Type-directed optional fields

[friedbrice]

This PR makes a backwards-compatible change to ToJSON and FromJSON that allows the instance to configure the optionality of record fields of that type when using Generic deriving or Template Haskell deriving.

Specifically, we make the following use-facing changes:

class ToJSON a where
    ...
    omitField :: a -> Bool
    omitField _ = False

and

class FromJSON a where
    ...
    omittedField :: Maybe a
    omittedField = Nothing

The new methods have default implementations that preserve the current behavior of Generically-derived and Template Haskell created instances.
Instance authors can use these methods to configure how to represent fields of this type as omitted from a JSON record.
As an added benefit, this PR eliminates the needs for special-case handling of Maybe and Option, significantly simplifying the Generic and Template Haskell code.

This PR solves #646 and #792, and it's similar to #839 by @phadej.
Please let me know what you think, if this PR is a suitable addition to Aeson, and if there's anything about it you'd like me to fix or change.

Thanks much!

[friedbrice]

These tests verify that the PR supports your desired behavior, @bergmark :)
(c.f. #839 (comment))

[phadej]

Looks good. Having omitField :: a -> Bool is indeed simpler that what I had in #839.

There probably should be something in ToJSON1 and FromJSON1 classes, as e.g. instance for Fix relies on them. How Fix Maybe should be serialised?

[friedbrice]

I was able to get the generic classes down to one method like you suggested. I still need to work on the ToJSON1 logic.

[phadej]

I marked this as draft.

Please ping when you seem happy again.

[friedbrice]

@phadej I'm pretty happy with it, now. I'm interested in your feedback :-)

[friedbrice]

Oh, also, @phadej, I do have some ideas about changes to To/FromJSON1, but I'd like to know more about what you have in mind before attempting that.

[phadej]

This looks good.

I have further things to do for this, but I think it's easier and fasterfor me to them myself rather than try to explain them.

Could you squash your changes into own commit (and rebase on recent master), so I can continue from there.

[parsonsmatt]

Some doc updates necessary, but this is super cool. ~268 lines of tests for new functionality means that this is really a +145 / -177 LOC change that adds a feature and removes some warts! Well done.

[parsonsmatt]

😄 Briliant

[friedbrice]

I was like, "whelp, we don't need this anymore"

[friedbrice]

Thanks for your help and ideas, btw, @parsonsmatt. The TH was trivially-easy, but the Generics threw me for a loop.

[friedbrice]

Could you squash your changes into own commit (and rebase on recent master), so I can continue from there.

Done. Thanks, @phadej !

[phadej]

@parsonsmatt

Some doc updates necessary,

If you think so please mark or even suggest amendments. I'm mostly happy with docs in this PR.

[friedbrice]

@phadej there were two places my docs had minor mistakes, but those were taken care of :-)

[phadej]

#1039 is merged. Thanks for making the initial version!