[openssl-project] to fully overlap or not to

[Andy Polyakov]

> Once again:
> - This PR *recognizes* the fact that special cases do in fact *exist*.

??? Is there application that heavily relies on some specific overlap
percentage?

> - The EVP layer has to be more aware of special cases anyhow.

Where does it come from? Why does it *have to* be more aware of special
cases? There is one special case that we support, fully overlapping
buffer, and that's it, it doesn't *have to* be more than that. Besides,
EVP is *generalized* layer and is not supposed to make overly specific
assumptions. While in order to support suggested cases and the suggested
way, you'll be forced to make an assumption about underlying cipher.
Specifically that it processes blocks in ascending memory order, i.e.
from lower to higher addresses. Once again, all our implementations do
that, but *generalized* EVP layer is not the right spot to assume that
they actually do.

> - The test coverage of error cases is not perfect, but I try to
>    improve that.
> - I don't see why we can't relax a check in cases that are known to
>    work, especially when test cases are added at the same time.

Because we don't have to. Because goal is not to provide maximum
flexibility, but maximum *robustness*. Well, it's not entirely true,
because robustness is not the only goal, but it's not inappropriate to
say that complexity has to be justified by something beside sheer why-not.