.visit(...) does not visit nodes that do not extend Printable

[jamestalmage]

This may make sense / not matter for processing ECMAScript related AST's, but it is a weird requirement to force on implementors of custom AST's.

This is a blocker for #57

[jamestalmage]

It appears this was intentional in @843b55b.

I have not yet figured out why visiting non-Printable nodes would be a bad idea, even in the ECMAScript context.

Assuming there are good reasons for it, perhaps we generalize path-visitor to take a custom filter function. The default would be to allow everything, but types.visit would expose an instance that filtered out non-Printable nodes.

Also (or possibly alternatively), it might be good to extend field definitions (i.e. def(...).field(...)) with a visitable flag, and make a types.getVisitableFields, or just pass a getVisitableFields implementation to the generalized path-visitor.

[benjamn]

I'm also having trouble remembering the original rationale… I think it was just for performance reasons (not visiting .loc, mostly), but that can be avoided in other ways.

[jamestalmage]

I think that Printable.check(...) line may also be what prevents it from trying to visit fields with primitive values.

[benjamn]

Digging deeper into the background, visitation used to be restricted to subtypes of Node, which is more restrictive than Printable. I switched to Printable in this commit in order to allow visiting non-Node printable things like comments (but still forbid things like SourceLocations).

I'm persuaded by your argument that we should relax the restriction even further. In order to be visitable, I think an object just needs to have a string-valued .type field, since that's what we look up in this._methodNameTable anyway.