Maxim Pugachev wrote:
> On Sun, Dec 13, 2015 at 10:38 PM, Ted Unangst <t...@tedunangst.com> wrote:
> > Maxim Pugachev wrote:
> >> Currently two checks in free() function confirm the correctness of
> >> freedsize argument. I think that it's better to check that provided
> >> freedsize fall into the same bucket that was recorded in kmemusage
> >> struct: it covers both cases.
> >
> > I don't know. This doesn't seem clearer to me. An error message with
> > bucket numbers isn't any more helpful.
>
> Ted, thank you for your comments!
>
> I can compare the buckets to generate a clearer message, but you're
> right, it'll look a bit complicated.
>
> I've read this code again and I think there is an error in a second
> check ("size too small"). For example, if freedsize = 2048 and size =
> 4096, then panic will not fire because "freedsize < size / 2" is
> false. But I guess it should, because BUCKETINDX(2048) == 11, but
> BUCKETINDX(4096) == 12.
>
> What do you think?
I remember thinking about this before. I forgot what I thought though.
Is there some case where the correct size is 1/2? I can't think of any.
