A one-element unpack target binds nothing

Found 2026-09-20 (frankH) censusing That Space Program under NilPy. tsp/stars.py:25 is (n,) = struct.unpack_from("<I", data, 6) and is that file's only wall.

def f():
    return [5]

(n,) = f()      # pxx: undefined variable (n)
print(n)
a, = f()        # pxx: undefined variable (a)  -- no parentheses, same failure
print(a)
(b, c) = [1, 2] # compiles
print(b + c)

CPython prints 5, 5, 3.

The two controls are the finding. The reading everyone reaches for first is that the PARENTHESES confuse the target scan; the bare a, = f() spelling rules that out, and (b, c) = f() compiling rules out the other side. What is left is the element COUNT.

When this is fixed, the fixture must carry BOTH spellings — parenthesised and bare — because a fix that special-cases the parenthesised form would pass a fixture written with only that one, and the bare form is the shape that proves the scan is arity-driven rather than punctuation-driven.

Resolution (2026-09-20, frankH)

Fixed. tupleTarget — set by the trailing COMMA and by nothing else — replaces nTargets > 1 at the two gates that decide whether a single value is INDEXED or stored whole, and PySkipOneTupleTargetGroup admits both parenthesised spellings. (c) = v was refused on the pin and at HEAD alike and is fixed with it: it binds the whole value, as c = v does, and it is the sibling that makes the comma readable as the discriminator — a fixture asserting (c,) without asserting (c) cannot show that the comma is what decides.

The first cut broke (b, c) = xs, which had always worked, and the reason belongs here rather than only in the commit: the target loop's nested-group arm ends in Continue, which jumps to the until and skips any tail. The rule had been written as a tail, so it existed for the plain-name spelling only. The remedy was to move it into the until EXPRESSION, which every arm reaches by construction. That is normalise-dont-special-case.md's second-path rule arriving inside ONE LOOP rather than across two files — grepping for the other file could not have found it, and the fixture did, because it asserts the neighbouring shapes and not only the two being fixed.

Fixture: test/test_nilpy_a_one_element_unpack_target_binds_the_element.npy, eleven rows against CPython 3.14.4, carrying both one-element spellings, both no-comma forms, both multi-element forms, the starred target, the nested group, and struct.unpack_from — the shape it was found on. Pinned control reds at the one-tuple with the original undefined variable (n). tsp/stars.py compiles.

Log: fixed and closed, commit dc17f567e. That sha is the FIX — the parser change, the fixture and the Makefile row — and the close rode in the same commit; there is no separate close to cite.