V3 patch causes EH failures in ILP32 mode

[Mark Mitchell]

B. Kosnik wrote:
> 
>>When I worked on an error-checking tool that had a -Weffc++-like feature 
>>(and said tool was actually endorsed by Scott Meyers, no less), we tried 
>>to figure out whether or not dynamic memory really was involved using 
>>various heuristics.  For example, "does the constructor call "new" and 
>>assign that to a data member?"  The -Weffc++ implementation in G++ is 
>>not as sophisticated.
> 
> 
> Well, since we're wishing..... yes, I really do wish -Weffc++ was more
> sophisticated. One, for the issue under discussion: the heuristics you
> mention sound pretty cool.

Yes....

I also forgot to say earlier that you should not make changes like this 
on release branches.  This patch is a good example of an unncessary 
change with unintended consequences; the goal for the release branch is 
stability, not cleanliness.  This patch didn't fix a bug, let alone a 
regresssion -- it should never have been applied.

Thanks,

-- 
Mark Mitchell
CodeSourcery, LLC
mark@codesourcery.com