Thomas will probably correct or expand on my answer once he sees this. :) But IIRC a dynamic require is really only a REPL thing. It can't work in a file because the compiler needs to know what m is, but it doesn't before actually running the code that require generates.

This is in a Rich comment form though, and thus in the REPL. I first evaluate the require, then fail to use the alias in the following evaluations.

Erm. That does surprise me. What also surprises me is that *ns* is nil, at least in a shadow-cljs REPL.

Disregard the whole previous comment - my env is busted.

OK, fixed. *ns* is still nil though. Given this code:

(comment
  (require '[clojure.string :as str])
  (str "a" "b")
  )

(require '[clojure.string :as str])
=> nil
(str "a" "b")
=> "ab"

Yo, sleep deprivation is no joke. The code above is the dumbest one I've written in a long time. At least it took me only a minute to see that.

Yeah, can reproduce what you see, so neither or us is doing anything wrong (or we both are):

(str/join ["a" "b"])
------ WARNING - :undeclared-ns ------------------------------------------------
 Resource: <eval>:1:2
 No such namespace: str, could not locate str.cljs, str.cljc, or JavaScript source providing "str"
--------------------------------------------------------------------------------

------ WARNING - :undeclared-var -----------------------------------------------
 Resource: <eval>:1:2
 Use of undeclared Var str/join
--------------------------------------------------------------------------------

I’m in good company (we would say in Swedish, dunno if it makes sense in English 😃 ).

There was a similar thread a year ago without any resolution: https://clojurians.slack.com/archives/C03S1L9DN/p1678325483467779

On my end, I'm attaching to shadow-cljs via nREPL and then run (shadow/repl build-id) to get to the CLJS REPL.

Using the REPL via npx shadow-cljs cljs-repl build-id works just fine:

shadow-cljs - connected to server
cljs.user=> 1
1
cljs.user=> (require '[clojure.string :as str])
nil
cljs.user=> (str/join ["a" "b"])
"ab"

@U05224H0W Any clue why aliases work when the REPL is started via cljs-repl but don't work if an nREPL connection is used?

Ah, I’m also using nREPL.

works fine for me?

hmm or not

It worked for me as well with the quickstart-browser project. I'm testing whether switching to deps.edn changes it, because that's what I was using for a test in my own project.

Nah, that still works. Not sure what the difference is.

I suspect I just broke something. REPL require has been broken so many times I can't even count it anymore 😛

Worthy of a test? :) No clue how feasible though, of course.

the reason this is so hard is basically the combination of REPL and hot-reload

ie. you do something at the REPL and a triggered hot-reload deletes everything you did

(as in created new aliases)

or you do something at the REPL that then causes a circular relationship between namespaces and breaks hot-reload 😛

this should of course still work, not sure why it doesn't

Mm, right. But what made you say "hmm or not"?

it worked on the first try, but out of habit I did a load-file and switched to the ns first

the issue appears to be that the alias is created in the wrong ns

Huh. I just updated to 2.28.2 (was using 2.27.5 before) and it now works for me just fine. Tried a few namespaces in the same REPL session, tried restarting the REPL session - all works. @U0ETXRFEW What version do you have?

Ah, it works only when run directly via the nREPL window. Doesn't work when I use a comment form.

I’m on 2.28.2 with the project in question.

> Ah, it works only when run directly via the nREPL window. Doesn’t work when I use a comment form. That’s extra weird. Maybe the nREPL window in Cursive is using user or something as ns and that makes it dodge the problem.

ok so the problem is relatively simple

require uses the ns of the current REPL "session", which if you just eval something and don't set it manually is cljs.user

so the alias is created in cljs.user

BUT regular eval uses the :ns supplied by nrepl OVER the current repl session

so in case of cursive the eval that arrives is

{:eof? false, :ns demo.browser, :form sg/query-ident, :source "sg/query-ident"}

(note that I'm testing in shadow-cljs project itself, doesn't really matter which ns is used)

but as you can see cursive supplied the :ns demo.browser here, although I never actually switched to it

I suspect your nrepl clients do the same

so the alias exists in cljs.user due to the require executing there, but not the ns you actually eval in

seems like that should be more consistent 😛

if you in-ns first it seems to be ok

Interesting, Calva used to do in-ns as part of any evaluation. But I changed that because sending the ns parameter along with the nREPL eval op should be quite equivalent. Not equivalent enough then.

no this is a bug

fix is incoming

fixed in 2.28.3

I can confirm that it is fixed in my project. Thanks, Thomas, you are the best! 🙏

It sort of depends. If you are just using clojurescript, then interop is fine. If you are using re-frame, then you'll want to hook into it's capabilities. And that would be similar for any other framework/library you might use.

I personally use https://github.com/oliyh/martian since the backend at my job exports OpenAPI 3.0 schemas

Fetch is fine.

thanks all for the answers, i will take a look into martian also