Retire the `__crtl_*` dodge-prefixes in crtl math

lib/crtl/src/math.c defines ten functions under deliberately wrong names, with the reason in the source:

/* NOT named `exp`: that name collides case-insensitively with Pascal Exp
   (two definitions -> silently broken call binding). C callers reach this
   through `#define exp(x) __crtl_exp(x)` in crtl math.h. */
double __crtl_exp(double x) { ... }

__crtl_exp, __crtl_log2, __crtl_log10, __crtl_sin, __crtl_cos, __crtl_tan, __crtl_sinh, __crtl_cosh, __crtl_tanh, __crtl_hypot, plus the matching #defines in lib/crtl/include/math.h.

The collision they dodge no longer fires — measured

pxxcio.pas no longer does uses math (it is uses platform, builtinheap), so the Pascal RTL is not in scope for an ordinary C program at all. Probed 2026-08-14:

extern double Power(double, double); (Pascal-only) does not resolve — the compiler warns it will come from the system C library;
a C body named exp with no <math.h> macro in the way binds ITSELF. All ten spellings do, each measured in its own program.

So the hazard the prefixes exist for appears to be gone, and this may be a pure deletion.

Why it is filed separately from the rule

[[feature-a-own-language-first-symbol-resolution]] names de-prefixing these as its acceptance test. The measurement above says the two are independent: the C-to-Pascal direction is already closed, while the rule is needed for the Pascal-to-C direction (a Pascal program that uses './math.c' loses its own Exp — see that ticket). Blocking a cheap cleanup behind a large resolution change would be wrong, hence this ticket.

If de-prefixing turns out to fail, that failure is evidence FOR the rule and belongs back on that ticket — say so there rather than re-prefixing silently.

Work

De-prefix the ten in lib/crtl/src/math.c; delete the ten #defines in lib/crtl/include/math.h (keep the NAN one — different mechanism).
Check nothing else references the old names (grep -rn __crtl_ across lib/crtl and test/).
Run the C corpus. A silent binding change is the failure mode, so compare RESULTS, not just exit codes — tools/gcc_diff_probe.sh is the instrument for the math ones.

Also — three stale docs, same visit

devdocs/dev/c-linking-and-crtl-autopull.md still describes math.c's sqrt/sin/pow as "a thin bridge to the Pascal RTL". Stale — they have real double-double C bodies now (math.c:959 etc.).
lib/crtl/README.md and lib/crtl/src/README.md both still say src/ is "reserved for the matching implementations once real candidates need them". It holds a 114-function correctly-rounded libm, stdio, string, and more.

Background for all of it: devdocs/dev/math-implemented-twice.md.

Gate

C tests green, the corpus unchanged in OUTPUT (not merely in exit status), plus self-host byte-identical since nothing in compiler/** should move.

Done 2026-08-16 — a pure deletion, and the feared case measured

All ten de-prefixed; the ten #defines deleted from lib/crtl/include/math.h (the NAN union kept — different mechanism). No compiler change was needed, as the ticket predicted.

The one thing worth recording is the case the source comment kept them alive for. lib/crtl/src/math.c's header said the prefixes "must stay" because "while pxxcio no longer imports math, a user program that does uses math itself would resurrect exactly that". That is a testable claim and it is false:

program mixed;
uses math, './cm.c';        { BOTH an `exp` and an `Exp` are visible }
begin
  writeln(cexp2x(1.0):0:15);   { the C exp  -> 2.718281828459045 }
  writeln(Exp(1.0):0:15);      { the Pascal -> 2.718281828459045 }
end.

Both sides answer correctly. So the collision is closed in the C-to-Pascal direction by pxxcio dropping uses math, not merely hidden by the prefixes.

Verified by RESULTS, not exit codes, since a silent binding change is the failure mode: gcc_diff_probe.sh at 0 NEW divergences on x86-64, and all eighteen cmath_* tests plus crtl_trig_huge pass — including the four correctly-rounded suites whose expectations are 80-digit decimal references, so a wrongly-bound call could not pass them by luck.

Test files and the three docs that described the dodge (lib/crtl/src/README.md, devdocs/dev/math-implemented-twice.md, the math.c header) are updated with the change rather than left to rot, and each says what to do if it regresses: record it on [[feature-a-own-language-first-symbol-resolution]], do not re-prefix silently. That ticket is unaffected — it is needed for the Pascal-to-C direction, which is still open.

Log

2026-08-16 — resolved, commit 615b4b5ab.

Retire the __crtl_* dodge-prefixes in crtl math