Review map-node vs namespaced-map-node

[lread]

Moved from TODO note in docs in user guide.

Our map-node and namespaced-map-node model the structure of the source code. This is convenient when navigating through nodes as one would through source.

But... when we want to logically treat any map as a map the difference is a bit awkward.

I was thinking a protocol for map nodes, but this would only be useful for the non-zipper (direct node) user:

• (get-map-children n)
• (replace-map-children n)?

So... not sure what might help. In a previous experiment, I modeled all maps the same, but this made dealing the space that can occur after the namespaced map prefix inconsistent with the rest of the whitespace treatment in rewrite-clj.

[lread]

To remain compatible with rewrite-clj v0, we'll need to keep node layout the same. Anything that we might invent to help treat all maps as maps can be added later. So I can move this issue to a lower priority for now.

[borkdude]

Another idea: make tag+ return namespaced keywords which can be used to build a hierarchy.

(isa? :rewrite-clj/string :rewrite-clj/token) ;;=> true

[lread]

Thanks for the new idea @borkdude!

For the uninitiated (and/or foggy-brained), in Clojure, we can define hierarchies via derive.

user=> (isa? :rewrite-clj/string :rewrite-clj/token)
false
user=> (derive :rewrite-clj/string :rewrite-clj/token)
nil
user=> (isa? :rewrite-clj/string :rewrite-clj/token)
true

So rewrite-clj would provide finer-grained type information via a new tag+ (see #114) but also define a queryable hierarchy for that new type info via Clojure's isa?.

Why do the keywords that are returned by tag+ need to be qualified with :rewrite-clj? Because we'd be registering their relationship via Clojure's derive whose registry is global and therefore we need the qualifier.

An alternative might be to use Clojure's make-hierarchy to create a rewrite-clj specific hierarchy, but then usage might be more awkward.