@cfleming Do you have any thoughts on this?

Sure, similar to that issue, if you file issues for specific problems I’ll look at them. Depending on what they are, I may or may not be able to fix them though.

As I mentioned in the next to last comment on that issue - you've mentioned before that you have no control over the order of orderEntry tags.

@cfleming Just created https://youtrack.jetbrains.com/issue/PY-37812

Right, that one I don’t think I can control.

@cfleming Seems like they think otherwise: https://youtrack.jetbrains.com/issue/IDEA-221903#focus=streamItem-27-3681725.0-0

"So it looks like this problem is caused by code in Cursive plugin which adds entries to com.intellij.openapi.roots.ModifiableRootModel in different order"

@U2FRKM4TW I actually don’t think that’s correct since I’m using their internal API to do that, but no matter because the new deps integration will be out shortly and I can control these things there. I’ve asked for some clarification on that issue.

Awesome, thanks!

I can’t easily use the .cpcache because I’d have to reproduce everything that the script does in Java/Clojure/Kotlin.

I did look at that, but I couldn’t find a Java implementation of the hashing command used (cksum).

And I can’t just invoke the script because I can’t get everything I need from the various script commands (I can’t remember off the top of my head what was missing).

@cfleming can you shell out to cksum from Java to use it?

Maybe. It’s still very brittle though, the script has a lot of non-trivial logic in it. If I reproduce it all, it’s very tied to a particular version of the script.

One thing I can do is just use a similar mechanism, but not try to use the same .cpcache that deps does.

One thing though, that bit hasn’t changed in this version, it worked the same way in the previous integration (in fact, that code didn’t change at all).

Some sort of workaround would be a big win here. REPL restarts can take 30s or more due to the cp calculation. Sounds like the solution is either 1) add the necessary commands to the clj script 2) implement a Cursive specific cache

@kenny Did the REPL startup take that long previously?

i.e. with the old deps integration?

Yes

Part of it has to do with S3 being slow, I think. Alex made it a bit better with the most recent release.

I remember timing it and Cursive seemed to be about 3-5s slower than resolving via the CLI. This project just takes a while to calculate though. Having to calculate it each repl start even though nothing changed is painful.

Ok, just trying to figure out if that’s a regression or not.

Don’t think so.

cksum is not at all important, was just picked b/c it's fast, easy, and ubiquitous. you could totally hash it however you like and store it in cpcache anyways

Then the Cursive caching and the command line caching would stomp each other. I’ll just store the cached version for Cursive elsewhere.

The only implementation of cksum I could find in Java is GPL.

But caching it independently is a better idea anyway.

You could use a known prefix or something to separate the cursive files in the cache dir

I would vote for shelling out too, just to minimize the differences in behaviour between cursive and the cli tools.

I just worked on some mobile app reverse engineering recently, so I had to deal with unknown protobuf structures as binary data, so I just wrapped xxd and protoc --decode_raw into functions:

(defn encode [data] (shell/run "xxd" "-g1" :in data))
(defn decode [data] (shell/run "xxd" "-r" "-g1" :in data :out-enc :bytes))

(defn protoc->raw [pb] (shell/run "protoc" "--decode_raw" :in pb))

where shell/run is just a little error handling and direnv support on top of clojure.java.shell/sh:

(defn run
  "Use `direnv exec` to ensure the `.envrc` & `shell.nix` environments are available"
  [& args]
  (let [cwd (.getCanonicalPath (io/file "."))
        {:keys [out err exit] :as res} (apply sh "direnv" "exec" cwd args)]
    (if-not (zero? exit)
      (-> (str "Error running `" args "`\n" out err)
          (ex-info res)
          throw)
      (if (string? out)
        (str/trim-newline out)
        out))))

it's so bad that direnv support is missing from intellij... makes it a lot harder to use nix under macos 😞

I also wish IntelliJ supported direnv!

IntelliJ (and specifically Cursive) does support https://github.com/ashald/EnvFile, is that useful?

(I don’t know exactly what direnv does)

It’s a similar idea, but not as extensive. Nor does it work with all direnv setups, if I’m reading it correctly.

Yeah. I looked at EnvFile when Cursive added support for it. EnvFile doesn't quite check all the boxes that direnv does.

Don’t know either. I’m on a fairly recent MBP so I wouldn’t expect it to be hardware related.