Regressions in 3.2

Regressions in 3.2

[Brad Lucier]

I thought I would try to say something funny, but in such a diverse group,
humor is not a sure thing.

So I'd just like to point out that there are 500+ regressions on
sparc-sun-solaris2.8 from the 3.1 prerelease:

http://gcc.gnu.org/ml/gcc-testresults/2002-05/msg00389.html

and it might make for smoother development if some of these were squashed
earlier in the development cycle.

Brad

Re: Regressions in 3.2

[Gerald Pfeifer]

On Mon, 13 May 2002, Brad Lucier wrote:
> So I'd just like to point out that there are 500+ regressions on
> sparc-sun-solaris2.8 from the 3.1 prerelease:
>
> http://gcc.gnu.org/ml/gcc-testresults/2002-05/msg00389.html

And several other nasty regression on mainline. For example c++/6611
  http://gcc.gnu.org/cgi-bin/gnatsweb.pl?cmd=view%20audit-trail&pr=6611
is a really bad one (for me at least).

Mark, how shall we deal with such regressions? Mark them high in GNATS
or wait until GCC 3.2 has branched?

(I think we really should try to get regressions fixed rather sooner
than later; stage 1 is not ment to be "Let's hack away now and fix on
the release branch".)

Gerald
-- 
Gerald "Jerry" pfeifer@dbai.tuwien.ac.at http://www.dbai.tuwien.ac.at/~pfeifer/

Re: Regressions in 3.2

[Mark Mitchell]

--On Tuesday, May 14, 2002 11:52:30 AM +0200 Gerald Pfeifer 
<pfeifer@dbai.tuwien.ac.at> wrote:

> On Mon, 13 May 2002, Brad Lucier wrote:
>> So I'd just like to point out that there are 500+ regressions on
>> sparc-sun-solaris2.8 from the 3.1 prerelease:
>>
>> http://gcc.gnu.org/ml/gcc-testresults/2002-05/msg00389.html
>
> And several other nasty regression on mainline. For example c++/6611
>   http://gcc.gnu.org/cgi-bin/gnatsweb.pl?cmd=view%20audit-trail&pr=6611
> is a really bad one (for me at least).
>
> Mark, how shall we deal with such regressions? Mark them high in GNATS
> or wait until GCC 3.2 has branched?

I'm not sure how to answer the GNATS question.  We really need a "release"
field to indicate for which release the bug is high priority.  I'd prefer
to keep high priority for the 3.1 branch, at this point.

On the other hand, you're definitely right; these bugs need to get fixed
-- and soon!  Independently of what we do with GNATS, we need to get them
fixed.  Please track down the patches that caused your problems, and ask
the person who made the change to fix the problem.  If that's not working,
I'll ask that we freeze the tree until the problems get fixed.

Let's make this clear: although the mainline is an appropriate place for
new, sometimes destabilizing work, it's not a place for regressions.  If
you've made a mess, it's your obligation to go clean up your mess as
quickly as possible.

-- 
Mark Mitchell                mark@codesourcery.com
CodeSourcery, LLC            http://www.codesourcery.com

Re: Regressions in 3.2

[David S. Miller]

   From: Mark Mitchell <mark@codesourcery.com>
   Date: Tue, 14 May 2002 10:18:12 -0700
   
   Let's make this clear: although the mainline is an appropriate place for
   new, sometimes destabilizing work, it's not a place for regressions.  If
   you've made a mess, it's your obligation to go clean up your mess as
   quickly as possible.

%99 of these problems existed long before DFA or any of the other
bigger changes have gone in.

I'm suspecting it is a bunch of "stuff that went into 3.1, but not
into the mainline where we should fix it 'properly'" and the
'properly' bit has slipped through the cracks.

I have this huge diff between the 3.1 branch and the mainline I did a
couple weeks ago before DFA and other big bits went in, and I still
have to go through the rest of it looking for problem causing
differences.

Frankly, I'm still spending the bulk of my time with 3.1 looking for
Sparc show stoppers I should fix in 3.1.1 To me my time is better
allocated to that and not the mainline.  I simply don't have the time
right now needed to look into the mainline regressions on Sparc.

I know that the Sparc backend is basically between the 3.1 branch and
the mainline, sans the DFA bits.  That was the first thing I verified
when I started diffing the mainline with the branch the other week.
No fixes have been lost in the Sparc backend.

I remember that noting in particular that there are a lot of
non-trivial differences in the c++ front end between the branch
and the mainline.  I am not very skilled in this area.  I also
note that the bulk of the Sparc regressions on the mainline are in the
libstdc++ and c++ torture testsuires.

Mark, it may be instructive for Jason or yourself to diff the branch
c++ frontend with the mainline and look for anything that sticks out
like a sore thumb.  Just a suggestion..

Re: Regressions in 3.2

[Mark Mitchell]

--On Tuesday, May 14, 2002 05:45:16 PM -0700 "David S. Miller" 
<davem@redhat.com> wrote:

>    From: Mark Mitchell <mark@codesourcery.com>
>    Date: Tue, 14 May 2002 10:18:12 -0700
>
>    Let's make this clear: although the mainline is an appropriate place
> for    new, sometimes destabilizing work, it's not a place for
> regressions.  If    you've made a mess, it's your obligation to go clean
> up your mess as    quickly as possible.
>
> %99 of these problems existed long before DFA or any of the other
> bigger changes have gone in.

That may well be; I don't mean to cast aspersions on any particular
patch or patches.  It's a general comment.

> I'm suspecting it is a bunch of "stuff that went into 3.1, but not
> into the mainline where we should fix it 'properly'" and the
> 'properly' bit has slipped through the cracks.

Could well be...

> Mark, it may be instructive for Jason or yourself to diff the branch
> c++ frontend with the mainline and look for anything that sticks out
> like a sore thumb.  Just a suggestion..

Indeed.  After 3.1 is out, I'll start on this kind of thing.

Thanks,

-- 
Mark Mitchell                mark@codesourcery.com
CodeSourcery, LLC            http://www.codesourcery.com

Re: Regressions in 3.2

[Brad Lucier]

>    Let's make this clear: although the mainline is an appropriate place for
>    new, sometimes destabilizing work, it's not a place for regressions.  If
>    you've made a mess, it's your obligation to go clean up your mess as
>    quickly as possible.
> 
> %99 of these problems existed long before DFA or any of the other
> bigger changes have gone in.

From my point of view, part of the problem was that there were so many
problems with the 3.1 sparc-solaris port that it wouldn't even bootstrap 6 weeks
before the proposed release date (this was not an isolated experience in
the months and weeks leading up to the release) and it took what appeared to
an outsider to be an enormous (and impressive) effort to clean up 3.1 to get
it into reasonable shape for release.

Now, unfortunately, a lot of that effort may have to be repeated on the 
mainline.

Brad

Re: Regressions in 3.2

[Phil Edwards]

On Wed, May 15, 2002 at 07:16:40AM -0500, Brad Lucier wrote:
> From my point of view, part of the problem was that there were so many
> problems with the 3.1 sparc-solaris port that it wouldn't even bootstrap 6 weeks
> before the proposed release date (this was not an isolated experience in
> the months and weeks leading up to the release) and it took what appeared to
> an outsider to be an enormous (and impressive) effort to clean up 3.1 to get
> it into reasonable shape for release.
> 
> Now, unfortunately, a lot of that effort may have to be repeated on the 
> mainline.

For the port-specific things, would it be feasible to just copy the 3.1
file back on top of the mainline files as a starting point?  I've never
worked hard on a port, so it's quite possible I'm talking out of my ass
here, but it seems like the 3.1 sparc code is now in much better shape
than what's on the mainline.

Phil

-- 
If ye love wealth greater than liberty, the tranquility of servitude greater
than the animating contest for freedom, go home and leave us in peace.  We seek
not your counsel, nor your arms.  Crouch down and lick the hand that feeds you;
and may posterity forget that ye were our countrymen.            - Samuel Adams

Re: Regressions in 3.2

[law]

In message <20020515154255.A2579@disaster.basement.lan>, Phil Edwards writes:
 > On Wed, May 15, 2002 at 07:16:40AM -0500, Brad Lucier wrote:
 > > From my point of view, part of the problem was that there were so many
 > > problems with the 3.1 sparc-solaris port that it wouldn't even bootstrap 6
 >  weeks
 > > before the proposed release date (this was not an isolated experience in
 > > the months and weeks leading up to the release) and it took what appeared 
 > to
 > > an outsider to be an enormous (and impressive) effort to clean up 3.1 to g
 > et
 > > it into reasonable shape for release.
 > > 
 > > Now, unfortunately, a lot of that effort may have to be repeated on the 
 > > mainline.
 > 
 > For the port-specific things, would it be feasible to just copy the 3.1
 > file back on top of the mainline files as a starting point?  I've never
 > worked hard on a port, so it's quite possible I'm talking out of my ass
 > here, but it seems like the 3.1 sparc code is now in much better shape
 > than what's on the mainline.
That's usually a bad idea -- for example, you'd stomp on the DFA
scheduler changes.

This is going to involve careful analysis of patches, both to the
backend and the front-end to identify and fix the regressions.

jeff

Re: Regressions in 3.2

[Phil Edwards]

On Tue, May 14, 2002 at 10:18:12AM -0700, Mark Mitchell wrote:
> I'm not sure how to answer the GNATS question.  We really need a "release"
> field to indicate for which release the bug is high priority.  I'd prefer
> to keep high priority for the 3.1 branch, at this point.

Bugzilla's got it.  I forget why we're waiting:  was Bugzilla put on hold
until post-3.1, or is there something we should be actually working on?
If the former, okay; if the latter, what?

-- 
If ye love wealth greater than liberty, the tranquility of servitude greater
than the animating contest for freedom, go home and leave us in peace.  We seek
not your counsel, nor your arms.  Crouch down and lick the hand that feeds you;
and may posterity forget that ye were our countrymen.            - Samuel Adams

Re: Regressions in 3.2

[Joe Buck]

> On Tue, May 14, 2002 at 10:18:12AM -0700, Mark Mitchell wrote:
> > I'm not sure how to answer the GNATS question.  We really need a "release"
> > field to indicate for which release the bug is high priority.  I'd prefer
> > to keep high priority for the 3.1 branch, at this point.
> 
> Bugzilla's got it.  I forget why we're waiting:  was Bugzilla put on hold
> until post-3.1, or is there something we should be actually working on?
> If the former, okay; if the latter, what?

The SC and RMS already blessed conversion to Bugzilla, so there's no
political approval issue.  It's just a matter of implementing it.

Re: Regressions in 3.2

[Craig Rodrigues]

On Wed, May 15, 2002 at 11:24:41AM -0700, Joe Buck wrote:
> 
> > On Tue, May 14, 2002 at 10:18:12AM -0700, Mark Mitchell wrote:
> > > I'm not sure how to answer the GNATS question.  We really need a "release"
> > > field to indicate for which release the bug is high priority.  I'd prefer
> > > to keep high priority for the 3.1 branch, at this point.
> > 
> > Bugzilla's got it.  I forget why we're waiting:  was Bugzilla put on hold
> > until post-3.1, or is there something we should be actually working on?
> > If the former, okay; if the latter, what?
> 
> The SC and RMS already blessed conversion to Bugzilla, so there's no
> political approval issue.  It's just a matter of implementing it.

Let's not beat that dead horse any more.  I think the carcass of that
discussion is now a fine purée.

There was some discussion of the plan here between Chris Faylor and
Dan Berlin:
http://gcc.gnu.org/ml/gcc/2002-04/msg00023.html

Chris and Dan are quite busy with their non-gcc lives, so that's why
it hasn't happened yet.  It will happen, I'm sure, sometime after gcc 3.1,
and the more help that Dan gets, I'm sure the faster it will go.

-- 
Craig Rodrigues        
http://www.gis.net/~craigr    
rodrigc@attbi.com

Re: Regressions in 3.2

[Phil Edwards]

On Wed, May 15, 2002 at 02:38:21PM -0400, Craig Rodrigues wrote:
> Let's not beat that dead horse any more.  I think the carcass of that
> discussion is now a fine purée.

Agreed.  I just wanted to check that we were waiting on some external event
(in this case, for new hardware to arrive), rather than lack of people
(because I have an interest in making this work).

-- 
If ye love wealth greater than liberty, the tranquility of servitude greater
than the animating contest for freedom, go home and leave us in peace.  We seek
not your counsel, nor your arms.  Crouch down and lick the hand that feeds you;
and may posterity forget that ye were our countrymen.            - Samuel Adams

Re: Regressions in 3.2

[Jan Hubicka]

> I thought I would try to say something funny, but in such a diverse group,
> humor is not a sure thing.
> 
> So I'd just like to point out that there are 500+ regressions on
> sparc-sun-solaris2.8 from the 3.1 prerelease:
> 
> http://gcc.gnu.org/ml/gcc-testresults/2002-05/msg00389.html
> 
> and it might make for smoother development if some of these were squashed
> earlier in the development cycle.

I've fixed at least the branch probability regression (about 6 in your list).

Concerning the libjava failures, in case they look to be CFG related, please
let me know.

Honza
> 
> Brad

Re: Regressions in 3.2

[John David Anglin]

As of last night, there were no regressions in 3.2 wrt 3.1 under hppa-linux
and hppa2.0w-hp-hpux11.11.  The profile arc problems have been resolved.
Jason's C++ patch backed out.  The problem with 950704-1.c is resolved
with changes to the pa backend.

I am pleased that all issues that have come up have been dealt with
quickly.

Dave
-- 
J. David Anglin                                  dave.anglin@nrc.ca
National Research Council of Canada              (613) 990-0752 (FAX: 952-6605)