rt_inflate_st: mut RtInflateState, src: pointer, len: int, out: pointer, cap: int) -> int {
+export function z_gunzip_in(z: ZInflate, src: pointer, len: int, out: pointer, cap: int) -> int {
   if len < 18 { return -1 }                 # 10 header + 8 trailer minimum
   if src[0] != 31 { return -1 }             # 0x1f
   if src[1] != 139 { return -1 }            # 0x8b
@@ -406,5 +426,5 @@ function z_gunzip(rt_inflate_st: mut RtInflateState, src: pointer, len: int, out
   }
   if (flg & 2) != 0 { pos += 2 }       # FHCRC: 2-byte header CRC
   if pos + 8 > len { return -1 }
-  return z_inflate(rt_inflate_st, offset(src, pos), len - pos - 8, out, cap)
+  return z_inflate_in(z, offset(src, pos), len - pos - 8, out, cap)
 }