Replace select!/try_select! with Future::{race,try_race}

[yoshuawuyts]

This changes the select and try_select macros into methods on Future. Resolves #387. Thanks!

Motivation

The output of join and try_join is polymorphic (n-sized tuple) which requires a macro. select and try_select don't require polymorphism which makes more sense for it to be a method.

Docs issues

Currently the only problem with this patch is that, uhh, the docs no longer build. There's something in the macro that's not working, and I'm not sure what. cc/ @stjepang maybe you have a clue?

Future work

I think we should consider adding join and try_join as methods too, in addition to the existing macros. That way there's polymorphic + non-polymorphic versions of both available, and the APIs become symmetrical again for the non-polymorphic case. Besides, it seems likely that joining 2 futures is the most common case, which can be a nice shorthand.

Screenshot

[yoshuawuyts]

okay yay, got the docs to compile! -- needed to set the params to std::future::Future. While not great, at least it works! 😁

[ghost]

I like having select as a function rather than macro.

This reminds me of cmp::min()/cmp::max()/Ord::min()/Ord::max(). Those functions take two parameters even though sometimes we want to take a minimum/maximum of more than two things. They work fine in practice.

I wonder what you think about having select and try_select as methods on FutureExt versus standalone functions in the future module. We currently have future::timeout() function that could've been Future::timeout() method.

Looking at the methods on Iterator and functions in the std::iter module, it seems to me some of the rules are this:

• Iterator constructors go into the std::iter module.
• Iterator transformers and consumers are methods on the Iterator trait.

However, since there are both cmp::min() and Ord::min(), it seems there is also a third rule:

• If a.foo(b) and b.foo(a) are equivalent, then foo is handy as both a standalone function and as a trait method.

By that logic, select should indeed be a trait method, although probably it wouldn't hurt to also have it as a standalone function. But I'm not insisting, just thinking out loud. Not sure what to make of this :)

Leaving down here just a few minor comments, otherwise 👍

[yoshuawuyts]

By that logic, select should indeed be a trait method, although probably it wouldn't hurt to also have it as a standalone function. But I'm not insisting, just thinking out loud. Not sure what to make of this :)

I think that's quite reasonable! Especially for join / try_join. But since they're macros, I wonder if select / try_select should also be macros? This is a bit tricky because there's not really any precedent in the stdlib for this either. Good questions tho; not sure either!

[yoshuawuyts]

Implemented all feedback!

[ghost]

Thanks for addressing the feedback! Feel free to merge when CI passes. Looks like Clippy is broken again :(

[yoshuawuyts]

Tests are currently broken, filed #431 to fix it.