Bill Sommerfeld wrote: > > is there some reason why you didn't define a xfrtn_t (since the function > signature of the free callback is different), and why you didn't use a > function signature of: > > boolean_t (*free_func)(caddr_t) > > (returning B_TRUE or B_FALSE as appropriate) > > the former seems appropriate from a cleanliness perspective (give the > compiler a better chance to catch type mismatches); the latter seems > simpler and my vague recollection was that it was slightly more > efficient as well.
The problem is that the db_frtn field of the dblk_t is defined as a frtn_t *. If I created a new structure type then I'd have to cast it anyway to assign it to that field, or change the typedef of dblk_t to union the two pointer types together and then #define something do that db_frtn still pointed at the frtn_t element. I could do this, but I was uneasy about changing the typedef of dblk_t. The fact that the definition frtn_t.free_func does not specify the arguments to the function seemed like a gift allowing me to keep the code changes to the minimum. The use of an incorrectly defined free_func that does not specify the second argument can also be made safe, as I stated, by the fact that if the boolean remains unmodified then the semantics are identical to desballoc() whereas a function that returns a void ending up being cast to one that returns a boolean would be bad. Paul
