Skip to content

Latest commit

 

History

History
89 lines (68 loc) · 3.65 KB

README.md

File metadata and controls

89 lines (68 loc) · 3.65 KB
Raw

mitte

An nREPL adapter for evaluating CLJS in MarkLogic

Requirements

  • Node.js (tested with v12)
  • Clojure (tested with 1.10.0)
  • MarkLogic Server (tested with 10.0-1)

Quickstart

  1. Ensure node.js is installed (v12 is recommended, nvm is handy for managing versions).
  2. Run clj -m mitte.core to start the nREPL server. A functional REPL will appear in the terminal but you will more likely want to connect with an editor (emacs/Cursive) using the nrepl port, which is printed and also written to the .nrepl-port file.
  3. Ensure that you have a local MarkLogic database running on port 8000 with an admin-local:admin user (or pass your own :user and :password in an options map in the next step).
  4. Evaluate (mitte.core/cljs-repl) to begin a MarkLogic javascript session and connect the current REPL to it. You can now eval!

Logging

Clojure print statements are forwarded to your original REPL window.

console.log statements are written to disk by MarkLogic to an ErrorLog file. The default path on my machine was:

/Users/<ME>/Library/Application Support/MarkLogic/Data/Logs/8000_ErrorLog.txt

Design Notes

js/cljs_evaluator.js is a minimal JavaScript client which is deployed to a local MarkLogic instance, and establishes a long-polling loop with a local JVM server dubbed "Mitte", which is run alongside the nREPL server. It is automatically installed and invoked on each call to mitte.core/cljs-repl (deployment is implemented in js/start_evaluator.js)

mitte.mitte-server implements the HTTP server handling the REPL loop.
Clients send GET requests to http://host:port/repl and await instruction, then respond via POST before the next GET. Unusually for JavaScript, the MarkLogic v8 context is fully synchronous.

To get the right behavior in this inverted GET design on the JVM side, we use a pair of blocking concurrent queues, each implemented with a java.util.concurrent.LinkedBlockingQueue to handle messaging. The /repl GET handler blocks until there's a JS form in the evaluation-queue, at which point it returns it as the body of the response to our client. Likewise, the POST handler places the body of the incoming request into the result-queue to be read by our nREPL server and returned to the waiting editor process.

When a session begins it is assigned a unique ID. When the server is restarted, requests referring to old sessions will be rejected.

mitte.repl-env implements a REPL environment implementing the necessary protocols to work with standard cljs.repl tooling.

The v8 environment is prepared in a multi-step process.

  • load cljs_evaluator.js script (no dependencies)
  • compile cljs.core, then load a few of the emitted files: goog/base, goog/deps, and <output-dir>deps.js
  • load closure_bootstrap.js to enable the Closure Library to load additional files on its own
  • now we are ready for ClojureScript - load cljs.core and mitte/marklogic-client, which implements some final loader behaviour

TODO

  • Stacktraces
  • Try using various MarkLogic api's from cljs
  • How to compile a project for deployment?
  • Odd error when (require ...) appears in a file (low priority, this is not supported syntax in cljs)

Quirks

  • JSON objects in MarkLogic are wrapped. To reach the object itself, use jsonThing.root.
  • Strings in JSON are also wrapped. Use jsonThing.root.aString.toString()
  • in the node.js api, db.documents.write writes a JSON doc to the database. in the javascript api, xdmp.documentGet does not get a document, it reads from the filesystem. Use fn.doc to read a JSON doc from the database.