Berin Lautenbach wrote:
Hmm. That's an interesting one. Have you seen a case where the
mp_attributes list is not already clear when the destructor is called?
Yes. That is why this fix has been appeared. But I can't remember those
particular case for now, after almost three years :) Sorry for late
report.
I can see how it could occur if the canonicalisation gets interrupted in
the midst of outputting Attribute nodes. This fix would work nicely,
and I think it should be safe as the code that normally cleans up
ensures mp_attributes is NULL, so it won't get called unnecessarily.
I reached the same conclusion when focusing on this issue today.
Good luck!
Vitaly
