non-global-extensibility

[greglook]

non-global-extensibility

[gfredericks]

I will probably be trying this out myself for a library I'm writing, so I'll point to some code once I have it.

[greglook]

Yeah, I've run into this too. I think that Fressian's model might be better for this - you explicitly specify a map of write handlers which are used to serialize, and on the other side you provide a map of read handlers from tags to reader functions.

[gfredericks]

ah right, so a map whose keys are classes?

[gfredericks]

how should that interact with the builtin functionality? e.g. what happens if I put clojure.lang.PersistentVector in the map?

[greglook]

I think you'd want the explicit handlers to override builtin serialization. Fressian also has a nice lookup-by-inheritance feature so you can use superclasses or interfaces to catch all the descendant types.

[gfredericks]

okay this sounds cool. incidentally if I end up doing this it will be for https://github.com/gfredericks/webscale

[gfredericks]

Alright, I think I got it working here.

Some design choices I had to make:

• name/format for the additional option: :class-lookup is a map from classes to (fn [value] {:tag ... :form ...})
• where to put the class/interface lookup algorithm: a new namespace puget.class-lookup
• how to cache results so we don't walk the inheritance hierarchy all the time: add a memoized function to *options* in the with-options macro
• how to incorporate it into the format-doc multimethod: check the class-lookup before the protocol, and dispatch to a modified version of the ::tagged-literal impl; this entails looking up the impl twice, but I don't think that's much different from what the protocol codepath is doing anyhow

Let me know how this sounds and I can add docs and make a PR.

[gfredericks]

(edited previous comment to point to a slightly different commit)

[gfredericks]

oh and tests. also I would add tests.

[greglook]

Released 0.9.0 with support for custom print handlers.