Conversation
|
@bozbez I have the airfoil validation tests fail due to memory leaks. But I am reasonably sure those leaks were already there. Why are these caught now? (op_realloc seems to be leaking memory, no idea what to do about it...) |
|
ASAN/UBSAN weren't previously used - we might want some make flag to enable/disable these since I'm not sure we want them on by default (might cause linking difficulties with non ASAN/UBSAN code). Looks like op_realloc allocates a new buffer if realloc doesn't maintain alignment but then doesn't clear up the old buffer. |
reguly
force-pushed
the
feature/global_indices
branch
from
4a241fe to
68973a4
Compare
bozbez
force-pushed
the
feature/global_indices
branch
from
cf590f6 to
ea2821b
Compare
bozbez
force-pushed
the
feature/global_indices
branch
from
cab68f1 to
4acf135
Compare
| to; /* set pointed to */ | ||
| int dim, /* dimension of pointer */ | ||
| *map; /* array defining pointer */ | ||
| idx_g_t *map_gbl; /* array with global indices (long type) */ |
There was a problem hiding this comment.
Is it essential to add another pointer?
Cant we cast and get it when needed [the only use case "if (new_map->map_gbl[i] >= to_set_global_size)"]
|
|
||
| op_map op_decl_map(op_set, op_set, int, int *, char const *); | ||
| op_map op_decl_map(op_set, op_set, int, idx_l_t *, char const *); | ||
| op_map op_decl_map_long(op_set, op_set, int, idx_g_t *, char const *); |
There was a problem hiding this comment.
Is it required to have two API calls, when we can overload the function?
For example,
There was a problem hiding this comment.
We can't overload in pure C without that _Generic, and in this case I think just having it simpler (and not requiring us to do anything fancy in Fortran either) is probably for the best
4 participants
We can start reviewing/testing this - the generic kway code path seems to have started working. @bozbez please take a look.