feat(visitor): allow to create custom object (instead of Hash) according to context

[kwatch]

(Redirected from https://bugs.ruby-lang.org/issues/12960?next_issue_id=12959 .)

New visitor class to create custom object intead of Hash. It creates custom object according to context (= last mapping key).

• Enables data access like ydoc.foo.bar instead of ydoc['foo']['bar'].
• Creates custom object instead of Hash, for example ydoc['teams'][0].is_a?(Team) and ydoc['teams'][0].members[0].is_a?(Member).

See document of CustomClassVisitor class for details.

You may feel that CustomClassVisitor is very experimental. I hope that at least changes of 'to_ruby.rb' will be imported to upstream, because implementation of CustomClassVisitor requires monkey-pathing to 'to_ruby.rb'.

[tenderlove]

I think I'm OK with the accept_key and accept_value change but not the addition of a new visitor class. I used the AST / Visitor pattern specifically so that people could implement (and maintain) their own visitors.

[kwatch]

Thanks a lot.

I think I'm OK with the accept_key and accept_value change but not the addition of a new visitor class.

I see. I'll revert the following commits.

• (#b5d3885) feat(psych): define 'Psych::Visitors::CustomClassVisitor' class
• (#b81d44e) feat(psych): support key '*' for default custom class
• (#34883b8) feat(psych): add test script

And I'll leave the rest commits.

• (#4213ebe) feat(psych): allow to generate custom Hash object
• (#c9061f6) feat(psych): allow custom Hash object for unknown tagged mapping
• (#95eab41) feat(psych): add hook point for mapping key and value
• (#080fd60) feat(psych): add hook points for mapping-like object
• (#b14d98f) feat(psych): add hook points for merge ('<<')

Is it OK? Give me your advice.

[tenderlove]

@kwatch yes, that makes sense. Please do it!