v3.0.0

[notgull]

• Replace libc in all backends with the rustix crate (Replace libc with rustix in some backends #108, Remove libc from dev deps #146).
• Use tracing instead of log for logging (Replace log with tracing #119).
• Breaking: Rework the API to use I/O safety. Note that this makes several previously safe functions unsafe. (Rework the API for I/O safety #123)
• Add support for the ESP-IDF platform. (Support for the ESP-IDF framework #128)
• Breaking: Make Event partially opaque, and create a new Events struct for holding events. (Extract the Events struct and make the Event struct opaque #133)
• Add support for running polling in Linux containers without eventfd available. (Add a pipe-based notifier to epoll #134)
• Specify the behavior when registered in multiple Pollers. (Specify behavior when registered in multiple pollers #136)
• Breaking: Use c_int from the standard library in polling::os::kqueue instead of defining our own. (Use std::os::raw::c_int and remove our own type alias #143)

[notgull]

@smol-rs/admins Any objections to me publishing this? I do not foresee needing to make any more breaking changes after this.

[zeenix]

None from me.

[taiki-e]

Would you add #143 to changelog?

[taiki-e]

LGTM aside from nits on changelog.

[fogti]

why is libc still in dev-dependencies?

Cargo.toml
53:libc = "0.2"

CHANGELOG.md
23:- Update use of `libc::timespec` to prepare for future libc version (#55)
24:- Update use of `libc::kevent` to prepare for future libc version (#56)
80:- Remove `libc` dependency on Windows.

src/os/kqueue.rs
40:    /// poller.add_filter(Signal(libc::SIGINT), 0, PollMode::Oneshot).unwrap();
63:    /// poller.add_filter(Signal(libc::SIGINT), 0, PollMode::Oneshot).unwrap();
66:    /// poller.modify_filter(Signal(libc::SIGINT), 1, PollMode::Oneshot).unwrap();
89:    /// poller.add_filter(Signal(libc::SIGINT), 0, PollMode::Oneshot).unwrap();
92:    /// poller.delete_filter(Signal(libc::SIGINT)).unwrap();

examples/wait-signal.rs
23:        let sigint = Signal(libc::SIGINT);

[fogti]

why is rustix pulled in with "std" as a feature?

[notgull]

why is libc still in dev-dependencies?

Good question, I've opened #146 to address it.

why is rustix pulled in with "std" as a feature?

This is necessary for std::os::fd::BorrowedFd to be used in rustix.

[fogti]

hmm why does this crate even have an std feature? it doesn't contain any working code if the feature isn't passed...

[notgull]

hmm why does this crate even have an std feature?

Good question. This was added in 272bb11 for reasons that elude me; maybe it was meant to port this crate to not require libstd at some point. Although std::os::fd::* is necessary to register sources, so it doesn't really seem all too relevant to me.

I can remove it.

[fogti]

Remove it.

[notgull]

Remove it.

Ok, done in #147

[fogti]

ok, ChangeLog should be updated to include #146 and #147.
was the "std" feature present in some proper release? Yes, since v2.3.0.

[notgull]

ok, ChangeLog should be updated to include #146 and #147. was the "std" feature present in some proper release? Yes, since v2.3.0.

#146 only affected dev dependencies, so it shouldn't be put in the user facing changelog. I added #147 though.