Re: c++/8931: g++ 3.2 fails to enforce access rules

Re: c++/8931: g++ 3.2 fails to enforce access rules

[Martin v. Löwis]

[8931]
> g++ 3.2 fails to enforce access rules

[Wolfgang]
> Since it is not a regression, it is not going to be fixed
> in any 3.2.* 

[Gaby]
> It would be really helpful if non-invasive bug fixes could make it to
> branch when it is not frozen.

While I sympathize with this request, I think clear policies are
needed; and "it is non-invasive" is not clear enough.

Instead, I would propose a policy under which this specific fix would
not be backported, as it is from the "accepts-illegal" class. People
running into this bug can use all well-formed code with a compiler
that only has accepts-illegal bugs, so having such bugs in a
maintenance release should be no obstacle for any user. At worst,
users will find that other compilers reject the code, at which point
they will have to fix their code, and be done with it.

More specifically:
- do backport:
  class is rejects-legal, wrong-code, ice-on-legal-code

- don't backport:
  class is accepts-illegal, ice-on-illegal-code, change-request, support
  severity is serious or non-critical (i.e. a work-around is possible)

I realise that this classification is both overlapping and
non-exhausting. If in doubt, don't backport.

As for rejecting backports because of severity: it would be good to
collect the possible work-arounds in case the work-around isn't
obvious (or it isn't obvious what the problem is), and the bug is
frequently reported.

Just my 0.02EUR,

Martin

Re: c++/8931: g++ 3.2 fails to enforce access rules

[Paul Jarc]

"Martin v. Löwis" <martin@v.loewis.de> wrote:
> People running into this bug can use all well-formed code with a
> compiler that only has accepts-illegal bugs, so having such bugs in
> a maintenance release should be no obstacle for any user. At worst,
> users will find that other compilers reject the code, at which point
> they will have to fix their code, and be done with it.

I don't think that's the worst that can happen.  In general, if the
source is not well-formed, how can we know that the generated code
will not be harmful?


paul

Re: c++/8931: g++ 3.2 fails to enforce access rules

[Gabriel Dos Reis]

Wolfgang Bangerth <bangerth@ticam.utexas.edu> writes:

| - given the really *large* number of open bug reports, I think the scarce
|   bug fixing resources gcc has serve the community better in the long term
|   if we let them focus on 3.3, rather than spending time backporting 
|   fixes. This way we might get 3.3 out earlier, which will certainly be 
|   better than any 3.2.2.

I'm not suggesting people spend their time backporting every
imaginable patch that happens to fix some bug on mainline.  There are
bug-fix patches that don't need any particular action than running
patch + regtesting.  I'm obvisouly talking of such patches.

I doubt we'll get 3.3 earlier just because we refuse to apply the same
patch (we do know passes regtesting) on mainline and branch.  

Or we could just make it clear that 3.2 branch is dead and have people
not  bothering about it.  That way, we could expect people focus
mainly on 3.3:  That would have the effect of saving any effort on 3.2
branch and make user clearly know that they should not expect anything
about 3.2.2.  That way, we could perhaps have 3.3 earlier.  It would
certainly be better than 3.2.2 since the latter would be non-existent.

-- Gaby

Re: c++/8931: g++ 3.2 fails to enforce access rules

[Wolfgang Bangerth]

> | Synopsis: g++ 3.2 fails to enforce access rules
> | State-Changed-From-To: open->closed
> |     Since it is not a regression, it is not going to be fixed
> |     in any 3.2.* and there is no value in keeping this report
> |     open.
> 
> It would be really helpful if non-invasive bug fixes could make it to
> branch when it is not frozen.
> 
> Setting the bar to only regression fixes is, IMHO, too high and
> renders the dot releases less useful and less attractive.  Indeed,
> I've seen lot of PRs being closed on the basis that they are fixed on
> mainline and since they are not regressions they won't be fixed in
> 3.2.x.  The net effect is that people would have to wait for some (long)
> undeterminated time before they had a compiler that fixes the bugs,
> and meanwhile we will be releasing compilers that could include
> those patches.  

I think I even concur, I am just executing the policies that have been 
set. However, in the discussion I would like some points to be kept in 
mind:
- if there are too many open reports in the database, it is difficult to
  manage and very annoying when one re-visits reports that are "half-open" 
  every so often. You do realize that we presently have about 1800 (!)
  non-closed reports and that it is easy to lose yourself into this 
  amount, right?
- given the really *large* number of open bug reports, I think the scarce
  bug fixing resources gcc has serve the community better in the long term
  if we let them focus on 3.3, rather than spending time backporting 
  fixes. This way we might get 3.3 out earlier, which will certainly be 
  better than any 3.2.2.
- if we allow other patches into the branch, it needs more testing; the 
  thing with limited resources applies here as well.
- someone will have to find the patch that fixed it on the mainline.

For this particular case I don't know how invasive the fix might be, so I 
can't comment on its impact on stability of the branch. 

W.

-------------------------------------------------------------------------
Wolfgang Bangerth              email:           bangerth@ticam.utexas.edu
                               www: http://www.ticam.utexas.edu/~bangerth

Re: c++/8931: g++ 3.2 fails to enforce access rules

[Gabriel Dos Reis]

bangerth@dealii.org writes:

| Synopsis: g++ 3.2 fails to enforce access rules
| 
| State-Changed-From-To: open->closed
| State-Changed-By: bangerth
| State-Changed-When: Fri Dec 13 12:49:13 2002
| State-Changed-Why:
|     Present mainline gives
|     tmp/g> /home/bangerth/bin/gcc-3.3-pre/bin/c++ -c x.cc
|     x.cc: In instantiation of `S<C<int> >':
|     x.cc:14:   instantiated from here
|     x.cc:10: error: `typedef int C<int>::Private' is private
|     x.cc:4: error: within this context
|     
|     Since it is not a regression, it is not going to be fixed
|     in any 3.2.* and there is no value in keeping this report
|     open.

It would be really helpful if non-invasive bug fixes could make it to
branch when it is not frozen.

Setting the bar to only regression fixes is, IMHO, too high and
renders the dot releases less useful and less attractive.  Indeed,
I've seen lot of PRs being closed on the basis that they are fixed on
mainline and since they are not regressions they won't be fixed in
3.2.x.  The net effect is that people would have to wait for some (long)
undeterminated time before they had a compiler that fixes the bugs,
and meanwhile we will be releasing compilers that could include
those patches.  

My two cents.

-- Gaby