Clojurians Log v2
Clojure programming
Channels
Apps
clojure-dev
dmiller 2023-01-04T20:11:57.384879Z
I've been trying to figure out why ClojureCLR has a failing test, coming from clojure.test-clojure.compilation. The test is the equivalent of
(list? (:arglists (meta #'when)))PersistentList .
However,
user=> (defn f [x] x)
#'user/f
user=> (class (:arglists (meta #'f)))
clojure.lang.PersistentVector$ChunkedSeqlist? defintely fails on that.
The :arglists value is computed by clojure.core/sigs and as I look at that, it is going to create PV$ChunkedSeq no matter what. So how does when and in fact everything I've tested in clojure.core have a PersistentList. Can it be that this is getting scrubbed by passing through AOT-compilation? In other words, the initialization code in the class file creates a PersistentList ? Because that's what I'm seeing in ClojureCLR on Framework 4.x, where we still have AOT-compilation. When I run on .Net Core and 6.0 and 7.0, there is no AOT, so when 's arglist is not a list.
Help!
2023-01-04T20:33:22.200329Z
Almost certainly it is aot compilation, the chunked seq won't be special cased in the compiler, so it will be serialized using pr-str to embed in the generated class file, and then the static init of the generated class file will use read-string to deserialize where it will come out as a list
2023-01-04T20:34:30.855289Z
The reason it is only sot compilation and not the regular compilation is in regular compilation the var metadata doesn't need to get serialized into bytecode
dmiller 2023-01-04T20:36:23.561699Z
On ClojureCLR, it comes out in the init method in the assembly as a call to PersistentList.create.
2023-01-04T21:33:26.104389Z
yeah, I was incorrect, about the pr-str/read-string, the compiler does handle seqs specially, but the way it does, they all get turned into PersistentList
borkdude 2023-01-04T20:22:28.548779Z
@dmiller FWIW, I stopped using list? for this exact reason and use seq? pretty much everywhere instead of it
Alex Miller (Clojure team) 2023-01-04T20:24:51.304579Z
Without spending some time on it, can’t definitively answer why, but i do think seq? is a better test here
dmiller 2023-01-04T20:26:03.916789Z
Believe me, I spent way too much time on this. I don't wish that on anyone else. Shall I post an issue on it, or do you want to?
Alex Miller (Clojure team) 2023-01-04T20:29:13.076799Z
An issue to do what?
dmiller 2023-01-04T20:30:34.153879Z
The test is ClojureJVM code. I just inherited it. I'll change it on ClojureCLR, no problem. Never mind.
borkdude 2023-01-04T20:30:52.582159Z
Alex Miller (Clojure team) 2023-01-04T20:31:03.798239Z
Yeah, I don’t think this is high priority to do anything about
dmiller 2023-01-04T20:32:21.729059Z
Understood. As I said, never mind. (I can't ignore it over here, so I'll make the change.)