Clojurians Log v2
clojure-dev 2023-08-02
just had a quick look at the new dev-methods branch. early comment/question: on the "arity" syntax Klass/meth-2 , what's gonna happen when the class is updated and the method cannot be disambiguated unambiguously? failure or reflection?
two cases: aot code and non-aot. AOT code will have compiled a thunk targeting the method found at the original compile time and so it will just continue to call that. The non-aot case will raise an ambiguity error from the compiler.
also, if possible consider keeping the methods dealing with disambiguation public 🙂 would be good to be able to reuse them from tooling (i'm thinking of t.a obv)
I would definitely like to facilitate tooling so maybe some thought can go into what would be useful to expose and how.
thank you. when the branch is more in a ready state if you could ping me I can have a deeper look and let you know my thoughts about it
and actually another comment :) I see a potential source of bugs when the signature syntax is used within a macro: imagine we have a
(ns foo (:import my.Klass))
(defmacro x [y] `(map Foo/meth-Klass ~y))Klass is resolved in the current namespace by the compiler and not by the reader, unlike for type hints, this means that the expansion on the macro implicitely depends on the right import being present at callsites of the macro. so this will fail to resolve Klass
(ns bar (:require foo))
(defn x [] (foo/x [1]))(ns baz (:import my2.Klass) (:require foo))
(defn x [] (foo/x [1])Fantastic catch! I'll look into this.
so would this be another source of state in the reader? source cannot read forms with reader resolved keywords, and it’s possible it won’t be able to read functions which use class imports?
(i mean this new shorthand for static methods on imported classes, not in general)
@dpsutton I'm not sure what you mean by "source of state in the reader" -- could you restate the question?
Yes. Give me 15 minutes and I’ll type of core clearly what I mean
@bronsa an open question at the moment is what to do in syntaxQuote if it encounters a signature containing a type that cannot be resolved. I'll noodle on this for a bit.
Ok an example showing what I'm thinking:
(ns foo
(:require [clojure.set :as set]))
(def x ::set/k)(require 'foo) (source foo/x) results in
user=> (source foo/x)
Execution error at user/eval148 (REPL:1).
Invalid token: ::set/ksource from working on any forms which use static methods in this new manner.@fogus I believe it would be reasonable to leave untouched. that's what happens for
`Foo/barI'm leaning that way too
@dpsutton that's a bug in source , should be of no consideration for this IMO
FWIW: the following is reported by source
(source inspected/mt)
(defn mt [strs]
(map String/toUpperCase-Locale strs (repeat Locale/ENGLISH)))I would have been surprised otherwise 🙂
yeah autoresolve keywords have a completely different resolution issue than the one proposed here
I believe an acceptable solution would be to extend syntax-quote to resolve those symbols at read time to e.g.
Foo/meth-my.Klassthinking about it a bit more, I believe syntax-quote should be extended to do this regardless, since it already does so for all other interop cases e.g.
user=> `(Double/fromString "")
(java.lang.Double/fromString "")