Re: Reintroducing critical warnings?

[Kjartan Maraas <kmaraas broadpark no>]

ma., 25.02.2008 kl. 13.47 -0500, skrev Owen Taylor:
> On Mon, 2008-02-25 at 09:30 +0000, Benjamin Otte wrote:
> > Havoc Pennington <hp <at> pobox.com> writes:
> > 
> > > Wait, that's the whole point is to crash the app 
> > > 
> > > The issue is that if it just prints stuff, people don't fix the bug
> > > (in part, perhaps, because nothing goes through bug-buddy). Maybe the
> > > fix is to bug-buddy the warning, but don't crash the app. Not sure how
> > > hard that would be to code.
> > > 
> > I guess this is the age-old question of "How should we treat a non-critical bug
> > during development?"
> > And I guess it is equivalent to "Should we enable gcc's -Werror switch for svn
> > builds?" which is the question I regularly clash about with Behdad.
> > 
> > In the end it's a tradeoff. A tradeoff between forcing users to make you care
> > about potentially harmless stuff that's not "a real bug" (to quote Behdad) and
> > missing real bugs because users don't care about bugs that don't bite them. It's
> > also a tradeoff between allowing you to be lazy and not fix warnings immediately
> > - or at least until the next gcc comes around that makes them real bugs - and
> > having to fix obscure warnings on weird platforms immediately because they break
> > the build or prevent the application from running.
> 
> Jumping in late on this conversation ... running from JHBuild I had to
> patch out the fatal warnings, because all sorts of apps I normally use
> *not from JHBuild* (Eclipse, Firefox, Pidgin, etc.) were dying from
> critical warnings caused by the application code, rather than bugs in
> the JHBuild built things.
> 
Filed a bug for pidgin:
http://developer.pidgin.im/ticket/4917

I don't see any warnings from firefox with a breakpoint in g_log(). Any
particular sites that cause this?

Don't have eclipse installed so I can't reproduce those.

> So, I'm not sure that the fatals warning has the intended effect of
> "catch problems", it may just have the unintended effect of driving
> people away from testing the unstable code at all.
> 
One could just clear the environment variables on the command line and
run the unstable stuff that way to avoid the problems. Or just file
bugs.

Cheers
Kjartan