Borrowing arguments ("Alan writes a web framework")

[alsuren]

This PR contains a write-up of my experiences from #78 (comment)

There is a bit of solutionising towards the end. I can rip that out if you want:

Alan longs to be able to say "this function takes an async function as a callback" (fn register_handler(handler: impl async Fn(state: &mut State) -> Result<Response, Error>)) and have Rust elide the lifetimes for him, like how they are elided for async functions.

It's also super-trimmed-down compared to my full experiences, and we could probably flesh out the morals a bit better.

[jbr]

A big plus one on this one. Not being able to describe the trait of async functions with non-static lifetimes in the arguments is by far the biggest pain in async rust right now. I'm excited to see how GATs improve this situation, but that's about solutions. Whole web framework designs might be different if mutable borrows could be passed into handler functions instead of owned types

[nikomatsakis]

Thanks! A few comments based on a quick read! The short version is that more examples would help bring the story to life for me.

[rylev]

There's definitely some really great insights here, but it looks like it's missing a bit of detail to really make the lessons obvious. Adding example errors and a bit more explanation of what exactly was confusing will go a long way.

[tanriol]

You're missing one more problem here: the final to_async_borrowing works nicely for async functions, but not for closures because the closure type inference is not powerful enough (rust-lang/rust#70263), or at least the developers haven't managed to trick rustc into handling it yet.

[nikomatsakis]

I feel like this is pretty good now! Let me review the outstanding comments, but I might just merge.