On Fri, Aug 28, 2009 at 19:15, Jeff King<peff@xxxxxxxx> wrote: > On Fri, Aug 28, 2009 at 07:00:59PM +0200, Alex Riesen wrote: > >> On Fri, Aug 28, 2009 at 17:02, Jeff King<peff@xxxxxxxx> wrote: >> > But that's the point: you can't do that without a race condition. Your >> > test gets a sense of the current time, then runs git, which checks the >> > current time again. How many seconds elapsed between the two checks? >> >> How _many_ do you need? > > I don't understand what you're trying to say. My point is that if you > are checking results to a one-second precision, you need to know whether > zero seconds elapsed, or one second, or two seconds, or whatever to get > a consistent result. Taking this particular case as an example, can't we just set the time of the _commit_ back in time? We can. And we don't need to know the difference precisely, it can be something like /[0-9]+ ago/, can't it? Ok, it is possible, that something goes terribly wrong and the test suite freezes for an extended period of time, so the pattern above does not apply anymore. In this case, wont you prefer the test suite to break? Ok, maybe not, if the freeze was an Ctrl-Z pressed at unlucky moment. Which involves an operator online and looking, and action and reaction will be both visible. So, yes, it is not absolutely safe, but this approach has no side effects on the working code. And very low probability of something go wrong. -- To unsubscribe from this list: send the line "unsubscribe git" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html