Track guessed as P from the test source. The ranker reads frontmatter, so an unset track parks a stub in Track T's queue regardless of what the body says -- correct the track: line if this is wrong.

origin/master has advanced 1 commit(s) since this sha. Re-verify at current HEAD before acting — the callback is tagged to the sha that was tested, which may no longer be the state of the tree.

regression: test-core#src:test/test_integer_longint_overload.pas red at 943c706936b3 (auto-filed by twatch)

Type: regression (auto-filed by Track T watcher, host plexus). Untriaged.
Found: 2026-08-20T09:25:51Z
Test source: test/test_integer_longint_overload.pas test/test_integer_longint_overload.expected

Repro

tools/testmgr.py --tier native --job 'test-core#src:test/test_integer_longint_overload.pas' at 943c706936b329e1777d68892c8e4eb444211ea8

Range

bad 943c706936b3, last good b24bb1474624, 3 commit(s) in range — the watcher narrows this by idle bisect; check tstate/TSTATE.md for the current range.

Log tail

ok: /tmp/testmgr-scratch-326413/test_int_longint_ovl26  [code=290947B  data=8676B  bss=53888B  procs=581]
--- test/test_integer_longint_overload.expected	2026-08-16 09:00:26.625847420 +0200
+++ -	2026-08-20 11:24:39.593906650 +0200
@@ -1,3 +1,3 @@
 FFFFFFFF FFFFFFFF FFFFFFFF
-FFFFFFFFFFFFFFFF FFFFFFFF
+FFFFFFFFFFFFFFFF FFFFFFFFFFFFFFFF
 longint longint int64 longword

Stub ticket: signal only. Track T agent (face 2) enriches or a dev track takes it from the repro line.

Triaged and fixed — one cause behind all three

Cause: f09524494 (bug-p-unary-minus-on-an-unsigned-operand-truncates-to-32-bits) made -x yield Int64 for every integer operand. That is FPC's rule for a negated VARIABLE — and it is NOT FPC's rule for a negated CONSTANT, which is folded first and typed by its VALUE: the smallest signed type that holds it. So -1 is a LongInt under FPC and became an Int64 here, and three tests saw it:

test	what moved
test_asm_emit	`-7` in an `array of const` became vtInt64, so the `vtInteger` reader skipped the entry and the `I=-7` line vanished
test_integer_longint_overload	`IntToHex(-1, 8)` bound the Int64 overload and printed 16 digits
test_strict_overload_width	the `literal` row moved from longint to int64

Measured against fpc -O- -Mobjfpc, which types every one of these LongInt.

Fix: ParseFactorCore's tkMinus arm asks ASTConstIntValue (new, in parser.inc) whether the operand is an integer constant expression. If it is, the AN_NEG node is typed by the folded value — tyInteger when it fits a signed 32-bit, tyInt64 otherwise — and only otherwise does the Int64 widening apply. The variable rule is untouched: -b on a Byte is still Int64.

Nine FPC-verified rows now agree exactly, including the range edges (-2147483648 is LongInt, -2147483649 is Int64), a named constant, and a constant expression (-(1+2)). Pinned in test/test_unary_minus_constant_keeps_longint.pas (13 assertions, FPC's own answers) alongside the three tests above.

2026-08-20 — resolved, commit 584c2703d.