@obj.Method as a call argument to a named method-pointer param: no overload match

• Type: bug (overload matching / parser). Compile error, not silent — pxx rejects a valid FPC program.
• Track: P (Pascal frontend overload matching of @obj.Method against an of object parameter type). The IR lowering for such a param already exists (IRLowerCallArg method-pointer-param path) — the arg just never reaches it because matching fails first.
• Found: 2026-07-15 while fixing [[bug-a-method-pointer-virtual-captures-static-address]] (the assignment form fn := @obj.Method works and is now VMT-correct; only the direct-argument form fails).

Repro

{$mode objfpc}{$H+}
type TB = class function M(a: longint): longint; virtual; end;
     TFn = function(a: longint): longint of object;
function TB.M(a: longint): longint; begin Result := a; end;
procedure Call(fn: TFn); begin Writeln(fn(5)); end;
var b: TB;
begin b := TB.Create; Call(@b.M); b.Free; end.

pxx: error: no overload of Call matches these arguments. FPC compiles and runs (prints 5). The assignment var fn: TFn; fn := @b.M; Call(fn); works.

Likely area

The overload matcher does not treat an @obj.Method (AN_METHODREF, typed tyPointer in a value/arg context — see the parser note at the @obj.Method build site) as assignment-compatible with a named of object record-type parameter. The IR side already materialises the {Code,Data} pair for a tyRecord method-pointer param (IRLowerCallArg), so this is purely a match-time gap: accept an AN_METHODREF argument for a method-pointer-typed (of object) parameter, mirroring the assignment-context acceptance.

Acceptance

The repro compiles and prints 5; a test/ regression covers @obj.Method passed as an argument (both virtual and non-virtual); the event-handler idiom (Register(@Self.Handler)) is exercised.

Log

• 2026-07-15 — resolved, commit c74f63f6.