On Sun, Oct 2, 2016 at 12:45 PM, Samuel Thibault <samuel.thiba...@gnu.org> wrote: > David Michael, on Sun 02 Oct 2016 12:18:50 -0700, wrote: >> On Sun, Oct 2, 2016 at 10:53 AM, Samuel Thibault <samuel.thiba...@gnu.org> >> wrote: >> > David Michael, on Sun 02 Oct 2016 10:22:12 -0700, wrote: >> >> Commit a87bf9a8eab3af79798131b60c1f7f92f995df8c breaks static linking >> >> (namely ext2fs.static) from missing pthread_setcancelstate. >> > >> > Ah? I don't understand how: this commit only makes libpthread use its >> > own internal __pthread_setcancelstate symbol. A pthread_setcancelstate >> > alias is still defined from pthread/pt-setcancelstate.c, how is it that >> > you don't get it? >> >> The following is the actual error when using that commit. It looks >> like pthread_setcancelstate is defined in libpthread2.a, and the >> linking command includes -lpthread, which includes -lpthread2. >> Changing error.c and pthread-functions.h doesn't seem correct since >> there are many other instances of that throughout glibc. > > Actually it *is* correct :) > When an application uses error(), it still shouldn't get the > pthread_setcancelstate symbol, and thus error.c should be using > __pthread_setcancelstate, to avoid pulling pthread_setcancelstate. > > That being said, glibc 2.22 didn't fix that yet, that'll be for glibc > 2.23 and later, better not try to backport this which might pose other > troubles. > >> /usr/i686-pc-gnu/sys-root/lib/libcrt.a(error.o): In function `__error': >> (.text+0x26f): undefined reference to `pthread_setcancelstate' > > So I'm surprised by this. Is -lpthread properly *after* libcrt.a?
Well, -lcrt doesn't appear explicitly in the command-line for either exec.static or ext2fs.static. The builds actually succeed if I add either -lc or -lcrt (or their full paths) before or after -lpthread, e.g. by `echo exec.static: -lc >> exec/Makefile`, so I'm not sure what is going on yet. Thanks. David
