Length of a dynamic array of Char returns 1

• Type: bug — Track P (Pascal lowering of Length in compiler/ir.inc).
• Found: 2026-08-30 by frankwasm, while A/B-ing Length across shapes to prove the feature-unicodestring-model item-5 change was neutral. The neutrality held (pinned == new on every row); this row was wrong on both sides, so it predates that work entirely.

Measured

{$mode objfpc}{$H+}
var w: array of Char; d: array of WideChar; b: array of Byte; i: Integer;
begin
  SetLength(w, 6); SetLength(d, 3); SetLength(b, 6);
  for i := 0 to 5 do w[i] := Chr(65 + i);
  WriteLn('char  len=', Length(w), ' high=', High(w));
  WriteLn('wide  len=', Length(d));
  WriteLn('byte  len=', Length(b));
  WriteLn('elems=', w[0], w[5]);
end.

The shape of the failure is what makes it worth a ticket rather than a note:

• The array is fine. High(w) is 5 and w[0]/w[5] read back the bytes that were written, so the allocation, the element size and the indexing are all correct. Only the Length answer is wrong.
• It is the ELEMENT TYPE, not dynamic-arrayness. array of WideChar and array of Byte, allocated and measured identically in the same program, are both right. Swapping Char for either fixes it.
• It is silent and it under-reports. for i := 0 to Length(w)-1 runs once instead of six times; for i := 0 to High(w) runs six times. Two spellings of the same loop over the same variable disagree, and neither errors.

What has already been eliminated

• IRCoerceCharArrayArg is NOT the cause. It is the obvious suspect — it sits in the argument loop directly above the tkLength arm and exists to turn a Char array into a string — but it exits on cpi < 0, and Length is a special with cpi = -Ord(tkLength), so it never runs for this call.
• It is not the item-5 shr 1. That is guarded on the argument being tyAnsiString, which a dynamic array is not, and pinned (which predates it) answers 1 too.

The remaining suspect, unverified: something on the Length path treats an array whose element is Char as a managed string and reads its block header, or converts it through the char-array-to-string wrapper, either of which would yield a one-character answer. Confirm with PXXDBG=a.ir:<proc> before writing a cause into this ticket — devdocs/dev/root-cause-over-microfix.md, and the standing rule that every wrong root cause in this repo was a plausible story nobody diffed against an oracle.

Related, but not this

• [[bug-p-a-char-array-is-not-a-string-in-any-direction]] (done) is the STATIC array[0..N] of Char case, and it is about assignment/comparison/Write, not Length. This is the dynamic array, and only Length.
• [[refactor-p-the-char-array-is-not-a-string-rule-is-spelled-five-times]] is the standing five-copies refactor over the same rule. If that consolidation is done first, check whether this falls out of it — a sixth site spelling the rule slightly differently is exactly the failure mode that ticket names, and that would make this a tickets-closed-per-change win rather than a microfix. Prefer that order if both are open.

Gate

make compiler/pascal26 + the program above matching FPC on all four lines.

Log

• 2026-08-30 — resolved, commit d79ee7c95.