CMake refactoring for loki_transform

[reuterbal]

This refactors the CMake integration of Loki. As this is intended to incorporate the changes in #149 and resolve the outstanding issues in there, this is filed as a PR against that branch.

The following changes are applied:

• A single entry point loki_transform is now the default invocation of loki-transform.py, replacing loki_transform_convert and loki_transform_command. This is also used by loki_transform_target to apply the actual transformations.
• The various utility macros to translate arguments of the CMake functions to command-line arguments for loki-transform.py have been consolidated into two routines to parse options and arguments. This is possible because validating arguments is now left purely to the Python layer, i.e., the loki-transform.py script. However, the CMake functions calling these translation macros still only accept the explicitly provided list of options and arguments, thus we don't risk silently ignoring options, e.g. when they have a typo or use an outdated interface.
• loki_transform_convert is marked as deprecated but retained as an entry point for backwards-compatibility, since this is used by regression tests. Once the regression test repositories have been brought forward, this entry point can be completely dropped
• loki_transform_plan will remain as a separate entry point since this invokes loki at configuration (and not at build time).
• loki_transform_transpile remains as well, since the interface in the CLI script is very different from convert. However, I have applied the same renaming of options there to align this further, in preparation for future integration into the convert command.

Overall, the loki_transform.cmake script has shrunk by almost 25% (mostly by moving purely utility routines to a separate file), but can be further reduced when the deprecated convert and transpile entry points are removed.

Tested locally for CLOUDSC, waiting for CI now to confirm this works also on ecWAM.
@mlange05: Please test this for ecphys!

[github-actions]

Documentation for this branch can be viewed at https://sites.ecmwf.int/docs/loki/166/index.html

[reuterbal]

PS: This also fixes an unrelated bug related to the member inliner, which wasn't applied even when --inline-members was specified. This was due to the fact that the option wasn't forwarded to the SCCBaseTransformation.

[codecov]

Codecov Report

Merging #166 (d3f80bc) into naml-convert-argument-array-shape (9cf5f51) will not change coverage.
The diff coverage is n/a.

@@                        Coverage Diff                         @@
##           naml-convert-argument-array-shape     #166   +/-   ##
==================================================================
  Coverage                              92.10%   92.10%           
==================================================================
  Files                                     90       90           
  Lines                                  16628    16628           
==================================================================
  Hits                                   15315    15315           
  Misses                                  1313     1313           

Flag	Coverage Δ
lint_rules	96.22% <ø> (ø)
loki	92.08% <ø> (ø)
transformations	91.25% <ø> (ø)

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

[mlange05]

This looks fantastic - many, many thanks for taking this over! I've flagged one showstopper that breaks regression (yes, we need to start using the "target" entry point in CLOUDSC!); and flagged a minor outdated comment. Should be good once the target mode is fixes and regression runs!

[mlange05]

I think ecphys is basically deprecated now too, no?

[reuterbal]

Thanks, good spot. Yeah, ecphys is already gone. Will change to convert.

[mlange05]

EC-phys regression tells me that there's still a dangling call to this, but it seems to be removed entirely. Was this intended to be moved to helpers, or is it entirely redundant now? In the latter case loki_transform_target needs a small update.

[reuterbal]

Thanks for testing and flagging! Yes, we need regression testing of this entry point asap. Please have another go, I have removed the dangling call which shouldn't have provided more than argument checking in this instance.

[mlange05]

Ok, again, many, many thanks for this (beyond initial scope, but dearly needed!). ECphys regression ran fine now, so GTG from me.