Front page | perl.perl5.porters |
Postings from May 2010
On Mon, May 10, 2010 at 05:21:52PM -0600, Alex Hunsaker wrote: > (Find at the bottom gcc, perl -V and uname output) > backtrace: > Program received signal SIGSEGV, Segmentation fault. > S_is_inplace_av (o=<value optimized out>, oright=<value optimized > out>, my_perl=<value optimized out>, o=<value optimized out>, > oright=<value optimized out>) > at op.c:8418 > 8418 (oright->op_type != OP_RV2AV && oright->op_type != OP_PADAV) > (gdb) bt > #0 S_is_inplace_av (o=<value optimized out>, oright=<value optimized > out>, my_perl=<value optimized out>, o=<value optimized out>, > oright=<value optimized out>) at op.c:8418 > #1 0x000000000042510d in Perl_peep (my_perl=0x8a3f90, o=0x8ccc50) at op.c:8923 > I get: > oright: (nil) > inplace_av: oright: (nil) > Segmentation fault > > Hrm Ok seems like gcc decided oright can never be NULL and so took out > those checks. If I do an optimize='-g' no -O2 it seems to work. Any > ideas? I am confident that it's not a misplaced __attribute__nonnull__ on oright. (These have been a problem before). It's gcc that has decided that oright can't be NULL, rather than us telling it. Your analysis looks sound to me. I can reproduce it on blead with gcc 4.5.0, with -Dusemultiplicity (but not with other related Configure options). I can't stop gcc doing it by adding -fno-inline to the compiler flags - its optimiser seems adamant that oright can never be NULL. Nicholas ClarkThread Previous | Thread Next