A local bound to both a Pascal class and its subclass loses subscripting

Measured 2026-09-09 (compiler 14e1b089f0c9)

import array
class Grid(array.array):
    pass

g = array.array("h", bytes(2))
print(g[0])          # CPython: 0   NilPy: TypeError: object is not subscriptable
g = Grid("h")

The failing subscript is the one written BEFORE the subclass binding, which is what says this is about the name's resolved type rather than about the subscript or about the subclass instance.

Controls, which is what narrows it

case	result
same name bound twice to the SAME Pascal class (`array.array("d")` then `array.array("h")`)	works, matches CPython
a `Grid` instance subscripted, ONE binding only	works
pure NilPy classes with `__getitem__`, rebound base -> subclass	works, matches CPython
Pascal class + subclass, two bindings	aborts

So: not rebinding in general (test_nilpy_rebind_across_unrelated_classes covers the unrelated case and passes), not subclassing in general, not the default property in general. It needs all three — a Pascal default indexed property, a NilPy subclass of that Pascal class, and two bindings of one name whose classes are related by inheritance.

Why it is prio 30

It is LOUD — an unhandled TypeError at run time, naming the operation. Nothing in the corpora does this; it was found by writing a test that reused a variable name, and the reuse was the test's own sloppiness. It is filed because the measurement is real and the narrowing is already done, not because anything is blocked on it.

Adjacent and probably the same machinery: the module-global binding resolver collapsed on 2026-09-09 (PyModuleBindingsOf) deliberately returns EVERY binding site rather than the first, because the three callers disagree about a name rebound at module scope. This is the LOCAL twin of that question and the answer here is currently "one binding decides every site".