Let's start with an example. What often only matters is how the code is used superficially, not what the macros generate in real life

Sure, here's varidic-proxy:

(defmacro ^:private variadic-proxy
  "Creates left-associative variadic forms for any operator."
  ([name fn]
     `(variadic-proxy ~name ~fn ~(str "A primitive macro version of `" name "`")))
  ([name fn doc]
     `(variadic-proxy ~name ~fn ~doc identity))
  ([name fn doc single-arg-form]
     (let [x-sym (gensym "x")]
       `(defmacro ~name
          ~doc
          {:arglists '([~'x] [~'x ~'y] [~'x ~'y & ~'rest])}
          ([~x-sym]
             ~((eval single-arg-form) x-sym))
          ([x# y#]
             (list '~fn x# y#))
          ([x# y# ~'& rest#]
             (list* '~name (list '~name x# y#) rest#))))))

(I tried dropping it in as a :macroexpand hook, and removeing the ^:private, but it doesn't seem to be working...)

So you want support for this macro for yourself, not for third party users, am I right?

(since it's private)

Mostly for users, because it defines all the public operators in primitive-math

+, -, rem, >=, etc

I suppose I could do declare , which would be friendly even to non-kondo users, but I wanted to see how ar I could get with hooks

ok, then, you can maybe expand in a hook to indeed, (declare +, -, ...)

or you could even just put declare inside of the code itself, just to satisfy kondo, it doesn't hurt anyone

Right, but as I mention in the parent comment, that won't trigger an error if someone attempts to use them as a HOF

are they macros?

(Which is probably rare, but who knows)

Yes

And not definlined

I think clj-kondo currently doesn't warn when a macro is used as a hof, but that might come sometime soon. In that case you could start doing this probably:

(declare ^:macro + ^:macro - ...)

Ahh, interesting. And in that case, no hooks are needed, I assume

correct

ok, lemme try that. thx!

Heh, I decided to write a little declare-macro macro, and immediately gave myself the same problem all over again.

Another way is to get rid of the macrology altogether and just copy paste the +, - etc macros. The low tech solution :)

I'm using a Lisp, dammit. I should be able to use macros everywhere!

More realistically, I'd prefer not to touch the core code of a library as old and widespread as primtive-math. It would probably be fine, but I don't really want to fool with it...

agreed: > the golden rule is to change nothing - unless we're adding something or fixing a bug (David Nolen, some years ago)

Always a good point.

Works in the REPL, btw

I think this has to do with configuration. E.g. this doesn't cause an NPE:

clj-kondo --config-dir /tmp --lint src/byte_streams/pushback_stream.clj

Interesting. Maybe some lingering reference in the imported primitive-math config.edn?

No, not that.

It seems to be related to the :lint-as entries

...

{:lint-as {byte-streams.utils/defprotocol+  clojure.core/defprotocol
           byte-streams.utils/deftype+      clojure.core/deftype
           byte-streams.utils/defrecord+    clojure.core/defrecord
           byte-streams.utils/definterface+ clojure.core/definterface
           clj-commons.byte-streams.utils/defprotocol+  clojure.core/defprotocol
           clj-commons.byte-streams.utils/deftype+      clojure.core/deftype
           clj-commons.byte-streams.utils/defrecord+    clojure.core/defrecord
           clj-commons.byte-streams.utils/definterface+ clojure.core/definterface}}

CLJ_KONDO_DEV=true clj -M:clj-kondo/dev --lint src/byte_streams/pushback_stream.clj
Execution error (NullPointerException) at clj-kondo.impl.namespace/var-classfile (namespace.clj:112).
Cannot invoke "clojure.lang.Named.getName()" because "x" is null

Full report at:
/var/folders/j9/xmjlcym958b1fr0npsp9msvh0000gn/T/clojure-10659668507427392406.edn

Seems to be related to the older byte-streams.utils/deftype+

seems to error here: https://github.com/clj-kondo/clj-kondo/blob/a749864535f0bb6f227310acf84f2fa7081b2684/src/clj_kondo/impl/namespace.clj#L112

probably should be some->> instead of ->> or so

I don't even know what this code does

hmmm...what's really wrinkling my brain is the code is identical in the non-depreacated ns, but that's not causing problems...somehow

ah, it lints if the var names differ only by case

and somehow a var name is missing there

The code between the old and new namespaces is 99% the same

do the protocol+ macros support the exact same syntax as the original or are there slight differences?

Within byte-streams, the deftype+ macros are identical. Will have to check the usages

Between projects tho, Zach frequently changed those potemkin macros, so each one has a slightly different functionality, fwiw

I'll try to debug some more

Could it be related to the both macro?

(Though that still wouldn't explain why it fails on the old ns, but not the new ns)

$ CLJ_KONDO_DEV=true clj -M:clj-kondo/dev --lint src/byte_streams/pushback_stream.clj
:ns byte-streams.pushback-stream :name PushbackStream
:ns byte-streams.pushback-stream :name Consumption
:ns byte-streams.pushback-stream :name ->Consumption
:ns byte-streams.pushback-stream :name trigger
:ns byte-streams.pushback-stream :name trigger
:ns byte-streams.pushback-stream :name put
:ns byte-streams.pushback-stream :name put
:ns byte-streams.pushback-stream :name expand-either
:ns byte-streams.pushback-stream :name expand-either
:ns byte-streams.pushback-stream :name walk
:ns byte-streams.pushback-stream :name walk
:ns byte-streams.pushback-stream :name prewalk
:ns byte-streams.pushback-stream :name prewalk
:ns byte-streams.pushback-stream :name both
:ns byte-streams.pushback-stream :name both
:ns byte-streams.pushback-stream :name nil
Execution error (NullPointerException) at clj-kondo.impl.namespace/var-classfile (namespace.clj:113).
Cannot invoke "clojure.lang.Named.getName()" because "x" is null

maybe it's this?

deftype+ (either [PushbackByteStream] [SynchronizedPushbackByteStream])

that seems to trigger the NPE

Yeah, but the identical code exists in clj-commons.byte-streams, too

I'm getting the same NPE over there

Oh? Let me check that...

Huh, now I see it, too

I think we should make clj-kondo more robust, it doesn't handle the absence of a var name properly here, but hopefully you know what to do?

No, actually

do you have a hook for both?

Not yet. The both/either macros are convoluted, iirc

And they're only internal

well, I think either that, or just ignore the contents of both and then declare what you need to signal the existence of vars to clj-kondo, if they are important?

I've considered just separating the classes entirely for a while now. Or adding a property to indicate which is which.

Sometimes the fact that Hickey forbade inheritance is super-annoying.

Anyway, I know where the problm lays, anyway

Thanks so much for looking into this!

Have a good day!

you too!