Subteam reports: 2015-08-07


#1

Libs team

Full report

First off, we had a great RustCamp last weekend! As many of us were involved with the event, it was a bit of a slow week.

Decisions from last week:

  • RFC PR #1184: RFC: Stabilize the #![no_std] attribute
    • Merged
  • RFC PR #1183: RFC: Allow changing the default allocator
    • Merged
  • RFC PR #770: io error handling design
    • Closed
  • RFC PR #980: read_exact
    • Merged
  • RFC PR #1194: RFC: Add item recovery collection APIs
    • To be merged, pending last bits of bikeshedding
  • PR #26818: Stabilize duration (with renamings)
    • Merge, need to decide about 1.3 backport

We’d like to call attention to the following two policy RFCs:

  • PR #1242: RFC: policy for rust-lang crates
  • PR #1224: Update the RFC process with sub-teams, amongst other things.

as well as an RFC relating to catch_panic and exception safety in Rust:

This week’s RFCs going into (or staying in) final comment period:

Lang team

Full report

The following RFCs are being promoted to final comment period:

  • PR #886: Permit #[must_use] attributes on functions as well as types. This allows for functions whose return value should not be ignored even if the type of that value is unexceptional (e.g., the ok() method of Result).
  • PR #890: Custom preludes. This RFC proposes an extension that allows crates to define their own preludes. This can be used to have common names available throughout a crate without forcing them to be explicitly imported everywhere. This is particularly useful when combined with the convention of having external crates define a pub mod prelude that downstream crates can import into their own local preludes. While clearly convenient, there were some concerns raised that this will give rise to multiple dialects of Rust.
  • PR #953: This defines traits to support += and other operators, closing a gap in our operator overloading support. The traits take the LHS via an &mut reference to permit in-place updates, take the RHS by value, and do not require that the Add trait also be implemented.
  • PR #1135: This PR permits raw fat pointers (e.g., *[i32] or *Trait) to be compared, just like raw thin pointer (e.g., *i32). The semantics are to compare both the pointer itself and any accompanying data (e.g., the length of the slice).
  • PR #1189: This PR simply corrects typos.

The following two RFCs have been accepted:

  • PR #1214: Clarify (and improve) rules for projections and well-formedness.
  • PR #1219: Allow aliasing imports when importing as a group.

In addition, I would like to call attention to the following RFC:

  • PR #1238: Nonparametric dropck. This RFC simplifies the dropck rules to close some soundness holes and make room for specialization. The change is expected to cause little to no breakage in practice, e.g., a crater run found no affected crates, but it nonetheless affects a core component of the language.

— nmatsakis

Compiler team

Full report

@arielb1 opened PR #27551, which changes how structs and enums are represented in the compiler, replacing various hashtables with a single AdtDef struct. This is a reimplementation of a similar PR by @aatch. In addition to cleaner code, it results in a small performance boost (approximately 5%).

There has been some progress towards removing drop flags. @pnkfelix landed his “nonzeroing move hints” branch (PR #26173). Unfortunately, some critical bugs were found shortly thereafter. The fix (PR #27413) is not yet ready.

— nmatsakis