On Fri, Jan 27, 2017 at 10:06 AM, Julian Seward wrote:
> On 27/01/17 16:36, Brian Hackett wrote:
>> Any thoughts or comments?
> It would be really good if testing for data races with TSan could be
> an integral, routine part of this development process. We've seen in
> the past that after-the-fact race detection and fixing is a slow and
> difficult process, and it would be much better to do this as you go
> along. Plus, having something that can validate intuitions about
> race-freedom (or otherwise) of code is in itself helpful for hacking.
Yeah, I agree completely about how hard it is to fix races in an
existing system. Using ProtectedData should help a lot with making
sure we have a coherent synchronization model in place, but using that
class *plus* TSan running in automation would be great.