Add support for multiple integers value for attribute and NonNull

[GuillaumeGomez]

For now this implementation doesn't work. Need to figure out what I did wrong.

[GuillaumeGomez]

gcc:

r:[irange] _Bool [0, 0] NONZERO 0x0
true_range:[irange] _Bool [1, 1]
false_range:[irange] _Bool [0, 0] NONZERO 0x0

libgccjit:

r:[irange] bool VARYING
true_range:[irange] bool [1, 1]
false_range:[irange] bool [0, 0] NONZERO 0x0

With this C code:

__attribute__((nonnull(1)))
int t(int *a) {
  if (!a) {
    return -1;
  }
  else {
      return *a;
  }
}

We need to understand why range isn't updated with gccjit:

gcc:

r:[irange] _Bool [0, 0] NONZERO 0x0

vs libgccjit:

r:[irange] bool VARYING

[GuillaumeGomez]

All fixed. Now remains to move the check earlier (in add_attribute methods directly) and to handle no_add_attrs.

[antoyo]

I'd like a more descriptive name than pp.

[antoyo]

Seems like the tabs got lost here.

[GuillaumeGomez]

Questions to ask David when we'll send the attributes patch upstream:

• Should we check attributes in the API?
• Is it valid to use warning/error in handle_attribute_* functions?
  • If yes, should we add warning/error in already existing handle_attribute_* functions?