owned this note   changed 3 years ago  
Published Linked with GitHub

Libs-API Meeting 2022-02-23

tags: Libs Meetings Minutes

Meeting Link: https://meet.jit.si/rust-libs-meeting-crxoz2at8hiccp7b3ixf89qgxfymlbwr
Attendees: Amanieu, David, Josh Triplett, Jane, Mara, Mark, The 8472

Agenda

Triage

FCPs

10 open T-libs-api FCPs:

  • 1 rust-lang/rfcs FCPs

    • [merge 3192] Add provide_any module to core - (4 checkboxes left)

  • 9 rust-lang/rust FCPs

    • [merge 89238] deprecate f{32,64}::DIGITS - (1 checkboxes left)
    • [merge 86845] Tracking Issue for owned locked stdio handles - (4 checkboxes left)
      • Cancel the FCP in favor of the change making the lifetimes 'static
    • [merge 90291] Loosen the bound on the Debug implementation of Weak. - (2 checkboxes left)
    • [merge 80697] Tracking Issue for const_intrinsic_copy - (1 checkboxes left)
      • Concern should be resolved
    • [merge 93263] Consistently present absent stdio handles on Windows as NULL handles. - (3 checkboxes left)
    • [merge 91789] Tracking Issue for try_reserve method on more containers - (4 checkboxes left)
      • Stuck on meaning of the usize for OsString/PathBuf.
      • Conclusion: We make guarantees about valid utf-8, that it matches str. No guarantees about invalid utf-8. Use the word 'bytes' in the docs.
    • [merge 92472] proc-macro: Stop wrapping `ident` matchers into groups - (3 checkboxes left)
    • [merge 93840] Stabilize Termination and ExitCode - (3 checkboxes left)
    • [merge 90829] Tracking issue for Vec::retain_mut and VecDeque::retain_mut - (3 checkboxes left)

pnkfelix (1), dtolnay (3), yaahc (3), joshtriplett (5), BurntSushi (11), m-ou-se (4), Amanieu (2), cramertj (1)

Nominated

  • 3 rust-lang/rust T-libs-api I-libs-api-nominated items
    • [93873] Reimplement `carrying_add` and `borrowing_sub` for signed integers.
      • Scott: Original impl was definitely wrong. The new ones seem accurate, but you can't chain them so they don't seem useful for the bigint purpose that the unsigned versions were designed for.
      • Mara: A bigint like Vec<u32> has 32-bit 'digits', but it doesn't make much sense to have all digits have their own sign.
      • Amanieu: it's for the very last digit only, which is the only signed one.
      • Should we just not have these?
      • If the goal is that llvm optimizes this to a specific instruction, we should probably have a codegen test.
      • Josh: We should rename this, because it isn't the same chainable operation as on unsigned.
      • David: Name seems fine.
      • Amanieu, Mara: Yeah, rename.
      • Needs better examples as well
      • Summary:
        • PR needs better examples
        • Tracking issue should list a requirement for a codegen test before stabilizing
        • PR should change the name (should be clear that output is overflow)
    • [94026] `Hasher::write` should clarify its "whole unit" behaviour
      • Discussed last time. Should've been untagged?
      • Josh: There's an interesting example (hash_hasher) of a hasher that just spits out the first 8 bytes, for e.g. a hashmap of uuids, that would be horribly broken for other use cases.
      • Mark: Add a separate method? Then hashers can implement them differently if they care about the issue.
      • h.write(&[a, b]); h.write(&[c]); vs h.write(&[a]); h.write(&[b, c]); - document that we don't guarantee they produce the same hash, and don't guarantee that they produce different hashes
      • Amanieu: Who is responsible for guaranteeing prefix-freedom? The hasher, or the impl?
      • Josh: If the hasher is responsible, that makes it easier for a fast non-DoS-resistant hasher to opt out
      • Do we need another method to do that?
    • [94079] [Draft] library: Move `CStr` to libcore, and `CString` to liballoc
      • Originally libc stuff was only available in std, not core. But a lot of target triples define a specific libc anyway, so there it should be fine. And on platforms without a libc, there are obvious definitions for types like c_int.
      • Using an extension trait to add CString methods to CStr is technically a breaking change. Lang item instead?
      • However, an extension trait does make CString less special, making it possible for someone to provide theire own CString type with the same interface (without pairing it with their own CStr type).

Waiting on team

Needs decision

None

Stalled Tracking Issues

Actions

  • Reply to all issues/PRs discussed in this meeting, or add them to the open action items.

Generated by fully-automatic-rust-libs-team-triage-meeting-agenda-generator

Select a repo