Re: libgtk2 png3 transition ... [2002-08-17]
On Sat, Aug 17, 2002 at 14:01:41 -0400, Jack Howarth wrote:
> I was surprised to see that all of gnome1 was being rebuilt against
> libpng3. Ryan Murray has a point with his concerns about breaking
> compatibility with other distros on that. However I guess it too late now.
> On the other hand, for these remaining packages which you listed that
> need to be rebuilt with libpng3... wouldn't it be far better to not keep
> rebuilding gnome1 packages in sid against libpng3.
It definitely wouldn't be, as halting the rebuild of GNOME1 against libpng3
now amounts to saying "we fscked up GNOME1 for sid and we're too lazy to fix
it".
> I would think the cleaner approach would be to replace any package in sid
> with its newer gnome2 version.
For many packages that's not an option. GNOME2's APIs are (usually for very
good reasons) quite different from GNOME1's. This means that for upstream
authors moving to GNOME2 entails real porting: a lot of work, for which they
may not have had time yet (e.g. gabber AFAICT), or which they have chosen to
do as part of a development cycle in which a lot of other changes are being
made as well (e.g. galeon).
> For example there is a gnumeric in experimental now.
Indeed. Gnumeric provides an excellent illustration as to why we should
commit to the libpng3 transition for GNOME1, or roll back to libpng2.
Sid currently has Gnumeric 1.0.9 packages. 1.0.x is the stable branch; it is
very firmly in maintenance mode, and I'm very happy we were able to ship
woody with a fairly recent version of it. It is very stable and even if
sarge's release cycle were ($DEITY forbid) to approach woody's in length,
I'm happy with it. Still, there are things to be improved: the stable branch
in CVS has some updated translations and a fix for a crash involving named
expressions, both of which will eventually end up in a 1.0.10 release. I'd
love to bring those improvements to sid, but I can't, due to the png3
transition not being finished. With sid as it stands now, I end up with a
gnumeric binary which will attempt to load against both png2 and png3, which
is one of the more reliable recipes for core dumps galore.
Yes, there are Gnumeric 1.1.x packages in experimental. They are in
experimental for good reasons, as they are of the "it builds, ship it"
variety; for instance, there are several major functionality regressions
compared to 1.0.x: there is no graphing support and the help function is
broken. Upstream is still actively developing 1.1.x and will hopefully start
stabilising things towards a 1.2.x stable release series within the next few
months. Right now, it has just undergone a major rewrite that dropped use of
libole for use of the shiny new libgsf and is switching to pango for
rendering of cell contents. I'm happy to have 1.1.x in experimental, it is a
good staging area to prepare for when 1.2.x arrives, but I'd hate to be
forced to move it to sid prematurely just because an essentially
straight-forward transition isn't completed.
> That would avoid polluting sid with bastardized libpng3-built gnome1
> packages. Granted we will have some breakage perhaps due to the alpha
> nature of some of those
I think you are severely underrating the amount of work (and time) needed
for upstream authors to go from GNOME1 to GNOME2.
> but it seems infinitely better than what we are doing now with this mass
> rebuild of gnome1 against libpng3!
That rebuild is peanuts compared to the GNOME1 -> GNOME2 transition. We're
in for the penny with the png2 -> png3 transition, we'd better be in for the
pound as well. Once png2 -> png3 is finished, we'll have the time to
properly reflect on how this kind of transition can be handled with less
intermittent breakage in the future.
Ray
--
LWN normally tries to avoid talking much about Microsoft - it is simply
irrelevant to the free software world most of the time.
http://www.lwn.net/2000/0406/
Reply to: