Re: Compilation time (was Re: GCC 3.3)

[Benjamin Kosnik <bkoz at redhat dot com>]

> I believe that there are many opportunities to improve the name lookup
> machinery with the effect of improving compile time.  Template
> instantiations in cc1plus is also a good candidate.

Gaby, I think the name lookup stuff you and Zack did was great. 

It would certainly be nice if template instantiation speed was made a
priority.... also somebody figuring out why 'extern template' seems to
slow down the compiler so much...

> | I also agree we should make compilation time a major goal for 3.4.

I agree, since speed has been swept under the rug for the 3.0.x, 3.1.x,
and 3.2.x releases, it might be nice to address it in a direct manner.
Matt, part of doing this might be getting the Apple regression tester to
actually spit out time data (even for things like bootstrap, and
individual timings of each language test run). That might make the speed
issue more visible.

> This also implies that we have a re-worked release process.

Again, I think the idea of "major goal" or "major goals" is one with
considerable merit. Tom Tromey, David Edelsohn, and I discussed this a
bit the last time this came up on this list. At that point, I believe
the plan was to suggest "major ideas for the next release" in the
immediate aftermath of a major release. Thusly, post 3.3 would seem to
be the place for this. 

It might also be a good place to think about formalizing some of the
best practices of the 3.2.x release cycle, such as announcements a month
ahead of time that branches will be frozen on a particular date (april
15, anybody) and that releases come two weeks afterwords (3.2.3 came
pretty close). I find silence and then out-of-the-blue messages of
things happening in four days less impressive, although understandable
considering the rigors of the job. I'm still in favor of doing quarterly
releases of the stable branch for at least a year after release, in much
this same manner.

Blah blah blah.... expect more disciplined resistance if any effort is
made to close the 3.4 branch soon after 3.3 is released.

-benjamin