Data races in cycles?

Lately I started getting some crashes from cycles standalone (3738dbe4), that smell like data races.
Building with thread sanitizer shown lots of data races. (I’m yet to do any detailed analysis.) Did anyone run into this kind of issues, or at least had a look at thread sanitizer output?

It seems tbb & tsan don’t really mix, so this may or may not be a false alarm.

Moreover my issue turned out not to be a data race …

What even is a data race?

BTW, the second call to reset_() in Session::run_cpu() is racy. It can swiftly bring down cycles stand-alone as soon as one tries to rotate the view in interactive mode.