Check if switch exists before creating

[tmattio]

Hi!

I noticed that @smorimoto worked on caching on a separate repo and tried to integrate https://github.com/avsm/hello-world-action-ocaml/blob/with-cache/.github/workflows/workflow.yml in Opium (rgrinberg/opium#183)

Unfortunately, it seems that caching the ~/.opam directory does not work with multiple compilers at the moment because ocaml-setup uses opam switch create even if a switch with the same name already exists.

This PR adds a check that will only create a switch if it does not already exist. The check is really basic, maybe there's a better way to do this with opam?

[dra27]

Changes need to be made in both the Windows and Unix install scripts (even if the former is not at present using caching).

Just for general reference, the venerable OCaml Travis CI Scripts are a good source of opam fu. The test you're proposing is very verbose to harden to deal with substrings and spaces - the better pattern is opam switch set "$1" || opam switch create "$1" "$1" (cf. https://github.com/ocaml/ocaml-ci-scripts/blob/master/.travis-ocaml.sh#L307)

[tmattio]

the better pattern is opam switch set "$1" || opam switch create "$1" "$1"

Very nice trick! I read Opam's manual in search of such a solution, but didn't figure this one out!

I looked at the Windows installation script and it does not seem to use opam switch create, so I don't think the issue would happen on Windows, even if it could be cached.

[smorimoto]

Yes, I did, but I don't recommend that users use it. Because at this time, there are very few ways for users to strictly cache using actions/cache.

[tmattio]

@smorimoto Interesting! What kind of issue could happen with this setup?

[smorimoto]

Well, this change is not bad for hackers. (So I'm not trying to argue against this PR.)

[dra27]

I looked at the Windows installation script and it does not seem to use opam switch create, so I don't think the issue would happen on Windows, even if it could be cached.

Oh, good point - it always init's!

[smorimoto]

A possible issue at this point is that actions/cache can sometimes corrupt the cache. We now have an idea to solve it and are already working on it, but if you remove the step runs opam upgrade --fixup from pushed setup right now, the build may fail. I haven't heard of it happening outside of the Rust project yet, but anyway the setup I pushed must be pretty fragile.

[smorimoto]

It's also fragile without opam lock.

[smorimoto]

This looks good to me. If @dra27 doesn't request additional changes to this, I will merge this and release the new patch version.

[dra27]

LGTM!

[smorimoto]

I'm out now, so I'll release a new version when I get home.

[smorimoto]

Thanks @tmattio!

[tmattio]

Great, thanks for your work on caching 😉

[smorimoto]

Well, that was important to our company, so anytime!