Maxime Villard <[email protected]> wrote: > Hi, > there's something terribly sad with the kmem API: kmem_free takes > a size argument. It has I think two major drawbacks: > > <...> > > Here is a patch which implements kmem_valloc. > > Comments?
I object to such addition; others already explained various aspects of this (it is not the first time we debate kmem(9) either), but one thing which might be worth to add: the underlying problem is also a lack of testing or at least double checking the error paths; they have a known attraction to bugs. You can have a missing mutex_exit() there and it is not something you can solve with a different API. However, we have sufficient means to detects these kind of bugs (DIAGNOSTIC, LOCKDEBUG). Another point: while defensive programming has certain merits, it also has very important drawback - it breaks the assumptions and decreases code clarity. Supporting malloc(NULL) or free(NULL) is an example of that since it introduces the ambiguity about the life cycle or validity of your pointer/object/whatever. It may also hide such bugs as double free if the pointer gets set to NULL after the first destruction (or it may free the wrong object and/or lead to memory corruptions). -- Mindaugas
