Seems we need to discuss on this topic. In this PR is attached files
for replacing GNU sdiff(1) with OpenBSD's sdiff(1). At first, as
already mentioned in PR, let state few things:
1. Although undocumented, sdiff supports all options supported by GNU
sdiff. Some options require GNU diff.
2. Offering sdiff(1) has regression tests and passed them all. While
our GNU sdiff(1) does not pass more than half of them (from 35 tests).
3. Offering sdiff(1) was written from scratch for the public domain.
Would be great to appraise these and others factors and take a
decision about replacing GNU diff(1) or not.

P.S. Sorry for base64 attachements, pressed "Send" too fast :)

-- 
Best regards,
Mindaugas