Messages in this thread | | | Date | Sun, 8 Aug 2010 12:31:40 -0700 | Subject | Re: [PATCH 0/6] improve list_sort test | From | Don Mullis <> |
| |
Artem Bityutskiy <dedekind1@gmail.com> writes: > Actually, your 'list_sort()' version does have a problem. I found out > that it calls 'cmp(priv, a, b)' with 'a = b' sometimes, and in these > cases 'a' and 'b' can point to something which is not a valid element of > the original list. Probably a senitel or something like that. > > It is easy to work around this by adding: > > if (a == b) > return 0; > > in the 'cmp()' function, but this is nevertheless a bug (not too bad, > though) and should be fixed.
Yes, invalid 'a' or 'b' pointers would be a bug. If providing a test case is hard, can you say what segment is pointed to? Into the stack? Into address ranges normal for elements, but not now on the list? Is there a pattern to the values returned? Is it perhaps always the first or last callback from a particular call to list_sort()?
That sometimes a==b is, on the other hand, by design:
/* * In worst cases this loop may run many iterations. * Continue callbacks to the client even though no * element comparison is needed, so the client's cmp() * routine can invoke cond_resched() periodically. */ (*cmp)(priv, tail, tail);
Adding a sentence to the function header comment reminding callers that they need to be able to handle a==b seems like a good idea.
| |