Re: -file-list-exec-source-files

[mec.gnu@mindspring.com (Michael Elizabeth Chastain)]

bob> The patch below checks symtab_to_fullname's return value against NULL.
bob> Even though this is the "trivial" fix, I believe it is the correct
bob> patch. When I added these calls recently, I blindly changed the old call
bob> to symtab_to_fullname, and the old call did not return NULL.

First, I had a formatting problem with the patch.  By the time
I got it, the patch had no tabs, and it looked like the tabs
were expanded to 4 spaces each (instead of 8 spaces).  No big deal,
I just applied the changes by hand.

I ran this patch in my test bed and it fixed the problem for me.  It
also has no regressions in any other tests.

As usual, my test bed is: native i686-pc-linux-gnu, red hat 8.0,
gcc 2.95.3 and gcc 3.3.3, dwarf-2 and stabs+.

I discovered a subtle interaction between my testbed, selftest.exp,
and this bug.  Unlike most people, I don't run the test suite in
the build directory.  My script runs the test suite in a fresh
'test' directory for each run.  The 'test' directory has a fresh
copy of the build/gdb/testsuite directory (which is actually just
a framework), but it does not have build/gdb itself.  And indeed
I remove everything in the build/ tree except for the testsuite/
subtree to save space.

When I make that removal, I get the crash.  No removal, no crash.

I think that's why symtab_to_fullname returns NULL for me.  I don't have
any more objfiles for gdb itself.  And it happens only in selftest.exp,
because selftest.exp is special.

I'm not qualified to say whether this is the *right* fix but it
fixes the problem for me.

Michael C

===

2004-06-28  Bob Rossi  <bob@brasko.net>

    * symtab.c (lookup_symtab): check return value of symtab_to_fullname