No basic improvements branch for 3.5 - work on tree-ssa

No basic improvements branch for 3.5 - work on tree-ssa

[Zack Weinberg]

Last release cycle there was a "basic improvements" branch maintained
throughout the bugfixes-only period.  This took changes suitable for
stage 2, which were merged into 3.4 after 3.3 branched.

I think this was mostly successful, but I'm not going to maintain
another such branch during this release cycle, because it would
increase the amount of work required to bring in the tree-ssa branch
at the beginning of the 3.5 development period.  Instead, I encourage
people to help out with the tree-ssa branch.

zw

Re: No basic improvements branch for 3.5 - work on tree-ssa

[Alexandre Oliva]

On Oct 20, 2003, "Zack Weinberg" <zack@codesourcery.com> wrote:

> I think this was mostly successful, but I'm not going to maintain
> another such branch during this release cycle, because it would
> increase the amount of work required to bring in the tree-ssa branch
> at the beginning of the 3.5 development period.

Why not just use the tree-ssa branch as the `basic improvements
branch' for this round?  It has been regarded as 3.5-to-be for quite a
while AFAIK.

-- 
Alexandre Oliva   Enjoy Guarana', see http://www.ic.unicamp.br/~oliva/
Red Hat GCC Developer                 aoliva@{redhat.com, gcc.gnu.org}
CS PhD student at IC-Unicamp        oliva@{lsd.ic.unicamp.br, gnu.org}
Free Software Evangelist                Professional serial bug killer

Re: No basic improvements branch for 3.5 - work on tree-ssa

[Zack Weinberg]

Alexandre Oliva <aoliva@redhat.com> writes:

> On Oct 20, 2003, "Zack Weinberg" <zack@codesourcery.com> wrote:
>
>> I think this was mostly successful, but I'm not going to maintain
>> another such branch during this release cycle, because it would
>> increase the amount of work required to bring in the tree-ssa branch
>> at the beginning of the 3.5 development period.
>
> Why not just use the tree-ssa branch as the `basic improvements
> branch' for this round?  It has been regarded as 3.5-to-be for quite a
> while AFAIK.

That's what I was suggesting.
zw

Re: No basic improvements branch for 3.5 - work on tree-ssa

[Richard Kenner]

    Last release cycle there was a "basic improvements" branch maintained
    throughout the bugfixes-only period.  This took changes suitable for
    stage 2, which were merged into 3.4 after 3.3 branched.

    I think this was mostly successful, but I'm not going to maintain
    another such branch during this release cycle, because it would
    increase the amount of work required to bring in the tree-ssa branch
    at the beginning of the 3.5 development period.  Instead, I encourage
    people to help out with the tree-ssa branch.

Unfortunately, we've found out that people usually work on what they
want to work on since this is basically a volunteer project.  Having
the basic improvements branch was useful in that it kept changes in
public view and not all in private trees.  It would seem to me that
all the previous arguments for it still apply: it makes it easier to
merge these change in during the start of the 3.5 development if they
are all in one place rather than in numerous private trees.  Another risk
of being in private trees is that by the time the patch is submitted, the
justifications for it will be less fresh in the mind of the author.

I don't understand the concerns you have with respect to the tree-ssa
experiment.

Re: No basic improvements branch for 3.5 - work on tree-ssa

[Richard Kenner]

    Why not just use the tree-ssa branch as the `basic improvements
    branch' for this round?  It has been regarded as 3.5-to-be for quite a
    while AFAIK.

Sort of, but that would put pressure on tree-ssa to actually make 3.5
or else cause a merge problem with the unrelated changes in that
branch.  I think it's much simpler for both efforts to not tie them
together like that in terms of schedules.

Re: No basic improvements branch for 3.5 - work on tree-ssa

[law]

In message <10310230257.AA14264@vlsi1.ultra.nyu.edu>, Richard Kenner writes:
 >    Why not just use the tree-ssa branch as the `basic improvements
 >    branch' for this round?  It has been regarded as 3.5-to-be for quite a
 >    while AFAIK.
 >
 >Sort of, but that would put pressure on tree-ssa to actually make 3.5
 >or else cause a merge problem with the unrelated changes in that
 >branch.  I think it's much simpler for both efforts to not tie them
 >together like that in terms of schedules.
Actually, we would welcome the additional folks working on the branch.  

Being open to a wider development audience would be a good thing at
this point IMHO.

jeff

Re: No basic improvements branch for 3.5 - work on tree-ssa

[Joe Buck]

On Wed, Oct 22, 2003 at 10:57:26PM -0400, Richard Kenner wrote:
>     Why not just use the tree-ssa branch as the `basic improvements
>     branch' for this round?  It has been regarded as 3.5-to-be for quite a
>     while AFAIK.
> 
> Sort of, but that would put pressure on tree-ssa to actually make 3.5
> or else cause a merge problem with the unrelated changes in that
> branch.  I think it's much simpler for both efforts to not tie them
> together like that in terms of schedules.

I don't see any reason to bump the minor version number if nothing is going to
change.  There's no question of tree-ssa not making 3.5, we would delay
3.5 until it does make it as far as I'm concerned.  tree-ssa is the
only hope I can see that GCC will ever be able to fix a critical failing
that produces inferior C++ code.  Without it, there's no point to a 3.5
release.

Re: No basic improvements branch for 3.5 - work on tree-ssa

[Arnaud Charlet]

> There's no question of tree-ssa not making 3.5, we would delay
> 3.5 until it does make it as far as I'm concerned.  tree-ssa is the
> only hope I can see that GCC will ever be able to fix a critical failing
> that produces inferior C++ code.  Without it, there's no point to a 3.5
> release.

Let's please keep things in perspective here: there are many other languages
besides C++ in GCC.
Not everyone is using C++, so there can be lots of good reasons for
making a 3.5 release.

Arno

Re: No basic improvements branch for 3.5 - work on tree-ssa

[Joe Buck]

On Thu, Oct 23, 2003 at 06:30:32PM +0200, Arnaud Charlet wrote:
> > There's no question of tree-ssa not making 3.5, we would delay
> > 3.5 until it does make it as far as I'm concerned.  tree-ssa is the
> > only hope I can see that GCC will ever be able to fix a critical failing
> > that produces inferior C++ code.  Without it, there's no point to a 3.5
> > release.
> 
> Let's please keep things in perspective here: there are many other languages
> besides C++ in GCC.

I would expect that the GCC problems I refer to hurt performance of all
languages that use struct/record types.

Other languages will benefit from tree-ssa even more, and front ends will
be easier to write.

Re: No basic improvements branch for 3.5 - work on tree-ssa

[Jeff Sturm]

On Thu, 23 Oct 2003, Joe Buck wrote:
> I would expect that the GCC problems I refer to hurt performance of all
> languages that use struct/record types.

Exactly.  Java will clearly benefit from tree-ssa.  Plus there is the new
fortran FE.  So, tree-ssa isn't just for C++.

Jeff

Re: No basic improvements branch for 3.5 - work on tree-ssa

[Richard Kenner]

    Actually, we would welcome the additional folks working on the branch.  

Of course, but that wasn't the proposal.

The proposal was to *merge* the tree-ssa experiment branch and people doing
*unrelated* things.

It's all well-and-good to suggest that people work on the tree-ssa experiment,
but since GCC is a volunteer project, we can't force that.  People may choose
to do other work and there is as much of a need for a place to do that
work as there was previously.

Re: No basic improvements branch for 3.5 - work on tree-ssa

[law]

In message <10310231114.AA14907@vlsi1.ultra.nyu.edu>, Richard Kenner writes:
 >    Actually, we would welcome the additional folks working on the branch.  
 >
 >Of course, but that wasn't the proposal.
 >
 >The proposal was to *merge* the tree-ssa experiment branch and people doing
 >*unrelated* things.
And we would still welcome that.

Jeff

Re: No basic improvements branch for 3.5 - work on tree-ssa

[Richard Kenner]

     >The proposal was to *merge* the tree-ssa experiment branch and people
     >doing *unrelated* things.

    And we would still welcome that.

That's not the issue. The issue is who takes responsibility for isolating and
merging those changes into 3.5 if there's a delay in the tree-ssa project or
if the experiment doesn't work out?

I don't see the advantage of having one branch that contains two major
unrelated pieces of work.

Re: No basic improvements branch for 3.5 - work on tree-ssa

[Richard Kenner]

    I don't see any reason to bump the minor version number if nothing is
    going to change.

Of course, but lots of people are doing lots of work.

    There's no question of tree-ssa not making 3.5, we would delay 3.5
    until it does make it as far as I'm concerned.  

I guess I wasn't clear.  I wasn't talking about it being ready in time
for the release since indeed I agree that if the experiment looks like it's
working, it would be worth delaying the release for it.

What I was talking about was when the branch would be ready to be
merged into the mainline.  If that was not right at the start of phase
1, it would then hold up lots of unrelated work unless somebody
extracted and merged them.  I don't see what we gain by merging these
two unrelated things.

    tree-ssa is the only hope I can see that GCC will ever be able to fix
    a critical failing that produces inferior C++ code.  Without it,
    there's no point to a 3.5 release.

That's a bit C++-centric, I'd say.  There's also C, Java, Fortran, and Ada.
It's certainly not inconceivable that a 3.5 release would make sense if it
produced significant enough improvements in those languages even if it
still had a performance problem with C++.

Re: No basic improvements branch for 3.5 - work on tree-ssa

[Richard Kenner]

    I would expect that the GCC problems I refer to hurt performance of all
    languages that use struct/record types.

    Other languages will benefit from tree-ssa even more, and front ends will
    be easier to write.

Nobody is trying to argue that it isn't worthwhile to persue, just
that the statement that nothing else is important enough to justify a
3.5 seems extreme.