esm_runscripts: wire up add_vcs_info / check_vcs_info_against_last_run (currently dead code)

[siligam]

Do you know why add_vcs_info and check_vcs_info_against_last_run (src/esm_runscripts/prepare.py) aren't run by default? They're commented out in every recipe YAML across configs/ and never called from Python — confirmed no *_vcs_info.yaml exists in a real Albedo experiment either.

Sounds like a good feature and it's a requirement from users for catalog/provenance metadata (see #1407, #1502, #1489).

[mandresm]

Those features were never fully tested. They were activated in production for a few days and it broke esm_runscripts, so we deactivate them. We never fixed it because we had to prioritize more urgent topics. Now they are a requirement for the catalogs, so we should consider fixing it. I suggest we add it to the less pressing list of things for the catalogs.

[pgierz]

I think part of the difficulty was that one couldn't always be sure users wouldn't manually mess with their <component>.model_dir between build-time and sim-time. There are several nuances that could be discussed.

There was long ago the idea to have the thing automatically do esm-master operations as part of the first run, so the yaml would be an entirely reproducible source of truth for your experiment (download, compile, simulate, archive) all from one thing. We obviously never got that all the way finished. If you look in the stubs for something like "First year, seeing if we need to compile", you'll see where it was originally going to get plugged in.

As a first shot: My recommendation would be to include an extra step in the prepexp/prepare part that, one the very first run, copies the model_dir into src, which we already have in the experiment tree. Then you at least know concretely where to look.