maketex: 50.32% -> 99.82% (rev1 symbols + Rev1 save/memcard differences; v1.00 byte-identical)

[SirEnkido]

What changed

main/dlls/engine/2/maketex (memory-card save/load) on GSAE01_rev1: 50.32% → 99.82%, in three steps:

1. Symbols only (config/GSAE01_rev1/symbols.txt): named saveGame_prepareAndWrite, loadMemCardImages, cardSetStatusNoCard2 from the v1.00 config → 50.32% → 87.90%
2. saveGame_prepareAndWrite: Rev 1 skips the slot rewrite when a callback is given (retail asm at 0x8007ED98 provably gates on cb == NULL) → 88.03%
3. loadMemCardImages: Rev 1 moved the comment writer into a separate global function (loadMemCardComments, defined after loadMemCardImages) and switched the check from gGameTextFontIsSjis to getCurLanguage() == LANGUAGE_JAPANESE, reordering the string pool; string-literal offsets are named macros → 99.82%

Steps 2–3 are wrapped in #ifdef VERSION_GSAE01_rev1 (the #else path is upstream's v1.00 code, whitespace-neutral).

Evidence

• The v1.00 object is byte-identical with and without the guards — sections, symbol table and relocations were compared against origin/main's maketex.c compiled with the same GC/2.0 cflags. The active GSAE01 target is untouched.
• ninja all_source progress build/GSAE01_rev1/report.json exits 0; independently re-audited (a whitespace regression in the SJIS block was caught and fixed in a follow-up commit)

⚠️ Please read — first version guards in game source

This PR introduces the first #ifdef VERSION_GSAE01_rev1 guards in src/ (5 of them; upstream has none outside the borrowed dolphin/pad/Pad.c VERSION_GCCP01). I want to be upfront about that rather than bury it:

• The define is emitted by upstream configure.py itself (-DVERSION_{version}), so the mechanism exists by design; docs/version_progress.md's "no version-specific source guards" describes what tools/version_progress.py does when projecting splits, and AGENTS.md has no rule against guards — but this is still a judgment call that's yours, not mine.
• A guard-free rewrite is not possible here: the retail Rev 1 asm provably changed the code (cb gate, factored-out helper, language check + reordered pool) while v1.00 has none of that — no shared helper can match both. (Sibling PR for engine/0 uses the same technique with the same byte-identical-v1.00 verification.)
• Fallback offered: if you'd rather not have guards, drop commits 2–4 and keep only commit 1 (6d72368d3b, symbols-only, 50.32% → 87.90%). Or tell me your preferred style and I'll restructure.

Other notes

• loadMemCardImages 99.17% / loadMemCardComments 99.03% residual: MWCC rematerializes the sMemoryCardFileNameString base once (lis/addi vs addi r3,r31,0xa0) — five source variants tried, none changed it; root cause is the pre-existing upstream names + offset addressing of a char array defined in src/track/intersect_memcard.c, not introduced here.
• Rev 1 .data for this unit stays low (gMemoryCardBannerAssetNames, 168 bytes) because the branch keeps the single v1.00-order byte array; deliberately did not add a sixth guard around data.
• loadMemCardComments is a newly coined name (no retail symbol); rename freely.

---

Produced by Claude agents following AGENTS.md, operated by @SirEnkido.

🤖 Generated with Claude Code