Re: [GSoC 2020] [Proposal draft #0] Initial JVM support for Guix

[Julien Lepiller]

Le 23 mars 2020 12:06:12 GMT-04:00, Leandro Doctors <address@hidden> a écrit :
>On Mon, 23 Mar 2020 at 12:39, Julien Lepiller <address@hidden>
>wrote:
>
>> Hi Leandro!
>
>Hi, Julien,
>
>Thanks a lot for your feedback! I add my comments below.
>
>
>> I'm glad you're interested in JVM languages.
>
>In fact, I am only interested in Clojure... However, given that
>Clojure is hosted on the JVM, it seems very difficult to me not to
>think in JVM terms when describing a "Clojure importer". This is the
>main reason behind the proposal name change from my previous emails.
>
>
>> I cannot find a way to I am currently trying to build a proper
>maven-build-system (it's a matter of days now).
>
>Again, I am only speaking in JVM terms because of Clojure's hosted
>nature.

Actually, I'm interested in android apps, but maven ended up being a 
requirement ^^"

>
>
>> Some of the issues we have with JVM languages and that you should
>take into account  for your proposal:
>[...]
>
>Thanks for this!
>
>
>> - maven allows programmers to upload a "source" jar, but it is meant
>for debugging, contains generated sources and might be incomplete.
>
>Even though I don't have data to back up this claim, in the
>Guix-Jupyter Scicloj video session [1] they mention that, in the case
>of Clojure, most packages usually include sources.
>
>[1] Guix-Jupyter Scicloj video session from January 9th, 2020.
>https://scicloj.github.io/posts/2020-03-07-guix-jupyter/

I'll check, thanks :)

>
>
>> - maven doesn't link to source repositories
>
>You're right.
>
>However, Clojars (the primary hub for Clojure packages) does. Maybe
>implementing Clojars support could be a first step?
>(I still have to assess this, as there are many Clojure packages that
>are published primarily in Maven Central...)
>
>
>> - how to find a source repository from a groupId and an artifactId is
>not clear to me.
>
>Considering the previous points, this seems mostly a Java-specific
>problem?
>
>
>> Considering that, I don't know if your project is doable. At least,
>all other importers are able to give you a working source, but if it's
>> really impossible, having a list of dependencies would already be
>great :)
>
>I agree with you in that this does seem non-trivial...
>Perhaps, this could be a research-oriented project? I mean, to see
>whether it is feasible to (eventually) support the JVM, and the major
>milestone being a minimalistic POC implementation for Clojars support?
>
>
>> We have a lot of java packages, but none of them are usable with
>maven yet. I've worked on those that are directly required to run
>> maven and its common plugins already, and will send patches as soon
>as everything works. There will need to be a huge work on
>> these packages to rebuild them, probably with the maven-build-system.
>
>I guess this means I should focus on Clojars?

From what you say, yes. A clojars importer would be great. If/when we find 
other ways to be able to import other JVM packages, we'll be able to improve 
upon that work :). Maybe the importer could work in that way:

Import from Clojars, if it fails, import from central, without source 
information. Basically, try multiple repos, from the most specific (that have 
source info) to more generic (that at least haveqversion and dependency 
information).

If you leave some space to be able to plug in other sources (eg. when we have 
scala, for sbt repos), I think you'll have a substantial contribution :). Does 
it sound doable?

>
>
>Best,
>Leandro