Convert feedstock to v1 recipe format by lgray · Pull Request #361 · conda-forge/numpy-feedstock

lgray · 2025-07-07T19:53:20Z

This PR converts numpy-feedstock to a v1 recipe and switches the conda build tool to rattler-build.
It has been partially generated with feedrattler v0.3.13 and portions fixed by hand.

Changes:

📝 Converted meta.yaml to recipe.yaml
Used a personal fork of the feedstock to propose changes
🔧 Updated conda-forge.yml to use rattler-build and pixi (optional)
🔢 Bumped the build number
🐍 Applied temporary fixes for python_min and python_version
🔄 Rerender the feedstock with conda-smithy
Ensured the license file is being packaged.

Please merge this as you like, if you're waiting on the spec or rattler-build to become more stable that's fine.
Mostly, I'm playing with it in my spare time with more difficult packages, it shouldn't hurt to see where there are rough edges.
If this is somehow a major draw to someone's time aside from my own, please let me know, I'm fine with doing something else.

conda-forge-admin · 2025-07-07T19:54:58Z

Hi! This is the friendly automated conda-forge-linting service.

I just wanted to let you know that I linted all conda-recipes in your PR (recipe/recipe.yaml) and found it was in an excellent condition.

rgommers · 2025-07-08T09:49:45Z

Thanks for working on this @lgray! I'm definitely keen to migrate to rattler-build, it's just so much nicer to work with. This looks like a significant improvement, and also fixes the linter complaints that are visible with conda-build (see for example #360 (comment)).

CI logs look quite good, one question about the warnings visible at the bottom for the Windows and macOS builds - are these expected?

⚠ warning Overlinking against "/usr/lib/libSystem.B.dylib" for "lib/python3.13t/site-packages/numpy/random/_common.cpython-313t-darwin.so"

lgray · 2025-07-08T12:39:24Z

Cool, nice to have that pleasant side effect concerning the blas tests!

The "overlinking against libSystem.B.dylib and python3XY.dll" complaints I've seen in other packages I've converted. They appear to be harmless, since logically your python module shall eventually need symbols from python and anything eventually needs symbols from core system libraries. It may just be indirect depending on what part of the package you're looking at in a specific shared lib. That's my personal understanding, at least.

If there's a serious performance reason (some memory mapping or locality issue) I don't understand that's critical here, we can dig. I'll go ahead and open an issue over in rattler-build, unless there already is one in which case I will link this PR.

lgray · 2025-07-08T12:44:33Z

Ah yeah the main developer sees it too for macos:
prefix-dev/rattler-build#1260

But seems fine with it?

recipe/recipe.yaml

h-vetinari · 2025-07-08T12:51:53Z

recipe/recipe.yaml

+
+context:
+  version: 2.3.1
+  dev: nodev


can we use something like "null" here instead of some arbitrary string?

I never tried null specifically but it complained loudly on empty strings. Shall give it a go.

Ah, yeah same complaint when using null:

Error: × expected a scalar value. ╭─[recipe/recipe.yaml:6:8] 5 │ version: 2.3.1 6 │ dev: null · ──┬─ · ╰── here 7 │ # numpy will by default use the ABI feature level for the first numpy version ╰──── help: `context` values must always be scalars (booleans, integers or strings) or uniform lists of scalars

What would you prefer instead of "nodev"?

I guess you may have meant "null" as a string as opposed to null the keyword? Still at that point it is some arbitrary string.

I could use a number? -1 works. I have done this for now and added a comment as to intended usage.

I guess you may have meant "null" as a string as opposed to null the keyword?

No, I meant the keyword. As you mention, "null" would just be yet another arbitrary string, and a very confusing one at that.

Given that we can't use empty strings, I'd like to switch this to something like rc_num. If it's -1, we ignore it, if it's >=0, we construct a rc${{ rc_num }} suffix.

Yeah but then you're doing if: (rc_num | int) > -1 whenever we need to do control flow.

if: rc_num != "-1"

Whatever you're cool with, but that will accept "-2" or "nosegoblins" as viable rc_num values.

I fully understand that, but we still have human reviewers in the loop here (and trivially, no tag for that would be found, so CI would be deep red). Making this fool-proof or "fuzzing" the recipe is not the right trade-off.

No prob! Will make the change when back at a computer with a keyboard.

recipe/recipe.yaml

lgray · 2025-07-08T14:30:33Z

I see in .ci_support that this deleted a bunch of lines like:

numpy:
- '2'

Does this need to remain or it's an obsolete artifact from the release of numpy 2?

h-vetinari · 2025-07-08T21:45:44Z

Does this need to remain or it's an obsolete artifact from the release of numpy 2?

It's an obsolete artefact of the fact that conda{,-smithy} sees that the recipe mentions numpy, thinks it depends on it, and thus needs to apply the values from the global pinning. It's harmless but good to remove these values from the variant configs.

lgray · 2025-07-09T16:26:09Z

Good - one of the rattler-build devs finds the overlinking complaints in windows benign.

lgray · 2025-07-11T13:54:48Z

Is there any further review for this PR?

h-vetinari · 2025-07-11T13:59:31Z

Can you reduce this to two commits please: one for the recipe transformation, and one for rerendering.

lgray · 2025-07-11T14:00:49Z

certainly!

h-vetinari

getting close, but a few nits remain

h-vetinari · 2025-07-15T02:09:32Z

recipe/recipe.yaml

+
+package:
+  name: numpy
+  version: ${{ version }}${{ "rc" + rc_num if rc_num != "-1" else "" }}


I'd like the precedence clearer here. Also, string concatenation in jinja should use ~ AFAIU; + is unreliable.

Suggested change

version: ${{ version }}${{ "rc" + rc_num if rc_num != "-1" else "" }}

version: ${{ version }}${{ ("rc" ~ rc_num) if rc_num != "-1" else "" }}

Ah, I did not know about the ~ operator in jinja! As it performs conversion to string, this would allow us to use an integer for rc_num in a pleasant way, as originally discussed. Up to you, it lints and re-renders fine locally.

done, using an int for rc_num.

h-vetinari · 2025-07-15T02:09:59Z

recipe/recipe.yaml

+        - pytz
+        - if: match(python, "<=3.11")
+          then: setuptools <60.0.0
+        - if: match(python, ">=3.12")


use else: here

h-vetinari · 2025-07-15T02:10:33Z

recipe/recipe.yaml

+        - if: linux
+          then: ${{ compiler('fortran') }}
+        - pkg-config
+


remove this newline, those two blocks form a unit

h-vetinari · 2025-07-15T02:11:41Z

recipe/recipe.yaml

+
+context:
+  version: 2.3.1
+  # -1 means "not a dev version", a number > -1 will emit postfix "rc${{ rc_num }}"


comment is out-dated

Suggested change

# -1 means "not a dev version", a number > -1 will emit postfix "rc${{ rc_num }}"

# "-1" means "not a dev version", a string != "-1" will emit postfix "rc${{ rc_num }}"

see commentary on being able to actually use numbers in a non-ugly way for rc_num

h-vetinari · 2025-07-15T02:12:16Z

recipe/recipe.yaml

+
+schema_version: 1
+


I think these lines can be deleted.

h-vetinari · 2025-07-15T02:12:54Z

recipe/recipe.yaml

+    - ${{ compiler('c') }}
+    - ${{ stdlib('c') }}


Let's use this opportunity to order the stdlib before the compilers

Suggested change

- ${{ compiler('c') }}

- ${{ stdlib('c') }}

- ${{ stdlib('c') }}

- ${{ compiler('c') }}

h-vetinari · 2025-07-15T02:14:16Z

recipe/meta.yaml

-    # some linux tests need a C/C++ compiler;
-    # extra f2py tests need a fortran compiler
-    - {{ compiler('c') }}        # [unix]
-    - {{ compiler('cxx') }}      # [unix]
-    - {{ compiler('fortran') }}  # [linux]
-    # For some cython tests
-    - pkg-config


these comments went missing

h-vetinari · 2025-07-15T02:15:35Z

recipe/recipe.yaml

+    - jakirkham
+    - msarahan
+    - pelson
+    - rgommers
+    - ocefpaf
+    - isuruf
+    - xhochy
+    - h-vetinari


Also, let's alphabetize that list in the rewrite commit.

…nda-forge-pinning 2025.07.15.07.42.52

h-vetinari

Thank you for your work and patience on this. :)

From reading discussions at numpy conda-forge/numpy-feedstock#361 (comment) and the GO-recipe https://github.com/conda-forge/go-feedstock/blob/main/recipe/meta.yaml#L51 this seems to be ok to overlink against.

…30301) * add go4 recipe * remove explicit python variants * skip win and osx builds for now * fix homepage link * clean up recipe - raising errors on overdepending/overlinking helped to figure out more reasonable dependencies - removing variants.yaml as they are anyways set by conda-forge pins - switching dependency from root to root_base - did a test run building a user analysis and running it on real data * add python to run dependencies again * add GO4License.txt to recipe * Add comment on ${{ compiler('cxx') }} resolution for the test environment. * try building on osx * only use libgl-devel host dependency on linux * remove globstar shell option for osx build * clean up and avoid using globstar as osx does not support it * try different sed commands for osx * try if cleanup is even necessary at all anymore * add osx libSystem.B.dylib to allowlist From reading discussions at numpy conda-forge/numpy-feedstock#361 (comment) and the GO-recipe https://github.com/conda-forge/go-feedstock/blob/main/recipe/meta.yaml#L51 this seems to be ok to overlink against. * remove the sed cleanups as they do not seem necessary any more * adjust test requirements for osx * try again on osx without compiler macro * clean up * further cleanup * patch compiler selection in Go4s Makefile.Linux Now it should read CXX from environment variables * adjust cmake invocation to conda-forge * build on of the test cases with cmake * try matching the library on both linux / osx * set extglob for test examples * remove special settings for dynamic_linking * try using CMAKE_ARGS.. but leads to errors * use CMAKE_ARGS * use CPU_COUNT parameter instead of trying to call nproc * remove manual pins The updated conda-forge-pinning solved the issue * patch lipthread linking * take DCMAKE_INSTALL_LIBDIR out of CMAKE_ARGS * adjust patches to docker building * handle linking issues only on linux * remove CMAKE argument with pattern mattching instead of array position * fix cmake argument expansion in example runs * remove optional hdf5 The linking issues on sysroot_linux seem to happen only in hdf5 dependencies * explicitly turn hdf5 feature off

lgray requested review from h-vetinari, isuruf, jakirkham, msarahan, ocefpaf, pelson, rgommers and xhochy as code owners July 7, 2025 19:53

h-vetinari reviewed Jul 8, 2025

View reviewed changes

recipe/recipe.yaml Outdated Show resolved Hide resolved

lgray mentioned this pull request Jul 8, 2025

Overlinking warnings against python and some system32 dlls in windows? prefix-dev/rattler-build#1772

Closed

h-vetinari reviewed Jul 8, 2025

View reviewed changes

recipe/recipe.yaml Outdated Show resolved Hide resolved

h-vetinari reviewed Jul 8, 2025

View reviewed changes

lgray force-pushed the convert_feedstock_to_v1_recipe_format branch from e9c8d59 to 4fe376a Compare July 8, 2025 13:17

h-vetinari reviewed Jul 8, 2025

View reviewed changes

recipe/recipe.yaml Outdated Show resolved Hide resolved

lgray force-pushed the convert_feedstock_to_v1_recipe_format branch 2 times, most recently from 80fdea0 to 59250ac Compare July 8, 2025 14:12

lgray force-pushed the convert_feedstock_to_v1_recipe_format branch 2 times, most recently from a3b1c7a to ca72588 Compare July 8, 2025 23:57

lgray force-pushed the convert_feedstock_to_v1_recipe_format branch from ca72588 to 80d7143 Compare July 11, 2025 14:04

h-vetinari reviewed Jul 15, 2025

View reviewed changes

lgray force-pushed the convert_feedstock_to_v1_recipe_format branch from 80d7143 to 5d278fd Compare July 15, 2025 13:33

lgray added 2 commits July 15, 2025 06:39

Convert to v1 feedstock and use pixi as conda install tool

017485c

MNT: Re-rendered with conda-build 25.5.0, conda-smithy 3.51.1, and co…

b97f7b6

…nda-forge-pinning 2025.07.15.07.42.52

lgray force-pushed the convert_feedstock_to_v1_recipe_format branch from 5d278fd to b97f7b6 Compare July 15, 2025 13:40

h-vetinari approved these changes Jul 15, 2025

View reviewed changes

h-vetinari merged commit c8f725c into conda-forge:main Jul 15, 2025
27 checks passed

HLinde mentioned this pull request Nov 24, 2025

Adding GSI Online Offline Object Oriented (Go4) analysis framework conda-forge/staged-recipes#30301

Merged

10 tasks

h-vetinari mentioned this pull request Dec 22, 2025

TEST: v2.3.x + blas variants #360

Draft

	version: ${{ version }}${{ "rc" + rc_num if rc_num != "-1" else "" }}
	version: ${{ version }}${{ ("rc" ~ rc_num) if rc_num != "-1" else "" }}

	# -1 means "not a dev version", a number > -1 will emit postfix "rc${{ rc_num }}"
	# "-1" means "not a dev version", a string != "-1" will emit postfix "rc${{ rc_num }}"


		schema_version: 1

Uh oh!

Conversation

lgray commented Jul 7, 2025

Uh oh!

conda-forge-admin commented Jul 7, 2025

Uh oh!

rgommers commented Jul 8, 2025

Uh oh!

lgray commented Jul 8, 2025

Uh oh!

lgray commented Jul 8, 2025

Uh oh!

Uh oh!

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

lgray Jul 8, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Uh oh!

lgray commented Jul 8, 2025

Uh oh!

h-vetinari commented Jul 8, 2025

Uh oh!

lgray commented Jul 9, 2025

Uh oh!

lgray commented Jul 11, 2025

Uh oh!

h-vetinari commented Jul 11, 2025

Uh oh!

lgray commented Jul 11, 2025

Uh oh!

h-vetinari left a comment

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

lgray Jul 15, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

h-vetinari Jul 15, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

lgray Jul 15, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

lgray Jul 8, 2025 •

edited

Loading

lgray Jul 15, 2025 •

edited

Loading

h-vetinari Jul 15, 2025 •

edited

Loading

lgray Jul 15, 2025 •

edited

Loading