8000 NEP: add NEP 56 on array API standard support in main namespace by rgommers · Pull Request #25542 · numpy/numpy · GitHub
[go: up one dir, main page]
Skip to content

NEP: add NEP 56 on array API standard support in main namespace #25542

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
merged 16 commits into from
Feb 28, 2024
Merged

NEP: add NEP 56 on array API standard support in main namespace #25542

Changes from 1 commit
Commits
Show all changes
16 commits
Select commit Hold shift + click to select a range
b871142
NEP: add NEP 56 on array API standard support in main namespace
rgommers Jan 5, 2024
c998fb2
DOC: rework for review comments: textual suggestions and easy changes
rgommers Jan 15, 2024
2262298
DOC: second batch of smaller textual changes
rgommers Jan 15, 2024
90df451
DOC: adopt suggestions on namespace and alias count/hiding
rgommers Jan 15, 2024
6bb9dd4
DOC: extend abstract with the main rationale and benefits
rgommers Jan 17, 2024
4803ba4
add section about which parts of the standard are not adopted
rgommers Jan 17, 2024
54a2723
Merge branch 'main' into nep-56-arrayapi
rgommers Jan 18, 2024
d9310c7
Add the in-place "no unsafe casting" item to the not adopted section
rgommers Jan 18, 2024
262b4a6
Merge branch 'main' into nep-56-arrayapi
rgommers Feb 13, 2024
808753c
Merge branch 'main' into nep-56-arrayapi
rgommers Feb 14, 2024
269d837
Merge branch 'main' into nep-56-arrayapi
rgommers Feb 15, 2024
191d4e1
NEP: update content mentioning array scalars
rgommers Feb 15, 2024
29eab50
NEP: improve backwards compatibility section and describe `copy=` better
rgommers Feb 16, 2024
60225ab
Merge branch 'main' into nep-56-arrayapi
rgommers Feb 20, 2024
115cf77
NEP: update description of `fft` changes
rgommers Feb 20, 2024
5188ca0
Merge branch 'main' into nep-56-arrayapi
rgommers Feb 28, 2024
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
Prev Previous commit
Next Next commit
Add the in-place "no unsafe casting" item to the not adopted section 
Also a few smaller textual changes.

[skip actions] [skip cirrus] [skip azp]
  • Loading branch information
@rgommers
rgommers committed Jan 18, 2024
commit d9310c73c7af3ad031f5677d6d6c428c9f596be1
26 changes: 20 additions & 6 deletions doc/neps/nep-0056-array-api-main-namespace.rst
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -14,8 +14,8 @@ NEP 56 — Array API standard support in NumPy's main namespace
Abstract
--------

This NEP proposes adding full support for the 2022.12 version of the array API
standard in NumPy's main namespace for the 2.0 release.
This NEP proposes adding nearly full support for the 2022.12 version of the
array API standard in NumPy's main namespace for the 2.0 release.

Adoption in the main namespace has a number of advantages; most importantly for
libraries that depend on NumPy and want to start supporting other array
Expand Down Expand Up @@ -227,9 +227,10 @@ their experience of using a stack of Python packages for scientific computing
or data science more seamless.

Finally, for authors of other array libraries as well as tools like Numba,
array API standard support should save them time. The design rules ([3]_), and
in some cases new APIs like the ``unique_*`` ones, are easier to implement on
GPU and for JIT compilers as a result of more predictable behavior.
API improvements which align NumPy with the array API standard will also save
them time. The design rules ([3]_), and in some cases new APIs like the
``unique_*`` ones, are easier to imp 8000 lement on GPU and for JIT compilers as a
result of more predictable behavior.


Backward compatibility
Expand All @@ -250,7 +251,6 @@ Raising errors for consistency/strictness includes:
2. Making ``cross`` error on size-2 vectors (only size-3 vectors are supported),
3. Making ``solve`` error on ambiguous input (only accept ``x2`` as vector if ``x2.ndim == 1``),
4. ``outer`` raises rather than flattens on >1-D inputs,
5. In-place operators are disallowed when the left-hand side would be promoted.

Dtypes of returned arrays for some element-wise functions and reductions
includes functions where dtypes need to be preserved: ``ceil``, ``floor``, and
Expand Down Expand Up @@ -493,6 +493,20 @@ to follow (at least at this time). These are:
already compliant), however there is no need to rush this change - doing so
for 2.0 would be unnecessarily disruptive.*

4. The requirement "An in-place operation must have the same behavior
(including special cases) as its respective binary (i.e., two operand,
non-assignment) operation" (excluding the effect on views).

*Rationale: the requirement is very reasonable and probably expected
behavior for most NumPy users. However, deprecating unsafe casts for
in-place operators is a change for which the impact is hard to predict.
Hence this needs to be investigated first, and then if the impact is low
enough it may be possible to deprecate the current behavior according to
NumPy's normal backwards compatibility guidelines.*

This topic is tracked in
`gh-25621 <https://github.com/numpy/numpy/issues/25621>`__.


Related work
------------
Expand Down
0