On Apr 23, 10:34 pm, Tim Abbott <tabb...@mit.edu> wrote:
> On Thu, 23 Apr 2009, mabshoff wrote:
Hi Tim,
> > Another thing: In 3.4.1 we downgraded GAP to 4.4.10 from 4.4.12 that
> > was upgraded in Sage 3.3 due to a significant number of bugs and
> > issues in GAP 4.4.12. How would you deal with something like that in
> > the packaged version of Sage? The whole point about shipping nearly
> > every dependency in Sage was that we can do things like that because
> > it is the only way Sage does work reliably and pass doctests. That
> > does not really work too well with a distribution like Debain with
> > tens of thousands of packages. While the number of packages depending
> > on GAP are probably close to zero in Debian for something like NTL
> > this might be an issue.
>
> Actually, NTL wasn't in Debian until I packaged it as part of my effort to
> get Sage into Debian. So at present all its dependencies are maintained
> by me. But I understand your point -- GMP, for example, has dozens of
> dependencies. Upgrades of popular libraries in distributions like Debian
> are often done with a great deal of staging and care so that problems are
> discovered before people upgrade in the first place.
Sure, NTL might not be the best example here, but say matplotlib. We
did not update to an svn release to make life harder for you, but
because we needed a patch that was upstreamed and not easily
rebasable. I think all the issue can and will be sorted out in the
near future after 4.0, but the update to pari-svn will happen. Indeed,
it is a surprise that it did not already happen and quite a few bits
in Debian outside Sage do use the pari library, i.e. clisp has an
optional pari module for example. And there is really no way around
that since the stable pari release is buggy in many ways and upstream
has also recommended to use svn. Indeed, in a recent email Karim
Belabas has actually called the stable pari "pari stale". I am quite
supportive of getting all issues you have resolved, but it seems
rather hard to get this fixed. I guess you could have a pari-2.3.4.deb
(just like there are two different python.debs AFAIK).
> As for the actual issue of downgrading packages, that can be difficult in
> a distribution. I have seen it done in Debian in cases where the new
> release is totally broken -- this is usually done by Debian releasing a
> version 4.4.12+really4.4.10 or similar. But perhaps instead the Debian
> maintainer will work with upstream to fix the bugs in the newer release.
ok.
> It's important to understand that distribution release cycles have
> relatively long freeze periods during which one only fixes bugs, which
> means one generally has quite a bit of warning when there is a problem
> that results in a doctest failure, and so one can explore a number of
> measures for trying to fix the issue before a release goes out.
Sure, we complained, it wasn't fixed. There didn't really seem any
compelling reason to have 4.4.12 over 4.4.10, so we downgraded
awaiting upstream to fix the problem.
> > Another problem is that often we have to put in fixes or use CVS for
> > non-Linux builds and with the Windows port this will become even more
> > extreme. So I truly don't see any reasonable hope we will ship clean
> > upstream anytime soon. Obviously if it is clean upstream and fixes in
> > patches in the spkg you can just ignore it.
>
> Fixes for non-Linux platforms like Solaris or Windows that don't change
> ABI should be fine in a stable release. There are a lot of Sage patches
> that add
>
> #ifdef CYGWIN
> ...
> #endif
>
> or similar that can be safely assumed on inspection to be harmless on
> Linux. If Sage only is applying patches like this, it is easy to just use
> the upstream release Sage is patching. That's why I stated the goal of
> cleaning out all ABI-changing patches, not cleaning out all patches
> altogether.
Good. I am not against this, I am just pointing out the other side.
> -Tim Abbott
Cheers,
Michael
--~--~---------~--~----~------------~-------~--~----~
To post to this group, send email to sage-devel@googlegroups.com
To unsubscribe from this group, send email to
sage-devel-unsubscr...@googlegroups.com
For more options, visit this group at http://groups.google.com/group/sage-devel
URLs: http://www.sagemath.org
-~----------~----~----~----~------~----~------~--~---
