Prepare 0.22.9 release

[hsyl20]

No description provided.

[hsyl20]

Somehow cabal-doctest doesn't find the new version of doctest. Isn't it because we pass --ignore-project?

[sol]

Somehow cabal-doctest doesn't find the new version of doctest. Isn't it because we pass --ignore-project?

Yes, that's expected. The cabal-doctest workflow will fail whenever you do a release. Build failures for cabal-coctest are informational for that reason, they don't block a merge. I don't have a better solution for this yet.

[hsyl20]

This MR now includes the fix for the testsuite to pass. See #448 (comment)

[sol]

With the right command line to run tests (I was using cabal test) I only get 1 failure with 9.10:

It's still not clear to me why this failure does not show up on CI. Is this maybe depending on the value of TERM or something? On the one hand, this would make sense, on the other hand you shouldn't look at TERM at all if the output is not going to a terminal.

[sol]

on the other hand you shouldn't look at TERM at all if the output is not going to a terminal.

The reason why ghci prints those links in the first place is because of

doctest/src/Language/Haskell/GhciWrapper.hs

Line 64 in f930eaa

evalThrow interpreter "GHC.IO.Handle.hDuplicateTo System.IO.stdout System.IO.stderr"

Redirection of stderr is only done after ghci has started. At that point GHC already decided that output is going to a terminal, hence we see the links.

A proper solution would be to do something along the lines of: https://github.com/hspec/sensei/blob/5c11026fa48e13ea1c351ab882765eb0966f2e97/src/Language/Haskell/GhciWrapper.hs#L63-L71

But things are actually even more complicated than that (not elaborating on it now, but if you want to work on it then please ask and I'll give the full story). That's why we are going with #448 for now.

[sol]

@hsyl20 we need to differentiate between:

1. Extracting comments with the GHCAPI.
2. Evaluating expressions with GHCi.

For (1) we want the exact same error behavior as GHC, including colors and hyperlinks.

For (2) we don't want colors and hyperlinks.

This code is concerned with (1), so we actually don't want this changes.

For the related test failure, this is actually an order dependency somehow. If you focus the single test, then that test succeeds. You could be anywhere, including in GHC itself. I don't have the time to investigate further right now. It doesn't happen on CI, so releases are not blocked by it.

[sol]

Please include only the version bump in this PR. (see my other comments)

[hsyl20]

I've removed the other commit. I won't have much time to work on this soon either.