Speed up GH build pipeline

[finestructure]

Unfortunately, we can't benefit from this new flag that uses pre-built SwiftSyntax libraries at the moment.

Due to our dependence on SwiftPM's branch release/6.1, we depend on SwiftSyntax release/6.1, which does not have a prebuilt library and we end up building it ourselves, as before 😞

[finestructure]

I've asked on the forums if prebuilts might become available for release/6.1. If not, that's another reason to try and drop SwiftPM as a dependency.

https://forums.swift.org/t/preview-swift-syntax-prebuilts-for-macros/80202/21

[finestructure]

Actually, this might work. We have good reason to fork SwiftPM already - there are two open PRs we need and they're unlikely to be be merged and cherrypicked to release/6.1 anyway:

• Mark a few PackageCollectionModel types as Sendable swiftlang/swift-package-manager#8635
• Widen swift-crypto version range swiftlang/swift-package-manager#8680

Especially the latter we need. We are currently not getting Postgres-NIO package updates due to that.

So since we need that fork, we might as well change SwiftPM's dependency on SwiftSyntax to a release version that has a prebuilt instead of the release/6.1 branch. For what we actually use from SwiftPM this will make no difference.

[finestructure]

I'm not sure if this is working at the moment. Not sure if it's still the wrong SwiftSyntax version or something else. We're getting

on this branch here, vs

on main. The release build is ~10% faster, true, but this might just be jitter?

[MahdiBM]

I just bumped into this PR and I want to shamelessly plug this article we published yesterday: https://swiftonserver.com/faster-github-actions-ci-for-swift-projects/

I believe it'll have a way bigger impact on your build times than using prebuilts, although they don't contradict eachother.
I'd say it'll lower both build times to less than 5 mins, possibly even around 3 mins.

[finestructure]

Ah, that looks very interesting! I’ve been meaning to look into caching and his seems to be exactly what we need :) Thanks @MahdiBM !

[finestructure]

Ok, this works great (down from ~17min, see above)

[finestructure]

Much better, now QueryPerformaceTests is the bottle neck

[finestructure]

End result: