serp-itertools

Drop-in facade for CPython's itertools, written in the Serpentine subset. Iterators are real lazy objects (classes implementing __iter__/__next__), so count() and cycle() are genuinely infinite, next(it) works, and for x in it: drains them exactly like CPython. Swap import serp_itertools for import itertools on the supported surface.

Verified against the real itertools with a randomized differential (8801/8801 checks) and byte-identical across CPython + both native backends.

Supported API

count(start=0, step=1), cycle(xs), repeat(obj[, times]), chain(a, b), islice(xs, stop | start, stop[, step]), compress(data, selectors), accumulate(xs), pairwise(xs), product(a, b), zip_longest(a, b, fillvalue), takewhile(pred, xs), dropwhile(pred, xs), filterfalse(pred, xs), batched(xs, n), permutations(xs[, r]), combinations(xs, r), combinations_with_replacement(xs, r).

Divergences from CPython

• Inputs are list[T] (the subset's iterable) with Copy element types; iterators snapshot their input at construction time.
• chain/product/zip_longest take exactly two inputs of the same element type (no variadic forms).
• permutations/combinations/combinations_with_replacement/batched yield list[T] rows instead of tuples (row arity is not statically known).
• repeat(x, times) uses times=-1 as the "not given" sentinel (infinite); CPython treats negative counts as empty.
• islice uses stop=-2 as the "one-argument form" sentinel and stop=-1 for CPython's None (unbounded).
• accumulate supports plain + accumulation only (no func/initial).
• zip_longest requires fillvalue; count is int-only.
• starmap, groupby and tee are not provided (*args application, nested group iterators, and tuple-of-iterators returns are out of subset).

Install

serp add serp-itertools
pip install serp-itertools