f2c transpile via convert

[MichaelSt98]

C transpilation via convert()

• tested with CLOUDSC branch: nams_f2c_transpile_via_convert
• do we already want to get rid of transpile()?

[github-actions]

Documentation for this branch can be viewed at https://sites.ecmwf.int/docs/loki/208/index.html

[codecov]

Codecov Report

Attention: 3 lines in your changes are missing coverage. Please review.

Comparison is base (b85aafe) 92.30% compared to head (b7a9007) 92.31%.

Files	Patch %	Lines
loki/transform/fortran_c_transform.py	70.00%	3 Missing ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##             main     #208      +/-   ##
==========================================
+ Coverage   92.30%   92.31%   +0.01%     
==========================================
  Files          95       95              
  Lines       17368    17368              
==========================================
+ Hits        16032    16034       +2     
+ Misses       1336     1334       -2     

Flag	Coverage Δ
lint_rules	96.22% <ø> (ø)
loki	92.31% <70.00%> (+0.01%)	⬆️
transformations	91.47% <ø> (ø)

Flags with carried forward coverage won't be shown. Click here to find out more.

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

[reuterbal]

Many thanks, very much looking forward to bringing this in!

However, I think we should use this opportunity to go a little further:

• Remove the transpile entry point from loki_transform.py. Going forward, we don't want to use this anymore so better get rid of it right away. The only place that uses this is CLOUDSC. To avoid breaking the regression tests, I would convert the loki_transform_transpile CMake wrapper to a compatibility layer that reshuffles the provided options, prints a big deprecation warning, and then calls loki_transform, similar to what loki_transform_convert does.

• Make the Fortran2CTransformation compliant with the Transformation manifests. I'm not sure what would be the minimal invasive way (we can have an offline discussion to talk through the different options), but one possibility would be to remove all the explicit recursion from the transformation, and change the invocation for the headers such that it is invoked for every module in those.

I would, however, not include anything beyond that in this PR to not expand the scope too much. For example, a subsequent PR should split the F2C Transformation to generate derived type accessors ("the header transformation") in a separate Transformation class.

[reuterbal]

Ah, here it is applied directly to a file, but the F2C Transformation is not marked as recursing to internal scopes - instead it does this internally. I think we should take this opportunity to make the Transformation compliant with the manifest parameterisation. I'll explain in the overall review comment what I mean.

Secondly, I'm not too thrilled by the fact that we provide files as headers (which, to me, implies a "read-only for enrichment" meaning) and then apply transformations to them. Since the purpose here is to deal with derived types, which we cannot capture in the dependency graph in the current Scheduler implementation, we may have to live with this for now, but that will be different with #213. To avoid that others copy this pattern, would you mind including a comment here that explains that this is temporary?
Longer term, this transformation step to generate accessor methods for derived types should become a separate Transformation that is applied to ModuleItem only.

[MichaelSt98]

transpile() is now removed from scripts/loki_transform.py as the F2C transpilation is integrated in convert().

Unfortunately it was not possible to "re-route" the CMake loki_transform_transpile to loki_transform/loki_transform_convert because the CONFIG argument can't be generically reconstructed.

Depends on CLOUDSC PR 69.

[reuterbal]

Thanks, very happy with this now! This is a great base for the further developments.

Once @mlange05 has had a look, I suggest the following (as acted out before) as a merge strategy:

1. Remove the commit that points to the custom CLOUDSC branch
2. Merge the PR with regression tests failing
3. Tests on CLOUDSC PR should be successful now
4. Merge the CLOUDSC PR
5. Run regression tests on Loki main again, should now be successful.

[mlange05]

Needs a quick rebase, but otherwise GTG I think. Merge strategy sounds good too @reuterbal .

[mlange05]

Why does the F2C transformation require an "idem" pass?