The Conduitpocalypse

At the end of last week, I made a number of breaking releases of libraries. The API impact of these changes was relatively minor, so most code should continue to work with little to no modification. I'm going to call out the major motivating changes for these releases below. If I leave out a package from explanations, assume the reason is just "upstream breaking changes caused breaking changes here." And of course check the relevant ChangeLogs for more details.

For completeness, the list of packages I released at the end of last week is:

• conduit-1.3.0
• conduit-extra-1.3.0
• network-conduit-tls-1.3.0
• resourcet-1.2.0
• xml-conduit-1.8.0
• html-conduit-1.3.0
• xml-hamlet-0.5.0
• persistent-2.8.0
• persistent-mongoDB-2.8.0
• persistent-mysql-2.8.0
• persistent-postgresql-2.8.0
• persistent-sqlite-2.8.0
• persistent-template-2.5.3.1
• persistent-test-2.0.0.3
• conduit-combinators-1.3.0
• yesod-1.6.0
• yesod-auth-1.6.0
• yesod-auth-oauth-1.6.0
• yesod-bin-1.6.0
• yesod-core-1.6.0
• yesod-eventsource-1.6.0
• yesod-form-1.6.0
• yesod-newsfeed-1.6.1.0
• yesod-persistent-1.6.0
• yesod-sitemap-1.6.0
• yesod-static-1.6.0
• yesod-test-1.6.0
• yesod-websockets-0.3.0
• classy-prelude-1.4.0
• classy-prelude-conduit-1.4.0
• classy-prelude-yesod-1.4.0
• mutable-containers-0.3.4

Switching to MonadUnliftIO

The primary instigator for this set of releases was moving my libraries over from MonadBaseControl and MonadCatch/MonadMask (from the monad-control and exceptions packages, respectively) over to MonadUnliftIO. I gave a talk recently (slides) at LambdaWorld about this topic, and have blogged at length as well. Therefore, I'm not going to get into the arguments here of why I think MonadUnliftIO is a better solution to this class of problems.

Unless I missed something, this change dropped direct dependency on the monad-control, lifted-base, and lifted-async packages throughout all of the packages listed above. The dependency on the exceptions package remains, but only for using the MonadThrow typeclass, not the MonadCatch and MonadMask typeclasses. (This does leave open a question of whether we should still define valid instances of MonadCatch and MonadMask, see rio issue #38.)

User impact: You may need to switch some usages of the lifted-base package to unliftio or similar, and update some type signatures. It's possible that if you're using a monad transformer stack which is not an instance of MonadUnliftIO that you'll face compilation issues.

Safer runResourceT

In previous versions of the resourcet package, if you register a cleanup action which throws an exception itself, the exception would be swallowed. In this new release, any exceptions thrown during cleanup will be rethrown by runResourceT.

conduit cleanups

There were some big-ish changes to conduit:

• Drop finalizers from the library, as discussed previously. This resulted in the removal of the yieldOr and addCleanup functions, and the replacement of the ResumableSource and ResumableConduit types with the SealedConduitT type.
• Deprecated the old type synonyms and operators from the library. This has been planned for a long time.
• Moved the Conduit and Data.Conduit.Combinators modules from conduit-combinators into conduit itself. This increases the dependency footprint of conduit itself, but makes it a fully loaded streaming data library. conduit-combinators is now an empty library.

Yesod: no more transformers!

The changes mentioned in my last yesodweb.com blog post have been carried out. The biggest impact of that is replacing HandlerT and WidgetT (as transformers over IO) with HandlerFor and WidgetFor, as concrete monads parameterized by the site data type. Thanks to backwards compat HandlerT and WidgetT type synonyms, and the Template Haskell-generated Handler and Widget synonyms being updated automatically, hopefully most users will feel almost no impact from this. (Authors of subsites, however, will likely have a more significant amount of work to do.)

That's it?

Yeah, this post turned out much smaller than I expected. There are likely breakages that I've forgotten about and which should be called out. I'll ask that if anyone notices particular breakages they needed to work around, to please either include a note below or send a PR to this blog post (link above) adding information on the change.