Allow 'raw' PKCS1 signatures

[joostrijneveld]

This provides more convenient compatability with other libraries that use such raw signatures in intermediate results. In particular, it matches the behavior of openssl rsautl -sign.

[coveralls]

Coverage increased (+0.02%) to 91.867% when pulling d8bc7ac on joostrijneveld:raw-rsa-signatures into 8affa13 on sybrenstuvel:master.

[coveralls]

Coverage increased (+0.02%) to 92.431% when pulling 69ccc2b on joostrijneveld:raw-rsa-signatures into 83e273b on sybrenstuvel:master.

[sybrenstuvel]

Please also update the documentation so that our users actually know about this feature and how to use it.

[joostrijneveld]

Good call. I've added a line to highlight this in the usage.rst documentation.

[joostrijneveld]

Not sure if you get notifications of new commits, so for completeness; I've added a test and a changelog entry :)

[0x2b3bfa0]

@sybrenstuvel: could you merge this PR?

[vstoykov]

Probably it should be rebased first because there are conflicts

[joostrijneveld]

@vstoykov Since the conflict is in a recently released CHANGELOG.txt update, that would mean starting a new release candidate (following up on 4.0). I'm not sure if I'm the right person to make that decision.

[sybrenstuvel]

There is one thing that bugs me, and that's the change in behaviour of _find_method_hash(). Now this no longer raises a VerificationError but just returns "RAW". This means that bad signatures can all of a sudden be accepted as correct. This is a deal breaker, as it makes this code insecure by default.

The documented behaviour of raising an exception in case of an error has been removed in this PR, without changing the documentation and adding huge big red letters that the API has changed. In any case I would reject such an API change, as such errors should never be silenced.

[sybrenstuvel]

Rejecting this PR as it silently ignores signature errors (making the code insecure-by-default), changes the semantics of functions without updating the docstring of those functions.

[joostrijneveld]

I agree, and would suggest to close this PR on that account; I had overlooked that marking signatures without an ASN1 tag as 'RAW' rather than raising an error was changing API behavior, and I see no obvious way to prevent doing so.