origin/master has advanced 3 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_nilpy_min_max_variadic.npy red at 9305672dbcd5 (auto-filed by twatch)

Type: regression (auto-filed by Track T watcher, host xeon). Untriaged.
Found: 2026-08-04T03:59:18Z
Test source: test/test_nilpy_min_max_variadic.npy

Repro

tools/testmgr.py --tier native --job 'test-core#src:test/test_nilpy_min_max_variadic.npy' at 9305672dbcd5f66880058f28120d237032e6fedc

Range

bad 9305672dbcd5, last good c51086424657, 3 commit(s) in range — the watcher narrows this by idle bisect; check tstate/TSTATE.md for the current range.

Log tail

pascal26:5: error: no overload of min matches these arguments
(tail)
pascal26:5: error: no overload of min matches these arguments
  argument types: (Integer, Integer, Integer)
  candidates:
    min(Int64, Int64, Int64, Int64, Int64)
  near:       >>>

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

Fixed 2026-08-04 (claude-AN — author of the offending commit)

Caused by 8a660bce7, "a def shadowing a builtin is registered at all". That commit made PyRegisterDefShells register a shell for a def whose name only exists in a UNIT — necessary, because otherwise def len(x) was never registered and no shadowing could work at all. But a shell exists from the top of the module, and this test encodes the rule that makes that wrong:

print(min(3, 1, 2))                       # CPython: 1 — the BUILTIN
...
def min(a, b, c, d, e) -> int: return 100
print(min(1, 2, 3, 4, 5))                 # CPython: 100 — the user's

Python rebinds a name when the def STATEMENT runs, so a call ABOVE the def still reaches whatever the name meant before. With the shell visible from the top, PyUserShadowsProc answered True for the earlier calls too, the overload demote dropped every pylib min, and min(3, 1, 2) was left with no candidate at all — "no overload of min matches these arguments … candidates: min(Int64 ×5)".

The test was right and the compiler was wrong; the expectation was not touched.

Fix: carry the def's POSITION

Reverting the shell would have re-broken len (without a shell the name is not visible at all — measured: len([1,2]) went back to the builtin). Both rules hold together instead:

ProcPyDefTok[], a parallel array beside ProcUnitIdx/ProcHdrTok (per [[project_tsymbol_field_landmine]] — a parallel array, not a TProc field), records the token index of every NilPy module-level def;
PyUserShadowsProc accepts a main-program candidate only when its ProcPyDefTok is 0 (everything that is not a NilPy module def) or at or before the current call site.

Deliberately NOT ProcHdrTok: that field is compared against CurBodyHdrTok under RequireForward and means a different thing.

Verified

test_nilpy_min_max_variadic diffs identical to CPython again, and so do all fifteen tests touched today, re-run individually against the oracle — including the len shadowing this had to keep working. tools/gate.sh quick GREEN, self-host byte-identical.

Note the loop that caught it: the shadowing commit gated green on quick, and test-core is a full-tier job, so the watcher found it and pinned the bad sha within the hour. Working as designed.

2026-08-04 — resolved, commit a6754ddf7.