Re: [Denemo-devel] Denemo broken ....

[Dominic Sacré]

On Wednesday 03 August 2011 18:45:29 R. Mattes wrote:
> On Wed, 03 Aug 2011 17:17:15 +0100, Richard Shann wrote
> > There is some doubtful looking code in main() which tries to save
> > the current score into a file called crashrecovery.denemo
> > this has never done anything useful for me, I prefer programs that
> > do not crash :)
> 
> Me too - but iff an application crashes I want a segfault. That way I
> can even get a core dump to debug after the fact. And all of that I
> loose just because of 'denemo_signal_handler'.  I _might_ tolerate
> such a beast iff it would sensibly try to do something useful - but
> calling (memory allocating) functions like g_print or the like after
> tripping over a dangling pointer, stomping all over it's own heap and
> while tumbling down possibly messing up it's symbol table will
> probably cause more havoc than a segfault :-)

I already disabled the SIGSEGV handler in the soc branch. It made 
debugging segfaults pretty much impossible unless Denemo was started in 
gdb. On the other hand, with regular JACK settings a client running in 
gdb has a good chance of immediately being zombified by the server...

If you ask me, the SIGSEGV handler should go away completely. Like 
Richard said, the code looks doubtful at best. Not to mention the fact 
that doing this kind of work in a signal handler results in undefined 
behavior according to the POSIX standard.


Dominic