Changed gromacs package for testing

[ianmkenney]

Moved from becksteinlab to bioconda package

[codecov-io]

Codecov Report

Merging #130 into develop will increase coverage by 0.12%.
The diff coverage is n/a.

@@             Coverage Diff             @@
##           develop     #130      +/-   ##
===========================================
+ Coverage    48.14%   48.27%   +0.12%     
===========================================
  Files           24       24              
  Lines         4831     4831              
  Branches       731      731              
===========================================
+ Hits          2326     2332       +6     
+ Misses        2352     2340      -12     
- Partials       153      159       +6

Impacted Files	Coverage Δ
gromacs/tools.py	54.01% <0%> (-0.73%)	⬇️
gromacs/utilities.py	53.05% <0%> (-0.33%)	⬇️
gromacs/config.py	65.8% <0%> (+0.64%)	⬆️
gromacs/core.py	57.75% <0%> (+3.01%)	⬆️

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update ebe740e...2f9a0f3. Read the comment docs.

[kain88-de]

defaults is always on by default.

[kain88-de]

you also need to add conda-forge for the version 2018 dependencies

[kain88-de]

maybe continuing to test gromacs 5.x from the beckstein lab repo would be good. there is still a large number of people using gromacs 5

[orbeckst]

The problem with the 5.x build is that it is currently tied to Python 2; we would need to build a Python 3 version (or get rid of any Python dependencies - which is the get_gmx script). We hope to not have to maintain the gromacs conda packages any more...

[orbeckst]

@ianmkenney said that the Bioconda gromacs 2018 core dumps on Travis – perhaps an issue with the instruction set not matching what is available on Travis??

@ianmkenney can you post the evidence that you have?

[kain88-de]

I can imagine where the core dumps come from. The travis/circleci build on conda forge was done with a random CPU and GROMACS always chooses the best available SIMD instruction set. On travis we can be randomly assigned an older CPU that doesn't support the instruction set. The only way around this is to specify a common instruction set like SSE3 during the cmake step. Making our own package on conda-forge can help to sort out these issues.

About maintenance. Using bioconda and conda-forge doesn't get rid of it completely but sharing responsibility is much easier.

[ianmkenney]

log with the error

I put gmx grompp --help and got an illegal instruction error

$ gmx grompp --help
/home/travis/.travis/job_stages: line 57:  2239 Illegal instruction     (core dumped) gmx grompp --help

[kain88-de]

yeah that is a instruction issue. But that is fixable upstream

[orbeckst]

WHat is the best approach to proceed? Raise an issue with bioconda and ask them to build Gromacs taking portability into account? E.g. build binaries with instruction set suffixes

• gmx_sse2
• gmx_avx
• etc

or just build gmx with sse2?

[kain88-de]

I'm submitted a small PR.

[kyleabeauchamp]

There is an ongoing project to upgrade Bioconda to Conda-build3 and also to figure out a way to build multiple variants of a package, with different optimization levels. Please take a look as Gromacs is a prime candidate for specifying multiple builds. AFAIK, I think the conda-forge folks are going to start assuming that SSE3 is available as the default.

bioconda/bioconda-recipes#2354

conda-forge/conda-forge.github.io#49

[kain88-de]

I would have gone for SSE3 but gromacs doesn't over it. I'm aware of the conda-build3 transition happening in conda-forge. It will still take a little while for them to make it though.

[kyleabeauchamp]

SSE2 SGTM then

[kain88-de]

@ianmkenney you should be able to try in about an hour.

[rhheilma]

Can I get off of this list, please? Thank you, Ryan

…

On Wed, May 9, 2018, 10:08 AM Ian Kenney ***@***.***> wrote: Moved from becksteinlab to bioconda package ------------------------------ You can view, comment on, or merge this pull request online at: #130 Commit Summary - Changed gromacs package for testing File Changes - *M* .travis.yml <https://github.com/Becksteinlab/GromacsWrapper/pull/130/files#diff-0> (11) - *M* environment.yml <https://github.com/Becksteinlab/GromacsWrapper/pull/130/files#diff-1> (4) Patch Links: - https://github.com/Becksteinlab/GromacsWrapper/pull/130.patch - https://github.com/Becksteinlab/GromacsWrapper/pull/130.diff — You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub <#130>, or mute the thread <https://github.com/notifications/unsubscribe-auth/AQC-MLSDZHUUasb2stb0hC6iyxcV1JJIks5twyJ6gaJpZM4T4rSA> .

[kain88-de]

@rhheilma you have to change your GitHub subscription setting yourself. We cannot do this for you.

[orbeckst]

The Python 3 PR #127 includes the new bioconda packages so we can fix everything there (and close this PR).

Thanks @kain88-de for fixing upstream.