← board

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 7 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_arr_of_ptr_elemrec_b354.pas red at 10dada0b7689 (auto-filed by twatch)

Repro

tools/testmgr.py --tier native --job 'test-core#src:test/test_arr_of_ptr_elemrec_b354.pas' at 10dada0b7689fee546516eec7ea90d1da4256053

Range

bad 10dada0b7689, last good d20300d288eb, 7 commit(s) in range — the watcher narrows this by idle bisect; check tstate/TSTATE.md for the current range.

Log tail

pascal26:51: error: "A": this value has no members (only records, classes, interfaces and variants do)
pascal26:51: error: "B": this value has no members (only records, classes, interfaces and variants do)
pascal26:51: error: "C": this value has no members (only records, classes, interfaces and variants do)
(tail)
pascal26:51: error: "A": this value has no members (only records, classes, interfaces and variants do)
  near: lst  i    >>> A    
pascal26:51: error: "B": this value has no members (only records, classes, interfaces and variants do)
  near: lst  i    >>> B    
pascal26:51: error: "C": this value has no members (only records, classes, interfaces and variants do)
  near: lst  i    >>> C

Diagnosis (Track A/P, 2026-08-25)

Self-inflicted by 582cdc934 (bug-p-a-member-on-a-computed-value-silently-reads- the-values-own-bytes). That fix put a recId = REC_NONE refusal inside the SHARED RequireRecMember, which is on all FOUR copies of member dispatch. Two of those are DEREF sites where REC_NONE is normal and harmless.

Measured, not reasoned: a temporary site-tag on the four call sites shows the refusal fires from pasparser_lval.inc:2710 (ParseSelectors), while the two tests the guard was written for hit pasparser_lval.inc:4204 and pasparser_expr.inc:1143 — the computed-receiver sites, and only those.

lst[i]^.A over an array[0..3] of PInner leaves the parser's running recName at REC_NONE (the leaking-LastTypePointerElemRec shape this very test was written for), yet the field still resolves correctly downstream — the pinned binary compiles it and prints 10 20 30. So the shared arm refused valid code, which is strictly worse than the silent path it replaced.

Reduction (/tmp/tt/r1..r3): the trigger is a decoy pointer type + a pointer LOCAL declared between the array alias and its use. Without the decoy the same program compiles either way.

Fix

REC_NONE means "no record id HERE"; only on a computed value does it also mean "and there never will be one". Split accordingly:

Verified: test_arr_of_ptr_elemrec_b354 prints 10 20 30; test_computed_member_fail and test_chained_helper_member_fail still refuse. tools/gate.sh quick GREEN (self-host fixedpoint converged).

Doctrine note for the next reader: the 582cdc934 write-up justified the shared placement with "every reader names a FIELD offset" and 141 lib units compiling unchanged. That sweep never compiled test/, which is where the counterexample lives — the widening was the guess, and Track T is what caught it.