[new release] eio (5 packages) (1.3)

[talex5]

Effect-based direct-style IO API for OCaml

• Project page: https://github.com/ocaml-multicore/eio
• Documentation: https://ocaml-multicore.github.io/eio/

CHANGES:

Bug fixes:

• posix: ensure spawn_unix wraps errors when calling openat (@dijkstracula posix: spawn_unix: ensure spawn_unix wraps its call to openat ocaml-multicore/eio#809).

• Use O_RESOLVE_BENEATH on FreeBSD (@talex5 Use O_RESOLVE_BENEATH on FreeBSD ocaml-multicore/eio#810, reported by @dijkstracula).
  FreeBSD needs -D__BSD_VISIBLE to be able to see this.
  Fixed the CI bug this revealed, which had started also affecting macos.

• Ignore ECONNRESET on close (@talex5 Ignore ECONNRESET on close ocaml-multicore/eio#787).
  FreeBSD returns ECONNRESET in certain (unclear) circumstances, but it does still close the FD successfully.
  If you care about this case, you should probably use shutdown instead and check for it there.
  Python and Ruby at least both explicitly check for and ignore this error too.

• On Windows, fix stdin broken-pipe and blocked domains (@bdodrem @talex5 On Windows, fix stdin broken-pipe and blocking domain. Issues #793 and #792. ocaml-multicore/eio#795).

  • Ensure blocking FDs are ready before trying to use them.
  • Replace eio_windows_cstruct_stubs.c by Unix functions added in OCaml 5.2,
    which correctly handle Window's strange use of EPIPE.

Documentation:

• Documentation fixes (@jonludlam Documentation fixes ocaml-multicore/eio#813).

• Minor documentation improvements (@talex5 Minor documentation improvements ocaml-multicore/eio#794).

Build fixes:

• Disable dune subst (@talex5 Disable dune subst ocaml-multicore/eio#789).

[talex5]

Test failures:

• revdeps failures look unrelated.
• ocaml-5.3-flambda: No fixed buffer available is because the CI machine didn't have enough fixed memory to run the test. This can be ignored (I'll update the test upstream to skip rather than fail in this case in future).
• opam-2.0-ocaml-5.3: ERROR while compiling conf-pkg-config.4 seems surprising, but unrelated to Eio.
• riscv64-ocaml-5.3: Linux >= 5.15 is required for io_uring support from the tests seems reasonable, assuming the test machine kernel really is that old.

I don't think any of these need to prevent merging.

[shonfeder]

Congrats on the new release and thanks for the update! Thanks also for reviewing the CI