Fix the build on FreeBSD (and OpenBSD?)

[jakepetroules]

For now, we use Dispatch for AsyncIO and process termination monitoring, since BSDs (including macOS) all use kqueue as the Dispatch backend anyways.

This likely also fixes the build on OpenBSD, but I haven't been able to verify that.

Closes #115

[jakepetroules]

@3405691582 does this build on OpenBSD?

[3405691582]

Some of the changes discussed on #115 are still missing, so with just the patch atop HEAD it doesn't build. If I apply those changes, namely those below, it seems to build:

• Sources/Subprocess/Platforms/Subprocess+Linux.swift needs to expose the extension and these changes
• Sources/Subprocess/Platforms/Subprocess+Unix.swift: the snprintf shim change
• Sources/_SubprocessCShims/include/process_shims.h: snprintf shim needs TARGET_OS_UNIX
• Sources/_SubprocessCShims/process_shims.c: snprintf shim needs TARGET_OS_UNIX
• Tests/SubprocessTests/ProcessMonitoringTests.swift: testInvalidProcessIdentifier needs an os(OpenBSD); in testDoesNotReapUnrelatedChildProcess the si_status extension doesn't seem to be active and I don't know why at the moment.

I tried to run the tests but ran into some other unrelated issues.

[3405691582]

FYI: Most recent revision builds successfully on OpenBSD!

[iCharlesHu]

Can you elaborate a bit more on why this change is needed?

[jakepetroules]

There's two threads:

• thread A: thread that calls registerFileDescriptor
• thread B: monitor loop

With the original code, the following ordering could occur:

• (thread A) add file descriptor N to epoll
• (thread B) process event for N
• (thread B) look up continuation <-- continuation is nil!
• (thread A) save continuation <-- too late

This change forces the ordering to become:

• (thread A) save continuation
• (thread A) add file descriptor N to epoll
• (thread B) process event for N
• (thread B) look up continuation <-- continuation cannot be nil

[iCharlesHu]

Thanks for the fix!