Well, ehm, let's start with the good news: this patch actually solves the
example main.cpp that I added to the bug report on January 15th. And it seems
to me to be a valid fix for the functions push_front(), push_back(),
emplace_front(), and emplace_back().
Here's the bad news: Unfortunately, this patch doesn't solve the actual
problem that std::list has: The pointers in __list_node_base still have the
type pointer to __list_node, although they can point to the anchor node which
is not a __list_node. The same is true for the __ptr_ member variables in
__list_iterator and __list_const_iterator. This leads to undefined behaviour
whenever these pointers are dereferenced although they currently point to the
anchor node, even if they are only dereferenced to access __prev_ or __next_.
I've slightly changed my example file to use splice() instead of
push_front(). And then the described problem reappears:
{F50161}
Please note that I've used a more recent compiler this time: g++
(Ubuntu/Linaro 4.8.1-10ubuntu9) 4.8.1. Please note further, that the patch I
have proposed on February 28th (in the bug entry, because I didn't know about
phabricator) does fix the problem for main2.cpp, too.
http://llvm-reviews.chandlerc.com/D3197
_______________________________________________
cfe-commits mailing list
[email protected]
http://lists.cs.uiuc.edu/mailman/listinfo/cfe-commits
