Users cannot omit fields of a record.

[vmchale]

I quite like Dhall from what I've seen so far. However, one of the pain points I noticed (compared to TOML) is that the user cannot simply omit a field of a record. If library authors were able to make values of type Maybe Text and have these omitted when writing the actual configuration, it would be much easier for users down the line.

Thanks for all the work on this language! I'm using it in a project and am excited to see where it goes :)

[ocharles]

As Optional is actually part of the core language this feels like it might be possible. Don't hold me to that, though 😄

[ocharles]

I should note that this is possible, in a sense, using the current version of Dhall. What changes here is that you can no longer provide a canonical type of the expression that your users give you. The process is essentially:

• Read Text
• Parse a Dhall 'Expr`
• Resolve imports
• Type check the result - only to make sure it is well formed
• Turn the Expr into whatever type you need by walking through it.

It's in the last step that you can start to do things like:

case expr of
  RecordLit fields -> do
    myFoo <-
      case Map.lookup "foo" fields of
        Just fooExpr -> Just <$>_ fooExpr
        Nothing -> return Nothing

Does that make sense? The difference here is that you aren't using Dhall.input, as Dhall.input tries to annotate the given input Expr with a specific type (it literally appends : T, for some T to drive the type-checking process).

[Gabriella439]

Yeah, the main issue here is that you would no longer be able to infer a canonical type for a expression

I recommend instead providing a record where all the Optional fields are empty that you can then override that using //, like this:

    let defaults = { bar = [] : Optional Bool, foo = [] : Optional Text }

in  defaults ⫽ { bar = [ True ] : Optional Bool, baz = 1 }

[ocharles]

Here is my above mentioned approach explored in the context of dhall-to-cabal: dhall-lang/dhall-to-cabal@e154561.

I must admit, I find it quite compelling, even at the cost of losing a canonical type.

[Gabriella439]

I think dhall-to-cabal is one case where you really want a canonical type. As I mentioned in #78, I think one of the most important outcomes of dhall-to-cabal is not only the executable, but a complete Dhall schema for a cabal configuration file.

[ocharles]

I've opened dhall-lang/dhall-to-cabal#12 to continue that discussion, rather than hijack @vmchale's original question.

[vmchale]

Great, thanks! This should work well enough for now :)