Support for named implementations of multimethods

[Dennis Schridde]

Hello!

https://clojuredocs.org/clojure.core/
defmethod#example-542692c7c026201cdc3269cd and https://clojuredocs.org/
clojure.core/defmethod#example-5db314cbe4b0ca44402ef7d4 suggest that a
definition like the following is allowed:

```
(defmulti my-multi (fn [,,,] ,,,))
(defmethod my-multi String my-string-multi [,,,] ,,,)
```

i.e. between the dispatch value (here: "String") and the argument list (here:
"[,,,]") it is allowed to fit in a name for this implementation of the
multimethod. I can confirm that this works and that the name shows up in
stacktraces as https://clojuredocs.org/uc/gstamp suggested.

Sadly this trips Cursive. It complains that "my-string-multi cannot be
resolved" and then does not detect the argument list, so that it will also
complain when the arguments are referenced from the function body.

Best regards,
Dennis

[Tamas (onetom) Herman]

What's the purpose of specifying a custom name?

Is it just about stacktrace readability, or are there any other benefits too?

I do use such facility in high-order functions, like middlewares, for the exact same reason.

Maybe we should provide some examples, which would highlight this benefit:

(defn mk-nudger [x]
    (fn nudger [delta] (+ x delta) (throw (ex-info "Bamm!" {}))))
(def nudge123 (mk-nudger 123))
=> #'user/nudge123
(nudge123 7)
Execution error (ExceptionInfo) at user/mk-nudger$nudger (user.clj:2).
Bamm!
clojure.lang.ExceptionInfo: Bamm! {}
    at user$mk_nudger$nudger__24196.invoke(user.clj:2)
    at user$eval24211.invokeStatic(user.clj:69)
    at user$eval24211.invoke(user.clj:69)

and without the `nudger` name:

(defn mk-nudger [x]
    (fn [delta] (+ x delta) (throw (ex-info "Bamm!" {}))))
(def nudge123 (mk-nudger 123))
(nudge123 7)
Execution error (ExceptionInfo) at user/mk-nudger$fn (user.clj:2).
Bamm!
clojure.lang.ExceptionInfo: Bamm! {}
    at user$mk_nudger$fn__24217.invoke(user.clj:2)
    at user$eval24224.invokeStatic(user.clj:69)
    at user$eval24224.invoke(user.clj:69)

-- 

  tom

[Dennis Schridde]

On Mittwoch, 15. Juni 2022 06:34:41 CEST Tamas (onetom) Herman wrote:
> What's the purpose of specifying a custom name?
> Is it just about stacktrace readability, or are there any other benefits
> too?

Easier-to-read stacktraces are the only benefit I know of.

> I do use such facility in high-order functions, like middlewares, for the
> exact same reason.

I use it when converting my structured log data into the format required by my
logging library (OpenTelemetry SDK):

(defmulti put-attribute-kv (fn [_ _ v]
(if-let [t (and (instance? List v)
(all-of-same-type v))]
[t]
(class v))))

(defmethod put-attribute-kv Keyword keyword-attribute [b k v]
(.put ^AttributesBuilder b (AttributeKey/stringKey (name k)) (name v)))

(defmethod put-attribute-kv String string-attribute [b k v]
(.put ^AttributesBuilder b (AttributeKey/stringKey (name k)) v))

,,,

(defn map->Attributes ^Attributes [m]
(.build
^AttributesBuilder (reduce-kv put-attribute-kv
(Attributes/builder)
m)))

In my case I could also determine which implementation the exception comes
from by looking at line numbers, but having the name in the stacktrace makes
it quite a bit more convenient.

--Dennis

[Imre Kószó]

Related issue https://github.com/cursive-ide/cursive/issues/899

[Imre Kószó]

What's the purpose of specifying a custom name?

From the GH issue it appears it also helps with multi-arity implementations https://github.com/cursive-ide/cursive/issues/899#issuecomment-639085186