Clojurians Log v2

Is there a shorthand for adding an entire namespace to sci opts? I thought that was possible but I don't see it here: https://github.com/borkdude/sci/blob/1dd8b77062f328f5cd87435858cf91f165bcadd4/src/sci/core.cljc#L168-L170

There isn't and there hasn't been one. You can use a combination of ns-publics and sci/copy-var

Not sure why I thought there was :thinking_face: But this will work as well, thanks

I'm not too keen on automatically doing this since I always feel there has to be some curation to keep GraalVM binary size in check :)

That or maybe problems with macros or other stuff. I usually build it up var by var

It seems like multimethods are shared between the host and the SCI environment. Is this intentional?

Not sure? Can you give an example?

(defmethod ig/init-key :some/key [_ _opts] 123)

  (sci/eval-string
   "(require '[integrant.core :as ig])
    (ig/init-key :some/key {})"
   {:namespaces {'integrant.core {'init-key ig/init-key}}})
;; => 123

I assume it works for any multimethod that you pass in

well, you are sharing the multimethod which gets information from your host environment. this is the same with any other closure

Interesting. I was expecting the state to be isolated. But if the state is stored in the mutlimethod var, then it makes sense it gets passed along

it might be different for defmulti's you define inside sci

yeah if it's created inside of sci, then it shouldn't leak out to the host

But defmethods are modified

I think that is true: https://github.com/borkdude/sci/blob/1dd8b77062f328f5cd87435858cf91f165bcadd4/src/sci/impl/multimethods.cljc#L72

the "global hierarchy" in sci is per ctx

but if you are passing in an external multimethod impl then it has access to your host's global hierarchy

I see in that code also that it closes over an atom which belongs to that multimethod only

so it's effectively just a closure you're passing into sci

That's troublesome for me then. I actually want to use sci to isolate code. So I might have to think of a different strategy

A different strategy would be to redefine those multi's in sci code itself

or create another defmulti in your host with a local hierarchy: > By default Clojure's multimethods dispatch off of a global hierarchy map.

like here: https://clojuredocs.org/clojure.core/defmulti#example-543d75b3e4b0a3cf052fe476

I want to use the multimethod of Integrant. I'm not sure if it's possible to override that

I'd probably have to create my own implementation

hmm, maybe the global hierarchy can be re-bound to some dynamic var? I'm not sure

no, it's not dynamic. https://github.com/clojure/clojure/blob/master/src/clj/clojure/core.clj#L1725 but you could patch that :P

haha

Sounds like a weird feature. being able to rebind the global hierarchy

Might as well define your own. Sadly Integrant doesn't support that

it should then really be called the default hierarchy

you could propose it to the core team

I honestly don't know if I like that idea myself

with-redefs might work here

note that this is not thread safe

I think I'd rather fork integrant to actually do what I want it to do

maybe you could also make just a shim function that calls defmethod but registers the value with a unique prefix for that sci ctx

The "value" being the keyword?

Or do you mean the function name?

Something like this:

(defmulti foo identity)

(defmacro reg-method [sci-ctx & body]
  `(defmethod foo ~sci-ctx [sci-ctx#]
     ~@body))

(reg-method "sci-ctx-1" (+ 1 2 3))
(reg-method "sci-ctx-2" (* 2 4))

(foo "sci-ctx-1") ;=> 6
(foo "sci-ctx-2") ;=> 8

but then the body of the reg-method macro will wrap your integrant calls

and you will pass in a function which wraps reg-method to sci so it seems to the sci program as if you're calling integrant normally

so integrant only supports one global system, like mount basically?

Yeah, there was an issue about adding an optional hierarchy, but that was never realized

Also Integrant development has stalled quite a bit

I see, so yeah, that would be the way to get out of this, an optional hierarchy or some hierarchy you can bind via a dynamic var, defaulting to the global one.

I'll go think about what would be the best option here. Thanks for the suggestions

Whenever I play around with SCI I always get fun problems to solve haha

The definitions are global, but you can run multiple systems no problem.

yeah right, you can run multiple instances of a global system

but not multiple different systems?