Better verifier results

[clementd-fretlink]

I gave a shot at improving the verifiers API, as per #11
I'll try to split up changes in separated commits. There is one commit for imports cleanup, don't pay it too much attention.

ToDo:

• split up ValidationError
• drop the monadIO constraint
• investigate the use of a monad transformer for effects and validation -> Later
• investigate letting verifiers access all caveats at the same time (through eg. a zipper) -> Later

First change: split up ValidationError into:

• VerifierResult (for… well… verifier results).
• VerifierError for caveats understood, but not discharged
• ValidationError for macaroon validation errors (either an issue with the macaroon itself, or with caveats discharge. Caveat discharge errors are accumulated in a NonEmpty (Caveat, [VerifierError]) case, which allows to know which caveats failed and why.

This allows to drop the Monoid ValidationError previously necessary for error accumulation. It was lossy, and some combinations didn't make sense (to the point that the first <> implementation was partial). The caveats validation is now a bit cleaner (even though keeping the caveats around makes it more complex than it could be).

Second change: relax the unused MonadIO constraint. This allows to have synchronous verification thanks to Identity (I've added a helper to make it even easier).

Third change: introduce a monad transformer to encapsulate both the effects needed for verifier computation and for the verifier results (maybe through a MonadError constraint). I'm not convinced it's really necessary, as it would increase complexity a lot, while providing little benefit IMO (macaroon validation errors should be treated right away, not carried on in the context).

[clementd-fretlink]

There are two things I'm not 100% sure of with the new VerifierResult ADT.

• should the error value be Text or can we make it parametric?
• should OutstandingCaveat carry the whole caveat or just its id? Currently it carries the whole caveat, and it requires a small cleanup function in tests. In application code, I think it's less an issue though.

[jtanguy]

As an aside, and related to #1, I wonder if the proposed structure can accommodate attaching discharge macaroons to the result, and recursively verify all the macaroons.

I think it is feasible without the need for a complete refactor.

[jtanguy]

I feel that OutstandingCaveat should bear the hole caveat.

For first-party caveats this is equivalent to storing just the id, but for third-party it could be useful to group the non-discharged caveats by origin

[clementd-fretlink]

We can leave 3rd party caveats for another PR I think

[clementd-fretlink]

I've renamed OutstandingCaveats? (squashed in the first commit), and made VerifierError parametric in a separate commit. Doing so make the types a bit more messy (it shows in the tests, where I've used type aliases and TypeApplications).

I've tried to keep the new extensions to a minimum, this PR currently adds:

• StandaloneDeriving to keep deriving Eq and Show on parametric result types
• ScopedTypeSignatures to add intermediate annotations on the caveats check function
• TypeApplications (only in the tests) to avoid splitting a small one liner.

[clementd-fretlink]

I've thought a bit more about a monad tranformer and zippers, I don't have a clear view of what's going to be possible yet, so I'll keep to the current scope for this PR.

I have a few ideas for both transformers and zippers, I can discuss them in relevant issues.

[clementd-fretlink]

I've opened issue #13 to sum up my ideas about zippers.

[clementd-fretlink]

@jtanguy so, what's left to do?

[jtanguy]

There is still an unchecked TODO in the PR description.
The zipper idea has been extracted to issue #13

What about the last item ? Is it tracked somewhere ? If so, we can merge this and then #14