Calva, Joyride, and Portal
An updated version of this post describes my latest Calva, Joyride, and Portal setup.
I've mentioned in several posts over the years that I switched my development setup from Emacs to Atom, initially with ProtoREPL and later with Chlorine, and then to VS Code, initially with Clover (a port of Chlorine) and more recently with Calva. There were several detours along the way, but that is the overall arc.
I've also mentioned a couple of times that I use Portal now, as an extension inside VS Code (after previously using Reveal and, before that, Cognitect's REBL).
I've also published my VS Code and Calva setup files on GitHub.
But I haven't really talked about what that experience is like on a day-to-day basis or any specifics of my integrated workflow.
My REPL
When starting a REPL, I use a number of aliases (which can be found in
my dot-clojure deps.edn
)
to add dependencies that are roughly equivalent to this:
org.clojure/tools.deps.alpha ; add-lib3 branch
{:git/url "https://github.com/clojure/tools.deps.alpha"
:git/sha "8f8fc2571e721301b6d52e191129248355cb8c5a"}
jedi-time/jedi-time {:mvn/version "0.2.1"}
cider/cider-nrepl {:mvn/version "RELEASE"}
djblue/portal {:mvn/version "RELEASE"}
party.donut/dbxray {:mvn/version "RELEASE"}
com.datomic/dev.datafy {:git/sha "4a9dffb"
:git/tag "v0.1"
:git/url "https://github.com/Datomic/dev.datafy"}
io.github.stuarthalloway/reflector
{:git/url "https://github.com/stuarthalloway/reflector"
:git/sha "93a0c19b4526c1180959e940202928d35e5c3cef"}
I have my own script to start various types of REPL but most of the time I
have a CIDER-enhanced nREPL server running with Portal middleware. See
my dot-clojure dev.clj
for full details. The TL;DR is that the script enables whatever is on the
classpath of the things it knows about but a basic version boils down to:
clojure -M:<dev-aliases> -m nrepl.cmdline \
--middleware '[portal.nrepl/wrap-portal cider.nrepl/cider-middleware]'
Portal
Inside VS Code, I use ctrl+alt+c ctrl+alt+c
to connect Calva to that running
nREPL server and then I have a key binding to open Portal (ctrl+alt+space p
),
using :launcher :vs-code
, so that it starts up inside VS Code, and finally
I arrange the workspace so I have Portal on the right and my editor(s) on the
left (when Portal opens, I press ctrl+alt+e ctrl+alt+right
to move it to
the right group, then ctrl+1
to put focus back on my code editor in the left
group). See my
key bindings
and my
Calva configuration
which has my custom REPL command snippets, activated via the ctrl+alt+space
prefix.
I do not bother having the REPL output window visible --
with the Portal middleware in place, any code you evaluate in VS Code/Calva
will cause the result to be tap>
'd and it will appear in Portal. This lets
me use Portal instead of the plain text REPL output window.
I start my REPL from a external terminal window and leave it running, because it
typically outlives my editor sessions. With the add-lib3 branch of tools.deps.alpha
I can add new dependencies without restarting my REPL.
My dev.clj
script will wire up clojure.tools.logging
(if it's on the
classpath) so that all log output is also tap>
'd into Portal.
Joyride
The final piece of the puzzle is
Joyride
which lets you script VS Code (and Calva) using ClojureScript -- powered by
@borkdude's excellent sci
.
I'm only just getting started with Joyride! I have a key binding that opens ClojureDocs.org in a browser (in VS Code) for the symbol my cursor is on and another that opens the Java documentation for a class name (or the class type of an expression I evaluate). You can see those two Joyride scripts in my VS Code/Calva setup repo.
I'm currently working on automating my remote debugging setup and that's what the rest of this blog post will cover!
Remote Debugging with Calva and Portal
There are quite a few moving parts in this section so I'm going to go through the code and configuration first and then talk about how it all fits together.
The idea behind this setup is that we can conditionally run an nREPL server and/or a Portal server in any process, triggered via JVM properties much like the built-in Socket server in Clojure. Then we start SSH tunnels over the VPN into the data center, connect a browser to the Portal server and connect Calva to the remote nREPL server, then debug the remote process by evaluating code within it and viewing the results in Portal.
Starting nREPL and Portal
We've chosen nrepl.server.port
and portal.server.port
as our two JVM
properties and we have the following code in a Polylith component
implementation:
(defn maybe-start-server
"Given a symbol identifying a function to start a server,
derive a property name for the port to use and see if we
can start the server."
[start-server-fn]
(let [[server-type] (str/split (namespace start-server-fn) #"\.")
property (str server-type ".server.port")]
(when-let [port-str (System/getProperty property)]
(try
(let [port (parse-long port-str)
start-server (requiring-resolve start-server-fn)]
(start-server {:port port})
(logger/info server-type "server started on port" port))
(catch Throwable t
(logger/error property
"property exists, but either"
"the port is invalid,"
"the library is not available,"
"or the server cannot be started:"
(type t) (ex-message t)))))))
In the interface for that component, we have:
(defonce ^{:doc "Optional nREPL server instance."} nrepl-server
(impl/maybe-start-server 'nrepl.server/start-server))
(defonce ^{:doc "Optional Portal server instance."} portal-server
(impl/maybe-start-server 'portal.api/start))
We add that interface namespace to the :require
clause of the ns
form
in the main namespace of each application for which we want to enable
remote debugging. Right now, we're doing that in just one application which
is internal-facing but we've run Socket servers in several of our applications
for years to enable debugging via the REPL so we'll probably add this to a
few more applications as we polish the workflow.
The script that starts each of our applications looks for a .jvm_opts
file
named for the application and adds those JVM options -- which is how we
specify the ports for the Socket server, the nREPL server, and the Portal
server.
Connecting VS Code to the Remote Server
As noted above, there are several steps needed here but Joyride can help automate them.
Here's the Joyride script I'm developing:
(ns remote-repl
(:require ["vscode" :as vscode]
[promesa.core :as p]))
(defn- start-tunnel [nrepl-port portal-port label remote-server]
(let [terminal (vscode/window.createTerminal #js {:isTransient true
:name label
:message (str label " Remote REPL...")})]
(.show terminal)
(.sendText terminal (str "ssh -N"
" -L " nrepl-port ":localhost:" nrepl-port
" -L " portal-port ":localhost:" portal-port
" " remote-server))))
(defn- start-browser [portal-port]
(vscode/commands.executeCommand "simpleBrowser.show" (str "http://localhost:" portal-port))
(p/do
(p/delay 2000)
(vscode/commands.executeCommand "workbench.action.moveEditorToRightGroup")
(p/delay 1000)
(vscode/commands.executeCommand "workbench.action.focusFirstEditorGroup")))
(defn- connect-repl []
(vscode/commands.executeCommand "calva.disconnect")
(vscode/commands.executeCommand "calva.connect"))
(defn- repl-setup [nrepl-port portal-port label remote-server]
(start-tunnel nrepl-port portal-port label remote-server)
(p/do
(p/delay 2000)
(start-browser portal-port)
(p/delay 1000)
(connect-repl)))
(repl-setup 6666 7777 "QA" "[email protected]")
This does the following:
- start a terminal in VS Code and run the
ssh
command to setup the tunnel with two ports mapped, - open a Simple Browser inside VS Code, pointing to the Portal server,
- move that browser to the right group (and set focus back to the left group),
- disconnect any existing REPL session and connect a new one.
The delay
calls are fairly arbitrary but allow for each preceding step to
complete before the next step starts (otherwise you get fairly unpredictable
behavior as all the steps tend to overlap!).
To avoid manually filling in the various REPL connection dialogs, we add this Calva setting:
"calva.replConnectSequences": [
{
"name": "QA nREPL Server",
"projectType": "deps.edn",
"cljsType": "none",
"nReplPortFile": ["development", "resources", "qa.nrepl"],
"menuSelections": {}
}
],
This adds QA nREPL Server
at the top of the list of REPL types that Calva
displays and references a file that contains the known port number in
development/resources/qa.nrepl
.
This reduces the connect sequence to:
- accept the suggested directory for the project
- select the
QA nREPL Server
REPL type
There's an open issue in Calva to provide an API for the whole connection sequence so this process can, hopefully, be simplified at some point!
Then we have a key binding to run this script:
{
"key": "ctrl+alt+b q",
"command": "joyride.runUserScript",
"args": "remote_repl.cljs"
},
At this point, my process is:
- connect my VPN to the data center
- open VS Code in the project
- press
ctrl+alt+b q
Joyride will start the SSH tunnel in terminal, open a browser in VS Code connected to the Portal server, and start the REPL connection sequence for me!
Debugging the Remote Process
At this point I can open my REPL scratch file and evaluate the whole file and then start evaluating expressions in the Rich Comment Form as needed:
;; scratch pad for debugging against Admin on QA/production
(ns sean-repl-scratch
(:require [next.jdbc :as jdbc]
[portal.api :as p]
[ws.admin-web.interface.system :refer [sys]]))
(set! *warn-on-reflection* true)
(defn- as-table [v] (with-meta v {:portal.viewer/default :portal.viewer/table}))
(comment
;; set the theme:
(p/eval-str (str
'(portal.ui.state/dispatch! portal.ui.state/state
portal.ui.state/set-theme!
:portal.colors/zerodark)))
;; start tapping...
(add-tap #'p/submit)
;; don't forget this:
(p/clear)
(remove-tap #'p/submit)
;; a useful datasource:
(def ds (-> sys :application :datasource))
;; =================== scratch code for debugging ===================
(as-table (jdbc/execute! ds ["describe table"]))
(as-table
(jdbc/execute! ds [(str "select * from table"
" where name like 'email%'")]))
)
This uses the Portal API to set the theme in the connected browser -- the
default for a headless Portal server is a light theme and I have VS Code
set to a dark theme so setting the browser to zerodark
makes things a lot
more readable for me.
I have expressions to add Portal as a tap>
listener, to clear Portal's
history, and to remove Portal as a tap>
listener. This is intended to
avoid leaving the results of my debugging session in memory and to ensure
that any tap>
calls in the production code don't continue to send results
to Portal when I'm no longer actively debugging things.
Our -main
function in the "admin web app" stores the running system
Component in a global Var
called sys
purely for REPL-based debugging sessions. It contains the
(running) web server and our application state, which in turn contains
a datasource (a HikariCP connection pooled datasource, specifically).
The as-table
function is a convenience to wrap values so that tap>
'ing
them into Portal will use the table viewer by default, instead of the
inspector viewer.
Portal supports datafy
/nav
and by using next.jdbc
to query
tables, I can then navigate through foreign key relationships in the data.
At some point soon, I hope to create some screencasts showing how this
works because it's harder to describe than to show!
Work in Progress
This is very much a work in progress. I have all of the above working
locally but it needs some polishing, and then I'll add the scripts and
configuration to my vscode-calva-setup
repo.
In addition, once Calva offers an API to start a REPL session without needing any dialogs or user input, this process can be streamlined further to potentially remove the REPL connect sequence and to allow for the automated process to automatically set the Portal theme in the browser.