Critical section functions as part of the public API

[edwardalee]

This PR removes the distinction between _lf_critical_section_enter() and lf_critical_section_enter(), using only the latter, and makes the function always available, not just for unthreaded targets. Similarly for lf_critical_section_exit() and lf_notify_of_event(). This enables programmers to use these functions. For example, in the MQTT reactors, this is useful when trying to deterministically assign a timestamp to an incoming message.

[erlingrj]

Thanks for addressing that. This looks better. I think it should work. The extra level of indirection was added to address the scenario where we have threading:false but tracing:true. In that scenario we want the C-runtime to use normal unthreaded API (e.g. disable interrupts) while the tracing files need the threaded API.

[edwardalee]

Thanks for addressing that. This looks better. I think it should work. The extra level of indirection was added to address the scenario where we have threading:false but tracing:true. In that scenario we want the C-runtime to use normal unthreaded API (e.g. disable interrupts) while the tracing files need the threaded API.

(Oops, hit the wrong button) If we later need a mix of unthreaded with tracing, I suggest we edit the tracing code to use the critical section mechanism.

[erlingrj]

Currently there is a mix of threaded and unthreaded in that case. It will work with linux/mac/win since the platform support implementing threaded/unthreaded is header-only. So trace.c can include the threaded version and reactor.c can include the unthreaded. If you look in trace.c you will see that it defines _LF_TRACE before including platform.h and then undefs it. This is to achieve this. The reason it is needed is that tracing creates threads and uses mutexes. We had some discussion of whether tracing could be implemented without any thread-support.

[edwardalee]

The test failure has nothing to do with this PR and has been showing up intermittently. I think it is due to excessively large debug output and I've separately pushed a change that should reduce the size. Merging this PR.