Re: GNOME Namespace Management - ARC & GNOME
- From: Mike Hearn <mike navi cx>
- To: desktop-devel-list gnome org
- Subject: Re: GNOME Namespace Management - ARC & GNOME
- Date: Thu, 16 Dec 2004 14:46:33 +0000
On Wed, 15 Dec 2004 16:57:11 -0500, Sean Middleditch wrote:
> How would they know? Why should they _have_ to know? If
> they install Linux Weekly Distro V56, then install FooApp, then upgrade
> to Linux Weekly Distro V57, FooApp should not break.
I think the exception to this is when the app is clearly breaking the
rules, or worked through sheer fluke.
As Owen said it's not black and white (and I'm glad to hear him say this
...), when changing stuff you have to weigh up the cost.
Cost is a function of how widespread the breakage is (though this is real
easy to underestimate), and what kind. Programs that rummage inside
internal structures to get what they want should probably break, but there
are a few cases eg with the __libc_stack_end mess I was investigating
lately where really the programs in question had little choice but to use
it even though it was internal. The interface was hidden, apps broke,
people got upset etc when really all that was needed was to make this
symbol public again (and in fact eventually this happened).
So even when apps are doing very unkosher things, if enough programs
deployed in the wild are doing it either by accident or because they
had no choice then this makes the cost of a change much higher even though
you're technically in the right. The ARGB visuals+gtk1.2/imlib situation
is similar here (widespread mistakes causing legitimate changes to break
Exactly where the line lies in this cost/benefit analysis for Sun and
GNOME is what we're trying to figure out. I think Mark is right in that
it's really much closer than we think, the difference being that GNOME
doesn't ship customers patches if things break, GNOME expects the
"customers" (read: users) to write their own or get them from elsewhere.
Or just live with the breakage.
ARCs definition of stability is probably just an extended version of
] [Thread Prev