*= on a sequence yields empty

xs = [1, 2]
xs *= 2
print(xs)            # CPython [1, 2, 1, 2]     pxx printed nothing at all

s = "ab"
s *= 3
print(s)             # CPython ababab           pxx printed an empty line

Silent. Found 2026-08-15 by a CPython differential sweep of slicing and mutation. xs = xs * 2 was correct all along, which is the tell.

Cause

x * n on a sequence is not a multiply: ParseTerm decides it with three special arms — PyMakeListRepeat, PyMakeBytesRepeat, and the str case, which tags the binop tyAnsiString so the IR lowers it to pystr_repeat. The three augmented-assignment sites (bare name, lhs-expression target, and the shared dotted-target tail in parser.inc) build an AN_BINOP from the token PyAugBinTok returns and consult none of them, so the sequence handle went into an arithmetic multiply.

PyAugMulNode is that decision in one place, called from all three. It answers -1 for a pair that really is arithmetic, so every other *= keeps its path. This is the same shape as |= on a set, whose fix note two hundred lines away says the same thing about the same three sites — devdocs/dev/normalise-dont-special-case.md.

Divergence recorded, not hidden

CPython's list *= MUTATES in place, so an alias taken beforehand sees the repeat; this REBINDS. += has the in-place form already (TPyList.extend) and *= has no in-place primitive to call yet — filed as [[bug-nilpy-augmented-sequence-repeat-rebinds-instead-of-mutating]]. An empty list was the strictly worse answer, so the rebind lands now and the aliasing half is its own ticket.

Gate

test/test_nilpy_augmented_sequence_repeat.npy (+.expected, in the Makefile), byte-identical to CPython: *= on a list, a str and bytes; the int and float forms untouched; a FIELD target and a SUBSCRIPT target (the other two augmented sites); a variable count; counts of 0 and 1; x = x * n still agreeing; and += on a list still being the in-place extend, alias included. gate.sh quick GREEN. No pin — frontend-only.