fix: restore 43U DOL hash on main #213

Merged
coah80 merged 13 commits from agent/restore0928/dol-hash into main 2026-09-28 04:26:31 +00:00
coah80 commented 2026-09-28 04:26:20 +00:00 (Migrated from github.com)

main hasnt actually built the right dol for a while, a clean build of 75c7a24c gives 0c106821… instead of 26116613…. the fork point f8d83e5f still matches, so a bunch of merges since then got through without a real hash check. objdiff scores by symbol name so it never caught any of this

what was broken and whats fixed here

  • 5 linked msl units (arith, bsearch, buffer_io, FILE_POS, mbstring) were built with exceptions off so they had no extab, turned -Cpp_exceptions on back on for them
  • mwld dead strips unreferenced funcs in mwcc built objects (they have a .comment section, the dtk split ones dont), so labs/itoa/strpbrk/strspn/strcspn/atol and a few data syms were getting dropped. added force_active in config.yml plus a FORCEACTIVE { $FORCEACTIVE } block in the ldscript template
  • #182 made gui::EventHandler novtable and rewrote HermiteIntp for EVERY unit, not just channelTitle. thats now all behind IPL_CHANNEL_TITLE_NOVTABLE, which puts the missing 0x18 vtables back in a ton of scene units
  • createNewRecord goes back to u16 addrType (thats the original mangled name, the enum version was 4 bytes longer)
  • function order fixes in TVRC, tiGUIManager, eggAudioMgr, AudioWavePlayer, plus iplDate's extra weak ctor and iplCSMtx string pooling put back to how it was at the fork point
  • eggAudioExpMgr goes back to Equivalent, mwcc emits two vtable thunks in the opposite order from the original and nothing i tried flips them

gates (fresh orchestrator build)

  • rm -rf build/43U && python3 configure.py --version 43U && ninja build/43U/ok passes
  • build/43U/main.dol sha1 26116613f624061ba99c8d1a299aaa6efa85670d, cmp against orig/43U/00000008.app is byte identical
  • full ninja passes
  • report: matched code 62.772488 -> 62.792923, linked code 51.134293 -> 51.116795 (eggAudioExpMgr unlinked), matched data 71.68328 -> 71.83301, linked data 63.94687 -> 63.940315, complete units 709 -> 708
main hasnt actually built the right dol for a while, a clean build of 75c7a24c gives `0c106821…` instead of `26116613…`. the fork point f8d83e5f still matches, so a bunch of merges since then got through without a real hash check. objdiff scores by symbol name so it never caught any of this what was broken and whats fixed here - 5 linked msl units (arith, bsearch, buffer_io, FILE_POS, mbstring) were built with exceptions off so they had no extab, turned `-Cpp_exceptions on` back on for them - mwld dead strips unreferenced funcs in mwcc built objects (they have a `.comment` section, the dtk split ones dont), so labs/itoa/strpbrk/strspn/strcspn/atol and a few data syms were getting dropped. added `force_active` in config.yml plus a `FORCEACTIVE { $FORCEACTIVE }` block in the ldscript template - #182 made `gui::EventHandler` novtable and rewrote `HermiteIntp` for EVERY unit, not just channelTitle. thats now all behind `IPL_CHANNEL_TITLE_NOVTABLE`, which puts the missing 0x18 vtables back in a ton of scene units - `createNewRecord` goes back to `u16 addrType` (thats the original mangled name, the enum version was 4 bytes longer) - function order fixes in TVRC, tiGUIManager, eggAudioMgr, AudioWavePlayer, plus iplDate's extra weak ctor and iplCSMtx string pooling put back to how it was at the fork point - eggAudioExpMgr goes back to Equivalent, mwcc emits two vtable thunks in the opposite order from the original and nothing i tried flips them gates (fresh orchestrator build) - `rm -rf build/43U && python3 configure.py --version 43U && ninja build/43U/ok` passes - `build/43U/main.dol` sha1 `26116613f624061ba99c8d1a299aaa6efa85670d`, `cmp` against `orig/43U/00000008.app` is byte identical - full `ninja` passes - report: matched code 62.772488 -> 62.792923, linked code 51.134293 -> 51.116795 (eggAudioExpMgr unlinked), matched data 71.68328 -> 71.83301, linked data 63.94687 -> 63.940315, complete units 709 -> 708
Sign in to join this conversation.