TMC PC port v0.1.4-experimental

0.1.4-experimental — 2026-05-03

Hyrule Town + South Hyrule field playthrough fixes, driven by a live GDB-under-the-game session against the issue tracker. Three crashes and one stuck-state bug closed; CI build stability improvements; matheo upstream merged.

Fixed (issue tracker)

• #16 Hyrule Town kinstone-bag crash chain. Five layered bugs all on the kinstone-fusion path that compound after picking up the bag: (1) gUnk_08001A7C and gUnk_08001DCC are packed 4-byte GBA-pointer tables but were declared as T*[] so arr[idx] read 8 bytes on x86-64 → garbage pointer → SIGSEGV the moment a kinstone fuser scripted up; (2) (u8*)(fuserProgress + (u32)fuserData) truncated a 64-bit pointer in common.c::GetFusionToOffer; (3) TextDispEnquiry's post-A_BUTTON modulo divided by zero on x86 (ARM silently returns garbage) after MemClear zeroed gMessageChoices.choiceCount; (4) kinstoneMenu.c::sub_080A4418 wrote to DMA3->sourceAddress (raw GBA hardware register address 0x040000D4, unmapped on PC); (5) KinstoneMenu_080A4468 dereferenced gPossibleInteraction.currentObject without checking it pointed inside candidates[]. Fixed each + added Port_PackedRomEntry() helper for the ~5 packed-pointer-table call sites. (commit b825b5fd)
• #19 South Hyrule field loading-zone crash. Two bugs on the spear moblin: (a) gUnk_080CC944 was the same packed-pointer-table pattern as #16, handing out garbage Hitbox pointers on x86-64; (b) definition->ptr.hitbox lives in read-only mmap'd ROM (gRomData), but the spear moblin's action routines write hitbox->offset_x/y/width/height every frame. Allocate a mutable Hitbox3D copy at init (AllocMutableHitbox() would zFree the const ROM pointer, so do it manually). (commit c749c609)
• #20 Peahat (and other gust-jar-killable enemies) corpse never despawns. GetNextFunction() returned 5 (OnGrabbed) for any entity with gustJarState & 4 set, before the health == 0 check. When a peahat is killed by gust jar, Peahat_OnGrabbed_Subaction5 sets health=0 but gj=0x04 stays set until ENT_COLLIDE toggles, which doesn't happen in that subaction's else branch. Result: dead-while-grabbed enemies loop in OnGrabbed forever — corpse stays, no death animation. PC port now prefers the health==0 dispatch so dead-while-grabbed enemies flow into GenericDeath → DEATH_FX → DeleteThisEntity. (commit c708678a)

Build / CI

• Linux runner pinned to ubuntu-22.04 (glibc 2.35) so the prebuilt tarball runs on Nobara 43, Fedora 43, Kubuntu 25.10, and any other distro that lags behind the Arch host's glibc 2.43. Addresses #2's class plus #17. (commits 91b5c9fd, a5f546da)
• Pre-generated USA map_offsets.h + gfx_offsets.h committed under a .gitignore exception so CI can build without needing a private ROM repository. Drops python build.py from the workflow in favour of direct xmake build tmc_pc + xmake build asset_extractor, which don't need a ROM at compile time. (commit fb36e928)
• Merged matheo/master twice today: ubuntu fmt12 build fix (#15 root cause); update-check infrastructure (port/port_update_check.{c,h}); Minish Woods Ezlo fight + lilypad rail data fallbacks; bootstrap-based asset management (port/port_asset_bootstrap.cpp); affineSet cleanup. (commits b5cf8067, d37edeb3)

Implicit fixes confirmed

• #23 Broken map (grey tiles) — works locally on the post-merge branch; one of the struct-alignment / BG-flush fixes resolved it. Awaiting Proton/Bazzite retest on this tarball.

Known issues (still open)

• #4 Pulled mushroom, #5 backflip at walls, #8 blue/red teleport icons, #9 Steam Deck packages, #11 boss asset rendering (deferred to next sprite-extraction pass).
• #21 Link's house glitched doors — likely same class as the door-priority issue carried from 0.1.2.
• #22 BGM doesn't mute on item-get — audio priority/ducking, same area as the festival-house BGM-reset fix in 0.1.2 but with a different interaction.
• #24 Tall-grass shoes overlay — known carry-forward.
• Inside-the-rolling-barrel scene, festival house facades, Minish Woods fog, mosaic effect — renderer-iteration deferred from 0.1.2.