Add support for YAML 1.2 non-finite (Infinite, -Infinite, NaN) numbers, Octal notation

[mjcarman]

The YAML 1.2 specification brings YAML into compliance with JSON as an official subset (via the JSON schema). It also fixes many of the ambiguities in YAML 1.1; e.g. the "Norway problem" (via the core schema).

I'd like to be able to use YAML 1.2 with Jackson. More specifically, I'd like to be able to validate documents using the YAML 1.2 core schema via networknt json-schema-validator, which depends on Jackson for its YAML support, which in turn depends on snakeyaml, which only supports YAML 1.1. Its sister (replacement?) project snakeyaml-engine is a YAML 1.2 parser. It's not clear to me if snakeyaml-engine also fully supports YAML 1.1, and thus not clear if Jackson could replace snakeyaml with snakeyaml-engine or would need to support both.

[cowtowncoder]

Jackson 3.x uses snakeyaml-engine.

[mjcarman]

Jackson 3.x uses snakeyaml-engine.

Apologies; that was a misunderstanding on my part. I'm acting as an intermediary for the developers on my team who are trying to use this and running into problems. After discussion with them, I think we've identified the problems (or at least the first layer of them).

1. You can configure the parser to use CoreSchema via LoadSettings, but Jackson is hardcoded to use JsonScalarResolver instead of the resolver for the schema (CoreScalarResolver).
2. YAMLParser._cleanYamlFloat() doesn't map .inf and .nan to Infinity and NaN (the string values recognized by Double.parseDouble())

Curiously, JsonScalarResolver recognizes a plain .inf and .nan as !!float despite the YAML 1.2 JSON Schema seemingly saying that they'd have to be explicitly tagged. Using them in a YAML file results in a NumberFormatException.

[cowtowncoder]

@mjcarman If might make sense to create 2 new issues if these could be handled separately. Failing test cases would be useful. Fixes, if any, should probably go in 3.x branch (for 3.2) unless fixes were low-risk, clear bug fixes.

[mjcarman]

@cowtowncoder To me they're tightly related, but you could consider item 2 to be a separate issue and a dependency for item 1. I'm willing to make an attempt at a PR, but before I spend time going down that path: what's the Jackson position on Inf and NaN? We've hacked around the problems above and hit a new wall:

tools.jackson.databind.exc.JsonNodeException: 'DoubleNode' method decimalValue() cannot convert value Infinity to java.math.BigDecimal: value non-Finite ('NaN')

Both .nan and .inf trigger an exception in NumericFPNode.decimalValue(). The code in jackson-databind passes through jackson-core, and ultimately returns !Double.isFinite(value). To me, it all strongly implies that Jackson neither supports nor wants/intends to support Infinity and NaN. That's a bit limiting for YAML, but entirely reasonable if Jackson views itself as being for JSON and only aims to support the subset of YAML which is compatible with it.

[cowtowncoder]

Quick answer: Jackson DOES support Infs and NaNs -- the main challenge is simply that while JDK float/Float/double/Double can represent these, BigDecimal cannot.
So improving YAML support would be welcome.

This is why NumericFPNode.decimalValue() fails: it must return BigDecimal.

There are some other nuances wrt JSON handling of not-a-number constants (as they are not allowed as Number tokens), but those should not matter to YAML handling.