Decompile 42 functions in world.c and type the chunk/triangle records by devcheckra1n · Pull Request #134 · Xeeynamo/ff7-decomp

devcheckra1n · 2026-08-31T02:04:09Z

42 functions in world.c, 144 down to 102.

two types came out of it. WorldTriangleRef {WorldChunkHeader* chunk;
WorldTriangle* tri; s8 unk8} lives at D_8010C83C and is what D_8010CA1C points
at, and WorldChunkNode {next; WorldChunkHeader* chunk} is the D_800E582C pool.
D_8010CA24 is a WorldChunkHeader*[20] with D_8010CA74 as the bump pointer over
it, so the bound that shows up as &D_8010CA74 in the asm is just
&D_8010CA24[20].

some existing declarations were wrong rather than missing:

WorldZolomSegment.x and .z are u16. retail reads them with lhu and never
sign-extends. they had only ever been written before, so nothing caught it.
func_800A19FC returns s32, not void. its only caller ignored the result.
D_80109D3C is a WorldChunkHeader*; D_800E5810 and D_800E580C are
WorldListNode*.

three things decided most of these, in case it saves anyone time:

an outer if plus an inner while on the same condition beats a do/while. gcc
drops the redundant guard, where an explicit do/while gets rotated and costs
a j and a duplicated load.
assigning a local inside the if rather than at its declaration keeps the
global's address inside the branch, which is where retail puts it.
p->r0 = p->r1 = c->r loads once and stores twice. as two statements it
reloads, since the first store may alias.

what didn't come out, left as INCLUDE_ASM:

func_800A5AD8 is one word short on a load-delay nop, and it compares
chunk->x and chunk->z as one 32-bit word. that needs a union on
WorldChunkHeader or it's a fake match, so it's out either way.
func_800A4138 and func_800B58F8 both have gcc hoisting the 2 of
GetGraphType() == 2 into an extra callee-saved register where retail
materialises it inline.
func_800A12AC and func_800AE4B8 are the right length with the right
instructions and differ only in prologue order.
func_800ADB30 wants D_8010AE24..30 as four independent symbols, which the
array typing can't produce.

func_800B6348, func_800B5DD8 and func_800B85D4 need structs that their
still-asm callers build, so they are left alone rather than guessed at.

Xeeynamo

Most changes are required based on the new style guide:
https://github.com/Xeeynamo/ff7-decomp/blob/main/STYLE.md

Xeeynamo · 2026-09-03T22:18:51Z

+    s32 col;
+    s16 idx;
+
+    if (D_800E5814 == 0 && D_800E5810 != NULL) {


Suggested change

if (D_800E5814 == 0 && D_800E5810 != NULL) {

if (D_800E5814 == 0 && D_800E5810 != NULL) {

return;

}

early return?

Xeeynamo · 2026-09-03T22:18:59Z

+    s16 idx;
+
+    if (D_800E5814 == 0 && D_800E5810 != NULL) {
+        if (TryAcquireLoadSlot(0) != 0) {


early return here too?

Xeeynamo · 2026-09-03T22:19:39Z

+    if (c != NULL) {
+        do {
+            if ((((c->z >> 2) * 9) + (c->x >> 2)) == arg0->unk4) {
+                func_800A5AD8(c);
+            }
+            c = c->next;
+        } while (c != NULL);
+    }


Suggested change

if (c != NULL) {

do {

if ((((c->z >> 2) * 9) + (c->x >> 2)) == arg0->unk4) {

func_800A5AD8(c);

}

c = c->next;

} while (c != NULL);

}

while (c != NULL) {

if ((((c->z >> 2) * 9) + (c->x >> 2)) == arg0->unk4) {

func_800A5AD8(c);

}

c = c->next;

}

Xeeynamo · 2026-09-03T22:19:58Z

+    if (chunk != NULL) {
+        do {


Suggested change

if (chunk != NULL) {

do {

while (chunk != NULL) {

this is also a while loop

Xeeynamo · 2026-09-03T22:20:27Z

+    WorldChunkHeader* chunk;
+
+    chunk = D_80109D3C;
+    if (chunk != NULL) {


Suggested change

if (chunk != NULL) {

isn't this if unnecessary for the match?

Xeeynamo · 2026-09-03T22:37:20Z

+    WorldChunkHeader* chunk;
+    SVECTOR* verts;
+
+    if (arg1 != NULL) {


early return

Xeeynamo · 2026-09-03T22:37:50Z

+        i++;
+    } while (i < 0x2B);


Xeeynamo · 2026-09-03T22:37:58Z


-INCLUDE_ASM("asm/us/world/nonmatchings/world", func_800B6C84);
+void StartStreamRead(void) {
+    if (D_80115A60 == 0) {


early return

Xeeynamo · 2026-09-03T22:38:56Z

+extern u8 D_8010CB24[];
+extern s32 D_8010D930;
+extern u8* D_8010D9A4;
+extern s32 D_8010D9A8;
+extern s32 D_8010D9AC;
+extern s32 D_8010D9B0;
+extern s32 D_8010D9B4;
+extern s32 D_80115A44;
+extern s32 D_80115A4C;
+extern s32 D_80115A54;
+extern u8 D_80115A6C[];
+extern u8* D_800C80BC;
+extern u32 D_8014A608;
+extern s32 D_8014A610;
+extern s32 D_800C74C4[][2];
+extern s32 D_80115A48;
+extern s32 D_80115A5C;
+void func_800B6AEC(void);
+extern u8 D_800C6940;
+extern u8 D_800C6A10[];
+extern MATRIX D_800C6808;
+extern MATRIX D_800C6828;
+extern MATRIX D_800C6848;
+extern s32 D_8010CB20;
+extern u8 D_800C6748[];
+extern s32 D_8010AD50;
 extern u8 D_8010D9BA[];
+extern u16 D_8009D2A6;


These are not sorted correctly

Xeeynamo · 2026-09-03T22:39:14Z

+extern s32 D_800C74C4[][2];
+extern s32 D_80115A48;
+extern s32 D_80115A5C;
+void func_800B6AEC(void);


devcheckra1n · 2026-09-04T02:40:44Z

all applied, pushed as 52473e2, 15 overlays match.

the D_8010B434 one was a good catch. 0x8010B47C - 0x8010B434 = 0x48, so exactly two 0x24 entries, DR_MODE at 0x0 and 0x18 unidentified after it. matched first try with &D_8010B434[i].mode at both call sites.

two things that would not take the rewrite as-is:

the triangle flag loop at 2848 is a plain for over WorldTriangle* now, so the walking pointer is gone. the flag itself has to stay a byte write though. retail does lbu / andi 0xbf / sb at +11, and textureAndLocationAndFlags being a u16 gives lhu / andi 0xbfff / sh instead. splitting 0xA into two bytes would fix it but it cuts through the 9 bit texture id, so i left the field alone and wrote the flag as a byte into it. happy to change the struct instead if you would rather have that.

the range condition at 2881 is half done. the outer test reads pos.vx >= 0x36000 && pos.vx <= 0x3FFFF and matches fine. the inner one on pos.vz costs a register in every readable form i tried; the subtraction lands in a0 instead of v0 and the following sltu follows it. so that one still has the unsigned subtract.

Three list initialisers: func_800A59A0 and func_800A60D8 thread the 0x40 WorldListNode entries at D_800E582C and the 0x20 WorldChunkHeader entries at D_80109A38, and func_800B392C threads the 32 0x74-byte records at D_8010CB24 backwards from its last element. func_800A0BE4 is the bump allocator over the buffer D_800D05DC points at, taking the panic path when the 0x20800-byte arena would overflow. func_800A7EA4, func_800B5274 and func_800B667C are subsystem resets. The rest: func_800A5B88 walks the loaded chunks for the one whose 9-column grid index matches, func_800AD928 reports whether any actor still has a script running, func_800AD970 packs an actor's position, type and direction into a two-word display record, func_800AF110 picks the light and colour matrices for the current mode, func_800B65E0 drives the AKAO fade commands, func_800B6C84 and func_800B7620 handle the streamed read for slot 2 and report its timing, and func_800B7AC0 copies a name out of the table func_800A40F0 indexes. D_8009A000, D_800E582C and D_80109A38 gain real types; the two ascending free-list loops derive their &array[1] base from the array itself, so the symbols splat named there need no declaration. func_800B6AEC, func_800B6DCC and func_800BBA34 gain prototypes because their addresses are taken.

…unk registration

…list handling

No func_ADDR left in the file's decompiled half. Most are clear from the code -- the fog ranges, the two bounded layer scrolls, the AKAO track and ambient calls, the stream state and its read. UpdateWorldMode, ResetEffectState and GetSavedParams describe what the code does rather than claim a purpose I could not pin down.

…order Rewrite the hand-expanded do/while loops as plain for and while, invert the guard conditions into early returns, drop the redundant chunk-search guard, give D_8010B434 a real struct so the draw-mode setup indexes it, and put the added externs in address order with the stray prototypes back in the prototype block.

Xeeynamo#142 named 13 of the same functions this branch did, and named 27 more that it calls; take upstream's names for all of them rather than renaming merged work. Xeeynamo#141 renamed DS_read to SysCdromStartLoadLzs.

devcheckra1n · 2026-09-06T16:22:39Z

@Xeeynamo ready to merge

… from parkedfuncs Three of the four functions parkedfuncs had beyond what PR Xeeynamo#134 shipped: InitSpritePrims needed plain array indexing over a hand-rolled byte offset, the skybox quad setup needed explicit address bindings for two new CVECTOR globals nothing had referenced before, and the shadow quad setup matched as-is. The fourth, WmLoadTxzFile (parkedfuncs called it ReadFileRetry), was a bad parked claim -- retail branches on useDsRead before calling a specifically-named SysCdromStartLoadLzs, not a function-pointer dispatch between two generic readers -- so it stays INCLUDE_ASM.

Xeeynamo · 2026-09-07T11:31:24Z

+    a = &D_800C6768;
+    b = &D_800C676C;
+    do {
+        p = (POLY_G4*)((i * 0x24) + (s32)D_800C6770);


Suggested change

p = (POLY_G4*)((i * 0x24) + (s32)D_800C6770);

p = &D_800C6770[i];

with the real D_800C6770 type, the code can be simplified

Xeeynamo · 2026-09-07T11:33:09Z

+    if (WmGetWmId() == mode) {
+        r = 0x20;
+        g = 0x20;
+        b = 0x30;
+    } else {
+        r = 0x40;
+        g = 0x40;
+        b = 0x40;
+    }
+    SetBackColor(r, g, b);


Suggested change

if (WmGetWmId() == mode) {

r = 0x20;

g = 0x20;

b = 0x30;

} else {

r = 0x40;

g = 0x40;

b = 0x40;

}

SetBackColor(r, g, b);

if (WmGetWmId() == mode) {

SetBackColor(0x30, 0x30, 0x30);

} else {

SetBackColor(0x40, 0x40, 0x40);

}

can be simplified, which also allows the removal of s32 r, g, b

Xeeynamo · 2026-09-07T11:36:30Z

+    s32 inside;
+    s32 zoff;
+
+    if (D_8010CA8C == 0 || D_8010CA8C == 2 || D_8010CA8C == 3) {


Early return, as per https://github.com/Xeeynamo/ff7-decomp/blob/main/STYLE.md

Xeeynamo · 2026-09-07T11:40:34Z

+void WmCreateShadowPacket(u8* prims, RECT* rect) {
+    POLY_FT4* p;


Suggested change

void WmCreateShadowPacket(u8* prims, RECT* rect) {

POLY_FT4* p;

void WmCreateShadowPacket(POLY_FT4* prims, RECT* rect) {

can you change the type and get rid of POLY_FT4* p;?

Xeeynamo · 2026-09-07T11:42:34Z

+    if (D_8010CB20 < arg0) {
+        D_8009A000[0] = 0x20;
+        D_8010CB20 = arg0;
+        D_8009A004[0] = 0x40;
+        D_8009A008[0] = arg0;
+    } else if (arg0 == -D_8010CB20) {
+        D_8010CB20 = 0;
+        D_8009A000[0] = 0xF1;
+        SystemAkaoExecute();
+        D_8009A000[0] = 0xBC;
+        D_8009A004[0] = 0;
+    } else {
+        return;
+    }
+    SystemAkaoExecute();


Suggested change

if (D_8010CB20 < arg0) {

D_8009A000[0] = 0x20;

D_8010CB20 = arg0;

D_8009A004[0] = 0x40;

D_8009A008[0] = arg0;

} else if (arg0 == -D_8010CB20) {

D_8010CB20 = 0;

D_8009A000[0] = 0xF1;

SystemAkaoExecute();

D_8009A000[0] = 0xBC;

D_8009A004[0] = 0;

} else {

return;

}

SystemAkaoExecute();

if (D_8010CB20 < arg0) {

D_8009A000[0] = 0x20;

D_8010CB20 = arg0;

D_8009A004[0] = 0x40;

D_8009A008[0] = arg0;

SystemAkaoExecute();

} else if (arg0 == -D_8010CB20) {

D_8010CB20 = 0;

D_8009A000[0] = 0xF1;

SystemAkaoExecute();

D_8009A000[0] = 0xBC;

D_8009A004[0] = 0;

SystemAkaoExecute();

}

did you try this?

InitSpritePrims and the walk-to-tail loops become plain for/while, the tpage vsync window check drops the unsigned-subtract trick for a readable range, and func_800AA640 drops a redundant cast. Kept the (s16) cast on ExpireChunks' age compare; removing it (as asked, with a question mark) breaks the match, verified by rebuild.

devcheckra1n · 2026-09-07T16:17:55Z

applied the rest, pushed as a248107.

InitSpritePrims and the walk to tail loops (AllocRegionNode, UpdateRegionLoad, WmGetElementWithBlockIdAndSetItFirst, IsRegionLoading, ExpireChunks, AllocChunk, WmAbortMapLoading) are plain for/while now, dropped the redundant if (x != NULL) { while (x != NULL) guards. the fade overlay vsync check lost the unsigned subtract trick for a readable range test. func_800AA640 dropped a redundant cast.

kept the (s16) cast on ExpireChunks' age compare though, tried removing it and it breaks the match on rebuild. figured you meant it as a question rather than a firm ask given the wording.

Xeeynamo

The comments regarding D_800C6770 have not been addressed yet.

…WmCreateShadowPacket

devcheckra1n · 2026-09-07T21:01:49Z

D_800C6770 typed as POLY_G4, both call sites simplified. also went back through the rest of the review and found four more that hadn't actually landed: WmRestoreEntityPosAndDirFromSavemap and SetCurrentTriangle are early returns now, WmSetGteColourSettings drops r/g/b for direct SetBackColor calls, WmCreateShadowPacket takes POLY_FT4* directly, ToggleAmbientSound duplicates the SystemAkaoExecute call into each branch instead of a shared tail call. pushed as c1ffd83 and 2ebfc7d, 17 overlays match.

Xeeynamo requested changes Sep 3, 2026

View reviewed changes

halkuncode mentioned this pull request Sep 5, 2026

WorldMap function labels #142

Merged

devcheckra1n force-pushed the worldfunc3 branch from 52473e2 to 973c5b2 Compare September 5, 2026 03:58

devcheckra1n added 10 commits September 5, 2026 09:04

world: decompile the camera transform, heading blend and actor placement

71dc6e6

world: decompile the akao cue, actor collision, minigame state and ch…

5e6e700

…unk registration

world: decompile the chunk/triangle record, draw-mode setup and free-…

7dcad70

…list handling

world: decompile the chunk cache, list searches and zolom segment lookup

ecd4f1a

world: name 22 of the decompiled functions

98bc3ec

world: reflow to the new 120 column limit

26507a6

world: adopt the names Xeeynamo#142 and Xeeynamo#141 landed

adee19b

Xeeynamo#142 named 13 of the same functions this branch did, and named 27 more that it calls; take upstream's names for all of them rather than renaming merged work. Xeeynamo#141 renamed DS_read to SysCdromStartLoadLzs.

devcheckra1n force-pushed the worldfunc3 branch from 973c5b2 to adee19b Compare September 5, 2026 13:15

Xeeynamo requested changes Sep 7, 2026

View reviewed changes

Xeeynamo reviewed Sep 7, 2026

View reviewed changes

devcheckra1n added 2 commits September 7, 2026 16:56

world: give D_800C6770 its real POLY_G4 type

c1ffd83

world: early returns, drop unnecessary locals, real POLY_FT4 type on …

2ebfc7d

…WmCreateShadowPacket

Xeeynamo approved these changes Sep 8, 2026

View reviewed changes

Xeeynamo merged commit cc83524 into Xeeynamo:main Sep 8, 2026
1 check passed

	p = (POLY_G4)((i 0x24) + (s32)D_800C6770);
	p = &D_800C6770[i];

		void WmCreateShadowPacket(u8* prims, RECT* rect) {
		POLY_FT4* p;

	void WmCreateShadowPacket(u8* prims, RECT* rect) {
	POLY_FT4* p;
	void WmCreateShadowPacket(POLY_FT4* prims, RECT* rect) {

Conversation

devcheckra1n commented Aug 31, 2026

Uh oh!

Xeeynamo left a comment

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

devcheckra1n commented Sep 4, 2026

Uh oh!

devcheckra1n commented Sep 6, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

devcheckra1n commented Sep 7, 2026

Uh oh!

Xeeynamo left a comment

Choose a reason for hiding this comment

Uh oh!

devcheckra1n commented Sep 7, 2026

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

devcheckra1n commented Sep 6, 2026 •

edited

Loading