On Wed, Oct 6, 2021 at 5:01 PM Thomas Munro <thomas.mu...@gmail.com> wrote: > BTW, on those two OSes there are some messages like this each time a > submake dumps its output to the log: > > [03:36:16.591] fcntl(): Bad file descriptor > > It seems worth putting up with these compared to the alternatives of > either not using -j, not using -Otarget and having the output of > parallel tests all mashed up and unreadable (that still happen > sometimes but it's unlikely, because the submakes write() whole output > chunks at infrequent intervals), or redirecting to a file so you can't > see the realtime test output on the main CI page (not so fun, you have > to wait until it's finished and view it as an 'artifact'). I tried to > write a patch for GNU make to fix that[1], let's see if something > happens. > > [1] https://savannah.gnu.org/bugs/?52922
For the record, GNU make finally fixed this problem (though Andres found a workaround anyway), but in any case it won't be in the relevant package repos before we switch over to Meson/Ninja :-)