Andrei Dziahel

> Interestingly, all PR caches from yesterday have already been evicted while main caches are still there. Oh, were they? That'd be really useful. Were the corresponding branches deleted or...

> Maybe we can do better with some combination of ghcup and `--system-ghc` or so. We could even set stack up with `ghcup`, and [configure stack in a way it'll...

> Since all caches are evicted after 7 days anyway, we could just run a daily workflow to restore (and save?) the current caches on main. This way they'd always...

> Not really, they use different GHC versions Why do we stick to 9.4.5 for Stack, BTW? 9.4.8 is in Stackage for a month or so

Excellent point on caching GHC and dependencies in non-nix builds separately

> do you see any way of "not storing the cache" with cache-nix-action? @wolfgangwalther perhaps, use [`restore` action](https://github.com/nix-community/cache-nix-action/tree/v4.0.3/restore#only-restore-cache) only? That's what the [`cache-nix-action` is using internally](https://github.com/nix-community/cache-nix-action/blob/v4.0.3/action.yml#L68) anyway.

> I start to wonder why do we bother with caching this stuff via github actions at all? Nix caching does shave off a minute or two per Nix-involving job,...

I mean, the parts you _have_ done already could be extracted to separate PR and merged, making this PR smaller and easier to review in the end.

The https://hackage.haskell.org/package/configuration-tools (source at https://github.com/alephcloud/hs-configuration-tools) could be the right tool for the job — it combines CLI options parser, *JSON instances, JSON/YAML config parsing (that's where it falls short for...

@mkleczek whoa, you've got yourself a nice `flake.nix` too! Seems to be way ahead of https://github.com/PostgREST/postgrest/pull/3105 of mine (and working, I presume?). Do you mind us borrowing a thing here...