Custom Converter fails when both decimal and doubles are present

[steffenskov]

Source/destination types

internal record Example(double Double, decimal Decimal);

Source/destination JSON

{"Double":-1.7976931348623157E+308,"Decimal":-79228162514264337593543950335.0}

Expected behavior

I expect to be able to deserialize the JSON back into an Example record, even when using this custom JsonConverter:

class CustomConverter : JsonConverter
{
	public override void WriteJson(JsonWriter writer, object? value, JsonSerializer serializer)
	{
		// Not relevant
	}

	public override object? ReadJson(JsonReader reader, Type objectType, object? existingValue, JsonSerializer serializer)
	{
		var token = JToken.ReadFrom(reader);
		var dbl = double.Parse(token["Double"]!.ToString());
		var dec = decimal.Parse(token["Decimal"]!.ToString()); // Throws FormatException
		return new Example(dbl, dec);
	}

	public override bool CanConvert(Type objectType) => true;
}

Actual behavior

A FormatException is thrown when attempting to parse the decimal, because the reader by default treats all floating point numbers the same via the FloatParseHandling property. Changing the FloatParseHandling property to Decimal just moves the problem to parsing the double instead.

Steps to reproduce

var example = new Example(double.MinValue, decimal.MinValue);
var json = JsonConvert.SerializeObject(example);
var deserialized = JsonConvert.DeserializeObject<Example>(json, new CustomConverter());

[steffenskov]

This seems related: #2839

[steffenskov]

I've narrowed it down to line 2204 in JsonTextReader.cs, where it inspects _floatParseHandling. Wouldn't a fallback mechanism here work?
e.g. attempting conversion to decimal, and if it overflows go to double instead of throwing?
I realize it should probably have a new FloatParseHandling enum value to avoid breaking existing implementations. Something like DecimalWithFallbackToDouble (or something more concise)

Just an idea.

[mtrackeros]

Typy źródłowe/docelowe

internal record Example(double Double, decimal Decimal);

Źródłowy/docelowy kod JSON

{"Double":-1.7976931348623157E+308,"Decimal":-79228162514264337593543950335.0}

Oczekiwane zachowanie

Oczekuję, że będzie można deserializować kod JSON z powrotem do przykładowego rekordu, nawet w przypadku korzystania z tego niestandardowego:JsonConverter

class CustomConverter : JsonConverter
{
	public override void WriteJson(JsonWriter writer, object? value, JsonSerializer serializer)
	{
		// Not relevant
	}

	public override object? ReadJson(JsonReader reader, Type objectType, object? existingValue, JsonSerializer serializer)
	{
		var token = JToken.ReadFrom(reader);
		var dbl = double.Parse(token["Double"]!.ToString());
		var dec = decimal.Parse(token["Decimal"]!.ToString()); // Throws FormatException
		return new Example(dbl, dec);
	}

	public override bool CanConvert(Type objectType) => true;
}

Rzeczywiste zachowanie

A jest zgłaszany podczas próby przeanalizowania ułamka dziesiętnego, ponieważ czytelnik domyślnie traktuje wszystkie liczby zmiennoprzecinkowe tak samo za pośrednictwem właściwości. Zmiana właściwości na po prostu przenosi problem do analizowania zamiast tego.FormatException``FloatParseHandling``FloatParseHandling``Decimal``double

Steps to reproduce

var example = new Example(double.MinValue, decimal.MinValue);
var json = JsonConvert.SerializeObject(example);
var deserialized = JsonConvert.DeserializeObject<Example>(json, new CustomConverter());