Files
neoLegacy-mac/update.md

5.6 KiB
Raw Permalink Blame History

Iggy 1.2.30 Reconstruction — Status Update

Generated 2026-07-19

Short answers to your questions

  • Is it normal source code yet? Have you gone from pseudocode → normal C? Yes. The deliverable in iggy-decompiled/src/reconstructed/ is clean, human-readable, buildable C11 — named functions, real control flow, #if-gated per-platform layouts, provenance headers. It is not Ghidra pseudocode. The Ghidra pseudocode lives in iggy-decompiled/raw/<platform>/ and is the input (ground truth), not the output.
  • Where am I? ~62 of the ~70 substantive translation units are reconstructed and committed; 9 empty/stub TUs are conclusively classified. What remains is the long tail: the ~8 largest files (the SWF loader + the AVM2 flash display list).
  • When done? The remaining 8 are the biggest and hardest in the whole codebase (swf_as3vm_flash_display alone is 16,440 lines — larger than everything else combined). No firm date: this is a marathon tail, each of those is a multi-round effort. By count we're ~88% done; by code weight closer to ~60–65%.

Pseudocode → normal C, concretely

Before (Ghidra raw/durango/…, machine-recovered pseudocode):

void as3vm_ByteArray__construct(longlong *param_1) {
  if (*(longlong *)(as3vm_Capabilities_get_version + *param_1) != 0) {
    *(longlong *)(*(longlong *)param_1[3] + 0x27) =
         *(longlong *)(as3vm_Capabilities_get_version + *param_1);
    ...

After (src/reconstructed/…, source-equivalent C):

void as3vm_ByteArray__construct(void *frame) {
  void *obj = THISOBJ(frame);
  /* pending buffer is a VM-struct field (single-deref), adopted into the object. */
  void **pending = (void **)AT(VM(frame), E_PENDBUF);
  if (*pending != NULL) {
    *(void **)AT(obj, OB_BA_BUFOBJ) = *pending;
    ...

Every reconstructed unit:

  • compiles clean under -std=c11 -O2 -Wall -Wextra -Werror for all four platforms (durango x64 / orbis x64 / ps3 PPC big-endian / vita ARM);
  • exports exactly the retail symbol set and imports exactly the retail import set (verified with llvm-nm per platform);
  • reproduces the retail struct offsets, constant tables, and byte-level data (verified with llvm-objdump -s against the shipped objects);
  • has a behavioral test suite (currently ~880 tests, green) and an independent fidelity review against all four platform decompiles before it's committed.

This is a faithful reconstruction, not the literal lost original source: it matches behavior, ABI, and data layout, and the deep/rare code paths are reconstructed to the recovered structure and reviewed, not stubbed.

Done so far (this stretch: 34 modules + 9 empty classifications)

AVM2 / AS3 virtual-machine core (complete): as3vm_names, as3vm_create, as3vm_gc, as3vm_class, as3vm_convert, as3vm_property, as3vm_interpreter (the bytecode dispatch loop), as3vm (pool + public ValueRef API), as3vm_verifier, as3vm_abc_decode (the ABC file parser), as3_component_helpers.

AS3 builtin classes: imp_array (Array/Vector), imp_math (Math), imp_string (String), classes_base_dummy (E4X XML stubs), classes_flash_dummy (flash.* stubs), swf_as3_bitmap_data (BitmapData), swf_as3_misc (Capabilities/Sound/Security/ByteArray/…).

Utility libraries: zlib (DEFLATE inflate), dtoa (David Gay float↔string).

SWF / runtime path: swf_playhead (timeline), iggy_api_functions (public C API), and iggy_main (the central init/runtime/error/RNG/translation keystone — in its final review-and-fix round right now).

(Plus the earlier foundation: rrMath, rrstring, arena allocator, unicode, fonts, sound/mixer, the whole swf_draw / triangulation / stroke render cluster, platform threads/atomics/time, etc.)

What's left — the hard tail (~8 units)

lines unit what it is
16,440 swf_as3vm_flash_display the AVM2 flash.display list (DisplayObject/Sprite/MovieClip…) — the giant
6,635 swf_decode SWF/.iggy tag decoder
6,323 swf_play movie playback + display-list execution
5,983 swf_as3vm AS3↔display-object glue
4,341 as3vm_imp_misc remaining AS3 builtins
3,728 swf_as3vm_flash_events flash.events (EventDispatcher…)
3,487 swf_as3_textfield TextField
3,144 swf_jpeg JPEG decoder

These are the pieces that turn "the VM and runtime exist" into "a SWF loads, ticks, and renders through the reconstructed runtime." They're being taken in import-closure order.

How each module is built (the pipeline, so you know the bar)

  1. RED tests authored from the four platform objects (exports, imports, owned data bytes, provenance, ABI) — fails until the source exists.
  2. Implement to green, faithfully from the raw, reconciled across all 4 platforms.
  3. Independent fidelity review vs all four decompiles — catches the harness-invisible defects (32-bit offset bugs, big-endian bitfields, skeleton pipelines).
  4. Fix the findings; re-review if the change was substantial.
  5. Commit only when 4-platform clean + reviewed.

That review step keeps catching real, otherwise-silent bugs — a skeleton verifier, an offset-broken parser, a swapped GC write-barrier, ungated 32-bit offsets, and (just now) a phantom struct field in iggy_main. Nothing ships on "the tests pass" alone.

One tracked follow-up

A narrow, ps3-only, test-invisible residual in iggy_api_functions (big-endian resource-flag bit positions) is documented in-source and captured as a background task chip; everything else on that module is source-equivalent on all four platforms.