DEP: Deprecate numpy.fix #30098

hellerve · 2025-10-30T01:20:34Z

This PR fixes #30096 (especially its follow-up comments) by deprecating numpy.fix.

Cheers

ngoldbaum

Thanks! Probably best to wait for someone else to weigh in before implementing my suggestion. Otherwise this all looks great.

numpy/lib/_ufunclike_impl.py

hellerve · 2025-10-30T02:39:55Z

It seems that my last error is due to this section: https://github.com/numpy/numpy/pull/30098/files#diff-6cdc3b490da76e0f287028f0c0fa780b24100106713a4216594f8d33d162fec6R36-R47

which is test data, but not picked up as tests, and a check complains about warnings.filterwarnings being used outside of tests (full message: AssertionError: ignore filters should only be used in tests; found in C:\hostedtoolcache\windows\Python\3.12.10\x64\Lib\site-packages\numpy\typing\tests\data\pass\ufunclike.py on line 37). Not sure how to resolve that best.

doc/release/upcoming_changes/30098.deprecation.rst

seberg

Nice, thanks. LGTM, two nits to change docs and maybe add to test_deprecations.py which would be nice (just to discover it, since this isn't C, I don't care too much about it).

numpy/lib/_ufunclike_impl.py

numpy/lib/tests/test_ufunclike.py

ngoldbaum · 2025-10-30T15:10:49Z

numpy/typing/tests/data/pass/ufunclike.py

+        "ignore",
+        message="The function numpy.fix is deprecated in favor of numpy.trunc",
+        category=DeprecationWarning,
+    )


@jorenham is there a proper way to call things that trigger deprecation warnings in the typing test data? this seems to run afoul of the test that checks the codebase for warning filters.

We could also patch that test to skip the typing test data. Maybe it doesn't matter if the typing tests trigger deprecation warnings? I don't actually know.

Completely fine to change ignore it in those tests in principle, I think. The only reason for the filter in other tests was that it is nicer to have another solution (e.g. pytest.warns or @pytest.filterwarnings).

(The test comes from a different time and age and the original purpose is moot, now we just have nicer spellings :))

I am unsure what this means. Do I get rid of these calls completely or should I look at how to change the test script? I can do the latter, but that seems like a separate PR to me, right?

@jorenham is there a proper way to call things that trigger deprecation warnings in the typing test data? this seems to run afoul of the test that checks the codebase for warning filters.

We could also patch that test to skip the typing test data. Maybe it doesn't matter if the typing tests trigger deprecation warnings? I don't actually know.

Yea there is the typing_extensions.deprecated decorator that can be used in the stubs. It can be applied to classes, functions, or individual overloads. See for example

numpy/numpy/_core/arrayprint.pyi

Lines 96 to 188 in d8eb225

# public numpy export

@overload # no style

def array2string(

a: NDArray[Any],

max_line_width: int | None = None,

precision: SupportsIndex | None = None,

suppress_small: bool | None = None,

separator: str = " ",

prefix: str = "",

style: _NoValueType = ...,

formatter: _FormatDict | None = None,

threshold: int | None = None,

edgeitems: int | None = None,

sign: _Sign | None = None,

floatmode: _FloatMode | None = None,

suffix: str = "",

*,

legacy: _Legacy | None = None,

) -> str: ...

@overload # style=<given> (positional), legacy="1.13"

def array2string(

a: NDArray[Any],

max_line_width: int | None,

precision: SupportsIndex | None,

suppress_small: bool | None,

separator: str,

prefix: str,

style: _ReprFunc,

formatter: _FormatDict | None = None,

threshold: int | None = None,

edgeitems: int | None = None,

sign: _Sign | None = None,

floatmode: _FloatMode | None = None,

suffix: str = "",

*,

legacy: Literal["1.13"],

) -> str: ...

@overload # style=<given> (keyword), legacy="1.13"

def array2string(

a: NDArray[Any],

max_line_width: int | None = None,

precision: SupportsIndex | None = None,

suppress_small: bool | None = None,

separator: str = " ",

prefix: str = "",

*,

style: _ReprFunc,

formatter: _FormatDict | None = None,

threshold: int | None = None,

edgeitems: int | None = None,

sign: _Sign | None = None,

floatmode: _FloatMode | None = None,

suffix: str = "",

legacy: Literal["1.13"],

) -> str: ...

@overload # style=<given> (positional), legacy!="1.13"

@deprecated("'style' argument is deprecated and no longer functional except in 1.13 'legacy' mode")

def array2string(

a: NDArray[Any],

max_line_width: int | None,

precision: SupportsIndex | None,

suppress_small: bool | None,

separator: str,

prefix: str,

style: _ReprFunc,

formatter: _FormatDict | None = None,

threshold: int | None = None,

edgeitems: int | None = None,

sign: _Sign | None = None,

floatmode: _FloatMode | None = None,

suffix: str = "",

*,

legacy: _LegacyNoStyle | None = None,

) -> str: ...

@overload # style=<given> (keyword), legacy="1.13"

@deprecated("'style' argument is deprecated and no longer functional except in 1.13 'legacy' mode")

def array2string(

a: NDArray[Any],

max_line_width: int | None = None,

precision: SupportsIndex | None = None,

suppress_small: bool | None = None,

separator: str = " ",

prefix: str = "",

*,

style: _ReprFunc,

formatter: _FormatDict | None = None,

threshold: int | None = None,

edgeitems: int | None = None,

sign: _Sign | None = None,

floatmode: _FloatMode | None = None,

suffix: str = "",

legacy: _LegacyNoStyle | None = None,

) -> str: ...

Ah, I see, that's what's missing. This PR also needs to update the type stubs for np.fix:

numpy/numpy/lib/_ufunclike_impl.pyi

Lines 16 to 35 in d8eb225

@overload

def fix( # type: ignore[misc]

x: _FloatLike_co,

out: None = None,

) -> floating: ...

@overload

def fix(

x: _ArrayLikeFloat_co,

out: None = None,

) -> NDArray[floating]: ...

@overload

def fix(

x: _ArrayLikeObject_co,

out: None = None,

) -> NDArray[object_]: ...

@overload

def fix(

x: _ArrayLikeFloat_co | _ArrayLikeObject_co,

out: _ArrayT,

) -> _ArrayT: ...

If I use that decorator locally (after importing of course 🤦 ), I still seem to be getting the errors on the deprecation message. So I suppose the question remains whether the checks should be dropped from numpy/typing/tests/data/pass/ufunclike.py and numpy/typing/tests/data/reveal/ufunclike.pyi?

numpy/lib/_ufunclike_impl.py

numpy/lib/_ufunclike_impl.pyi

hellerve · 2025-10-30T20:31:39Z

So, it seems like the easiest way to deal with this deprecation is to

remove the examples from the doc string (those get checked and currently fail), and
remove the typing tests.

Both of these mean losing fidelity, but I’m not well-versed enough in the codebase to propose alternatives, unfortunately.

jorenham · 2025-10-30T20:46:12Z

2. remove the typing tests.

You could also smack a # type: ignore[deprecated] onto them, AFAIK.

hellerve · 2025-10-30T21:02:47Z

You could also smack a # type: ignore[deprecated] onto them, AFAIK.

I might be misunderstanding, but doesn’t that then defeat the prupose of the tests?

jorenham · 2025-10-30T22:47:18Z

You could also smack a # type: ignore[deprecated] onto them, AFAIK.

I might be misunderstanding, but doesn’t that then defeat the prupose of the tests?

it would only suppress the deprecated errors; so any other mypy errors would still be reported

rgommers

This deprecation looks justified (despite us not removing this for 2.0 because of concerns of widespread usage), but a blocking comment for merge with a few comments:

This should be proposed on the NumPy mailing list before it can proceed, like all deprecations.
This should not be merged this late in a release cycle, so please make it 2.5.0 not 2.4.0
This cannot be merged before downstream users of our nightlies have been updated. E.g., I see usages in scikit-image and in PyWavelets. PRs to replace those usages would be nice to help this move along.

hellerve · 2025-11-03T15:19:23Z

You could also smack a # type: ignore[deprecated] onto them, AFAIK.

I might be misunderstanding, but doesn’t that then defeat the prupose of the tests?

it would only suppress the deprecated errors; so any other mypy errors would still be reported

Looks like that has the same issue, since now the assertion error is: AssertionError: ignore filters should only be used in tests; found in /numpy/venv/lib/python3.11/site-packages/numpy/typing/tests/data/pass/ufunclike.py on line 37. So neither using warning filters nor ignore filters seems to work here.

ngoldbaum · 2025-11-03T15:20:41Z

This should not be merged this late in a release cycle, so please make it 2.5.0 not 2.4.0

Fair enough - I cleared the 2.4.0 milestone.

@hellerve - if you're interested, it might be a fun exercise to contribute to a bunch of other popular projects as well if you could take a shot at doing the migration in a few popular packages. In addition to PyWavelets and scikit-image, which both looks like they use fix in a way that can be straightforwardly replaced by trunc, It looks like pandas has two uses that can be replaced by trunc and explicitly tests support for fix in one spot. That test will likely need to handle the new deprecation warning.

It also looks like astropy has a few uses, but only to test or support wrapping of NumPy. That code might also need to handle the deprecation warning. Probably worth raising an issue over in astropy.

It looks like jax implements np.fix - maybe they care about this change. Pythran also has some tests that will need updating.

SciPy and matplotlib don't use it, as far as I can see.

ngoldbaum · 2025-11-03T15:22:40Z

Looks like that has the same issue, since now the assertion error is: AssertionError: ignore filters should only be used in tests; found in /numpy/venv/lib/python3.11/site-packages/numpy/typing/tests/data/pass/ufunclike.py on line 37. So neither using warning filters nor ignore filters seems to work here.

IMO it's fine to add an exception to the test that checks for warning filters to ignore code in the typing/tests/data directory. You'd have to add a new if block here:

numpy/numpy/tests/test_warnings.py

Lines 67 to 74 in 1e424da

    
           if base / "testing" in path.parents: 
        
               continue 
        
           if path == base / "__init__.py": 
        
               continue 
        
           if path == base / "random" / "__init__.py": 
        
               continue 
        
           if path == base / "conftest.py": 
        
               continue

hellerve · 2025-11-03T15:23:33Z

This should not be merged this late in a release cycle, so please make it 2.5.0 not 2.4.0

Fair enough - I cleared the 2.4.0 milestone.

@hellerve - if you're interested, it might be a fun exercise to contribute to a bunch of other popular projects as well if you could take a shot at doing the migration in a few popular packages. In addition to PyWavelets and scikit-image, which both looks like they use fix in a way that can be straightforwardly replaced by trunc, It looks like pandas has two uses that can be replaced by trunc and explicitly tests support for fix in one spot. That test will likely need to handle the new deprecation warning.

It also looks like astropy has a few uses, but only to test or support wrapping of NumPy. That code might also need to handle the deprecation warning. Probably worth raising an issue over in astropy.

It looks like jax implements np.fix - maybe they care about this change. Pythran also has some tests that will need updating.

SciPy and matplotlib don't use it, as far as I can see.

I feel totally up to it (more than for the mailing list proposal, anyway 😅). I’ll take a look and start contributing PRs. Is there a script you use to check for usages or just go through a set of known repositories and check?

hellerve · 2025-11-03T15:24:10Z

Looks like that has the same issue, since now the assertion error is: AssertionError: ignore filters should only be used in tests; found in /numpy/venv/lib/python3.11/site-packages/numpy/typing/tests/data/pass/ufunclike.py on line 37. So neither using warning filters nor ignore filters seems to work here.

IMO it's fine to add an exception to the test that checks for warning filters to ignore code in the typing/tests/data directory. You'd have to add a new if block here:

numpy/numpy/tests/test_warnings.py

Lines 67 to 74 in 1e424da

if base / "testing" in path.parents:

continue

if path == base / "__init__.py":

continue

if path == base / "random" / "__init__.py":

continue

if path == base / "conftest.py":

continue

As part of this PR or a different one (since it’s more or less unconcerned with this particular deprecation)?

rgommers · 2025-11-03T15:31:42Z

Is there a script you use to check for usages or just go through a set of known repositories and check?

I just search known downstream projects first to get a sense, and then I may check what https://github.com/search?type=code says.

ngoldbaum · 2025-11-03T15:36:02Z

As part of this PR or a different one (since it’s more or less unconcerned with this particular deprecation)?

Doing it in this one is fine. I doubt it will come up again before this is merged.

ngoldbaum · 2025-11-03T20:33:02Z

I just pinged the mailing list: https://mail.python.org/archives/list/[email protected]/thread/LYRHETXPN32S7FNHKNLJFDIYWGQHLODJ/

hellerve · 2025-11-03T21:15:19Z

Alright, looks like we got the tests green, and I also adjusted the code comments to make the target 2.5.0. I will go through the important downstream projects I can find in the coming days and submit PRs.

hellerve · 2025-11-09T13:37:47Z

I did an (admittedly small) amount of digging for references. I unfortunately don’t have good insight into what repositories are considered "important", and Github search limits me to 5 largely unordered pages of results.

I did submit 4 PRs just now, among others to pywt and scikit-image. I also checked a few others but couldn’t find much. (No guarantees, though)

ngoldbaum · 2025-11-10T16:51:08Z

It looks like there are also a couple other uses in pandas that can be straightforwardly updated, if you want another easy contribution:

pandas/_libs/tslibs/timestamps.pyx
3495:                np.fix((153 * month - 457) / 5) +

pandas/core/arrays/datetimes.py
2290:            + np.fix((153 * month - 457) / 5)

There's another use in the tests that's testing explicitly for support of np.fix and probably shouldn't be updated.

hellerve · 2025-11-11T09:44:28Z

It looks like there are also a couple other uses in pandas that can be straightforwardly updated, if you want another easy contribution [...]

Contributed a PR there too.

seberg · 2025-11-11T10:07:23Z

Thanks @hellerve, it's cool to see that you updated things there!

There was a small unintended regression in np.fix but I guess we'll wait for one release to actually add the deprecation.
Because of that, I think we should split out the change to np.trunc into a separate PR to be merged quickly. I can just do that, but if you are interested in that, that would also be great!

hellerve · 2025-11-11T13:02:24Z

I’ll solve the merge conflicts once #30168 has also been merged (I’m assuming it will be merged before this PR is even considered).

charris · 2025-11-17T20:20:47Z

Needs rebase. #30168 is in.

…gs in np.fix tests

hellerve · 2025-11-18T21:41:02Z

@charris Thanks for the heads-up! Done!

ngoldbaum changed the title ~~Deprecate numpy.fix~~ DEP: Deprecate numpy.fix Oct 30, 2025

ngoldbaum reviewed Oct 30, 2025

View reviewed changes

numpy/lib/_ufunclike_impl.py Outdated Show resolved Hide resolved

ngoldbaum added this to the 2.4.0 release milestone Oct 30, 2025

ngoldbaum added the 07 - Deprecation label Oct 30, 2025

jorenham reviewed Oct 30, 2025

View reviewed changes

doc/release/upcoming_changes/30098.deprecation.rst Show resolved Hide resolved

seberg approved these changes Oct 30, 2025

View reviewed changes

numpy/lib/_ufunclike_impl.py Outdated Show resolved Hide resolved

numpy/lib/tests/test_ufunclike.py Show resolved Hide resolved

ngoldbaum reviewed Oct 30, 2025

View reviewed changes

numpy/lib/_ufunclike_impl.py Outdated Show resolved Hide resolved

jorenham reviewed Oct 30, 2025

View reviewed changes

numpy/lib/_ufunclike_impl.pyi Show resolved Hide resolved

This comment has been minimized.

Sign in to view

rgommers requested changes Nov 2, 2025

View reviewed changes

ngoldbaum removed this from the 2.4.0 release milestone Nov 3, 2025

jorenham mentioned this pull request Nov 7, 2025

TYP, DEP: numpy.fix pending deprecation #30168

Merged

hellerve mentioned this pull request Nov 11, 2025

ENH,MAINT: rewrite np.fix to use np.trunc internally #30197

Merged

jorenham added this to the 2.5.0 Release milestone Nov 17, 2025

hellerve added 12 commits November 17, 2025 22:04

API: deprecate numpy.fix (see numpy#30096)

f11ac52

STY: fix long line and use warnings module instead of suppress_warnin…

b2b990f

…gs in np.fix tests

TST: fix mypy tests for numpy.fix deprecation (see numpy#30096)

2833278

STY: fix import order in ufunclike test

28c432d

DEP: fix numpy.fix deprecation notice

9daeafd

TST: Add deprecation text for numpy.fix

7000427

DEP: Fix deprecation wording in numpy.fix

5d4bc07

TST: add deprecation warning to typing stubs for numpy.fix

2ee780c

TST: remove unused import

7bdae5e

TST: fix numpy.fix tests for deprecation

c4c7b40

TST: add a blacklist for typing/tests/data in test_warnings.py

e49f717

DEP: fix deprecation target version for numpy.fix

44b7c57

hellerve force-pushed the 30096-deprecate-np-fix branch from d9398a1 to 44b7c57 Compare November 17, 2025 21:11

hellerve added 2 commits November 17, 2025 22:24

TST: re-add lost overload to type of fix

959f4e2

TST: fix message in deprecations test for numpy.fix

1bcbd32

	# public numpy export
	@overload # no style
	def array2string(
	a: NDArray[Any],
	max_line_width: int \| None = None,
	precision: SupportsIndex \| None = None,
	suppress_small: bool \| None = None,
	separator: str = " ",
	prefix: str = "",
	style: _NoValueType = ...,
	formatter: _FormatDict \| None = None,
	threshold: int \| None = None,
	edgeitems: int \| None = None,
	sign: _Sign \| None = None,
	floatmode: _FloatMode \| None = None,
	suffix: str = "",
	*,
	legacy: _Legacy \| None = None,
	) -> str: ...
	@overload # style=<given> (positional), legacy="1.13"
	def array2string(
	a: NDArray[Any],
	max_line_width: int \| None,
	precision: SupportsIndex \| None,
	suppress_small: bool \| None,
	separator: str,
	prefix: str,
	style: _ReprFunc,
	formatter: _FormatDict \| None = None,
	threshold: int \| None = None,
	edgeitems: int \| None = None,
	sign: _Sign \| None = None,
	floatmode: _FloatMode \| None = None,
	suffix: str = "",
	*,
	legacy: Literal["1.13"],
	) -> str: ...
	@overload # style=<given> (keyword), legacy="1.13"
	def array2string(
	a: NDArray[Any],
	max_line_width: int \| None = None,
	precision: SupportsIndex \| None = None,
	suppress_small: bool \| None = None,
	separator: str = " ",
	prefix: str = "",
	*,
	style: _ReprFunc,
	formatter: _FormatDict \| None = None,
	threshold: int \| None = None,
	edgeitems: int \| None = None,
	sign: _Sign \| None = None,
	floatmode: _FloatMode \| None = None,
	suffix: str = "",
	legacy: Literal["1.13"],
	) -> str: ...
	@overload # style=<given> (positional), legacy!="1.13"
	@deprecated("'style' argument is deprecated and no longer functional except in 1.13 'legacy' mode")
	def array2string(
	a: NDArray[Any],
	max_line_width: int \| None,
	precision: SupportsIndex \| None,
	suppress_small: bool \| None,
	separator: str,
	prefix: str,
	style: _ReprFunc,
	formatter: _FormatDict \| None = None,
	threshold: int \| None = None,
	edgeitems: int \| None = None,
	sign: _Sign \| None = None,
	floatmode: _FloatMode \| None = None,
	suffix: str = "",
	*,
	legacy: _LegacyNoStyle \| None = None,
	) -> str: ...
	@overload # style=<given> (keyword), legacy="1.13"
	@deprecated("'style' argument is deprecated and no longer functional except in 1.13 'legacy' mode")
	def array2string(
	a: NDArray[Any],
	max_line_width: int \| None = None,
	precision: SupportsIndex \| None = None,
	suppress_small: bool \| None = None,
	separator: str = " ",
	prefix: str = "",
	*,
	style: _ReprFunc,
	formatter: _FormatDict \| None = None,
	threshold: int \| None = None,
	edgeitems: int \| None = None,
	sign: _Sign \| None = None,
	floatmode: _FloatMode \| None = None,
	suffix: str = "",
	legacy: _LegacyNoStyle \| None = None,
	) -> str: ...

	@overload
	def fix( # type: ignore[misc]
	x: _FloatLike_co,
	out: None = None,
	) -> floating: ...
	@overload
	def fix(
	x: _ArrayLikeFloat_co,
	out: None = None,
	) -> NDArray[floating]: ...
	@overload
	def fix(
	x: _ArrayLikeObject_co,
	out: None = None,
	) -> NDArray[object_]: ...
	@overload
	def fix(
	x: _ArrayLikeFloat_co \| _ArrayLikeObject_co,
	out: _ArrayT,
	) -> _ArrayT: ...

Uh oh!

DEP: Deprecate numpy.fix #30098

Are you sure you want to change the base?

DEP: Deprecate numpy.fix #30098

Conversation

hellerve commented Oct 30, 2025 • edited by ngoldbaum Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

ngoldbaum left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

hellerve commented Oct 30, 2025

Uh oh!

Uh oh!

seberg left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

ngoldbaum Oct 30, 2025

Choose a reason for hiding this comment

Uh oh!

seberg Oct 30, 2025

Choose a reason for hiding this comment

Uh oh!

hellerve Oct 30, 2025

Choose a reason for hiding this comment

Uh oh!

jorenham Oct 30, 2025

Choose a reason for hiding this comment

Uh oh!

ngoldbaum Oct 30, 2025

Choose a reason for hiding this comment

Uh oh!

hellerve Oct 30, 2025

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

hellerve commented Oct 30, 2025

Uh oh!

jorenham commented Oct 30, 2025

Uh oh!

This comment has been minimized.

hellerve commented Oct 30, 2025

Uh oh!

jorenham commented Oct 30, 2025

Uh oh!

rgommers left a comment

Choose a reason for hiding this comment

Uh oh!

hellerve commented Nov 3, 2025

Uh oh!

ngoldbaum commented Nov 3, 2025

Uh oh!

ngoldbaum commented Nov 3, 2025

Uh oh!

hellerve commented Nov 3, 2025

Uh oh!

hellerve commented Nov 3, 2025

Uh oh!

rgommers commented Nov 3, 2025

Uh oh!

ngoldbaum commented Nov 3, 2025

Uh oh!

ngoldbaum commented Nov 3, 2025

Uh oh!

hellerve commented Nov 3, 2025

Uh oh!

hellerve commented Nov 9, 2025

Uh oh!

ngoldbaum commented Nov 10, 2025

Uh oh!

hellerve commented Nov 11, 2025

Uh oh!

seberg commented Nov 11, 2025

Uh oh!

hellerve commented Nov 11, 2025

Uh oh!

hellerve commented Oct 30, 2025 •

edited by ngoldbaum

Loading