Implement most of RFC 2930, providing the ReadBuf abstraction

[beepster4096]

This replaces the Initializer abstraction for permitting reading into uninitialized buffers, closing #42788.

This leaves several APIs described in the RFC out of scope for the initial implementation:

• read_buf_vectored
• ReadBufs

Closes #42788, by removing the relevant APIs.

[rust-highfive]

r? @kennytm

(rust-highfive has picked a reviewer for you, use r? to override)

[jyn514]

@drmeepster do you mind squashing the commits? 41 is a lot to read through.

[beepster4096]

Alright I squashed it.

[beepster4096]

Hopefully, weird errors in files I didn't touch don't show up this time.

[beepster4096]

uh I think this PR is cursed

[jyn514]

uh I think this PR is cursed

We debugged this locally, here's an MCVE:

> cat doctest.rs
//! ```
//! extern crate proc_macro;
//!
//! # fn main() {}
//! ```
> rustdoc +rustc2 --test doctest.rs 

running 1 test
test doctest.rs - (line 1) ... FAILED

failures:

---- doctest.rs - (line 1) stdout ----
error: expected one of `;` or `as`, found `<eof>`
 --> doctest.rs:2:14
  |
2 | extern crate p
  |              ^ expected one of `;` or `as`

Notice how the error unexpectedly cuts off after crate p. So it's at least feasible it's related to this PR - either Rustdoc has UB or one of the changes introduced a logic error. It's strange that this only happens for rustdoc and not rustc.

[beepster4096]

The remove unresolved import commit passing CI means that the error is caused by the changes to BufReader

[djc]

Very excited, thanks for working on this! I don't have a lot of experience with larger std changes like this, but I feel like it would be easier to land this if you focused on getting an initial PR merged that only includes the ReadBuf implementation (plus tests), and leave usage of the new type to future PRs (can still keep your changes, of course!).

[kennytm]

r? @KodrAus

[bstrie]

r? @sfackler

[joshtriplett]

In the interests of collecting possible naming considerations in one place:

• The RFC uses assume_init.
• assume_init seems evocative of an unsafe API: you're not making something initialized, you're making an assumption.
• add_init or similar would more clearly indicate the addition to the initialized region, rather than replacing it.
• Something like assume_added_init or add_assumed_init would have the benefits of both, but be more verbose.

[nrc]

As far as I can tell, parts of this are insta-stable, and thus need an FCP

@joshtriplett which parts will be insta-stable?

[joshtriplett]

We reviewed this in today's @rust-lang/libs-api meeting, and confirmed that there are no insta-stable bits of this. We do see two potential issues with this that should be covered and addressed in the tracking issue:

• assume_init naming
• &mut ReadBuf allows replacing the whole ReadBuf, which would allow pointing it to a different buffer of memory. Some callers may not expect that, and that may be error-prone for some high-performance use cases. We should consider having an opaque type wrapping a &mut ReadBuf (name subject to bikeshedding), adding a method to ReadBuf to return one, and then using that as the argument to read_buf.

However, since this isn't insta-stable, we can address both of those through further development on nightly.

@rfcbot cancel

@bors r+

[rfcbot]

@joshtriplett proposal cancelled.

[bors]

📌 Commit cd23799 has been approved by joshtriplett

[Mark-Simulacrum]

FWIW, I think it would be nice to squash the commits here, but it doesn't seem worth blocking over and it looks like the feature letting upstream maintainers push to the branch isn't turned on for this PR.

[bors]

⌛ Testing commit cd23799 with merge 3b263ce...

[bors]

☀️ Test successful - checks-actions
Approved by: joshtriplett
Pushing 3b263ce to master...

[rust-timer]

Finished benchmarking commit (3b263ce): comparison url.

Summary: This benchmark run did not return any relevant changes.

If you disagree with this performance assessment, please file an issue in rust-lang/rustc-perf.

@rustbot label: -perf-regression

[RalfJung]

Ah, I am so happy to see this comment disappear. :)