Add Reader::skip_exact method

[canndrew]

This implements the fix for #19022

[rust-highfive]

Thanks for the pull request, and welcome! The Rust team is excited to review your changes, and you should hear from @nikomatsakis (or someone else) soon.

[nodakai]

Oops, please have a look at my half-baked and stale PR #18137

I didn't have progress in it because I was thinking too much about #13989 (comment)

My current plan is to make skip() a "pure virtual (C++ terminology)" and add an auxiliary method skip_by_read() for its default implementation. Pros: each implementation of the Read trait can choose the most efficient implementation for its skip(). If repeated read() folowed by discarding a buffer is the only way to do so, we can simply implement skip() on top of skip_by_read(). Cons: because Rust trait doesn't have the notion of private/protected methods, exposing an auxiliary method skip_by_read() is a bit ugly.

[nikomatsakis]

I'm not familiar enough with the direction that this API is taking to judge this particular pull request. @alexcrichton or @aturon, would one of you take this up please?

[aturon]

cc @sfackler @erickt

[canndrew]

Here's a thought. A good optimisation for the default implementation might be

fn skip_exact(&mut self, n: uint) -> Result<()> {
    if n < THRESHOLD {
        let buf: [u8, ..THRESHOLD] = unsafe { mem::uninitialized() };
        try!(self.read_at_least(n, buf[..n]));
    }
    else {
        try!(self.read_exact(n));
    }
    Ok(())
}

ie. read to the stack if we're only reading a small amount (I suspect people will often only be skipping a few bytes here or there). Not sure how to pick a good value for THRESHOLD though.

[nodakai]

All you need is read_at_least() into an on-stack array in a loop: nodakai@9fd9953

[aturon]

@canndrew I'm sorry for the long delay getting back to you on this. My basic feeling is that this is a relatively rare thing to need to do, and looping with read_at_least with an on-stack array is a simple enough work around, that we don't want to expand the API at this time. (We are also in the process of a complete overhaul of std::io, so I'd like to wait for expansions at least until after that's complete).

[alexcrichton]

To echo @aturon and my similar comment in @nodakai's PR, I'm going to close this for now in expectation of the upcoming std::io RFC.