WideChar variable → string: concat segfaults, arg mis-resolves (assign fixed)

• Type: bug (Track A/P — string concat + arg lowering / overload match). Sibling of [[bug-pascal-widechar-var-to-string-segfault]] (assign case, fixed b1cbd204).
• Status: done
• Found: 2026-07-17, context sweep right after fixing the assign case — a textbook demonstration of the shape/context whack-a-mole that [[refactor-centralize-managed-string-pchar-conversion]] exists to end.

Repro

procedure show(const s: AnsiString); begin writeln('arg=', s); end;
var s: AnsiString; w: WideChar;
begin
  w := WideChar($41);
  s := w;            { assign  — FIXED, b1cbd204 (prints A) }
  s := 'x' + w;      { concat  — FIXED, this session (single-sided string+ordinal) }
  show(w);           { arg     — STILL OPEN: compile error "Mismatch in MatchProcCall" }
end.

Remaining: only the ARG/overload case. show(w) (widechar var to a const AnsiString param) fails overload resolution — MatchProcCall does not treat a tyUInt16 arg as string-compatible. Lower-severity than the crashes (a compile error, not a segfault); needs the same widechar-string-compat rule at the call site, or the centralization refactor.

Root (same family, different sites)

WideChar collapses to tyUInt16 with no subtype marker (see the assign bug). The assign path was taught to wrap a tyUInt16 RHS in WrapWideCharToUTF8, but the same conversion is re-decided independently at:

• string concat lowering (+ with a string and a non-string operand): the tyUInt16 operand is not converted → treated as a pointer/wrong width → crash.
• argument matching (MatchProcCall): a tyUInt16 value is not considered string-compatible, so show(w) fails overload resolution (FPC accepts widechar→string implicitly).

Each site re-implements "is this a char-like value that must become a string?" — exactly the scatter the refactor targets.

Fix

• Narrow: apply the same tyUInt16 → WrapWideCharToUTF8 conversion at the concat site, and make MatchProcCall treat a tyUInt16 arg as string-compatible (wrapping at the call site). Point-fixes; more moles will follow.
• Systemic (preferred): [[refactor-centralize-managed-string-pchar-conversion]], extended to cover the WideChar→string conversion alongside PChar→string. One MaybeConvertToString(node) keyed on the operand's char-likeness, called from assign / concat / arg / return. Kills PChar and WideChar scatter together. Fourth instance this session — the evidence for centralizing is now overwhelming.

Acceptance

• All three lines above work (or arg errors cleanly, never the concat segfault); a test/test_*.pas regression per context.
• Gate: make test + self-host byte-identical.

Log

• 2026-07-18 — resolved, commit 3abb5529.