Re: Simple exception use-case

[Phil Muldoon <pmuldoon at redhat dot com>]

Phil Muldoon wrote:
> Ok if I break main, then run, and then finally add the catch throw and 
> catch catch, it works. That's another on my list to investigate. But 
> while doing this I stumbled onto something interesting. If you look at 
> these two test-runs, one from upstream GDB one from a patched Fedora 
> GDB you'll see that the Fedora GDB did not loose control of the 
> inferior. This test is by no means conclusive, but it intrigues me. 
> For the purpose of this use-case and estimation I was going to 
> concentrate purely on upstream, but I think I'll take a look at Fedora 
> GDB and investigate the patch load. There may be a patch that already 
> fixes possible longjmp or other inferior control issues. If so, can 
> look at the estimation to convert this patch upstream, tweak it - 
> whatever - over writing new support. At the very least I'll 
> investigate why this one use-cases diverges on two different GDBs.
Replying to myself. If I build the Fedora RPM without any patches at 
all, the behaviour is the same (ie stepping when in the libstdc++ code 
does not lose control of the inferior). The only difference in the 
Fedora build, and my build from upstream CVS is that Fedora GDB has a 
configure option (among others):

--with-separate-debug-dir=/usr/lib/debug

If I build upstream gdb-6.8 with that option,  GDB does not lose control 
of the inferior when stepping through the catch libstdc++ code. If I 
build without that option, stepping after a:  "catch catch" and control 
is lost. Oh well, it was worth a morning investigating it.  I certainly 
learnt a lot ;) But this is a separate (debuginfo) issue I think (?)

Regards

Phil