std.posix: unlock nonblocking connect() use case

[iamtimmy]

Match Windows WSAEALREADY with Linux EALREADY
return gracefully when socket reports it is connected

for easy reference the msdn page and the man page.

The msdn page also says that WSAEINVAL should be handled the same due to ambiguities. The WSAConnect api doesn't mention the same in the documentation. Is that a reason to change to WSAConnect?

[squeek502]

Is it feasible to add a test for what this fixes in std/posix/test.zig?

[squeek502]

Ah, the tests in net/test.zig might be a better reference (and maybe the added test should be put there instead?):

Actually, there might be an existing (but disabled, see #18315) test in place that might be fixed with your changes:

zig/lib/std/net/test.zig

Lines 380 to 406 in 4f3b59f

	test "non-blocking tcp server" {
	if (builtin.os.tag == .wasi) return error.SkipZigTest;
	if (true) {
	// https://github.com/ziglang/zig/issues/18315
	return error.SkipZigTest;
	}
	const localhost = try net.Address.parseIp("127.0.0.1", 0);
	var server = localhost.listen(.{ .force_nonblocking = true });
	defer server.deinit();
	const accept_err = server.accept();
	try testing.expectError(error.WouldBlock, accept_err);
	const socket_file = try net.tcpConnectToAddress(server.listen_address);
	defer socket_file.close();
	var client = try server.accept();
	defer client.stream.close();
	const stream = client.stream.writer();
	try stream.print("hello from server\n", .{});
	var buf: [100]u8 = undefined;
	const len = try socket_file.read(&buf);
	const msg = buf[0..len];
	try testing.expect(mem.eql(u8, msg, "hello from server\n"));
	}

(would need to be verified that it'll no longer be flaky, though; ideally that'd be something like reproducing the failure, making the failure consistent if possible, and then showing that the changes fix the failure)

[iamtimmy]

I moved the test to test/net.zig and based on that other non-blocking tcp server test I managed to complete the test for this pr.

I think the issue with the non-blocking tcp server test might be related to a specific MacOS case in accept

[iamtimmy]

CI Failed. Turns out Linux and BSD refuse the connection to a nonexistent socket...
I fixed the formatting and made that part of the test only run on Windows.

Any ideas for a scenario that works on mac and linux?

[squeek502]

Is there a reason WSAEISCONN/ISCONN shouldn't return an error (something like error.IsConnected)? Is there no use case for distinguishing between success and WSAEISCONN/ISCONN?

[iamtimmy]

I don't know of one, I opted for this because it makes total sense to me. Could opt for an error as well.

[squeek502]

Will need to wait for a Zig core team member to weigh in.