← board

A set LITERAL keeps duplicates; set().add() removes them

s = {1, 2, 2, 3}
print(len(s))          # CPython: 3    pxx: 4



s = set()
s.add(1)
s.add(2)
s.add(2)
print(len(s))          # CPython: 2    pxx: 2      <- correct

A set is backed by TPyList and add appends only when the element is absent — that IS the set contract as designed (see the set branch of PyAnnTypeTermAt). The literal {...} does not go through add; it builds the list directly, so duplicates survive. So the design is intact and one construction path skips it.

Silent, and the shape it breaks is the one sets are used for: {...} as a membership/dedup set, then len() or an in scan over it.

Found by the data-structure sweep against CPython (aliasing, copy, nested lists and dicts, tuple unpacking, swap, unpacking in a for-loop, join, reversed, nested indexing) — everything else there matched, apart from [[bug-nilpy-sorted-over-tuples-or-lists-fails]].

Gate

make test-nilpy + self-host byte-identical, plus set literals with duplicate ints, strings and mixed types, and a literal followed by add of an element it already contains.

Log