Allow accepting missing fields of other types than Maybe

[lortabac]

When deriving FromJSON instances generically, there is no way to provide a default value to a missing field of a record if the field is not a Maybe.

This seems related to #614, in that the proposed solution may help fixing both issues.

[tolysz]

I would say it is related to: #323
To make aeson full flexible we need 2 extra contructors: one for missing, one for rawNumber.

[lortabac]

@tolysz Thanks. It seems that your proposal could fix this issue. However I am not sure it is a good idea to modify Value.

Since the problem is specific to records, what I would like to have is some way to specify whether a record field is allowed to have a default type or not (and what the default value should be) according to its type.

[tolysz]

I can only say it is a good idea ;) It is transparent to virtually all applications.

I am porting my old patches to the new version of aeson and it will just work (The generic is almost done(example work), the TH is a bit pain, but it will be done soon).
Adding the extra constructors does not change the output, just allow to work with JSON more easily.

[lortabac]

@tolysz It works, but I am not sure it is the most appropriate solution. Since this issue only concerns records, I would prefer a solution that only concerns records.

If you add new constructors to Value you end up with values that do not fit into any JSON type. For ex. how would you encode Missing in isolation? Or in a list? Would it make sense?

[Lysxia]

genericToJSON/genericParseJSON could also take a list of field encoders/decoders to customize the behavior on various fields (referred to by name or by type). I have an example of this approach to generic deriving in generic-random (customizing random generators instead of JSON parsers). It is technically a breaking change (aeson's generic type classes would need to carry an additional parameter) but seems feasible without breaking too much in practice.

[tolysz]

@lortabac I prefer a situation where I can encode all possible scenarios, rather than a limited subset, even if some would be invalid (a stricter validation would remove those invalid cases).
I would encode Missing as mempty or null or error "you should never encode it" (but I love robustness and hate errors, so I chose the first or the second. The 2nd produces a valid JSON ;) but you should never see it).

My proposed approach composes well and it is really useful, but I do get that people are afraid of changing the holly library. Good thing is that I do not see any other useful constructors ;) but the new one is essential if we go via the Value.
Direct encoding solvers problems with how we encode floating point numbers (as you can choose anything you want for your custom type ~ including an invalid string), but I like to fix the Value as well.
Missing is essential, signal the level up to remove the object key ~ there is no (negative) encoding.

Anyway, this will be coming back, let me update the old patches. You can look at them. Each time I improve something there.

[torgeirsh]

What's the latest news on this? Is it possible to derive FromJSON and have the instance allow Maybe fields to be missing, or do such instances have to be written manually?

[lortabac]

@torgeirsh You can already do that with Generic instances. If a Maybe field is missing, it defaults to Nothing.

In this issue I propose to extend this behavior to other types than Maybe, which is currently not possible.

[torgeirsh]

@lortabac I can't get it working:

data MyType a = MyType {
        field1 :: Int,
        field2 :: a
} deriving (Show, Generic, FromJSON)

*Main> parseEither (parseJSON :: Value -> Parser (MyType (Maybe Int))) (Object $ fromList ["field1" .= 1])
Left "Error in $: parsing Main.MyType(MyType) failed, key \"field2\" not present"

Do you mean something else by "Generic instances"?

[Lysxia]

@torgeirsh That's a somewhat different issue, see also #690 and the extra documentation added here. Don't hesitate to ask for further clarification though!

[torgeirsh]

@Lysxia Thanks, I can see that I have the FromJSON version of that issue.

[georgefst]

Any update here? I actually have a datatype where certain fields have a type isomorphic to (). I would like them to be populated with the same value, regardless of whether the field is there or not. (NB. in the presence of type families, this isn't as silly as it sounds.)

I've written a manual FromJSON instance, but it involves a lot of rather cumbersome boilerplate.

[bergmark]

Sorry, no work has been done on this as far as I know.

Seems like @Lysxia's suggestion is a promising approach:

genericToJSON/genericParseJSON could also take a list of field encoders/decoders to customize the behavior on various fields (referred to by name or by type). I have an example of this approach to generic deriving in generic-random (customizing random generators instead of JSON parsers). It is technically a breaking change (aeson's generic type classes would need to carry an additional parameter) but seems feasible without breaking too much in practice.

[anka-213]

The workaround I've used is to transform the value with this function prior to parsing, so that it can work with any parser that can accept null (instead of just Maybe).

import qualified Data.HashMap.Strict as HM

addNullField :: Text -> Value -> Value
addNullField s (Object o) = Object $ HM.insertWith (\_new old -> old) s Null o
addNullField _ v = v

I would use it like this:

data Foo = Foo 
  { someField :: SomePossiblyNullType
  }  deriving (Show, Generic)

instance FromJSON Foo where
  parseJSON = genericParseJSON defaultOptions . addNullField "someField"

This also works for polymorphic types:

data Foo a = Foo 
  { someField :: a
  }  deriving (Show, Generic)

instance FromJSON a => FromJSON (Foo a) where
  parseJSON = genericParseJSON defaultOptions . addNullField "someField"

-- >>> decode "{}" :: Maybe (Foo (Maybe Int))
-- Just (Foo {someField = Nothing})

-- >>> decode "{\"someField\": 3}" :: Maybe (Foo (Maybe Int))
-- Just (Foo {someField = Just 3})