Refactor to work with matchms 0.6.0

[florian-huber]

Started simple and became somewhat bigger with the recent changes of matchms.

• Include Python 3.8 (and CI for this).
• Remove double CI runs (as with matchms before).
• Integrate Spec2Vec and Spec2VecParallel and make them inherit from BaseSimilarity.
• Remove now redundant parallel integration test.
• Add tests for Spec2Vec.pair() and .matrix() methods.

[florian-huber]

Not yet sure why that is, but in the CI verify runs it seems to install the released spec2vec 0.2.0 from nlesc instead of the locally published package. And it does so only for Python 3.7.

[sverhoeven]

The build and verify jobs do not have same matrix.

Also the upload artifact name does not include matrix identifiers making it possible to get wires crossed.

Let me do some local testing:

• try to make build artifact with a identifier for current workflow run in version of build package
• try to add --use-local to https://github.com/iomega/spec2vec/blob/refactor_environment/.github/workflows/conda_build.yml#L200-L205
• try to add --only-deps to https://github.com/iomega/spec2vec/blob/refactor_environment/.github/workflows/conda_build.yml#L200-L205 and do conda install $BUILDDIR

[sverhoeven]

Luckliy a local conda build has the Spec2Vec class extending BaseSimilarity.

• Tried conda install --channel bioconda --channel conda-forge --channel nlesc --channel $BUILDDIR -v --use-local spec2vec still installs spec2vec from nlesc channel.
• Tried conda install --channel bioconda --channel conda-forge --channel nlesc --channel $BUILDDIR -v --only-deps spec2vec does not install matchms, so will not work.
• Tried channel priorities with conda install --channel /tmp/ff/noarch --channel bioconda --channel conda-forge --channel nlesc --strict-channel-priority --override-channels -v spec2vec, but it got stuck resolving.

Seems https://anaconda.org/nlesc/spec2vec/0.2.0/download/noarch/spec2vec-0.2.0-pyhec00094_0.tar.bz2 and GH workflow /home/runner/work/_temp/spec2vec/_build/noarch/spec2vec-0.2.0-pyhec00094_0.tar.bz2 have the same version + build string so for conda it does not matter which is installed.
Changed conda/meta.yaml with perl -pi -e 's/number: 0/number: 999/' conda/meta.yaml

-number: 0
+number: 999

The install with conda install --channel bioconda --channel conda-forge --channel nlesc --channel $BUILDDIR -v spec2vec got install from local channel. spec2vec ff2/noarch::spec2vec-0.2.0-pyhec00094_999 instead of spec2vec nlesc/noarch::spec2vec-0.2.0-pyhec00094_0. I repeated command a couple of time (even with channel reorder) and it kept installing the 999 package.

So I think adding a step before https://github.com/iomega/spec2vec/blob/refactor_environment/.github/workflows/conda_build.yml#L125 to change the build number in conda/meta.yml should fix it.

[florian-huber]

Thanks a lot @sverhoeven . That looks better than the hard coded fix I had added in the end (installing matchms first, and then spec2vec).

[florian-huber]

It also worked for me, locally. But here it still fails and picks spec2vec from nlesc.

[sverhoeven]

Grr, can you add listing for /home/runner/work/_temp/spec2vec/_build and ${RUNNER_TEMP}/spec2vec/_build/noarch in build and verify job respectivly? I want to check that the 999 package is available in verify job.

[florian-huber]

Seems to me like the file in /noarch is present as expected: spec2vec-0.2.0-pyhec00094_999.tar.bz2.
The "solution" of first installing matchms and then spec2vec did work, so as a backup we could move back to that. (it feels a bit weird though to hard code this into the conda recipe, but OK...)

conda install --channel bioconda --channel conda-forge --channel nlesc matchms
conda install --channel bioconda --channel conda-forge --channel $BUILDDIR spec2vec

[edit]:
We could also add matchms to the local build folder so that we would mimic a local nlesc channel and could only run conda install --channel bioconda --channel conda-forge --channel $BUILDDIR spec2vec ?

[sverhoeven]

Agree, let's rollback the build string 999 change and go with conda install matchms + conda install spec2vec then we don't have to depend on the conda resolving algorithm.

[sverhoeven]

Great to see the CI green again.

Just have some minor improvements.

[sonarqubecloud]

Kudos, SonarCloud Quality Gate passed!

0 Bugs
0 Vulnerabilities (and 0 Security Hotspots to review)
0 Code Smells

100.0% Coverage
0.0% Duplication

[florian-huber]

Thanks a lot for the help, Stefan!