Register format decimal128

[ralfhandl]

Format for 128-bit decimal floating-point numbers as proposed in proposal in #3190 (comment).

Fixes #603 with proposal for representing NaN, -INF, and INF.

[ralfhandl]

@baywet @handrews please review

[baywet]

should we only have string here? or at least warn about a loss of accuracy in case of JSON serialization?

[ralfhandl]

No, it's a well-known problem for some serializers and parsers, and already called out in https://www.rfc-editor.org/rfc/rfc8259#section-6

[handrews]

@ralfhandl is this string, number as in you can use either, or do you mean string number as in "a string containing a number"? If the latter, I would just say string. If the former, I suggest the comma for clarity 🧐

[baywet]

this is a good callout for clarity. If we change this, we should also change it in the infamous decimal format

[darrelmiller]

@ralfhandl Are you referring to this line:

implementations will approximate JSON numbers within the expected precision

I don't think that RFC 8259 wording is particularly clear about potential loss of precision when storing decimals as numbers.

[handrews]

@darrelmiller @ralfhandl we would probably want to reference https://datatracker.ietf.org/doc/html/draft-bhutton-json-schema-00#section-4.2.1 which includes:

Whitespace and formatting concerns, including different lexical
representations of numbers that are equal within the data model, are
thus outside the scope of JSON Schema. JSON Schema vocabularies
(Section 8.1) that wish to work with such differences in lexical
representations SHOULD define keywords to precisely interpret
formatted strings within the data model rather than relying on having
the original JSON representation Unicode characters available.

[ralfhandl]

@handrews It's "type": ["string", "number"], we use content negotiation to allow clients select whether the server will send JSON numbers or wrap the numbers as JSON strings.

[baywet]

Can you update the type for decimal as well please for consistency?

[ralfhandl]

@baywet Done in separate PR #3210

[darrelmiller]

1. No objection to adding support for both string and number
2. We should add a warning about the risks of using number and point to related spec
3. For decimal serialized as strings, call out how things such as inf and Nan are supported.

[ralfhandl]

Strange: I can only re-request a review from a single reviewer, trying to re-request for a second reviewer removes the re-request for the first reviewer 🤔