[PATCH] [analyzer][Review request] Better modelling of memcpy by the CStringChecker (PR16731)

[Anna Zaks]

  You have a bunch of new false negatives added in invalidateDestinationContents. My understanding is that these regressions are due to us wanting to suppress the warning that we issue in this case:

  int *p = (int*)malloc(4);
  memcpy(p, src, sizeof(int)*4);

  Currently, we issue a leak warning on 'p'. But with your patch we don't? (Please, add the test case that shows the improved behavior.)

  Looks like for dest, we want all the subregions to escape but the top region not to escape. Can't this be expressed in your infrastructure?

http://llvm-reviews.chandlerc.com/D1887