should be different forms

IIRC, aliases are resolved to namespaces when the form is read, but c-set alias is defined in the ns later, on eval

woah, first one works? TIL

Not following about different forms?

When I copy/paste your second block of code in my REPL it works fine: (doesn't work indeed)

clj꞉myapp.core꞉> 
(comment
  (->> ["1" "2"]
       (reduce (fn [acc v]
                 (require '[clojure.set :as c-set])
                 (c-set/union acc #{v}))
               #{})))
nil  ;; output

Is that because it's in the REPL and not from file?

That shouldn't make a difference. Did you perhaps first run the do block?

Also, you should get #{2 1} or something like that back...

(I probably evaluated the comment 😉)

😂

This is known as the gilardi scenario: https://technomancy.us/143

Encouraging title!

tl;dr you cannot introduce new aliases and use them unless you're in a top level form or top level do

but you can get around this by just using (resolve 'clojure.set/foo) and then use that

Awesome. And why is it called Gilardi scenario? There is nothing on the Internet, that I have found, that tells me that!

After the man, the legend, @U06C3GHJR who probably described it first?

I can't find any references on that, but I always assumed that to be the case :)

Ah yes, here it is confirmed: https://clojurians.slack.com/archives/C03S1KBA2/p1604003087391900

OMG. This is so awesome.

If someone runs into troubles unpacking the blog post about the Gilardi scenario, like I did, here is how to make the OP code work:

(comment
  (->> ["1" "2"]
       (reduce (fn [acc v]
                 (require '[clojure.set :as c-set])
                 ((resolve 'c-set/union) acc #{v}))
               #{})))

Why not use requiring-resolve at that point? :)

I'm new to this. 😃 Yeah, that works too:

(comment
  (->> ["1" "2"]
       (reduce (fn [acc v]
                 ((requiring-resolve 'clojure.set/union) acc #{v}))
               #{})))

That works now because require is still synchronous in joyride :)

In the future you may have to do some promesa wrapping around this

Asked the question here for easier googling by future me: https://ask.clojure.org/index.php/11922/how-to-require-a-symbol-from-within-a-function Please answer 😃 🙏

I got another solution as an answer to that question. It was a bit unclear from the prose of the answer if it worked, but I tried it and it does work. A bit to my surprise...

That's a clever trick ;)

Hmm, not sure if it's supposed to work everywhere though

Can you explain the clever trick for me? 🥺

Hiya! Here's my take: Macro expansion happens early in the process of evaluating the form. Macros take arguments and return "expanded" code. In this case, the expanded code is "nil", but in the process of producing that nil, there is a side-effecting function call to require (the function). The clever trick hooks into the evaluation machinery at a time early enough ("macro expansion time") to allow the alias reference to be properly resolved when it is compiled a bit later. Where the clever trick is a little bit iffy is that the task of understanding Clojure code is simpler if macros are purely functional: transforming "arguments in" (which may include "code as data" (e.g., &body)) to "code out". Calling functions for side effects inside a macro is not in keeping with a purely functional (easier to understand and test) style. However, in this case, the workaround macro is very simple and is intentionally used only for its single side effect. The potential for producing confusing/hard to debug behavior is very small.

Awesome. Thanks, @U06C3GHJR! It makes sense and I now learned about a part of the process that I had the wrong ideas about.

I now wonder why the answerer said this when I informed about my use case: > Sorry, but my hack with a pseudo-macro will not be useful in this case, even though it gives the good result in the REPL.

I tested it, and it does work in the Joyride script where I ran into the Gilardi scenario to begin with. clj-kondo doesn't like it though, but that is probably a configuration issue.

I'm surprised to be honest that it works in SCI. A better solution is to just use require + resolve imo

And like I said, require is going to be async so behavior is going to change here

You might as well prepare yourself for that with promesa right now

Even though it's still sync, treat the result from require as a promise

Can I do that change now, you mean, and be future compatible? Maybe move this part of the discussion to #joyride...

Ok ping me there

I'm not planning to use the pseudo macro thing for this. But I am curious about why it wouldn't work. Given what I now know about it, thanks to @U06C3GHJR, it seems it should work? At least in real Clojure and ClojureScript. SCI might be a different story, even though it is amazingly true to the ”real” things.

In ClojureScript it won't work

In bb it works because require is synchronous there

So it just depends on how you implement require

Thanks! Now I can rest about this. 😃

(:macro (meta #'or)) ;; => true

this flag should be true for any macro https://github.com/clojure/clojure/blob/b1b88dd25373a86e41310a525a21b497799dbbf2/src/jvm/clojure/lang/Var.java#L243-L245

Cool, thanks!

you can also use isMacro method of Var it defined just three lines below my link but then you make your code depend on internal implementation details

Aren't you also relying on implementation details when using the metadata?

I tried both, but just toyed with them. Hylang is not really Clojure, its just a Lisp syntax for Python, everything else is Python, so the way you do things is the same as Python. If you're used to Python semantics and programming constructs over those of Clojure, its a good choice if you just want Lispyness. The Lisp has some limits, for example let doesn't exist, you only get function scope and global scope, no let scope, just like in Python but unlike in other Lisps. libpython-clj is my preferred one, but that's because I much prefer Clojure proper, and that lets you use Python libs from Clojure proper. Its straightforward, everything I tried worked, but I suspect some more esoteric Python libs might have issues, not sure.

Since you said you prefer Python data structure, and want easy integration with Python, HyLang is your best bet.

thank you didibus i tried only libpython-clj, one test project, worked nicely clojure like. but i was curious if i can give the code i made with libpython-clj to a python programmer, to use it as library

its ok i guess trying them both is the best way thank you again

I'd be surprise if you could, I think with libpython-clj you only get one way interop, Clojure can use Python, but not the other way around.

In theory, there's also GraalVM polyglot stuff, and if you run Python there, it could make use of other things like Clojure, but you have to run everything with GraalVM including the Python, so it is not standard CPython

it is not documented because it is private. And I don't know how private macro can be used in user code

Oh. Well I simply found it and didn't even try to call. But even then I could copy the source code. And what I wanted to know if it would be safe to use it.

I don't know why it wouldn't ) it is just 4 lines of code so even in case of problems debugging should not be a massive headache

Maybe. I was expecting the answer to be something similar to "the class may not be loaded if nobody imported it" or something like that. Because it's possible that the class may be marked as if it doesn't exist before anybody loads it. But I don't need this function now. Simply curious.

you can use class without explicit import. Upon symbol lookup clojure will try to load it for you

So does it mean that the macro will load the class?

Byte code will be loaded, yes.

I decided to use it. I need to parse some EDN and I'm migrating from joda time. This is my internal representation that I then use to bind serialization/deserialization methods.

(def serializers (merge {"mm/instant" [java.time.Instant java-time/instant]
                         "mm/local-date-time" [java.time.LocalDateTime java-time/local-date-time]}
                        (when-class "org.joda.time.DateTime"
                          {"clj-time/date-time" [org.joda.time.DateTime time-coerce/to-date]})))

https://www.maria.cloud/ offers a very nice experience. Not sure how it's implemented, but might be worth checking out. But sorry, I think this is just clojurescript.

@U0AQ1R7FG You can achieve this with SCI + your library in the configuration. Let me know if you need any help with that. Possibly you can also use nextjournal's clojure-mode for the code editing.

And when combining all of this with clerk, it would be quite cool. Maybe a more out of the box experience would help. cc @U5H74UNSF

This comes pretty close: https://twitter.com/hashbrown490/status/1530668983552077824 (read the thread, it's done with SCI btw)

@U0AQ1R7FG An example: https://nextjournal.com/try/borkdude/confetti It would be cool if you could have your own library in there (sicmutils is included here and you can play around with it) and just front-end, no backend, so it can be hosted on github pages (although maybe you could have the front-end load notebooks from a gist or so)

@U04V15CAJ thank you, this is very helpful!

java.util.Queue is an interface, you can't instantiate it. You have to find a concrete class implementing that interface that satisfies your needs and instantiate that instead.

also, you do not need to instantiate with a type parameter

generics (the <AsyncContext> part) are erased in Java, meaning everything is really a plain Object under the hood

Thanks folks! Indeed. (new java.util.concurrent.ConcurrentLinkedQueue) this was what I needed! Have a nice day! @U2FRKM4TW @U050ECB92 I assume then there is really no syntax for type parameter in Clojure ?

correct

and it is a fiction at the VM level

(It's a Java-the-language thing)

Got it 🙂 thanks again! Learned something new.

Tbf they're working on making generics real in the VM level

Yes, correct.

how does it know the line and column but not the source?

#:clojure.error{:phase :execution,
                   :line 306,
                   :column 7,
                   :source "NO_SOURCE_FILE"}
       

licenses=> (def x 1)
#'build.licenses/x
licenses=> (meta #'x)
{:line 16,
 :column 1,
 :file "NO_SOURCE_PATH",
 :name x,
 :ns #object[clojure.lang.Namespace 0x6634558a "build.licenses"]}

just typing in the repl will do that

and if you repeat the definition you’ll see the line number increases

I think the main REPL uses a LineNumberingPushbackReader, IIRC...

https://github.com/clojure/clojure/blob/master/src/clj/clojure/main.clj#L368-L374 -- so it depends on how you start the REPL, but the default is to use LNPR.