On Thu, 3 Apr 2025 10:01:07 -0700 (PDT) Jeremy Drake wrote: > On Thu, 3 Apr 2025, Johannes Schindelin wrote: > > > I still have a question that I would like to be answered in the commit > > message, too: > > > > If `signal_arrived` is only initialized in `fixup_after_fork()` but user > > callbacks that use this are called by `atforkchild()`, why did this not > > trigger _all the time_ before your reordering of the calls? > > Based on my recollections, Takashi probably knows better > > 1) there has to be a pthread_atfork child callback registered > 2) this callback has to call raw_write > 3) raw_write now calls cygwait (which is now reenterancy-safe due to other > fallout from this) > 4) cygwait allows signals to be processed, so needs the signal-handling > stuff to be properly initialized. > > I'm guessing, if raw_write doesn't need to wait (ie, there's room in the > pipe for the write) it doesn't hit the signal stuff.
Thanks for the explanation. Actually, cygwait() waits for a mutex at the beginning of raw_write(). This is introduced by the commit 7ed9adb356df, so the bug does not affect before that commit. > But I get your request for explaining the scenario in the commit message. I'll add the descriptions requested by Johannes before push. Thanks! -- Takashi Yano <takashi.y...@nifty.ne.jp>
