feature/issue-175-auto

[njlr]

• Introduce Auto module
• Re-enable many tests

This PR introduces an Auto module that is agnostic to the JSON backend used. It does this by composing Encoders / Decoders from Thoth.Json.Core using reflection. There is some Fable specific stuff to pave over the limitation of reflection for some targets.

The majority of the old tests are re-enabled and unchanged.

LMK what you think!

[njlr]

Fixed merge conflicts

[MangelMaxime]

Thank you for the update, I just got back from vacation will have a look to make a new Thoth.Json release this week and take this chance to look at the Auto API support once again.

[MangelMaxime]

Hello @njlr,

I am finally starting to have a look at this PR seriously.

Looking at the code it seems like there are a lot of changes compared to the originals versions, do you remember what motivated these changes?

For example, you seems to use this pattern a lot:

#if FABLE_COMPILER
                let fail (innerType: Type) (x: string) : obj =
                    Decode.fail x |> box
#else
                let private failGenericMethodDefinition =
                    getGenericMethodDefinition "Fail"

                let fail (innerType: Type) (x: string) : obj =
                    failGenericMethodDefinition
                        .MakeGenericMethod([| innerType |])
                        .Invoke(null, [| x |])
#endif

instead of probably the DecoderCrate

    type private DecoderCrate<'T>(dec: Decoder<'T>) =
        inherit BoxedDecoder()
        override __.Decode(path, token) =
            match dec path token with
            | Ok v -> Ok(box v)
            | Error er -> Error er
        member __.UnboxedDecoder = dec

    let boxDecoder (d: Decoder<'T>): BoxedDecoder =
        DecoderCrate(d) :> BoxedDecoder

    let unboxDecoder<'T> (d: BoxedDecoder): Decoder<'T> =
        (d :?> DecoderCrate<'T>).UnboxedDecoder

Is it because it improves something over the crate?

[njlr]

Looking at the code it seems like there are a lot of changes compared to the originals versions, do you remember what motivated these changes?

Sorry, it has been a while!

IIRC, the thinking is to move all reflection and dynamic casts to the encoder/decoder generation step and out of the encode/decode stage. In theory this offers better performance for the most common use-case: generate an encoder/decoder once and then call it many times.

The Auto module uses reflection to invoke the core API in an automatic way, but does not contain any special encoder/decoder types.

[MangelMaxime]

Ok thank you for the answer.

At least now, Thoth.Json has benchmark setup so it should be possible to make a quick comparaison between both approach an see which one works the best at least for .NET.