* fix: restore MSL extab flags and force-active stripped symbols
* fix: limit ChannelTitle novtable header changes to IPL_CHANNEL_TITLE_NOVTABLE
PR #182 made gui::EventHandler novtable and rewrote HermiteIntp for every
unit, which dropped __vt__Q23gui12EventHandler from linked scene units and
moved EventHandler::onEvent. Restore the pre-#182 definitions unless the
macro is defined.
* fix: restore u16 addrType in cdb::Manager::createNewRecord
The original symbol takes u16 (mangled UsUs); the enum parameter made
createNewRecord 4 bytes longer. Match the caller in iplNwc24Manager with a
u16 local assigned through if/else.
* fix: emit TVRC loop handlers after TVRCSendStartAsync
The original object places both alarm handlers at the end of .text.
* fix: construct scene::Date::mscMaxDate with utility::Date
The MaxDate wrapper emitted a weak MaxDate ctor that the original does
not have, shifting .text by 0x10 after iplDate. Go back to a plain
utility::Date followed by the 0x10-byte padding array.
* fix: define textinput PaneComponent dtor after PaneManager dtor
Matches the original object's function order, which also pulls the weak
GUIComponent dtor into place.
* fix: define EGG audio Arg and SoundHeapMgr ctors out of line
Emits IAudioMgr::Arg::Arg first and SoundHeapMgr's weak dtor ahead of its
ctor, matching the original eggAudioMgr function order.
* fix: define WaveFile dtors in AudioWavePlayer after the ctor
The original emits WavePacket's weak dtor first, then the WaveFileAiff and
WaveFileWav dtors. Declaring the dtors after getBit keeps the vtables in
AudioWaveUtility.
* fix: restore fork-point iplCSMtx string pooling
Moving "MTX34" into a static const array put it in .sdata2 and left the
pooled .sdata copy to be dead-stripped, shifting both sections.
* fix: share tiGUIManager's named 0.0f constant
drawLine_ and PaneComponent::contain now load lbl_81694EE8, as in the
original, instead of emitting their own 0.0f ahead of it in .sdata2.
* fix: force-active unreferenced iplDate and WDScan data
MWLD dead-strips scene::padding (the 0x10 bytes after Date::mscMaxDate)
and the unreferenced WDScan .sdata2 bytes lbl_81695063/lbl_81695067,
which shifted .bss and .sdata2.
* fix: unlink eggAudioExpMgr
MWCC emits the @52 calc thunk ahead of the @52 dtor thunk, but the
original has them the other way round, which swaps two vtable entries.
Link the original object until the order is matched.
* fix: shrink Calendar .data padding now that EventHandler has a vtable
With gui::EventHandler no longer novtable, its 0x18-byte vtable fills most
of the gap after __vt__Q33ipl5scene8Calendar; keep one 4-byte pad.
Personal fork so AI-assisted decomp work has a home without disturbing
upstream. README explains the fork; AGENTS.md documents the method and
forbids ever opening a PR against koopthekoopa/wii-ipl.
Matching progress: +9 exact functions, 0 regressions, main.dol still
byte-identical (26116613f624061ba99c8d1a299aaa6efa85670d).
iplESMisc gains CheckTmdParentalControl, CheckSafeDeleteTitle,
NumPrivateContents, checkContentsNum and GetValidTicketIndex, plus the
other functions matched in the previous pass.
The load-bearing finding, written up in AGENTS.md: MWCC emits a
translation unit's string literals in source-token order. A missing
function, or an inverted if/else, shifts every later string offset and
breaks functions that were never edited. Fixing that pool ordering lifted
TMDFile::Open from 99.92 to 99.96 without touching Open.
Adds tools/decomp-assist/ with pool_diff.py, ctxdiff.py, disasm_fn.py,
export_pyghidra.py and the Ghidra decompilation dump.