Re: GCC 4.2.2 Status Report

[Daniel Berlin]

On 9/4/07, Mark Mitchell <[EMAIL PROTECTED]> wrote:
> Summary
> ===
>
> The GCC 4.2.1 release was July 18, so our target for a 4.2.2 release is
> September 18th.  I plan to build RC1 this Sunday, September 9.  If all
> goes well, we'll have 4.2.2 out around the 18th; if not, we'll delay a
> bit from there.
>
> One critical issue: has GCC 4.2.x been fully converted to GPLv3, at this
> point?  If not, we'll have to wait until that is done before we can
> release, per the FSF's instructions.
>
> Quality
> ===
>
> Here are the open regressions:
>
> Priority   #
>   ---
> P1 26
> P2108
> P3  3
> Total 137
>
> Many of the P1s are ICEs, so at least users know the compiler is broken...
>
> We still have the nasty aliasing problems:
>
> PR32182 [4.2 Regression] -fstrict-aliasing optimizations cause co...

It's not clear from the PR that this is either an aliasing bug, and not either
1. a C++ FE bug
or
2. an invalid testcase
> PR32328 [4.2 Regression] -fstrict-aliasing causes skipped code

Hard to fix without regression 28778, i'm still working on how to do it.
>
> and various other such problems.  We also have:
>
> PR32327 [4.2 Regression] Incorrect stack sharing causing removal ...
>
> though Diego's last comment seems to indicate that's something of a
> could-happen bug at the moment.
>
> In short, I don't see anything here that would prevent a release,
> though, of course, I'd certainly be happier to get the number of
> regressions (and, particularly, P1 regressions) down.
>
> Previous Report
> ===
>
> http://gcc.gnu.org/ml/gcc/2007-07/msg00704.html
>
> --
> Mark Mitchell
> CodeSourcery
> [EMAIL PROTECTED]
> (650) 331-3385 x713
>

Re: GCC 4.2.2 Status Report

[Richard Guenther]

On 9/5/07, Daniel Berlin <[EMAIL PROTECTED]> wrote:
> On 9/4/07, Mark Mitchell <[EMAIL PROTECTED]> wrote:
> >
> > We still have the nasty aliasing problems:
> >
> > PR32182 [4.2 Regression] -fstrict-aliasing optimizations cause co...
>
> It's not clear from the PR that this is either an aliasing bug, and not either
> 1. a C++ FE bug
> or
> 2. an invalid testcase

While I cannot exclude (1), the testcase is certainly valid.

Richard.

Re: GCC 4.2.2 Status Report

[Joseph S. Myers]

On Tue, 4 Sep 2007, Mark Mitchell wrote:

> One critical issue: has GCC 4.2.x been fully converted to GPLv3, at this
> point?  If not, we'll have to wait until that is done before we can
> release, per the FSF's instructions.

Apart from anything else, we are still awaiting new wording for the 
various exceptions in use so installed headers and runtime libraries can 
be converted - I don't know if that's critical for this release, but 
exceptions for projects looking to release soon are supposed to be a 
higher priority 
.

-- 
Joseph S. Myers
[EMAIL PROTECTED]

Re: GCC 4.2.2 Status Report

[Daniel Berlin]

On 9/5/07, Richard Guenther <[EMAIL PROTECTED]> wrote:
> On 9/5/07, Daniel Berlin <[EMAIL PROTECTED]> wrote:
> > On 9/4/07, Mark Mitchell <[EMAIL PROTECTED]> wrote:
> > >
> > > We still have the nasty aliasing problems:
> > >
> > > PR32182 [4.2 Regression] -fstrict-aliasing optimizations cause co...
> >
> > It's not clear from the PR that this is either an aliasing bug, and not 
> > either
> > 1. a C++ FE bug
> > or
> > 2. an invalid testcase
>
> While I cannot exclude (1), the testcase is certainly valid.
>
>
Can you please update the PR to this effect?
It's certainly not clear from reading the comments :)
>

Re: GCC 4.2.2 Status Report

[Richard Guenther]

On 9/5/07, Daniel Berlin <[EMAIL PROTECTED]> wrote:
> On 9/5/07, Richard Guenther <[EMAIL PROTECTED]> wrote:
> > On 9/5/07, Daniel Berlin <[EMAIL PROTECTED]> wrote:
> > > On 9/4/07, Mark Mitchell <[EMAIL PROTECTED]> wrote:
> > > >
> > > > We still have the nasty aliasing problems:
> > > >
> > > > PR32182 [4.2 Regression] -fstrict-aliasing optimizations cause co...
> > >
> > > It's not clear from the PR that this is either an aliasing bug, and not 
> > > either
> > > 1. a C++ FE bug
> > > or
> > > 2. an invalid testcase
> >
> > While I cannot exclude (1), the testcase is certainly valid.
> >
> >
> Can you please update the PR to this effect?
> It's certainly not clear from reading the comments :)

I thought comment #13 was enough ;)

"It looks like 12.6.2/5-6 specify it enough to make the testcase valid.  The
BaseClass is only once initialized by the most derived object initializer
specification."

Richard.

Re: GCC 4.2.2 Status Report

[Mark Mitchell]

Joseph S. Myers wrote:
> On Tue, 4 Sep 2007, Mark Mitchell wrote:
> 
>> One critical issue: has GCC 4.2.x been fully converted to GPLv3, at this
>> point?  If not, we'll have to wait until that is done before we can
>> release, per the FSF's instructions.
> 
> Apart from anything else, we are still awaiting new wording for the 
> various exceptions in use so installed headers and runtime libraries can 
> be converted

Personally, I don't see how that's a problem, as long as the license on
the files with exceptions is still GPLv2 + .  That's not a change for users.  The files with exceptions
might not be compatible with GPLv3 by themselves -- but the exception
allows them to be linked into complete GPLv3 programs, since it allows
combination with anything.

However, I'll err on the side of caution and ask the FSF (via the SC
list) whether they have any objection to us going ahead.

Thanks,

-- 
Mark Mitchell
CodeSourcery
[EMAIL PROTECTED]
(650) 331-3385 x713

Re: GCC 4.2.2 Status Report

[Joe Buck]

Joseph S. Myers wrote:
> > Apart from anything else, we are still awaiting new wording for the 
> > various exceptions in use so installed headers and runtime libraries can 
> > be converted

On Wed, Sep 05, 2007 at 08:54:08AM -0700, Mark Mitchell wrote:
> Personally, I don't see how that's a problem, as long as the license on
> the files with exceptions is still GPLv2 +  before>.  That's not a change for users.  The files with exceptions
> might not be compatible with GPLv3 by themselves -- but the exception
> allows them to be linked into complete GPLv3 programs, since it allows
> combination with anything.

Mark,

There are no GPLv2-only files in the repository and there never were.  The
license up until this summer was GPLv2 or any later version.  So there
is no issue of license incompatibility.  Any of the files with exceptions
can be linked with GPLv3 code with no problem, and even without an
exception there's no problem.  An external GPLv2-or-any-later-version
front end can also be linked with gcc 4.3 and distributed.

There is only a license incompatibility issue for code that is GPLv2
only (without the "any later version" clause) and the FSF has never
accepted such code.

Re: GCC 4.2.2 Status Report

[Richard Kenner]

> The files with exceptions might not be compatible with GPLv3 by themselves

Why?  I thought GPLv2 and GPLv3 are "compatible".

Re: GCC 4.2.2 Status Report

[Joe Buck]

On Wed, Sep 05, 2007 at 12:18:36PM -0400, Richard Kenner wrote:
> > The files with exceptions might not be compatible with GPLv3 by themselves
> 
> Why?  I thought GPLv2 and GPLv3 are "compatible".

They are not; each requires that the work as a whole be licensed the same
as the individual file.  However, we never had any GPLv2-only files.  A
GPLv2-or-any-later-version file is compatible with GPLv3.