Support GHC-9.4

[TeofilC]

This is an early attempt to get this working with GHC-9.4.1.

At the moment we need to use some patched versions of libraries taken from yesodweb/yesod#1769

There's also an issue with changes to the constraint solver and TypeErrors (upstream issue: https://gitlab.haskell.org/ghc/ghc/-/issues/21677). I've worked around it by applying some TypeApplications.

This is mostly just here to save work for others who want to try using 9.4

[TeofilC]

Now that 9.4 has been released I'll try to tidy this up in the next couple of days

[TeofilC]

Currently everything builds except servant-auth-swagger and servant-auth-docs as they require doctest to build and afaik that's not been updated to work with 9.4. All the tests that can be compiled pass though.

I feel like the changes to cabal.project should be dropped before merging, but I've left them there for now to ease testing.

[alpmestan]

@TeofilC Thanks a lot for this. I just allowed CI to run for this branch. I'll try to keep an eye on this PR, let me know when you consider this ready (i.e the doctest business is sorted out etc).

Cheers!

[TeofilC]

Still waiting on some things but I've updated the overrides and listed the allow-newers

[benz0li]

Cross reference: #1608

[TeofilC]

Just a quick update on this. We are still waiting on sol/doctest to support 9.4 as well as the postgresql packages to bump bounds.

[TeofilC]

I've rebased on top of master and updated the cabal.project to use sol/doctest HEAD, which supports 9.4. I have the doctests passing locally now. We are waiting on some hackage releases/ version bumps in order to be able to drop stuff from the cabal.project.

It looks like there is a genuine failure on the ghcjs 8.6 test suite because it doesn't seem to support the type applications syntax that I'm using to fix stuff for 9.4. There's several ways we could fix this:

1. drop support for ghcjs 8.6
2. hide the type applications behind CPP, so, it only occurs for GHC 9.4+
3. try to figure out some way of avoiding this issue without using type applications

[tchoutri]

After some thought, I vote in favour of 2.

[TeofilC]

I've added a commit that implements that option.

[maksbotan]

I've merged servant-multipart and servant-js PRs, please update this one.

[TeofilC]

I'll take a look at the doctest failures tomorrow

[maksbotan]

Looks like CI for 9.4 failed

[TeofilC]

The doctest failures are caused by GHC's error output changing.

It used to produce message that start with Could not deduce but now they start with Could not solve.

I'm going to update the tests to just expect Could not as that's common to both and still ensures that some sort of error's been triggered. See 86c61c6

[maksbotan]

Thanks!

Published Hackage revisions:

• servant-client-core
• servant-client
• servant-auth-client
• servant-auth-docs
• servant-auth-server
• servant-auth-swagger
• servant-docs
• servant-foreign
• servant-http-streams

New versions:

• servant-server 0.19.2
• servant 0.19.1

[benz0li]

Thank you @TeofilC