FPC seed drift: `PyMakeTruthy` forward is in the wrong file

Type: regression, cold-start bootstrap path — Track A (compiler/**)
Filed: 2026-08-01 by Track T from the fpc-bootstrap canary.
Advisory, per testmgr.fpc_canary_job(): a red here is a notice for Track A, not a gate on anyone's push. Nothing day-to-day uses the FPC seed — which is exactly why it rots silently.
Culprit: 01b0b32f4 "fix(N): bool() must consult __bool__/__len__, via the shared truthiness rule". Broken at 2a3fd4404bdb, still broken at f5de12d6e.

The failure

$ fpc -Mobjfpc -O2 -Tlinux -Px86_64 -FU/tmp/x -FE/tmp/x -o/tmp/c compiler/compiler.pas
parser.inc(10074,19) Error: Identifier not found "PyMakeTruthy"
compiler.pas(1031) Fatal: There were 1 errors compiling module, stopping

Why — an ordering problem, not a missing forward

The forward exists, in the wrong file:


used at	`parser.inc:10074` — included by `compiler.pas:86`
forward at	`pyparser.inc:181` — included by `compiler.pas:100`
defined at	`pyparser.inc:1827`

That forward fixes the use-before-definition within pyparser.inc; its own comment says so and cites the earlier drift ticket. But the new use is in parser.inc, 14 includes earlier, where the identifier is not yet known.

The project already has the right home for this. compiler.pas:68-72:

{ PXX_NEED_FORWARDS := FPC or PXX_REQUIRE_FORWARD. PXX without it prescans and
  skips this; placed before symtab.inc so it is above every use it covers. }
{$ifdef FPC}{$define PXX_NEED_FORWARDS}{$endif}
{$ifdef PXX_NEED_FORWARDS}{$include forwards.inc}{$endif}

forwards.inc exists precisely for declarations FPC needs and PXX does not, and it is included before everything. It does not currently mention PyMakeTruthy.

Fix — MOVE the forward, do not add one (verified)

The obvious patch does not work. Adding it to forwards.inc while leaving pyparser.inc:181 in place gives:

pyparser.inc(181,10) Error: Function is already declared Public/Forward
                            "PyMakeTruthy(LongInt):LongInt;"

— the duplicate-forward shape FPC rejects and PXX tolerates, already on record as drift shape 2 in the earlier seed-drift ticket. So:

add to compiler/forwards.inc: function PyMakeTruthy(n: Integer): Integer; forward;
delete compiler/pyparser.inc:181 (the now-duplicate forward).

Verified on this box — with exactly those two edits, fpc … compiler.pas exits rc=0, the cold-start path is restored. Reverted afterwards; nothing committed to compiler/** from Track T.

Safe for PXX: it prescans and skips forwards.inc entirely (PXX_NEED_FORWARDS is FPC-only), so removing the in-file forward costs it nothing.

The recurring pattern is the real story

This is the third instance today of one shape: a helper is added and used above (or before) its declaration, PXX resolves it because it prescans, FPC does not, and only the advisory canary notices.

morning: 28b15ad3f "restore FPC bootstrap — helpers were declared AFTER their first use"
earlier: the regression-fpc-seed-drift-b1976-stale batch
now: this one

The mechanism to prevent it already exists and works; what is missing is that a cross-file use has to be registered in forwards.inc, and nothing enforces that at the time of writing. A cheap guard would be worth more than the next three fixes: the canary already runs in native/limited/full, so the feedback exists — it is just advisory, so it is easy to walk past.

FIXED by Track A — `9d2d98856` (verified by Track T, 2026-08-01)

"fix(A): forward-declare PyMakeTruthy in parser.inc — FPC seed build was red".

Verified on this box at HEAD: fpc -Mobjfpc -O2 -Tlinux -Px86_64 … compiler/compiler.pas reports 0 errors. The cold-start path is restored.

Track A forward-declared it in parser.inc, above the use, rather than moving it to forwards.inc as this ticket proposed. Both work and the canary is green either way, so this is not a correction — but the difference is worth recording, because the two placements are not equivalent going forward:

forwards.inc is included FPC-only (compiler.pas:72, PXX_NEED_FORWARDS), so declarations there cost PXX nothing and sit above every use by construction.
a forward in parser.inc is compiled by both front ends and only covers uses below that point in that file.

Neither is wrong. The one that matters is that the next cross-file helper hits the same wall, which is the open half of this ticket and is not addressed by either placement: nothing enforces registering a cross-file use at the time of writing, and the canary that catches it is advisory by design.

Log

2026-08-01 — resolved, commit 9d2d98856.

FPC seed drift: PyMakeTruthy forward is in the wrong file