PR6678?

[Paolo Carlini]

Benjamin Kosnik wrote:

>>interesting to compare such behavior with that of 3.0.4: if wecan
>>assess that this is a regression from 3.0.4 we can raise the priority
>>of the PR to HIGH and almost certainly someone will look into it
>>before 3.1.1.
>>    
>>
>This should be bumped up to HIGH now. I don't think it is a regression,
>but it is serious. I noticed something similar when I did the valgrind
>fixups.
>
Done!

Paolo.