run test-suite (and repack netlib) for blis by h-vetinari · Pull Request #62 · conda-forge/blas-feedstock

h-vetinari · 2021-01-02T14:21:46Z

No description provided.

conda-forge-linter · 2021-01-02T14:21:50Z

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) and found it was in an excellent condition.

h-vetinari · 2021-01-02T14:44:25Z

Getting an error

  File "/opt/conda/lib/python3.8/site-packages/conda_build/render.py", line 475, in _simplify_to_exact_constraints
    raise ValueError("Conflicting exact pins: {}".format(exact_pins))
ValueError: Conflicting exact pins: [['3.9.0', '3_h92ddd45_netlib'], ['3.9.0', '3_h92ddd45_netlib'], ['3.9.0', '6_blis']]

Based on the number of resolution steps before, I'm guessing this happens for blas-devel. Let's see if d80fe2b helps.

conda-forge-linter · 2021-01-07T21:44:44Z

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

I was trying to look for recipes to lint for you, but it appears we have a merge conflict.
Please try to merge or rebase with the base branch to resolve this conflict.

Please ping the 'conda-forge/core' team (using the @ notation in a comment) if you believe this is a bug.

conda-forge-linter · 2021-01-07T22:00:07Z

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) and found it was in an excellent condition.

isuruf · 2021-01-07T22:19:49Z

FWIW, I don't like repackaging the netlib variant of lapack as blis. With the current scheme, it's obvious that we are using BLAS from blis and LAPACK from netlib. With the new scheme, it's not obvious whether LAPACK comes from netlib or flame or other source.

h-vetinari · 2021-01-07T22:25:25Z

FWIW, I don't like repackaging the netlib variant of lapack as blis. With the current scheme, it's obvious that we are using BLAS from blis and LAPACK from netlib. With the new scheme, it's not obvious whether LAPACK comes from netlib or flame or other source.

It's a point I expected to discuss with you. ;-)

But right now I was just trying to figure out how to run the test suite for blis here... Not sure if conda could be taught to correctly resolve the exact subpackage pins mixed with other variants, but so far, the only thing I've gotten working is a repack.

I had an intermediate version (lost since the rebase) of renaming the lapack(e)-outputs for blis only, e.g. it could be something like lapack(e)-3.9.0-7_netlib_repacked_for_blis, which would still work with the *blis selector. WDYT?

h-vetinari · 2021-01-07T22:37:00Z

Also, I now have a running test suite for blis on windows, and the results are bad: 13% tests passed, 90 tests failed out of 103. This probably explains why I'm having ~440 errors in the scipy test suite for blis on windows.

h-vetinari · 2021-01-08T15:09:50Z

@h-vetinari: Also, I now have a running test suite for blis on windows, and the results are bad: 13% tests passed, 90 tests failed out of 103. This probably explains why I'm having ~440 errors in the scipy test suite for blis on windows.

On second thought - all the BLAS* tests pass, all the LAPACK* tests fail - turns out the integration isn't working yet after all...

…da-forge-pinning 2021.01.08.18.10.03

h-vetinari

@isuruf, this should be ready for a first round of discussions - not saying repacking is a must, but I believe (with the build string distinction) it's definitely feasible without ambiguity.

h-vetinari · 2021-01-09T01:38:32Z

-        - {{ pin_subpackage("blas-devel", exact=True) }}
+        - {{ pin_subpackage("blas", exact=True) }}


@isuruf, not sure I understood your intention from #64 here correctly, but blas-devel depending on itself sounds fragile and circular. My understanding would be that it builds on blas and adds the "development headers" (or whatever is the equivalent)...?

Yes that's a mistake and should be removed. blas should depend on blas-devel.

I thought it was the other way around? Kinda like blas-devel > blas (in terms of dependencies)

At least that's how I've understood the -dev[el] packages in various distros, for example.

h-vetinari · 2021-01-09T01:39:08Z

-        - liblapacke {{ version }} *netlib                   # [blas_impl == 'blis']
-        - {{ pin_subpackage("libcblas", exact=True) }}
        - {{ pin_subpackage("libblas", exact=True) }}
-        - {{ pin_subpackage("blas-devel", exact=True) }}


Same here, not sure why blas should need e.g. mkl-devel?

h-vetinari · 2021-01-09T19:07:04Z

Ping @isuruf :)

h-vetinari · 2021-01-14T20:09:09Z

@isuruf
WDYT about linux-64/liblapack-3.9.0-8_netlib_repacked_for_blis.tar.bz2?

h-vetinari · 2021-01-26T19:28:05Z

@isuruf
I'd like to merge this - I believe I have addressed your concerns, but would like to have confirmation from (or discussion with) you to be sure.

isuruf · 2021-01-26T19:30:04Z

I'm not a fan of this. I believe we can test this by creating the symlinks for blis in the testing script. I'll have a look, but will not be soon.

h-vetinari · 2021-01-26T19:34:01Z

ok, thanks for the response. If you can outline your plan (e.g. symlink what to where), I can give it a shot. In the meantime, I'm placing this PR in draft mode.

isuruf · 2021-01-26T19:36:53Z

Thanks @h-vetinari for all your efforts.

I think what we need to do is in test_blas.sh call build_pkg.sh for blis four times with PKG_NAME set for blas, cblas, lapack, lapacke.

h-vetinari · 2021-01-26T19:39:54Z

OK, I'll think about it and might give it a shot. While we're at it, could you briefly comment on this? Circular dependencies sound bad to me, and should probably be fixed ASAP independently of the blis-stuff.

h-vetinari · 2021-03-28T11:56:43Z

Obsoleted by #65

h-vetinari requested review from isuruf, jakirkham, ocefpaf and pelson as code owners January 2, 2021 14:21

h-vetinari marked this pull request as draft January 2, 2021 14:21

h-vetinari force-pushed the master branch from 1ba98d7 to ac83538 Compare January 2, 2021 15:35

run test suite for blis

7d28d2d

h-vetinari force-pushed the master branch from 6df4464 to 9c051bb Compare January 7, 2021 22:00

h-vetinari added 6 commits January 9, 2021 02:02

repackage lapack(e) for blis

7be81a4

unify expected-file tests

61bd55e

update comment to better reflect actual use

2fed195

fix blas-devel self-dependence

69d28bf

bump build number

03534d1

distinguish build strings for netlib_repacked_for_blis

8cf57bc

h-vetinari force-pushed the master branch from bf99678 to 8cf57bc Compare January 9, 2021 01:03

MNT: Re-rendered with conda-build 3.20.5, conda-smithy 3.8.6, and con…

58fde3a

…da-forge-pinning 2021.01.08.18.10.03

h-vetinari marked this pull request as ready for review January 9, 2021 01:36

h-vetinari changed the title ~~WIP: try running test-suite for blis~~ run test-suite (and repack netlib) for blis Jan 9, 2021

h-vetinari commented Jan 9, 2021

View reviewed changes

h-vetinari marked this pull request as draft January 26, 2021 19:34

h-vetinari closed this Mar 28, 2021

h-vetinari mentioned this pull request Mar 28, 2021

WIP: Build all blas-metapackages in this feedstock (also for netlib) #68

Draft

		- {{ pin_subpackage("blas-devel", exact=True) }}
		- {{ pin_subpackage("blas", exact=True) }}

Uh oh!

Uh oh!

Conversation

h-vetinari commented Jan 2, 2021

Uh oh!

conda-forge-linter commented Jan 2, 2021

Uh oh!

h-vetinari commented Jan 2, 2021

Uh oh!

conda-forge-linter commented Jan 7, 2021

Uh oh!

conda-forge-linter commented Jan 7, 2021

Uh oh!

isuruf commented Jan 7, 2021

Uh oh!

h-vetinari commented Jan 7, 2021

Uh oh!

h-vetinari commented Jan 7, 2021

Uh oh!

h-vetinari commented Jan 8, 2021

Uh oh!

h-vetinari left a comment

Choose a reason for hiding this comment

Uh oh!

h-vetinari Jan 9, 2021 • 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!

isuruf Jan 26, 2021

Choose a reason for hiding this comment

Uh oh!

h-vetinari Jan 26, 2021

Choose a reason for hiding this comment

Uh oh!

h-vetinari Jan 9, 2021

Choose a reason for hiding this comment

Uh oh!

h-vetinari commented Jan 9, 2021

Uh oh!

h-vetinari commented Jan 14, 2021

Uh oh!

h-vetinari commented Jan 26, 2021

Uh oh!

isuruf commented Jan 26, 2021

Uh oh!

h-vetinari commented Jan 26, 2021

Uh oh!

isuruf commented Jan 26, 2021

Uh oh!

h-vetinari commented Jan 26, 2021

Uh oh!

h-vetinari commented Mar 28, 2021

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

h-vetinari Jan 9, 2021 •

edited

Loading