[patch, 3.4] Backport fix for PR c++/22618

[Volker Reichelt]

On  6 Dec, Richard Guenther wrote:
> On 12/6/05, Volker Reichelt <reichelt@igpm.rwth-aachen.de> wrote:
>> The following patch backports Mark's patch for accepts-invalid bug
>> in PR c++/22618 to the 3.4 branch.
>>
>> Bootstrapped and regtested on x86_64-unknown-linux-gnu.
>> Ok for the 3.4 branch?
> 
> Will this fix some rejects-valid or wrong-code issues?  If not, we
> should rather not backport fixes that may turn (wrong) working code
> into non-working code.  People have complained before if compiler
> behavior changes this way during a release series.
> 
> Richard.

I don't think supporting broken code helps anybody. If the code
is broken, the code has to be fixed *anyway* once people switch to
GCC 4.x or an older compiler (the bug is a regression, otherwise we
wouldn't backport the fix) or a completely different compiler.

Not backporting means that the compiler stays broken and allows buggy
code to be written that has to be fixed later. This is an additional
burden that we can avoid by backporting the patch.
We do have a complaint that the compiler is broken now, namely
PR c++/22618.

Btw, we already backported the fix, namely from mainline to the
4.0 branch.

Regards,
Volker