single branch builds for mumps-seq, mumps-mpi #36

minrk · 2018-11-13T12:36:37Z

So the cf automation tools can update both flavors.

I'd like to resolve #34 here as well. Until/unless I hear otherwise,
I will implement option 4 there: produce a mumps-seq and mumps-mpi package,
where mumps-seq gets _seq suffixes on its libs and its headers go in include/mumps_seq. mumps-mpi remains unchanged. This follows a pattern established by the debian packaging of mumps and mumps-seq (there mumps is mumps-mpi, but I don't want to do that here, since it will break everything currently depending on serial mumps).

The mumps package will be deprecated but not removed so that nothing breaks.

TODO:

single recipe with multiple outputs
rename serial mumps to mumps-seq
add _seq to mumps-seq libraries
install libseq headers in include/mumps_seq
add mumps-include package for the common headers
run actual build tests for mumps-seq

eliminates mumps-mpi branch

conda-forge-linter · 2018-11-13T12:36:42Z

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.

if no outputs match the top-level name

conda-forge-linter · 2018-11-13T15:56:20Z

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

I wanted to let you know that I linted all conda-recipes in your PR (recipe) and found some lint.

Here's what I've got...

For recipe:

The outputs section contained an unexpected subsection name. skip is not a valid subsection name.

- call the seq package mumps-seq (no more mumps package) - mumps-include package for shared headers used for both mumps-seq, mumps-mpi, unix-only - _seq.a suffix for mumps-seq libs (matches debian packaging) - libseq headers in include/mumps_seq (matches debian packaging)

conda-forge-linter · 2018-11-13T15:57:23Z

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.

conda-forge-linter · 2018-11-15T09:12:48Z

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

I wanted to let you know that I linted all conda-recipes in your PR (recipe) and found some lint.

Here's what I've got...

For recipe:

Selectors are suggested to take a <two spaces>#<one space>[<expression>] form. See lines [48]

conda-forge-linter · 2018-11-15T09:13:46Z

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.

conda-forge-linter · 2018-11-15T09:57:19Z

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

I wanted to let you know that I linted all conda-recipes in your PR (recipe) and found some lint.

Here's what I've got...

For recipe:

Selectors are suggested to take a <two spaces>#<one space>[<expression>] form. See lines [46]

so ~dp0 is the recipe dir, not where we want it

conda-forge-linter · 2018-11-15T12:10:27Z

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.

- no headers in mumps-mpi (mumps-headers covers that) - mpi.h, mpif.h from libseq go in include/mumps_seq (no others from libseq)

minrk · 2018-11-22T14:39:23Z

ok, I believe this is entirely ready to go now

basnijholt · 2018-11-22T14:41:25Z

Great work!

Hopefully this will lead to fewer problems when using kwant + Fenics.

minrk · 2018-11-22T14:42:22Z

Should we do a local test build of kwant with the new mumps-seq to make sure this works?

it should have no hash inputs

minrk · 2018-11-22T15:13:34Z

I did a local build of kwant but found an issue. This is creating two builds of mumps-include which are pinned exactly as a dependency but have different hash inputs. Fixing the build string on mumps-include should do it.

mumps-seq doesn’t need this because it’s a new package but once a package has made the mistake of adopting the blas_openblas feature, it cannot drop it until conda-forge moves to blas 2.0 metapackage

use x.x.x pinning instead of exact seems to cause weird package-not-found errors on Windows (surely a conda-build bug)

minrk · 2018-11-23T12:16:44Z

I was able to use the latest commit (18edb09) and
conda-forge/kwant-feedstock#58 to build mumps-mpi, mumps-seq, and kwant with mumps-seq, and I was able to create an env with fenics and kwant together and everything seemed happy, so I think this should be all set.

minrk · 2018-11-26T12:58:27Z

@basnijholt any more testing you want to do before landing this one?

basnijholt · 2018-11-26T13:24:39Z

Thanks a lot for testing! I have complete faith in that it will work 👍

I'll merge this now.

minrk added 4 commits November 13, 2018 13:17

import files from mpi branch

1130a23

use multiple outputs to get mpi variants

197ef1c

eliminates mumps-mpi branch

dummy mpi dependency to ensure the full matrix is rendered

67336c3

MNT: Re-rendered with conda-smithy 3.1.12 and pinning 2018.11.13

3430bdd

need dummy dependencies to get the build matrix out of conda-smithy

c93a474

if no outputs match the top-level name

minrk force-pushed the single-branch branch from 7a5d4dd to 74ca577 Compare November 13, 2018 15:57

minrk changed the title ~~[WIP] single branch builds for mumps, mumps-mpi~~ [WIP] single branch builds for mumps-seq, mumps-mpi Nov 13, 2018

specify build script explicitly for mumps-seq

ae45409

minrk force-pushed the single-branch branch from 1feb743 to 6289a56 Compare November 15, 2018 09:13

minrk force-pushed the single-branch branch from 6289a56 to 2304111 Compare November 15, 2018 09:56

minrk added 2 commits November 15, 2018 13:08

Maybe calling it bld-seq.bat will help?

a527e32

bld-seq.bat isn't copied to checkout anymore

ff2e2a2

so ~dp0 is the recipe dir, not where we want it

minrk force-pushed the single-branch branch from 2304111 to ff2e2a2 Compare November 15, 2018 12:10

minrk force-pushed the single-branch branch from 85fb359 to 1eaa4ab Compare November 20, 2018 09:18

This was referenced Nov 20, 2018

restore mumps_mpi.h header in serial mumps #37

Merged

Rebuild for Python 3.7, GCC 7, R 3.5.1, openBLAS 0.3.2 conda-forge/ipopt-feedstock#24

Merged

minrk force-pushed the single-branch branch from 251a00c to 7bb56cc Compare November 20, 2018 15:33

minrk added 2 commits November 20, 2018 20:58

ensure headers are installed in the right places

44c94e7

- no headers in mumps-mpi (mumps-headers covers that) - mpi.h, mpif.h from libseq go in include/mumps_seq (no others from libseq)

Merge branch 'master' into single-branch

1543f47

minrk force-pushed the single-branch branch from 7bb56cc to 1543f47 Compare November 20, 2018 19:59

minrk changed the title ~~[WIP] single branch builds for mumps-seq, mumps-mpi~~ single branch builds for mumps-seq, mumps-mpi Nov 21, 2018

minrk mentioned this pull request Nov 21, 2018

mumps and mumps-mpi cannot be installed side-by-side #34

Closed

This was referenced Nov 21, 2018

build examples using test.source_files #38

Closed

after installing fenics=2017.2, mumps-mpi breaks other packages that depend on mumps conda-forge/fenics-feedstock#62

Closed

force build string on mumps-include to include only build number

d80d8cd

it should have no hash inputs

ensure mumps-mpi keeps its blas feature

6b0e794

mumps-seq doesn’t need this because it’s a new package but once a package has made the mistake of adopting the blas_openblas feature, it cannot drop it until conda-forge moves to blas 2.0 metapackage

This was referenced Nov 22, 2018

build with incoming mumps-seq package conda-forge/kwant-feedstock#58

Merged

build shared libs #39

Merged

revert mumps-include build string

1907e6b

use x.x.x pinning instead of exact seems to cause weird package-not-found errors on Windows (surely a conda-build bug)

minrk force-pushed the single-branch branch from f27961c to 1907e6b Compare November 22, 2018 22:11

use existing make PLAT=_seq to add _seq to libs

18edb09

basnijholt merged commit 0989da7 into conda-forge:master Nov 26, 2018

minrk deleted the single-branch branch November 26, 2018 14:33

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

single branch builds for mumps-seq, mumps-mpi #36

single branch builds for mumps-seq, mumps-mpi #36

minrk commented Nov 13, 2018 •

edited

conda-forge-linter commented Nov 13, 2018

conda-forge-linter commented Nov 13, 2018

conda-forge-linter commented Nov 13, 2018

conda-forge-linter commented Nov 15, 2018

conda-forge-linter commented Nov 15, 2018

conda-forge-linter commented Nov 15, 2018

conda-forge-linter commented Nov 15, 2018

minrk commented Nov 22, 2018

basnijholt commented Nov 22, 2018

minrk commented Nov 22, 2018 •

edited

minrk commented Nov 22, 2018

minrk commented Nov 23, 2018

minrk commented Nov 26, 2018

basnijholt commented Nov 26, 2018

single branch builds for mumps-seq, mumps-mpi #36

single branch builds for mumps-seq, mumps-mpi #36

Conversation

minrk commented Nov 13, 2018 • edited

conda-forge-linter commented Nov 13, 2018

conda-forge-linter commented Nov 13, 2018

conda-forge-linter commented Nov 13, 2018

conda-forge-linter commented Nov 15, 2018

conda-forge-linter commented Nov 15, 2018

conda-forge-linter commented Nov 15, 2018

conda-forge-linter commented Nov 15, 2018

minrk commented Nov 22, 2018

basnijholt commented Nov 22, 2018

minrk commented Nov 22, 2018 • edited

minrk commented Nov 22, 2018

minrk commented Nov 23, 2018

minrk commented Nov 26, 2018

basnijholt commented Nov 26, 2018

minrk commented Nov 13, 2018 •

edited

minrk commented Nov 22, 2018 •

edited