[Haskell-cafe] Re: Crypto-API is stabilizing

[Heinrich Apfelmus]

Thomas DuBuisson wrote:
> Sorry, the example was all messed up, even if it did communicate what
> I wanted its just so broken I must fix.
> 
> Slightly contrived example:
> 
>    buildAgreementMessage :: (Monad m, CryptoRandomGen g,
> ASymetricCipher k) => g -> k -> m (B.ByteString, (k,k), g)
>    buildAgreementMessages g k = do
>        ((p,q),g') <- eitherToFail (buildKeyPair g)
>        let pBS = encode p
>            msg = runPut $ do
>                        putByteString agreementHeader
>                        putWord16be (B.length pBS)
>                        putByteString pBS
>        return $ (sign msg k, (p,q), g')
> 
> Again, this is simply trying to re-enforce the fact that buildKeyPair
> (formerly 'generateKeyPair') does have a place.

Granted.

However, the key feature of your example is that a new key is derived 
from an old key, i.e. the function used is

     type BuildKeyPair g k = CryptoRandomGen g => g -> ((k,k),g)

     buildKeyPair' :: k -> BuildKeyPair g k

Thanks to the additional argument, this can be added to the  Key  record

     data Key = Key { cipher :: BuildKeyPair g k , ... }

In other words, the  Key  can also store a method to generate new keys 
with the same cipher algorithm.


All examples that use  buildKeyPair  and type classes can be 
reformulated in terms of  Key  with this additional field. That's 
because  buildKeyPair  actually expects a type argument; the  cipher 
filed merely shifts that argument to the value level.


Regards,
Heinrich Apfelmus

--
http://apfelmus.nfshost.com