[PATCH] Change the way labels are propagated when comparing memory through libc functions

[Lorenzo Martignoni]

  I agree that we don't know which approach is the best; however, my feeling is that introducing extra dependencies is not a good idea, especially if the return value is a pointer or could be used as an index (e.g., what we do for strchr and strlen). The thing that would make most sense to me would be to propagate the labels in the same way we would do if we compiled and instrumented the functions.

  Maybe we could just take the following shortcut: move these functions to a different file (without the calls to dfsan_*) and then compile the new file with the instrumentation turned on. What do you think?

http://llvm-reviews.chandlerc.com/D2252