Re: [RFC] fullname attribute for GDB/MI stack frames

[Bob Rossi <bob@brasko.net>]

On Thu, May 05, 2005 at 11:15:26AM -0400, Dennis Brueni wrote:
> On Wed, 4 May 2005 20:05:06 -0400, Bob Rossi wrote:
> > 
> > This may or may not be what you mean. However, if you do
> > -file-list-exec-source-files, GDB prints the filename and fullname for
> > every file that it knows about. The purpose of this is to let the
> front
> > end know about each and every file that GDB knows about. As I've
> stated
> > before, I believe most if not all FE's depend on the fullname to be a
> > unique key.
> 
> I think you have just struck on the right way to write the tests
> for fullname with respect to stack frames.  Forget regex's, we
> want the value of the fullname to exactly match the same fullname
> attribute pumped out by -file-list-exec-source-files.  This would just
> take one utility function.  The test for -file-list-exec-source-files
> could then be the central point for verifying the "correctness" of
> fullname.  I'm willing to redo my stack/fullname tests if this idea
> is amenable to the group.

The value of the fullname will always match the output of
-file-list-exec-source-files. You use the same function
"source_full_path_of" and pass in the same symtab that
-file-list-exec-source-files uses. In fact, that function reuse's
symtab->fullname if it has already been calculated and is still on the
disk.

I personally think that at best, Eli's originally regex will need to be
used. For instance, I don't believe that GDB should change in order to
get the proper regex in this testcase. I believe that we should use a
regex that matches the current output of GDB. If it is determined that
GDB is broke, it can be fixed in another patch.

Thanks,
Bob Rossi