Boost library proposal for Natty

[Scott Kitterman]

On Monday, October 11, 2010 12:30:46 pm Clint Byrum wrote:
> On Oct 11, 2010, at 7:30 AM, Scott Kitterman wrote:
> > Boost1.42 is default in Debian Sid/Squeeze and what we have in Main for
> > Maverick.  Boost1.43 and 1.44 have been released upstream, but are not
> > packaged in Debian.  Additionally, we carried boost1.40 in Universe in
> > Maverick so that we would have a boost that built the openmpi parts of
> > the library.
> > 
> > My proposal for Natty is that we do ~nothing.  The Debian default of 1.42
> > is unlikely to change before Natty feature freeze and so we should plan
> > on this being the Ubuntu default for Natty.  We should also keep a full
> > (with MPI) version of boost in the archive.  I would propose this stay
> > at 1.40 unless a newer version appears in Debian in the relatively near
> > future.
> 
> Can you explain a bit why 1.42 doesn't have MPI? I understand they
> changed some things in recent releases with regard to OpenMPI, but
> not exactly what has changed.

openmpi is in Universe and we don't want it in Main (there's nothing driving 
it there and supporting it in Main because it's used to build boost would be 
too much for too little gain).  This is nothing to do with upstream, but 
strictly how we manage the Ubuntu archive.

> > There are still some packages using 1.40 that could migrate to 1.42. 
> > This transition should be finished so that dropping 1.40 when the time
> > comes is easy.
> > 
> > Comments?  Suggestions?
> > 
> > I'd like to get this out of the way now so that we know that all the
> > syncs are built against the version of boost we want to keep for Natty. 
> > In the event that Squeeze is released before Natty feature freeze and
> > Debian updates to a newer boost for default, then we'll look at the
> > situation and re-evaluate.
> 
> Thinking forward to the future, I'd like to see the newest version
> of libraries like boost get into the earliest possible release of
> Ubuntu. My reasoning here, is that if boost 1.44 is only added
> during 11.10, then bugs reported in that release will not likely
> have time for an upstream release which has a chance of getting
> into 12.04, which will likely be the next LTS. Whereas if we ship
> 1.44 early, in Natty, then there will be a full 6 month cycle of
> Ubuntu to have the library in users' hands, which gives us time to
> evaluate and push bugs upstream well before the LTS cycle even
> begins.

In general, I think that this makes sense.  In the case of boost, I'm not so 
sure.  The boost libraries are large and would take more than a little effort 
to package new versions.  Also the new versions have some new libraries and I 
would not like to get out of sync on package splits or naming.

> I also think its a good idea for us to push a little beyond Debian
> when they are frozen. The proper way to do this, IMO, would be to
> work hard to get these new releases into Experimental and feed bug
> reports back. The more this is done, the better off Debian unstable
> will be when squeeze is finally released, right?

True.  If the Debian Boost maintainers are willing to entertain boost1.44 in 
Experimental and someone is inclined to help them, I think it makes sense.  I 
would not, however, switch a default boost that's newer than Debian's.  If we 
get boost1.44 in the archive we can use it to replace boost1.40 to be in 
Universe with MPI (which reinforces the idea that we want to get packages 
still using 1.40 shifted to 1.42 to make this transition easy if we get a 
chance to do it).

Scott K