Fix the build and tests with exceptions 0.9.*

[gelisam]

The builds for PRs #52 and #53 both fail with the same error:

During the build phase, there is a warning which wasn't there before, saying that the generalBracket method is missing from a MonadMask instance. Neither PR touches that part of the code, and yet previous builds don't include the warning.

This is only a warning, but alas, both warnings become errors later on, when the unit tests are executed. In fact, each test fails separately with the same error message, indicating that it must be hint which is somehow trying to load its own code and failing because of a warning.

I did write both PR #52 and PR #53, but I don't think it is my changes which have caused this. One reason is that the two PRs touch different parts of the codebase and yet fail in the same way. The other is that while PR #53 configures Travis so that it also attempts to build with ghc-8.4, it doesn't make any change which should impact the build for the other versions of ghc, and yet the build fails for those versions of ghc as well.

My conclusion is that something in the build environment must have changed between the last (successful) build and these ones.

[gelisam]

I can reproduce this locally, although with difficulty. I cannot reproduce the problem when running the tests with ghcid --comman="stack ghci hint:test:unit-tests" --test=main, stack test, nor cabal new-test, but I can reproduce the problem with cabal test. I deleted my ~/.cabal and ~/.ghc folders and carefully run the same commands as Travis, namely cabal update, cabal install --only-dependencies --enable-tests, cabal configure --enable-tests, and finally cabal test, but I don't know if all those steps are necessary to reproduce the problem.

Since I managed to reproduce the problem locally, I don't think that Travis changed their build environment; rather, I think this is the usual issue with cabal builds, namely that someone publishing a new version of a package somewhere can change our build plan and cause the build to fail mysteriously even though no code has changed. I bet a lower-bound is missing somewhere, not necessarily in our list of dependencies but perhaps in theirs.

[mvdan]

The joy of builds that aren't stable and reproducible.

I'll look into this tomorrow - thanks for bringing it up.

[mvdan]

I deleted my ~/.cabal and ~/.ghc, ran cabal install and cabal test, and it succeeded. Though, my dependencies originate from the Arch Linux packages, not from cabal. This makes me think that some dependency's minor or patch version broke the build.

I ran the Travis steps, and it still ran fine. So it appears the exact commands are not to blame (and I wouldn't see why, as they've been the same for a long time).

[mvdan]

My best guess is that this library's 0.9.0 update broke the build: https://hackage.haskell.org/package/exceptions-0.9.0

It defines generalBracket, CI is using 0.9.0, and I'm still on 0.8.3.

[mvdan]

Bingo - removing the lib from my system and re-running install and test, and the tests are now failing.

[mvdan]

Success: https://travis-ci.org/mvdan/hint/builds/348662873

I'm going to retitle this issue to remove this temporary dependency restriction for the upcoming release.

[mvdan]

More information on the API breakage:

https://hackage.haskell.org/package/exceptions-0.9.0/docs/Control-Monad-Catch.html#v:generalBracket

commercialhaskell/stackage#3315

[gelisam]

Please also add the upper-bound for published versions of hint, using hackage's revision mechanism. On a blank system (such as the docker images I have created for #56), cabal install hint successfully installs a broken hint which always fails with a generalBracket-related message.

[mvdan]

You're right, I forgot about the stable version. Done: https://hackage.haskell.org/package/hint-0.7.0/revisions/

[gelisam]

I have investigated exceptions-0.9.0 and have discovered an important regression. I sent them a fix and they published exceptions-0.10.0, which breaks the API in a slightly different way than exceptions-0.9.0 did.

I can write a MonadMask instance which will work with exceptions-0.10.0, but it won't work with older versions of the exceptions library. Are you okay with requiring exceptions >= 0.10.0?

[mvdan]

It doesn't seem like we have much of an option in the long term. If anyone is stuck with exceptions 0.8.0, they are likely also stuck with an older version of GHC, in which case using a slightly older version of Hint should be fine.

[gelisam]

The main reason for using exceptions-0.8.3 instead if exceptions-0.10.0 at the moment would not be because of an old ghc, but because a dependency has an `exceptions <= 0.9` upper bound. I have spent the week at work updating dependencies to their latest version in the hope that they have already updated their upper bounds, and sending them a PR otherwise. It will take a while for the entire ecosystem to update their upper bounds.

…

On Thu, Mar 15, 2018, 12:49 PM Daniel Martí ***@***.***> wrote: It doesn't seem like we have much of an option in the long term. If anyone is stuck with exceptions 0.8.0, they are likely also stuck with an older version of GHC, in which case using a slightly older version of Hint should be fine. — You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub <#54 (comment)>, or mute the thread <https://github.com/notifications/unsubscribe-auth/AAC_aKvRl3W9395lf7QymBSzj4NnB_1oks5tept_gaJpZM4SY0ei> .

[mvdan]

Is there any such library that won't compile with Hint 0.10.x? Besides hint, that is :)

The set of libraries using exceptions has already been split in two; it's just a matter of what side we want to be on. I'd say go for the newer for the upcoming Hint release, unless there is a good reason not to.

[gelisam]

I'd say go for the newer for the upcoming Hint release

Do you have a timeline on when you plan to perform this release? I still want to implement this, but I haven't had a lot of time to put on this and I don't want to hold you back.

[gelisam]

Never mind, I found the time for it this morning :)

[mvdan]

I wanted to do the release sometime in April or May, but I was briefly on holiday and now travelling :) I'll look at your patch now.