goldeneye-pc-port/docs/HANDOFF-ai-xmacro.md

RESOLVED — historical. D7 closed at the compile+link milestone. Kept for the AI X-macro analysis. Not active work.

HANDOFF — AI X-macro system (D7), the last build blocker

Status

Full ge007 build: 224 objects compile; only chraidata.c fails. All other .c files build clean. Findings D1–D6 are resolved and recorded in docs/PCPortResearch.md §11. D7 (this file) is the remaining blocker.

Build command:

export PATH="/c/msys64/mingw64/bin:$PATH"
cd build-pc && cmake --build . --target ge007 2>&1 | grep -A3 -i "chraidata\|aicommands\|bondaicommands"

The failing file and its includes

src/game/chraidata.c includes (lines 14–17):

#include <ultra64.h>
#include <bondgame.h>
#include <bondaicommands.h>   // <- pulls in the whole AI X-macro system
#include "chrobjdata.h"

bondaicommands.h includes:

• bondconstants.h (which at line 17 includes <CPPLib.h>, and at line 731 includes "aicommands.def" to build the AI_CMD enum)
• aicommands2.h (line 864) — a pre-generated header (from tools/cmdbuilder.c), already preprocessed.

So the failing path is the raw aicommands.def include (bondconstants.h:731) plus the bondaicommands.h macros (isSubroutine, SWITCH, CALL, …) that chraidata.c uses directly.

Exact error chain (verified by rebuild)

Trigger: CALL(GAILIST_PLAY_IDLE_ANIMATION) at chraidata.c:114 (and similar at 123, 126, …). The chain:

1. CALL(...) expands to code that calls isSubroutine(<ID>).
2. isSubroutine(ID) (bondaicommands.h:275–284) contains:

   #define isSubroutine(ID) ((ID == GAILIST_PLAY_IDLE_ANIMATION) |\
                               (ID == GAILIST_BASH_KEYBOARD) | \
                               (ID == GAILIST_ATTACK_BOND) | \
                               (ID == GAILIST_RUN_TO_BOND) | \
                               (ID == GAILIST_STARTLE_AND_RUN_TO_BOND) | \
                               (ID == GAILIST_WAIT_ONE_SECOND) \
                               IF(NOT(DEFINED(SETUPSUBROUTINES(ID))))\
                               (\
                                   | SETUPSUBROUTINES(ID)\
                               ))
   

3. DEFINED(x) (CPPLib.h:279–280):

   #define DEFINED(x) _DEFINED(x)
   #define _DEFINED(x) IS_PROBE(CAT(_DEFINED, _##x##_))
   

   With x = SETUPSUBROUTINES(ID), the ## pasting _##x##_ tries to paste _ against SETUPSUBROUTINES(ID) — a macro call (multiple tokens), not a single pp-token. GCC hard-errors:

   bondaicommands.h:281:63: error: pasting ')' and '_' does not give a valid
   preprocessing token
   

4. CPPLib.h:287 has a special case that was meant to handle this:

   #define _DEFINED_SETUPSUBROUTINES(ID) PROBE(~),
   

   i.e. the intent is that _##SETUPSUBROUTINES(ID)##_ → _DEFINED_SETUPSUBROUTINES(ID)_ → the defined macro above. But GCC's ## requires each side to be a single pp-token, so it never reaches that special case. IDO's preprocessor evidently tolerated / expanded this differently.

Secondary errors (same root family)

• chraidata.c:183/202/249/640: macro 'SWITCH' requires 49 arguments, but only 20/14/26/26 given. The SWITCH macro (bondaicommands.h, ~line 776) has 49 fixed parameters (1 VAR + 16×3 case slots) and uses IF_VA(NOT(IS_EMPTY(..))) to detect which cases are present. Game code calls it with fewer args, relying on IDO leniency. GCC enforces the exact arg count.
• aicommands.def:514:270 / 550:21 / 2846:21: "expected expression before ','" / "expected '}' before 'AI_Return'" / "expected '}' before 'AI_IFICouldSeeBond'". These are downstream of the isSubroutine/DEFINED failure (the CALL expansion aborts mid-macro, leaving dangling tokens). Note the reported line numbers (514/550/2846) do not match the raw aicommands.def lines (those are comment/brace lines) — they are positions in the expanded macro text, not the source file. Don't chase them as source lines.

Key facts / ground truth

• aicommands.def call counts: 255 _AI_CMD(, 119 _AI_CMD_POLYMORPH(, 2 _AI_DEBUG(.
• _AI_CMD_POLYMORPH definitions across the tree:
  • aicommands.def:97 — 4-arg (PRE, NAME, CMD_PARAMS_POST, POLY_PARAMS) but inside a #if 0 block (lines 88–100) → inactive.
  • bondconstants.h:721 — 5-arg (CMD, A, P, Q, D) under #ifdef __sgi (not us).
  • bondconstants.h:726 — variadic (...) under #else (this is the one active for GCC when building the AI_CMD enum at bondconstants.h:731).
  • chrai.c:170 — 5-arg (CMD, A, P, Q, D).
  • chrai.c:915 — 5-arg (C, N, P1, P2, D).
• aicommands.def:85–87: #ifndef __GNUC__ #include <CPPLib.h> #endif → for GCC CPPLib.h is not included from aicommands.def itself; it comes in via bondconstants.h:17.
• aicommands.def:11888: #undef _AI_CMD_POLYMORPH (late in the file).
• aicommands2.h is pre-generated (header says "Do not edit … generated by cmdbuilder from src/aicommands.def"). It is included by bondaicommands.h:864 and deb_loadallmodels.c:8. It is not the failing path.
• isSubroutine is the only macro that feeds a macro call (SETUPSUBROUTINES(ID)) into DEFINED(). That is the unique trigger for the ## pasting error.

Recommended next steps (in order)

1. Fix the DEFINED(SETUPSUBROUTINES(ID)) pasting — the single root cause of the CALL/isSubroutine errors. Preferred: a port/ shim that redefines DEFINED (or _DEFINED) after CPPLib.h is included, special-casing the SETUPSUBROUTINES(ID) argument so it does not ##-paste a multi-token macro-call. Sketch:

   /* port/shim/... after <CPPLib.h> */
   #undef DEFINED
   #define DEFINED(x) _PORT_DEFINED(x)
   #define _PORT_DEFINED(x) /* if x is SETUPSUBROUTINES(ID) -> PROBE(~), else fall back */
   

   The tricky part: detecting "x is exactly SETUPSUBROUTINES(ID)" in the preprocessor. Alternative: redefine isSubroutine in the shim to compute the SETUPSUBROUTINES presence without DEFINED() (e.g. via a per-setup #define HAS_SETUPSUBROUTINES 0/1 flag set before the include). Investigate which is less invasive before committing.
   • Constraint: bondaicommands.h and CPPLib.h are game files — do not edit them. All fixes go in port/ (shim headers found first via include path) or, if truly unavoidable, a -DPORT-gated src/ change with a PC port: comment.
2. Fix the SWITCH 49-arg requirement — make a shim SWITCH variadic (or pad the call sites). Verify the IF_VA/IS_EMPTY case-detection still works under GCC once variadic.
3. Re-verify the aicommands.def:514/550/2846 errors disappear once (1) and (2) are in — they are likely downstream of the CALL expansion abort, not independent bugs.
4. Rebuild and confirm chraidata.c compiles → full ge007 links.
5. Record the resolution in docs/PCPortResearch.md §11 D7 (replace the "OPEN" / "TBD" text) and delete this handoff file.

Do NOT

• Do not edit tools/, Makefile, rsp/, ld/ (N64 build, non-negotiable).
• Do not edit bondaicommands.h, bondconstants.h, CPPLib.h, aicommands.def, aicommands2.h unless a -DPORT-gated, layout/semantics- preserving change is truly unavoidable (and then flag it PC port:).
• Do not chase the aicommands.def:514/550/2846 line numbers as source lines (they are expanded-macro positions).