Avoiding the automatic installation of Janet dependencies that are only necessary for development.
Development Dependencies in Janet
I now use Jeep, an alternative project manager for Janet, to achieve the same outcome as the approach suggested in this article.
Janet’s package manager, JPM, can be used to install the dependencies of a Janet project that are specified in
project.janet under the
:dependencies key. If the dependency itself has dependencies, JPM will install these as part of the installation process. That’s conceptually simple but it means that if you need dependencies for the development of your project (e.g. a testing library like my Testament), these will be installed for all consumers of your project. That seems like a waste.
Other dependency managers like Ruby’s Bundler or Node’s NPM avoid this problem by providing a means to identify a dependency as being for development. In my latest Janet library, Argy-Bargy, I tried to see if I could do something similar. Here’s what I came up with:
(declare-project :name "Argy-Bargy" :description "A fancy command-line argument parser for Janet" :author "Michael Camilleri" :license "MIT" :url "https://github.com/pyrmont/argy-bargy" :repo "git+https://github.com/pyrmont/argy-bargy" :dependencies  :dev-dependencies ["https://github.com/janet-lang/spork" "https://github.com/pyrmont/testament"]) (declare-source :source ["src/argy-bargy.janet"]) # Development (def project-meta (dyn :project)) (task "dev-deps"  (if-let [deps (project-meta :dev-dependencies)] (each dep deps (bundle-install dep)) (do (print "no dependencies found") (flush))))
This solution has two components.
First, we can add whatever keys we want to the
declare-project call so let’s create a key
:dev-dependencies and add a tuple of URLs just like we would with
Second, we’ll create a task that we can run to install these dependencies. Unfortunately, at the time of writing, we can’t access the values defined in
define-project (I filed an issue about this) so we need to create a
project-meta binding that’s accessible from our task’s environment. Then we basically do the same thing that JPM does when we run
Finally, eagle-eyed readers will notice that my
project.janet file has one additional task:
(task "netrepl"  (with-dyns [:pretty-format "%.20M"] (import spork/netrepl) (eval ~(netrepl/server "127.0.0.1" "9365"))))
This creates a
netrepl task we can use to start a netrepl server. Why is
eval in there? The problem is that the binding
netrepl/server depends on the Spork library being in the
project.janet environment. However, since we’re not importing it, we have a problem. My solution is to quasi-quote the function call and then evaluate the result at runtime. ✺