Fix incorrect JsonProvider numeric inference

[panesofglass]

See #1221

Hypothesis is that commit 8d05e4b is not quite correct.

[panesofglass]

Apologies for the formatting changes. I didn't do anything to change the formatting. Should I try to revert that?

[panesofglass]

I think this is supposed to infer decimal, but I could be wrong.

[panesofglass]

Appears these new lines are the issue: https://github.com/fsharp/FSharp.Data/blob/8d05e4bf32fe750cb0f1bad977cbd5b97b39ab17/src/Json/JsonInference.fs#L34-L35

What's wanted here? I would expect the format of the number to override the use of Math.Round(v) = v as a means of inference. If not, what is required to make this work correctly?

[panesofglass]

@ovatsus @baronfel I don't know the right way to fix this. Help is appreciated.

[baronfel]

Personally I think the single flag for inferTypesFromValues is a bit wide for coercing both strings and numeric types. I'd be in favor of expanding the number of static/configuration parameters for this (maybe just a split for string-inference and number-inference) as well as expanding the number of JsonValue cases to include Integer and LongInteger or something semantically similar, since the only time where we have enough information is at JsonValue.Parse time. Armed with JsonValue.Integer and JsonValue.LongInteger we could easily handle type mapping to int/int64 for json provider even without inferTypesFromValues set.

[dsyme]

Close/open to re-run CI

[dsyme]

Closing this out as the repo has diverged a long way since the proposal was made

[panesofglass]

We ultimately never upgraded and are stuck on FSharp.Data 3.3.3. Has this been fixed in a different PR by any chance?