On Tue, May 03, 2016 at 04:39:45PM +0700, Duy Nguyen wrote: > On Tue, May 3, 2016 at 11:45 AM, Jeff King <peff@xxxxxxxx> wrote: > > On Mon, May 02, 2016 at 11:26:02AM -0700, Junio C Hamano wrote: > > > >> Caller supplied buffer would be the way to go when multiple threads > >> could be showing errors and warnings, right? > >> > >> It would not make too much of a difference for die(), though. > > > > I think it can matter for die(). This is happening above the pluggable > > die_routine() layer, so it's possible for an async task to call die() > > which will end in pthread_exit(), while the main program is calling > > die() to end in regular exit(). > > > > I imagine it's exceedingly unlikely, and of course both threads are > > dying anyway, but it could result in some pretty weird memory errors. > > It's probably worth taking precautions against, especially because it's > > to easy to do so. > > So, sum it up, the new fmt_with_err() will take a buffer and size. > die_errno() sticks to its static buffer as before. error_errno() and > warning_errno() can allocate 1k on stack to be more > multithread-friendly. Let me know if some other change is needed. Yes, though note that die_errno() currently has a stack buffer, not a static one. It should remain that way (and error_errno and warning_errno should follow suit) to be friendly to multi-threading. -Peff -- 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