Rats, I said I would do this yesterday. Here it is: a new test script
for PR gdb/1250.
http://sources.redhat.com/gdb/bugs/1250
This is the bug where gdb loses its marbles when backtracing through a
function which calls a noreturn function such as 'abort'. The calling
function has no epilog after the call to 'abort', so when gdb looks at
the frame for that function, gdb sees the first instruction of the
*next* function and uses information for that function instead.
This happens in gdb.base/corefile.exp but I think it is nice to have a
specific test for it.
I tested this on native i686-pc-linux-gnu with gdb HEAD, gcc 2.95.3 and
v3, dwarf-2 and stabs+.
It PASSed with gcc 2.95.3 because gcc 2.95.3 does not optimize away the
epilog. I think that this is okay. The gdb user really just wants to
put breakpoints on things like 'abort' and 'exit' and have it work, and
if it works because the compiler is simple, that is okay.
It KFAILed with all the gcc 3.3's that I used.
I would like to commit this to HEAD, wait a few days or a week, and then
commit it to gdb_6_0-branch.
OK to commit?