The base64 encoding should have a checksum, period. The checksum should
be computed on the underlying binary stream from which the base64
representation is derived.
The whole point of base64 is that it doesn't matter what binary
representation you use for the transported characters such as 'A': the
XXXX spec tells you exactly which six bits you get for each character.
The transported characters could be in ASCII or EBCDIC or anything else.
This property does not extend to the other encodings.
It's a separate problem (and a separable issue) as to whether any other
content encodings should provide checksums. I'd argue mildly against
them, since I believe that they're not useful as end-to-end checks, due
to charset changes and MTA munging (even as simple as adding extra CRLFs
to message ends). I don't care about them that much. But they're
SEPARATE from the proposed checksum on the binary stream underlying the
base64 representation.
I object to mixing up the two concepts. In particular, I object to
holding the obvious and clear benefits of base64-checksumming as a
hostage to the (to me, ill-defined) benefits of checksumming all
encodings.
I would press for a base64 checksum as part of the encoding itself, much
as Henry Spencer's revisited-uuencoding has done it, immediately. I
would attempt to let the other checksummers slug it out independently.
Craig