IntToStr of a QWord above 2^63 renders it negative

uses SysUtils;
var q: QWord;
begin
  q := $8000000000000000;
  WriteLn(q);              { 9223372036854775808  — correct }
  WriteLn(IntToStr(q));    { -9223372036854775808 — WRONG }
end.

fpc 3.2.2 prints 9223372036854775808 for both. Every QWord with the top bit set is affected, through every operator that produces one:

Cause

lib/rtl/sysutils.pas declares exactly one overload:

function IntToStr(value: Int64): AnsiString;

FPC declares IntToStr(Value: QWord): string alongside it. Without the QWord arm the argument converts to Int64 and the top bit becomes a sign.

This is a missing overload, not an overload-resolution bug — verified separately: with both a Sig(Int64) and a Sig(QWord) declared in user code, pxx picks the QWord arm for a QWord argument, and even for a q shl 1 expression. So adding the declaration is the whole fix.

Note the asymmetry that makes this easy to miss: StrToQWord, StrToQWordDef and TryStrToQWord all exist in the same unit. Only the rendering direction is missing. And WriteLn(q) is correct, so the value is right everywhere except the one path most code uses to build a string.

Fix

Add to lib/rtl/sysutils.pas, interface and implementation:

function IntToStr(value: QWord): AnsiString;

rendering unsigned. FPC also has UIntToStr as an explicit spelling; worth adding at the same time since it costs nothing once the digit loop exists.

Check whether Format's %d/%u and IntToHex have the same gap before closing — same unit, same shape of omission.

Track

Track B: lib/rtl/sysutils.pas is B's file and the fix is an RTL declaration. Filed from Track A, which found it but does not own the file. Build with $(PXX_STABLE); no compiler change is involved.

Found by

An integer-arithmetic differential (34 programs). It surfaced disguised as a shift bug — IntToStr(u64 shl 63) was the failing row — and only isolating IntToStr(u64) on a plain variable showed the shift had nothing to do with it. WriteLn(u64 shl 63) was correct all along.

Log

• 2026-08-27 — resolved, commit 0b2f731b6.

Resolution (2026-08-27, frankB)

Fixed as one of a three-ticket cluster — see [[bug-b-inttostr-of-a-qword-prints-it-signed]] for the full write-up. In short: lib/rtl/sysutils.pas had no unsigned decimal renderer at all, so one (UIntToStr(QWord)) was added and IntToStr(QWord) plus Format's now-split %u arm both route through it, instead of each carrying its own signedness decision. Regression: test/lib_qword_render.pas + .expected in make lib-test, 29 rows byte-identical to fpc 3.2.2.