Add support for decoding missing record fields to Nothing

[jvliwanag]

Decodes missing fields in a record as Nothing.

Adds a helper type class DecodeJsonField

Fixes #92

[thomashoneyman]

This is an interesting suggestion. From my reading, we're essentially moving from using getField for each field in a record:

purescript-argonaut-codecs/src/Data/Argonaut/Decode/Decoders.purs

Lines 203 to 213 in 9ce8e17

	getField
	:: forall a
	. (Json -> Either JsonDecodeError a)
	-> FO.Object Json
	-> String
	-> Either JsonDecodeError a
	getField decoder obj str =
	maybe
	(Left $ AtKey str MissingValue)
	(lmap (AtKey str) <<< decoder)
	(FO.lookup str obj)

to attempting to use getFieldOptional for each field, falling back to getField:

purescript-argonaut-codecs/src/Data/Argonaut/Decode/Decoders.purs

Lines 215 to 224 in 9ce8e17

	getFieldOptional
	:: forall a
	. (Json -> Either JsonDecodeError a)
	-> FO.Object Json
	-> String
	-> Either JsonDecodeError (Maybe a)
	getFieldOptional decoder obj str =
	maybe (pure Nothing) (map Just <<< decode) (FO.lookup str obj)
	where
	decode = lmap (AtKey str) <<< decoder

I'm intrigued, but this would be a significant change so I'd like to leave this open for a time to allow some discussion. Thank you for taking the time to write this up!

[jvliwanag]

Anything we can do to see this go through? Would love to use argonaut records directly for simple web api’s.

[thomashoneyman]

I'll leave this open for a little while to let other maintainers weigh in, but if I haven't merged it in a week please ping me and I will @jvliwanag -- thanks!

[garyb]

I'm fine with it - I probably wouldn't choose to have it work this way personally if I was designing a library from scratch, as I don't think you can distinguish between the field missing and having no value present now, right? But Maybe collapsing happens with the encoding used elsewhere so it's already established that this library doesn't care about that kind of distinction.

[jvliwanag]

Are we good with this @thomashoneyman ? :)

On a side note, I do prefer Nothing be encoded as missing fields. I'll open a separate issue on that.