Re-tracked P -> A on 2026-08-20. The stub guessed P from the test source (compiler/compiler.pas), but the subject is not the Pascal frontend: it is the compiler failing to reproduce itself when built in --threadsafe mode. That is core self-host/codegen — Track A.

origin/master has advanced 4 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:compiler/compiler.pas@2 red at 57b9b7148d32 (auto-filed by twatch)

• Type: regression (auto-filed by Track T watcher, host plexus). Untriaged.
• Found: 2026-08-20T03:46:12Z
• Test source: compiler/compiler.pas tools/progress.sh

Repro

tools/testmgr.py --tier native --job 'test-core#src:compiler/compiler.pas@2' at 57b9b7148d3290ac089cd9360c6e4553a4b44bfb

Range

bad 57b9b7148d32, last good 003d733936aa, 7 commit(s) in range — the watcher narrows this by idle bisect; check tstate/TSTATE.md for the current range.

Log tail

/tmp/testmgr-scratch-3997779/pascal26-threadsafe-self /tmp/testmgr-scratch-3997779/pascal26-threadsafe-next differ: byte 97, line 1
(tail)
ok: /tmp/testmgr-scratch-3997779/pascal26-threadsafe-self.3999198.tmp  [code=8637667B  data=227176B  bss=211128564B  procs=2930]
ok: /tmp/testmgr-scratch-3997779/pascal26-threadsafe-next.3999198.tmp  [code=8637667B  data=227040B  bss=211128564B  procs=2930]
/tmp/testmgr-scratch-3997779/pascal26-threadsafe-self /tmp/testmgr-scratch-3997779/pascal26-threadsafe-next differ: byte 97, line 1

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

Verified at HEAD — 2026-08-20 (sha 9156a7c37)

It still reproduces. Run by hand rather than through a tier, so it costs ~90s instead of a suite (the two compiles the Makefile job does, nothing else):

./compiler/pascal26 --threadsafe compiler/compiler.pas /tmp/ts-self
/tmp/ts-self         --threadsafe compiler/compiler.pas /tmp/ts-next
cmp /tmp/ts-self /tmp/ts-next        # differ: byte 97, line 1

Measured, and this is the part that narrows it:

Same source, same --threadsafe flag on both compiles, identical code size — the whole delta is 32 bytes of data, and the only variable is whether the compiling binary was itself built threadsafe. So this is not "threadsafe output is wrong"; it is that a threadsafe-built compiler EMITS 32 data bytes fewer than a non-threadsafe-built one does, for identical input. One of the two is wrong and the flag is changing compiler behaviour it should not change.

Note ts-next's data size (227040) is exactly the default build's, which suggests ts-self is the odd one out — i.e. the non-threadsafe binary is emitting 32 bytes the threadsafe one does not. Confirm that direction before hunting: 32 bytes reads like four pointers or one small table/interned literal. PXXDBG and a data-section diff of the two outputs will say what the bytes are; do not theorise it.

Scope note it is worth being precise about: the ordinary self-host gate did NOT catch this and is not wrong — make compiler/pascal26 and tools/selfhost_fixedpoint.sh prove the fixedpoint at the DEFAULT flags, and both are green at this sha (see the sibling ticket). This is the same class as the -O0-only self-compile failure CLAUDE.md's claims section warns about: the fixedpoint holds at one build configuration.

Root cause found and fixed — 2026-08-20

Not a --threadsafe bug. The compiler's emitted output depended on argv[0]: ParseUsesUnit interned the resolved unit path — which begins with ExeDir, derived from ParamStr(0) — through InternStr, and InternStr appends its text to the emitted string pool. Every binary the compiler produced therefore carried the path the compiler had been invoked as.

The plain self-host chain passed only by coincidence (both generations run from $(TESTTMP) under equal-length names); the threadsafe chain starts from ./compiler/pascal26 and continues from $(TESTTMP)/pascal26-threadsafe-self, two different ExeDirs, so it was the one that showed the 32-byte data delta.

Full diagnosis, fix and verification: [[bug-a-the-compilers-output-depends-on-argv0]].

Verified for THIS job: ts-self and ts-next byte-identical, both data=226912B; tools/gate.sh quick GREEN.

• 2026-08-20 — resolved, commit 3b0a886e9.