make the type of extract richer

[divarvel]

See #387 for the whole context. The main idea is to make extract return a more detailed errors (currently, it returns a Maybe).

Following discussion with @ocharles, and based on @sellout's comments in the issue, I went with a Validation (NonEmpty ExtractError), where ExtractError is either InvalidType or a freeform text for user-supplied errors. Reading the tests, the example provided by @sellout seems at odds with the contact that extract should only fail on type errors.

For now it does not track line number or anything else, but thanks to the applicative instance, it accumulates errors for lists and records. Instead of providing a monad instance through a newtype, I added a validate function which is essentially >>=.

I've added dependencies to either (for Validation) and to semigroupoids (for Data.Functor.Alt, as NonEmptyList a only forms a semigroup).

ToDo

• add a proper error message for the errors that are not type mismatches
• make tests pass (there is an error message mismatch)

[Gabriella439]

Looks great so far! Just a few small suggestions

[divarvel]

As suggested, i've special-cased the error message for a single error (so that it doesn't change current behaviour). I've added a small message when there are multiple errors, and added an error message for extraction errors.

Once we've found a satisfying solution with validate, I can squash my commits.
I'll also need to put proper bounds on the new dependencies, but I'm not sure how. I've used stackage for too long now

[Gabriella439]

It looks like the stack-lts6.yaml's extra-deps section needs newer versions of some dependencies to support the either/semigroupoids dependencies

[divarvel]

I can drop semigroupoids and use ealt instead of <!>.

[divarvel]

I've downgraded deps in stack-lts-6.yaml, but there are still issues with transformers-compat.
I've tried removing semigroupoids as well, but it still fails. I'm not sure what I can try next.

[Gabriella439]

@divarvel: I'll take a look at it tonight to see if I can get it building with stack's LTS6 resolver

[divarvel]

Thanks!

[Gabriella439]

Alright, I have a pull request up against your branch that I believe should fix the LTS 6 build failures: divarvel#1

[divarvel]

Thanks a lot! It looks like it's fixed on LTS-6. I've rebased on latest master, so unless something else is missing, I can squash the commits.

[Gabriella439]

@divarvel: You're welcome! There's no need to rebase or squash ahead of time because the repository settings enforce that we "Squash and merge" anyway when merging pull requests