On Wed, Nov 27, 2019 at 10:13:16PM +0700, Doan Tran Cong Danh wrote: > gmtime/localtime is considered unsafe in multithread environment. > > git was started as single-thread application, but we have some > multi-thread code, right now. > > replace all usage of gmtime/localtime by their respective reentrant ones. I think this is a good change. A minor point, but I think it may be simpler if the first four were just a single patch. There's no rationale given at all in the 3rd and 4th ones. Which is because you already explained it in patch 1, but that won't help somebody who digs up the commit via "git blame". So I think they either ought to be one patch, or they should repeat the rationale (I'd probably go with the first, but I could live with the second). > On Windows, we may be taxed, since gmtime_r and localtime_r is our compat > functions, because we memcpy from returned data of gmtime/localtime. > > To address that, I made patch #5 and included it together with this series. > I'm not sure how much portable it is. > It seems to run fine with my Windows 7 x86 and Windows 10 x64 VM, (inside > git-sdk for Windows, if it matters). > I'm Cc-ing j6t and Dscho on that patch to get their opinions. That makes sense from my reading of gmtime_s(), but there may be something more subtle going on with compatibility, so I'll leave that to the experts. -Peff
