(My second-last paragraph was redundant, saying the same as the bit before it.)
- Julian
I (Julian Foad) wrote:
> * This test looks includes a case where text and a property are set
> to two versions of 'evil' text thay yield the same MD5 sum. I
> initially thought this case would ensure test coverage for the unusual
> (MD5 collision) case for all functions under test, but it does not.
> The 'evil' text forms only a substring of the property-list rep, and
> the MD5 sums of the full reps no longer collide.
>
> Also it's confusing because the test sub-case 'evil text' (i=3)
> doesn't actually end up executing the code path for finding identical
> MD5s of the reps, because the special text is just the value of one
> prop and not the complete rep of the properties list.
>
> I propose the attached patch, for a start. It doesn't do anything to
> resolve the change (regression) of the exact behaviour of non-strict
> mode comparisons, which is the subject of another thread. It only
> addresses the issues listed in this email.
>
> Thoughts?
>
> - Julian
Received on 2015-09-25 20:06:44 CEST