Did you also un-revert the "support for temporary destructors" patch? Without
this, the test you mentioned has to fail, because it has no way of detecting
that the destructor does not return. With that patch applied, all tests pass
(at least for me :) ). The patch is incomplete, and it could fail, but this
will only happen for more complex expressions. To deal with that, I have tried
a bit different approach. It is still based on the "saving the conditional
expression value in the environment" idea, but I'll be saving it differently,
not in the processBranch terminators. I'll attach that as a separate patch once
i'm confident about it.
I didn't get around to trying the trivial inlined function. I will reply to
the rest of the comment after that.
Right now, I have a different question, which is somewhat unrelated, but it
is stopping me from doing large-scale tests of my changes. E.g., for code
patterns like this
const std::initializer_list<A> &a = { A(), A() };
I end up with having destructor for 'const A [2]' in the CFG. This is then
causing assertion failures all around the analyzer, since it assumes that only
CXXRecordDecl's can have destructors. So, my question is: Is the presence of
this element in the CFG intentional or is it a bug. I.e., should I fix the
CFGBuilder to not generate the node, or change the analyzer to deal with it.
Right now, I'm inclined towards the first option, as I don't see the purpose in
having a destructor for the array (and i have the destructors for the contained
elements in the cfg anyway), but I wanted to double check that. Do you know the
answer, or should I direct it to someone else?
http://llvm-reviews.chandlerc.com/D1259
_______________________________________________
cfe-commits mailing list
[email protected]
http://lists.cs.uiuc.edu/mailman/listinfo/cfe-commits
