NoQ added a comment.
That'd be a hell for you because when the container is updated you won't be
able to easily find iterators all that iterate over it. Normally what you want
to do is keep mapping iterators to container regions, and when the region dies,
"freeze" the data (make sure it can no longer be mutated, through an assertion
or, ideally, via the type system) and //re-map// the iterator to data directly.
----
I guess you need to track dead containers somehow anyway because you want to
find bugs that consist in using iterators from different containers:
void foo(vector<int> x, vector<int> y) {
vector::iterator i = x.begin(), e = y.end(); // Both containers are
garbage-collected after this statement.
vector::iterator k = find(i, j, 123); // Bug: 'i' and 'j' aren't
from the same container.
}
For that purpose i'm fine with tracking dead regions (as long as they aren't
marked live, just stored). Or, for enforcing correctness, you could "anonymize"
them (i.e., in the container data add an immutable int field that says "this is
container #5" and use these ids for your check, then recover the original
container region from the identifier in the bug visitor); but i don't think
it's necessary.
https://reviews.llvm.org/D48427
_______________________________________________
cfe-commits mailing list
cfe-commits@lists.llvm.org
http://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
