Try using cabal-cache for github CI

[Mikolaj]

Here's the link:

https://hackage.haskell.org/package/cabal-cache

We could stop caching store/ and instead cache something safer, more selective and not causing our test to fail due to the dreaded

Error:     Ambiguous module name ‘System.Directory’:
      it was found in multiple packages:
      directory-1.3.7.1 directory-1.3.7.1

[Mikolaj]

See haskell/ghcup-hs#770. And also HLS uses it.

[ulysses4ever]

So, this tool has two modes of synchronization: S3 and local dir. We won't use S3 I assume. The local dir mode, from skimming their docs (which don't say much how they relate to caching the store, unfortunately!), feels like reinventing the store outside cabal purview. This is really sad, if you think about it: we're kind of saying that we don't trust our store, so we "simply" use a standalone reimplementation…

[Mikolaj]

Oh noes, @jneira would not betray us by running cabal-cache in HLS CI if it was such a blatant expression of a lack of trust in how cabal handles store/, would you @jneira?

[Mikolaj]

BTW, @hasufell seems to use a real s3 for that: https://github.com/haskell/ghcup-hs/blob/ec8333b22340b465e12610a614a5a59e94e5c930/.github/scripts/common.sh#L24

I don't know what advantages over a file handled by GHA there are, though, and where we'd get an s3 for us from.

[hasufell]

BTW, @hasufell seems to use a real s3 for that: https://github.com/haskell/ghcup-hs/blob/ec8333b22340b465e12610a614a5a59e94e5c930/.github/scripts/common.sh#L24

I don't know what advantages over a file handled by GHA there are, though, and where we'd get an s3 for us from.

I can provide a bucket with credentials for you.

[jneira]

Sorry, but i am not involved in HLS development anymore, since several months ago, so i am afraid i can say much about HLS ci. 🤷

[Mikolaj]

@jneira: oh, ok, that's fine; sorry to bother you :)

@hasufell: so what's the benefit of cabal-cache exactly? and is there any special benefit from using s3 as opposed to the github filesystem (in our case)? if there is a link with a ready explanation, please feel free to send me to RTFM

[hasufell]

@jneira: oh, ok, that's fine; sorry to bother you :)

@hasufell: so what's the benefit of cabal-cache exactly? and is there any special benefit from using s3 as opposed to the github filesystem (in our case)? if there is a link with a ready explanation, please feel free to send me to RTFM

github filesystem is limited in space.

[ulysses4ever]

github filesystem is limited in space.

Well, if this is the only benefit, I'd like to see an evidence that we actually run into problem with not enough space.

I can provide a bucket with credentials for you.

This is generous, thank you. I wouldn't want to buy into a solution with so high (low?) bus factor.

[hasufell]

I retract my offer. I don't have any stakes in this decision and don't have capacity or interest to assist.