Am I correct in thinking that deps.edn provides no way to build an artifact?

Or even to specify co-ordinates for one?

I think you understand correctly. Or put in a better way I understand the same.

Have you seen ghadi's depstar?

https://github.com/healthfinch/depstar

I have, but that just creates a jar

Also, is Clojure itself an implicit dependency? The examples don’t add it like lein does.

That does an Uber jar so I think it gets the deps of the clj runner I think

It removes itself though

Depstar itself I mean

Oh, I meant in general - if I run a REPL, the Clojure dep comes from the runner?

As far as I can tell yes. I've never seen deps.edn specify a clojure version

Interesting. A lot of this is going to be really weird for Cursive support 😐

It's been rough in CIDER so far as well. It feels a bit like boot. It's not super obvious how to best approach it

Yeah, the loose approach to project identity is tricky.

I think @dominicm has done some work towards that end but it's still a little brittle

Kinda like boot you need to know what aliases or tasks to invoke

Yeah, boot is tricky for Cursive too 🙂

@cfleming clj -Sdescribe lists the deps.edn files that are read in.

You can also get that from tools.deps.alpha as clojure-environment (I think).

My bad: clojure.tools.deps.alpha.reader/clojure-env

The difficulty is there’s no consistent identifier for a module

(it shells out to clj -Sdescribe tho')

Do you inject your dependencies like cider or do you run another process? I know cursive is big on static analysis

I think we're gonna have to merge edn files from emacs lisp

@dpsutton I don’t inject any dependencies.

Most of Cursive’s functionality is static in the editor.

For clj-new I ended up having to ask clj what deps.edn files it would load, reading them, resolving them, making a classpath, and loading those files...

environment   (clojure-env)
        all-deps      (read-deps (:config-files environment))
...
  (-> (deps/resolve-deps all-deps resolve-args)
      (deps/make-classpath (:paths all-deps) {})
      (str/split (re-pattern java.io.File/pathSeparator))
      (->> (run! pom/add-classpath))))

Where are git deps cloned to?

Is there a local .deps directory or something?

Ah, ~/.gitlibs

Is it possible to use file:/// URLs with :git/url?

You can use :local/root

I can’t get it to work.

(instead)

Right, but then I can’t pin a version.

(although I’m mostly trying to avoid having to publish to github just to play around 🙂 )

So :local/root deps just stay where they are on disk, with no version. :git/url is cached to ~/.gitlibs per :sha. :mvn/version is cached to ~/.m2 with a real version.

@cfleming re prior question, the default Clojure version is specified in the install-level deps.edn (which is merged with ~/.clojure/deps.edn and local deps.edn). You can specify it in local deps.edn too if you like.

I’m thinking about the case when a Cursive user imports a deps project and maybe doesn’t even have the CLI installed.

Where is the install-level stuff put? I guess that’s OS/packager dependent?

yeah, but clj -Sdescribe will tell you that

you can use the tools.gitlibs library to determine where the gitlibs are being cached https://github.com/clojure/tools.gitlibs/blob/master/src/main/clojure/clojure/tools/gitlibs.clj#L23

The problem is that IntelliJ, at least on OSX, doesn’t typically know what the users’s shell path is.

Will any of this work at all if the user doesn’t have the clj CLI stuff installed?

no

That’s a bummer.

I mean, some of it can be used directly via the lib but the lib really works in tandem with the scripts

Hmm

the scripts are the bridge from the user’s environment and config files into the lib

I’ll have to see if I can detect the install, and consider what to do if it’s not there or I can’t find it.

the scripts are the only layer that know about (for example) classpath caching

but they use the lib to build the cache

So there’s no way to programmatically determine the classpath from Clojure/Java?

Without exec’ing, I mean?

I guess yes, but with no caching.

Hmm

there is no simple answer to that

the lib can compute a classpath given a configuration

but the configuration is inherently a product of the environment

The environment will be different when scripts are invoked by IntelliJ

to some degree, yes. and I think the question is how much you let the user know that :)

I think the user expectation will be that they get the same effect as running clj in the root of the project

I’ve tried to provide a lot of tools to replicate some or all of what’s in the script layer from inside the lib (Sean is using many of those in the boot tooling)

right, but even running clj in the root of the project will be different when run from IntelliJ.

Well, the environment will be different, I have no idea if the differences will be significant.

the only things that are critical are current working directory and path

there are potentially some environment variables that can affect things but none of them are likely to be in use

path will be the thing that’s different.

can you explicitly add an option to point to the bin dir for clojure?

In general OSX doesn’t give apps access to env vars, and doesn’t execute shell init scripts anyway.

I think I’ll have to.

on os x, if you install via brew (recommended) clojure will be in /usr/local/bin

well, on linux too if you follow the script

windows is … coming

and the install level deps.edn will be at /usr/local/lib/clojure/deps.edn

that last is linux, mac will vary per brew install

mac will be at something like /usr/local/Cellar/clojure/1.9.0.375/deps.edn when installed via brew. maybe there’s a symlink that

’s version independent, not sure

Ok, thanks.

I have tried to do what I can to make things discoverable and computable. if there’s things I can do to help more, I’m open to suggestions

With boot-tools-deps I actually pull the install-level deps.edn from the Clojure brew-install GitHub repo, into the JAR of the library. Looking back, that was probably a bad idea 🙂

I probably ought to change it to use clojure-env instead!

it’s a bad idea :)

@cfleming If it helps, I've been using Cursive for deps.edn projects by first doing clj -Spom and then using the resulting pom.xml

I also imagine Cursive might be able to use the output of clojure -Spath, as it gives you the classpath, including paths for any :local/roots or :git/url / :sha deps

So...say I have a colleague who does not trust brew install and would like to use the linux script for installing clojure...is it possible to do that?

(answer is no, we tried and install -D does not unfortunately work on mac)

The install really only does 5 steps, should be easy enough to do them manually.

@dominicm yep he probably will do that

you’ll notice the brew install just runs an install.sh in the tar

so you can manually do basically the same thing

with some effort and messiness, install.sh for mac and linux-install.sh could become the same script. I have so far not been able to convince myself that gives me any great value though and there are some useful reasons for them to be different.

@alexmiller thanks for answering, was wondering about that as well, good to know

so the difference about install on Macosx is only with -d Create directories. Missing parent directories are created as required. No -D (capital).

BSD vs GNU install 😔 if we want to work on BSD we should support both really.

I've seen some applications probe by using --help and grep

https://www.freebsd.org/cgi/man.cgi?query=install&apropos=0&sektion=0&manpath=FreeBSD+11.1-RELEASE+and+Ports&arch=default&format=html for reference

I think for mac is even a bit different from BSD, I did not see -D at all

Hum, weird, annoying, standards, yay.