Re: Core files

[Todd Dukes <tjdukes flash net>]

Owen Taylor wrote:
> 
> Currently, GDK calls g_error() ( => abort() ) for
> X IO errors and X errors.
> 
> For X IO errors, this makes no sense, since
> an X IO errors almost certainly have no relation
> to anything the client does. They are going to
> occur either because the server died, or becuase
> somebody did an XKillClient.
> 
> So, for X IO errors, I'm about to apply the
> following patch which makes GDK die nicely
> in that case.
> 
> The question is what to do about X errors. On
> one hand, they typically do indicate something
> going wrong, but on the other hand:
> 
>  1) They occur asynchronously in general, so a
>     core isn't very useful unless the program
>     was run with --sync
> 
>  2) They may occur for unpredictable reasons
>     (i.e., the plug part of a plug/socket dies
>     at the wrong time)
> 
>  3) Enlightment does the wrong thing when you
>     Annihilate a window and destroys the window
>     instead of XKillClient()'ing the window.
>     So this currently causes GTK+ to core dump.
> 
> So, I'd like to make the core-dumping not the
> default. Given that, there are two options
> 
>  a) Never dump core, always exit() with a error
>     on stdout.
> 

The code writes to stderr, is that what you meant?


>  b) Dump core only if GTK+ is compiled with
>     --enable-debug.
> 
> Opinions?
>                                         Owen
> 

How about an option where a .gdk_errors file is appended
to or created in the user's home directory rather than
sending to stdout or stderr. This way it would be easier
for them to forward them to someone if gdk exits unexpectedly.
( You have to remember I'm on Solaris. :) It is unbelievably
easy to crash thier X server )


thanks,
Todd.
-- 
-----------------------------------------
Todd & Lisa Dukes
tjdukes@flash.net