This is a follow-up to #119:

A lot of tests in sv-comp wrap calls to malloc into calls to the helper function ldv_malloc. We only have one abstract heap object for all memory allocated via ldv_malloc (as the malloc for these calls is always in the same place), which might be a big source of imprecision.

The easiest solution would be to treat ldv_malloc specially just like we do malloc, but according to @sim642 there are different versions of ldv_malloc with some of them also simulating failing malloc calls while normally we can assume them to succeed in sv-comp.

Another approach would be to have some limited form of context-sensibility specifically for memory allocated inside functions such as ldv_malloc.