Entry 25

Dungeons, fields, battle and text checked on every volume

2026-09-23worldbattleuivolumestest

Filestools/dungeon.pytools/field.pytools/battle.pytools/font.pytools/test_dungeon.pytools/test_field.pytools/test_battle.pytools/test_font.pydocs/engine/dungeon.mddocs/engine/field.mddocs/engine/battle.mddocs/engine/font.mddocs/content/game-data.mddocs/disc/volumes.md

21 ran the area generator on all four volumes. This entry does the same for the dungeon and field generators of 19, the battle rules of 22 and the text rendering of 23: each volume's own code in eemu against the Python reimplementation. A helper agent did the work, one system at a time. I re-ran the whole suite and read two of the differences myself. The results are in the "Other volumes" sections of the dungeon, field, battle and text pages, and summarised on the four discs.

First, the harnesses

None of the four harnesses ran on the other volumes as they stood. Each failed for reasons that were not about the game:

• Structure layouts. Mutation shrank DUNGEON from 0xd3910 to 0xce0 bytes by moving its per-floor arrays into new[] allocations. The size was read from WORLD_MAN::GO's allocation, and the new pointer offsets from the constructor. ccSkill's creator and target moved by 0x10. The field and text structures did not change.
• Wrong or missing carried names. MUT's skillTbl points 0x11b0 bytes into the table, and its MailTbl points into code. skillTbl is missing on OUT and QUA, and enemyTbl on MUT. The tools now find every table through the code that uses it:
  • battle.py through each table's accessor;
  • dungeon.py through MakeFloor's jump tables;
  • font.py through Extract's and Disp's address loads. On Infection this reproduces the symbol table's addresses and DWARF sizes, and tests assert it.
• Emulator reach. OUT's WORLD::GetHeight reached sceVu0OuterProduct on stack garbage, and eemu does not interpret lqc2. The height feeds only z, which is not compared, so it is now stubbed on every volume.

Then, the results

0 mismatches everywhere:

Real differences, each now read from the executable rather than keyed on the volume:

• Dungeons (MUT on):
  • The floor count is a field, 15 for area 125 instead of the constant 10.
  • EditDungeon grows to 90 entries, adding areas 118 and 119.
  • SetDungeonTypeFromField uses 21's substitute records. It swaps the saveData+0x6772 test for bit 62 of saveData+0x5ec8, which sends story areas down the random-area path with plain types.
• Fields:
  • OUT and QUA compute distances with sqrt.s rather than sqrtf. No case differs.
  • From MUT on, area 100 is no longer protected.
• Battle (MUT on):
  • Exdefense now keeps only the bits whose defence is not lowered, rather than all or nothing (MUT gcmn:0x005933f8).
  • An area skill no longer stops when its aimed target is dead (read, not run).
  • There are 21 party members.
  • The tables are rebalanced: stronger summons, 127 enemy rows changed, Kite starting at level 30, name fixes.
• Text (OUT on): ccKanjiStrlen counts every %x pair as a glyph. I read the branch myself:
  • INF SLUS_202.67:0x0015f1d0 adds 1 only for %# (and %%);
  • OUT SLUS_205.63:0x0015e260 adds 1 for any other pair.
• Unchanged: ChooseRoomSize, FOOT, MakeRoom, SetAllGim, every field constant and object table, the fonts, the trims, the colours, expCalcTbl and the erosion table.

Each test file gained a TestOtherVolumes class that runs a few cases per extracted volume and skips otherwise. The full suite, 14 files, passes with each file under two minutes.