WIP: Allow bit-reproducible results with 1 main and 1 runner thread

A variety of mostly minor changes which allow reproducible results with a single runner. A couple one-line changes, where the original code is safe but hinders this goal, are enabled only by a DETERMINISTIC_SCHEDULE preprocessor flag. If you'd like to avoid preprocessor proliferation, these changes can be removed easily enough (but are probably worth documenting).

Resolves #58.

Title changed from Allow bit-reproducible results with 1 main and 1 runner thread to WIP: Allow bit-reproducible results with 1 main and 1 runner thread

Should this be turned into an engine_policy ?

That's a nice idea, yes. Especially if (at another time, I'm not sure I can justify it right now) it were deterministic with more than just a single runner. I'll take a look at that. I also need to fix a deadlock due to a brain fart in scheduler_enqueue and check the performance is still OK.

My suggestions it mostly that we don't want to hinder the performance in the case where we have more than one threads and bit-reproducible results are not possible anyway.

Added 10 commits:

• 197f17a1...a7baf53d - 9 commits from branch master
• b383d69e - Allow bit-reproducible results with 1 main and 1 runner thread

Status changed to closed

I'll close this for now. There's an occasional deadlock I need to fix, and I'd like to investigate the option of a deterministic scheduler with multiple runners for when consistency is more important than absolute performance.