MLX compatibility: Manipulation functions

Open

#462 opened on Aug 12, 2026 11 comments 0 reactions 0 assignees View on GitHub

Nobody has claimed this yet.

Assessment

Difficulty: 4/5
Estimated time: 3-5 days
Newbie friendliness: 48/100

Issue type: Bug
Clarity: Mostly clear
Activity status: Active
Tech stack: python
Domain: data

Research direction

Start by running the named compatibility tests: test_broadcast_arrays, test_moveaxis, test_unstack, and the expand_dims, repeat, reshape, and roll tests. Read the implementations and test expectations for each failed or partial case. Done means the MLX compatibility behavior matches the Array API requirements listed in the table, including tuple returns, tuple axes, accepted repeat inputs, copy handling, and keyword-only arguments.

Written by the indexing model from the issue text.

Description

Array API	MLX Analog	Status	Notes	Test Node	Result
broadcast_arrays(*arrays)	broadcast_arrays(*arrays, stream=None) -> list[array]	incompatible	Array API 2025.12 requires a tuple of arrays; MLX returns a list.	test_broadcast_arrays	Failed
broadcast_shapes(*shapes)	broadcast_shapes(*shapes)	compatible	MLX returns the expected broadcasted shape.	TestBroadcastShapes::test_broadcast_shapes	Passed
broadcast_to(x, /, shape)	broadcast_to(array, shape, *, stream=None)	compatible	Required broadcasting behavior passes. MLX adds the optional `stream` extension.	test_broadcast_to	Passed
concat(arrays, /, *, axis=0)	concat(arrays: list[array], axis: int \| None = 0, *, stream=None)	compatible	Required behavior passes. MLX additionally accepts `axis=None`, positional `axis`, and `stream`.	test_concat	Passed
expand_dims(x, /, axis)	expand_dims(a, /, axis: int \| Sequence[int], *, stream=None)	partial	Tuple-axis coverage passes, but MLX incorrectly rejects a valid negative insertion axis at the lower boundary.	TestExpandDims::test_expand_dims; TestExpandDims::test_expand_dims_tuples	Failed / Passed
flip(x, /, *, axis=None)	flip(a, /, axis: None \| int \| Sequence[int] = None, *, stream=None)	compatible	Passes for `None`, integer, and tuple axes. MLX additionally permits positional `axis`.	test_flip	Passed
moveaxis(x, source, destination, /)	moveaxis(a, /, source: int, destination: int, *, stream=None)	incompatible	Array API accepts integers or tuples for `source` and `destination`; MLX accepts only scalar integers and rejects valid calls such as `moveaxis(x, (), ())`.	test_moveaxis	Failed
permute_dims(x, /, axes)	permute_dims(a, /, axes: Sequence[int] \| None = None, *, stream=None)	compatible	Required axis permutation behavior passes. MLX additionally permits omitted or `None` axes.	test_permute_dims	Passed
repeat(x, repeats, /, *, axis=None)	repeat(array: array, repeats: int, axis: int \| None = None, *, stream=None)	partial	Array API allows scalar or 1D integer-array `repeats`, while MLX documents only scalar integers.	test_repeat	Failed
reshape(x, /, shape, *, copy=None)	reshape(a, /, shape: Sequence[int], *, stream=None)	partial	Tested behavior passes, but MLX lacks the standard `copy` keyword and therefore cannot accept all conforming calls.	test_reshape	Passed
roll(x, /, shift, *, axis=None)	roll(a, shift: int \| tuple[int], axis: None \| int \| tuple[int] = None, /, *, stream=None)	partial	Tested behavior passes, but MLX makes `shift` and `axis` positional-only. Array API permits keyword `shift` and requires `axis` to be keyword-only.	test_roll	Passed
squeeze(x, /, axis)	squeeze(a, /, axis: None \| int \| Sequence[int] = None, *, stream=None)	compatible	Integer and tuple-axis behavior passes. MLX additionally allows omitted or `None` axes.	test_squeeze	Passed
stack(arrays, /, *, axis=0)	stack(arrays: list[array], axis: int \| None = 0, *, stream=None)	compatible	Required behavior passes. MLX additionally accepts positional `axis`, `axis=None`, and `stream`.	test_stack	Passed
tile(x, repetitions, /)	tile(a, reps: int \| Sequence[int], /, *, stream=None)	compatible	Required tiling behavior passes. The parameter-name difference is irrelevant because it is positional-only.	test_tile	Passed
unstack(x, /, *, axis=0)	unstack(x, /, *, axis: int = 0, stream=None) -> list[array]	incompatible	Array API requires a tuple of arrays; MLX returns a list, including `[]` instead of `()` for an empty axis.	test_unstack	Failed

Dominant language: Python
Stars: 131
Forks: 49
Avg merge: 2d 8h
Merged PRs (30d): 6

Contributor guide

Open the contributing guide

First steps

Read the whole issue, then the project's contributing guide.
Comment on the issue to say you are picking it up — it saves two people doing the same work.
Fork the repository and make your change on a branch.
Open a pull request that references the issue number.

Similar issues

validate-structure inspecte challenge/work/.terraform et rend faux les labs qui ont un provider en cache Open

bug

Difficulty 2/5 1-3 hours Newbie friendliness 75/100

stephrobert/dsoxlab#238 · Opened Sep 24, 2026
Uniform probability densities use the upper bound as an interval width Open

Difficulty 2/5 1-3 hours Newbie friendliness 75/100

nasa/fmdtools#40 · Opened Sep 24, 2026
_common_folder() treats a single file as the common folder; single-file archives can't be installed (4.2.8) / crash (master) Open

Difficulty 2/5 1-3 hours Newbie friendliness 75/100

sublimehq/package_control#1780 · Opened Sep 24, 2026
Set 'Sample' input in 'autorun.py' to utilize 'Sample' Tab in eLog Open

Difficulty 2/5 1-3 hours Newbie friendliness 65/100

pcdshub/mfx#309 · 2 comments · Opened Sep 24, 2026
niri config file addition does not respect existing optional include Open

Difficulty 2/5 1-3 hours Newbie friendliness 70/100

nwg-piotr/nwg-displays#145 · Opened Sep 24, 2026

More Python issues