← board

regression CASCADE: 12 jobs newly red at 4e27dc2be114 (auto-filed by twatch)

Repro (start with a suspect, or any listed job)

tools/testmgr.py --tier native --job '<job>' at 4e27dc2be11471f92064756e17d426617d0d284b

Newly red jobs

Cascade stub: one signal for one event. Track T agent (face 2) or the owning dev track triages the root; individual tickets only for whatever remains red after the root is fixed.

Triage — 2026-08-19, against a self-host fixedpoint at 5b1539fc7

Each of the 12 was reproduced and rewritten individually, not sed-ed as a set. They are two different problems that shared one error message.

Half 1 — test fixtures importing Pascal units. FIXED here.

test/nilpy_units/{fmtprobe,samenamebase,mixinproto,mixinproto2,cprobe_unit,basehook}.pas are Pascal units, and these .npy files imported them by bare name. That is exactly the spelling the new rule closes, and the compiler names the replacement in its own diagnostic. Rewritten to import '<name>.pas' as <name>, each with a comment saying why:

file verified
test_nilpy_array_of_const_unit.npy x:2
test_nilpy_class_named_after_its_imported_base.npy override: derived / inherited: pascal-side / base: base pascal-side
test_nilpy_multiple_inheritance_imported_base.npy all four lines
test_nilpy_qualifier_vs_cproc.npy main / bye
test_nilpy_subclass_unit_base.npy override: KeepCase / inherited: keepcase
test_nilpy_two_imported_bases_fail.npy reaches its own wall (below)

Verified by running each one's exact Makefile assertion — the recipe's printf comparison and -Futest/nilpy_units flag, not just the compile. All seven PASS (renamed_class_attrs included, see below).

The negative test was the one that needed care. test_nilpy_two_imported_bases_fail.npy asserts that two imported bases are REFUSED. It was still "failing to compile", so the ! in front of it was satisfied — but for the wrong reason: it now died at the import wall, and the paired grep -q 'names TWO base classes whose bodies are not in this file' caught it. That grep is the only thing standing between this shape and a test that passes while asserting nothing. After the rewrite it reaches the real refusal at line 18 and the grep matches.

The shape to check for, generally. A negative test whose only assertion is a leading ! asserts "something went wrong", not "the right thing went wrong" — so any rule change upstream of it can move the failure to a new wall and the test stays green while asserting nothing. Whenever a refusal test sits downstream of a changed rule, check that it still dies at ITS OWN message; the paired grep -q is what makes that checkable, and a _fail/_reject test without one cannot be checked at all. Third instance of this family on 2026-08-19.

Two jobs in the red set needed no edit at all and were failing as recipe-block collateral:

lib/rtl/configparser.pas needs no edit (raised as a possible Track B question): configparser is already in the curated PyRtlUnitServesPython list, so it still resolves by bare name. It was only a downstream source in a job line.

Half 2 — the six examples/tk/*.npy. NOT a test bug; blocked.

These are not fixtures that happen to import a Pascal unit. lib/pcl/tkinter.pas was written to BE Python's tkinter, and says so in its own header — the missing entry in the curated list is the defect. Rewriting the examples would put the spelling every real tkinter program uses out of NilPy's reach. Filed as bug-n-tkinter-is-missing-from-the-python-serving-unit-list (Track A, compiler/parser.inc) with the one-line fix measured: with tkinter added to the list, all six compile and no example or test needs an edit.

Not applied here — it is a shared parser.inc edit and the A/P slot is held.

One thing lost, and it is a Track U question

There is no spelling for a from-import of a Pascal unit: from 'basehook.pas' import ConfigBase is refused with expected a module name after from. So test_nilpy_subclass_unit_base.npy lost one of its three assertions; the other two (dotted base, unqualified base) still hold and its header records what went and why. Whether the quoted form should be accepted after from is noted in the Track A ticket.

Unblocked and moved to backlog/ by the coordinator, 2026-08-28. Its declared blocked-by names a ticket that has since been resolved, so this was sitting in blocked/ — which ready/next never scan — while it was actually rankable. Nothing about the work changed; only the record was stale. Found by a sweep (see chore-t-nothing-re-checks-a-blocked-by-edge-after-its-blocker-closes); 14 tickets repo-wide carry at least one blocked-by naming a closed ticket, five of them fully unblocked.

RESOLVED, 2026-08-30 (frankA) — verification only, no code change

All 12 jobs of the cascade are green, plus the two that were failing as recipe-block collateral. Nothing needed fixing: half 1 was fixed in the 2026-08-19 triage, and half 2's blocker (bug-n-tkinter-is-missing-from-the-python-serving-unit-list) has since been resolved — tkinter is in PyRtlUnitServesPython (compiler/pasparser_proc.inc:2847, which is where parser.inc's copy went in the P/A split). The ticket was open only because nothing re-checked it after the blocker closed.

Verified against a self-host fixedpoint at 22c67e5ea61e, by running each job's exact Makefile assertion — the -Futest/nilpy_units flag and the expect_same.sh comparison — not merely the compile.

job verdict
examples/tk/tkinter_facade.npy compiles (compile-only per recipe: needs an X display)
examples/tk/field_class_identity.npy compiles (compile-only)
examples/tk/callbacks.npy compiles (compile-only)
examples/tk/import_in_body.npy in a suite left / before / after both
examples/tk/shadow_format_except.npy module function / Tap BPM: 92.5 / caught: clipboard
examples/tk/facade_and_paths.npy all six lines
test_nilpy_array_of_const_unit.npy x:2
test_nilpy_class_named_after_its_imported_base.npy all three lines
test_nilpy_multiple_inheritance_imported_base.npy all four lines
test_nilpy_qualifier_vs_cproc.npy main / bye
test_nilpy_renamed_class_attrs.npy all six lines
test_nilpy_subclass_unit_base.npy override: KeepCase / inherited: keepcase
test_nilpy_two_imported_bases_fail.npy (collateral) refused, and at its own message
test_nilpy_kwargs_by_name.npy (collateral) all four lines

The negative test was checked the way this ticket's own triage said to check it. two_imported_bases_fail is satisfied by any failure, so ! alone would pass on a compiler that died at the import wall instead of the real refusal. Its paired grep -q 'names TWO base classes whose bodies are not in this file' matches, so it still dies at ITS OWN wall.

Three "output mismatches" appeared in the first sweep and were my harness, not the tests: I extracted TESTTMP from the Makefile without expanding its $(shell ...) calls, so the binaries were written to a literal path and the runs produced a bash error. Re-run with a real directory, all three pass. Recording it because a mismatch on a just-unblocked ticket is exactly the shape one would be tempted to file.

The one lost assertion is tracked, not dangling

The triage flagged that from 'basehook.pas' import ConfigBase has no spelling (expected a module name after from), costing test_nilpy_subclass_unit_base.npy one of its three assertions. Still true at this binary — and it was escalated properly rather than guessed: answered in decided/decide-should-from-accept-a-quoted-foreign-file, and carried by feature-n-from-accepts-a-quoted-foreign-file [N p45] and feature-n-a-quoted-from-import-reaches-another-language. The assertion comes back with that feature, not with this ticket; the test's header already records what went and why.

blocked-by: still names the closed blocker — left as the historical record of why this sat in blocked/; the coordinator's 2026-08-28 note above is the live statement.

Log