← board

The mismatch warning fires even when crtl owns the symbol

Symptom

#include <math.h>
#include <stdio.h>
int main(void){ printf("%g\n", floor(2.5)); return 0; }



warning: C declaration of 'floor' disagrees with the Pascal routine 'Floor' on
         the result type (float vs non-float) — binding to the C declaration
warning: C declaration of 'ceil'  disagrees ... — binding to the C declaration
warning: C declaration of 'nan' does not match the Pascal routine 'NaN' which
         takes 0 parameter(s), not 1 — binding to the C declaration

Three warnings for a five-line program. A C program that does not include <math.h> is clean, so this is on every math-using C compile — which is most of the corpus (lua, quickjs, cJSON, zlib, tcc).

Why it is noise rather than a signal

The warning is accurate and its resolution is correct. What it is missing is that the disagreement is now INTENTIONAL and already resolved properly:

Two languages, two contracts, each implemented on its own side. "Binding to the C declaration" is exactly right here, and it binds to crtl's body — verified: that regression test now shows libc.so.6 present and libm.so.6 ABSENT.

nan is the same shape and predates my change: crtl defines nan(const char *) and the Pascal NaN is paramless.

Suggested fix

Do not warn when the C side has a DEFINITION for the symbol — only when the mismatch would silently route a call somewhere surprising. The warning exists for the case where a Pascal routine quietly captures a C call ([[bug-c-pascal-math-names-hijack-libc-through-pxxcio]] is the real hazard, and notably it is SILENT because those collisions have matching arities). Firing when crtl owns the body inverts that: it is loud where things are fine and quiet where they are not.

If suppressing entirely is too broad, gate it behind a flag, or downgrade to a note.

Gate

The five-line program above compiles with no warnings; make test and the C suites stay green; the genuinely dangerous same-arity hijack case still warns or is otherwise caught (test/cmath_no_pascal_hijack.c).

No longer reproduces (verified 2026-08-11, claude-an-1)

The ticket's repro (#include <math.h> + floor(2.5)) compiles with no warning at all on pinned (v256) and HEAD, and runs correctly (2). The "2-3 'disagrees with the Pascal routine' warnings on every math-using compile" are gone. Fixed before v256; closing as no-longer-reproducing.

Log