ClientIO isn't useful.

[TravisWhitaker]

For better or for worse, the set of exceptions that may be thrown in IO is always open. This makes types like ExceptT e IO a essentially useless. runClientIO can actually return an IO that throws any exception, and checking for Left ClientError is just extra noise. I always use this function when interacting with this library (and all other libraries that use this pattern):

runThrowClientIO :: ClientIO a -> IO a
runThrowClientIO m = do
    res <- runExceptT m
    case res of Left e  -> throwIO e
                Right a -> pure a

Library users can be allowed to migrate away from ClientIO if they choose in a largely backwards-compatible way. The trick is to define a class like this:

class MonadIO m => MonadClient m where
    throwClientError :: ClientError -> m a

instance MonadClient IO where
    throwClientError = throwIO

instance MonadClient ClientIO where
    throwClientError = throwError

Functions in the library can then be lifted from ClientIO a to MonadClient m => m a without breaking too much existing code. Users are then free to choose ClientIO, IO, or to provide an instance for whatever monad they're already working in.

[lucasdicioccio]

Yeah I sort of agree with you and was adding ClientIO when I wanted to sort of get a better control on how to cleanly handle cases with http2-thrown exceptions in a single stream in the client code itself (before that, I could live with IO-about-everything but found unsatisfying the lack of guidance in the docs). I've yet to make up my mind about the pattern, the typeclass-style could definitely work.

Allow me a bit more time please: I'll be spending my OSS-time budget working on my gRPC libs in the near term.

[TravisWhitaker]

If you're open to the transition I'd be happy to open a PR implementing the MonadClient-driven approach. Just wanted to see how you felt about it before proceeding with that work.

[lucasdicioccio]

I’ll consult with some colleagues and let you know soon :-). Thanks for proposing your help, I appreciate that a lot! — Lucas

On 6 Nov 2019, at 02:01, Travis Whitaker [email protected] wrote:

If you're open to the transition I'd be happy to open a PR implementing the MonadClient-driven approach. Just wanted to see how you felt about it before proceeding with that work.

— You are receiving this because you commented. Reply to this email directly, view it on GitHub https://github.com/lucasdicioccio/http2-client/issues/71?email_source=notifications&email_token=AABUDUX4GFHOXXRK6SAGLG3QSIJOTA5CNFSM4JGRSLY2YY3PNVWWK3TUL52HS4DFVREXG43VMVBW63LNMVXHJKTDN5WW2ZLOORPWSZGOEDE4CQQ#issuecomment-550093122, or unsubscribe https://github.com/notifications/unsubscribe-auth/AABUDUWGS25VSOFLSDD4YRTQSIJOTANCNFSM4JGRSLYQ.

[ProofOfKeags]

Hi. @lucasdicioccio gave me the go ahead to work on maintenance for this library and I'm very interested in getting rid of ClientIO altogether. I don't have a great idea of the extended dependents of this library and as a result don't have a great idea if the backwards compatibility is a real concern or not. Especially if we can just bump major versions and rip the bandaid off altogether. @TravisWhitaker If you are still interested in this change happening I'd love to discuss further, otherwise I'll figure out what to do at some point.

[TravisWhitaker]

@ProofOfKeags Great, I'd be happy to help out any way I can. I'm definitely in favor of just eliminating ClientIO with a major version bump.

[ProofOfKeags]

The first thing I need to figure out is how to release 0.10.0.1. I ostensibly have the hackage permissions to do so at this point but I have never actually published to hackage before. After that, I think eliminating ClientIO is the next priority.

[SkamDart]

@ProofOfKeags Is removing ClientIO still on the roadmap? Happy to work with @TravisWhitaker or yourself to get this into the next release.

[ProofOfKeags]

I have bandwidth to review and merge etc but not to do it myself so yes-ish.