← board

origin/master has advanced 2 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-nilpy#src:test/test_nilpy_delitem_dunder.npy red at 954727cee668 (auto-filed by twatch)

Repro

tools/testmgr.py --tier full --job 'test-nilpy#src:test/test_nilpy_delitem_dunder.npy' at 954727cee6680daf514fcd5bb929814a1ca3c522

Range

bad 954727cee668, last good 29d980110b58, 15 commit(s) in range — the watcher narrows this by idle bisect; check tstate/TSTATE.md for the current range.

Log tail

pascal26:67: error: Nil Python: del is supported on a dict subscript, a list index, a list slice, or a class with __delitem__ (del d[k], del l[i], del l[a:b], del c[k])
(tail)
pascal26:67: error: Nil Python: del is supported on a dict subscript, a list index, a list slice, or a class with __delitem__ (del d[k], del l[i], del l[a:b], del c[k])
  near:  del o    >>>

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

FIXED 2026-08-09, Track A+N — a refusal claimed a del TARGET

Triage

Not a NilPy del bug and not caused by the pylib work in flight beside it. Verified with the control that removes the variable rather than by reading: HEAD and pinned BOTH fail, at different lines (HEAD 67, pinned 61 — pinned predates the whole __delitem__ feature), so it is pre-existing.

del o[7] on class OnlyDel — a class declaring only __delitem__.

Cause

The del lowering deliberately does not re-implement the postfix grammar: it parses c[k] as an ordinary expression and REWRITES the node it gets back. So it depends on WHICH node the subscript arms build, and it handles two shapes — an AN_CALL to __getitem__ when the class has one, and a plain AN_INDEX when it does not.

The not-subscriptable refusal added for [[bug-nilpy-subscript-read-without-getitem-yields-garbage]] replaced that second shape. A getter-less class now yields a run-time TypeError call node instead of an AN_INDEX, so the rewrite found nothing to rewrite and the statement failed to compile. The refusal is right for a READ; a del target is not a read.

That is the same trap as the __setitem__ arm fixed earlier today (bug-nilpy-setitem-without-getitem-write-does-not-compile): the subscript protocol has three members and the gate keeps being written in terms of one of them. __getitem__ present, __setitem__ present, __delitem__ present are three independent facts, and each non-read use has to say so.

Fix

PyInDelTarget, set around the single PyParseBoolExpr in the del parser and restored after. The refusal arm skips itself while it is set, so the plain AN_INDEX survives for the rewrite. Scoped rather than global state that lingers, and it is one flag rather than teaching the del lowering a third node shape — the rewrite's whole point is not to know about grammar.

Gate

test/test_nilpy_delitem_dunder.npy GREEN. Extended it so both halves of the flag are pinned: del o[7] must work AND v = o[7] on that same class must still raise TypeError. With only the first, inverting the flag would pass. Matches CPython. Self-host fixedpoint byte-identical.