Implement event loop based on QNX ionotify

[sfod]

Implement CRT event loop based on QNX event mechanism.

• Add a callback to the aws_io_handle struct. This callback is called by I/O operations to provide results to event loop. For backward compatibility, it is hidden by ifdef constructions.
• Use ionotify, a special function provided by QNX to handle I/O events, to implement CRT event loop.

By submitting this pull request, I confirm that my contribution is made under the terms of the Apache 2.0 license.

[codecov-commenter]

Codecov Report

Attention: Patch coverage is 39.21569% with 31 lines in your changes missing coverage. Please review.

Project coverage is 80.04%. Comparing base (c345d77) to head (5d9f58f).

Files with missing lines	Patch %	Lines
source/posix/pipe.c	18.75%	13 Missing ⚠️
source/posix/socket.c	44.44%	10 Missing ⚠️
source/s2n/s2n_tls_channel_handler.c	0.00%	4 Missing ⚠️
source/socket_channel_handler.c	69.23%	4 Missing ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##             main     #669      +/-   ##
==========================================
- Coverage   80.39%   80.04%   -0.35%     
==========================================
  Files          28       28              
  Lines        5992     6048      +56     
==========================================
+ Hits         4817     4841      +24     
- Misses       1175     1207      +32     

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

[bretambrose]

That's a lot of conditional compilation. Is that intended to be final?

[sfod]

Short answer: it's totally possible to completely remove all these compile-time conditions from existing sources.

Now, some details.

• All changes in the source/posix/ dir can be easily moved to the new source/qnx/ dir. This way, we'll have two almost identical posix implementations - one with events with feedback, one with no feedback.

• socket_channel_handler.c could also be split into two versions easily by moving it into source/something/ dir.

• As for io.h. It's a public header, so it's trickier. Some cmake magic is definitely can be done to use different headers depending on the platform, but I didn't look into it yet.

To be honest, I want to keep all these checks, because otherwise every fix to posix implementation or socket_channel should be duplicated in the qnx implementation.

Apart from this, do you think it's a viable solution to add feedback functionality to I/O events?

[bretambrose]

Some quick Initial feedback:

The conditional compilation should be removed from the headers and source. We dont have any backwards compatibility guarantees with the C libraries across different versions.

Additionally, I'm really concerned about the massive duplication between the qnx and posix folders from a maintainability standpoint.

[bretambrose]

Don't do it here, but this reminds me that I think we need to start tagging event loop logging with event loop type (iocp, kqueue, epoll, gcd, inotify) for easier disambiguation, especially if some platforms might have multiple types available (mac eventually)

[sfod]

I can actually introduce a new subject, AWS_LS_IO_EVENT_LOOP_IONOTIFY - "event-loop-ionotify". It won't take much effort. Unless you have something else in mind, then I'll leave it as is.

[bretambrose]

Under what circumstances would this happen and what does it mean? All we seem to do is log it.

[sfod]

Oh, this one is tough. I added a comment describing my interaction with this flag. I suggest we keep this log, maybe it'll reveal something.