Allows build on Solaris (#1877)

[psumbera]

This allows build on Solaris. Unfortunately at this moment tests are failing like this:

     Running unittests src/lib.rs (target/debug/deps/quinn-94a8797a83ab1c78)

running 18 tests
test work_limiter::tests::limit_work ... ok
test tests::local_addr ... ok
test tests::close_endpoint ... ok
test tests::handshake_timeout ... ok
test tests::zero_rtt ... FAILED
test tests::export_keying_material ... FAILED
test tests::rebind_recv ... FAILED
test tests::stress_stream_receive_window ... FAILED
test tests::echo_v4 ... FAILED
test tests::stress_both_windows ... FAILED
test tests::read_after_close ... FAILED
test tests::echo_v6 ... FAILED
test tests::stress_receive_window ... FAILED
test tests::echo_dualstack ... FAILED
test tests::ip_blocking ... FAILED
test tests::multiple_conns_with_zero_length_cids ... FAILED
test tests::stream_id_flow_control has been running for over 60 seconds
test tests::two_datagram_readers has been running for over 60 seconds
^C

[psumbera]

(Note that there might be some problems with tokio and/or mio as mio on Solaris is using poll() interface).

[djc]

Can you add a separate commit that bumps the MSRV?

[djc]

Thanks!

[Ralith]

Thanks! What happened to the FFI definitions drafted in libc that were thought to be necessary?

Unfortunately at this moment tests are failing like this

That's unfortunate. Is there an upstream issue open for the behavior in tokio or mio that you suspect is responsible?

Regardless, the most important question is whether the quinn-udp tests pass. Do they?

[psumbera]

Thanks! What happened to the FFI definitions drafted in libc that were thought to be necessary?

I'm keeping libc changes separately in my ws:

--- a/quinn-udp/Cargo.toml
+++ b/quinn-udp/Cargo.toml
@@ -19,8 +19,8 @@ default = ["log"]
 log = ["tracing/log"]

 [dependencies]
-libc = "0.2.113"
-socket2 = { workspace = true }
+libc = { git = "https://github.com/psumbera/libc.git", branch = "quinn4" }
+socket2 = { git = "https://github.com/psumbera/socket2.git", branch = "quinn" }
 tracing = { workspace = true }

 [target.'cfg(windows)'.dependencies]

Note that use of modified socket2 is also because of libc.

This is because I had to include to libc changes the use of __xnet_* socket functions:
rust-lang/libc@8281157
and this required me to rebuild whole Rust and socket2 had to use this modified libc as well.

Unfortunately at this moment tests are failing like this

That's unfortunate. Is there an upstream issue open for the behavior in tokio or mio that you suspect is responsible?

With use of __xnet_* functions in libc it's better now:

running 18 tests
test work_limiter::tests::limit_work ... ok
test tests::local_addr ... ok
test tests::close_endpoint ... ok
test tests::echo_v6 ... FAILED
test tests::ip_blocking ... ok
test tests::two_datagram_readers ... ok
test tests::export_keying_material ... ok
test tests::rebind_recv ... ok
test tests::multiple_conns_with_zero_length_cids ... ok
test tests::stream_id_flow_control ... ok
test tests::read_after_close ... ok
test tests::echo_v4 ... ok
test tests::zero_rtt ... ok
test tests::handshake_timeout ... ok                                                                                              test tests::echo_dualstack ... ok                                                                                                 test tests::stress_both_windows has been running for over 60 seconds                                                              test tests::stress_receive_window has been running for over 60 seconds                                                            test tests::stress_stream_receive_window has been running for over 60 seconds                                                     ^C                                               

Regardless, the most important question is whether the quinn-udp tests pass. Do they?

I had to disable some tests the same way NetBSD and OpenBSD do and it now passes like this:

running 3 tests
test gso ... ignored
test basic ... ok
test ecn_v6 ... ok

[Ralith]

I'm keeping libc changes separately in my ws:

If the definitions introduced/altered in the libc PR are necessary for the quinn-udp changes to work, then please vendor them in this PR, as mentioned in #1877 (comment). Once the libc change is merged and released, we can drop the vendored definitions.

[djc]

Thanks! Do you know of any good ways to test Solaris in CI?

[psumbera]

Now with libc 0.2.158 everything should be ok.

[psumbera]

Thanks! Do you know of any good ways to test Solaris in CI?

Unfortunately no. There is Solaris CBE release which could be used (https://blogs.oracle.com/solaris/post/announcing-the-first-oracle-solaris-114-cbe). But I don't know how this could be used for CI...

[djc]

The vmactions org (whose Action we use for FreeBSD) appears to also have a Solaris action:

https://github.com/vmactions/solaris-vm

[mxinden]

For what it is worth, here is the equivalent FreeBSD workflow mentioned above:

quinn/.github/workflows/rust.yml

Lines 12 to 33 in b9ce9c8

	test-freebsd:
	# see https://github.com/actions/runner/issues/385
	# use https://github.com/vmactions/freebsd-vm for now
	name: test on freebsd
	runs-on: ubuntu-latest
	steps:
	- uses: actions/checkout@v4
	- name: test on freebsd
	uses: vmactions/freebsd-vm@v1
	with:
	usesh: true
	mem: 4096
	copyback: false
	prepare: |
	pkg install -y curl
	curl https://sh.rustup.rs -sSf --output rustup.sh
	sh rustup.sh -y --profile minimal --default-toolchain stable
	echo "~~~~ rustc --version ~~~~"
	$HOME/.cargo/bin/rustc --version
	echo "~~~~ freebsd-version ~~~~"
	freebsd-version
	run: $HOME/.cargo/bin/cargo build --all-targets && $HOME/.cargo/bin/cargo test && $HOME/.cargo/bin/cargo test --manifest-path fuzz/Cargo.toml

[Ralith]

Thanks! CI coverage would be a great follow-up.