Consider adding `io::read_repeating`

[tokio-rs]

Consider adding `io::read_repeating`

[alexcrichton]

I'm having trouble just grokking the 6 (!) type parameters here... Perhaps this could be implemented externally first, see if it gains traction, and then maybe move in?

[cramertj]

Yeah, it's definitely hard to read. I'm not sure how to make that easier. The basic idea is just that you're reading again and again into a user-provided buffer and passing a reference to the data read into a closure which can handle the data appropriately, either performing some side-effect or producing a Future of some data.

[cramertj]

I'd be happy to spin it into a gist/PR/its own crate if you think that would make the intent clearer. I don't think many people would download a crate just for this function, both because of overhead and because of the difficulty of discovering the functionality.

The goal is just to provide an alternative of sorts to Decoder that handles the incoming bytes in a buffer directly, rather than spinning it out into a Bytes.

[alexcrichton]

I agree that discoverability is a problem yeah but it's not really tenable for us to just add every combinator we can think of to tokio-io? I feel like we'd just need some threshold for inclusion beyond "a PR was sent" although I'm not quite sure what precisely such a threshold would be...

[cramertj]

it's not really tenable for us to just add every combinator we can think of to tokio-io

Yeah, I totally get that. I see io::repeating_read(read, [0; 1024], process_buf) as equivalent to read.framed(Bytes).map(process_buf), but it allows the users to provide their own buffer and so avoids the unnecessary buffer heap allocation. If that's behavior we want to encourage, then I think repeating_read should be included. If the opinion is that these intermediate allocations don't matter enough to merit the additional method and ergonomics cost, then I'm happy to close this, as I agree that it's better to provide a single way to do things where possible. However, if we want to encourage zero-alloc parsing of the incoming buffer, then tokio-io needs something like repeating_read.

[cramertj]

I've written a prototype implementation of this here (FWIW there are only 4 type parameters 😄).

[alexcrichton]

Ah so this is sort of where the Decoder and Encoder style traits are intended for "easy mode" in terms of being a good entry point but not a one-size-fits-all solution. I think most of the time we try to make it go further in terms of "one size fits all" then it becomes very difficult to work with, so we opted to just have an "easy, but not always production ready" mode and then leave other abstractions up to users

[carllerche]

Yes, tokio-io shouldn't include all possible permutations. It is totally fine to move helpers to external crates.

[carllerche]

Closing this was leaning towards "won't merge" and there has been no followup discussion. Please open a new issue if you want to continue this discussion.