从Clojure REPL重新启动HTTP服务器的工作流程

Question

A Clojure-based project of mine uses the netty (required by aleph ) web server. I start the server, along with other components, in a web.clj file like this:

(ns myproject.web)

(def server (atom nil))

(defn initialize []
  (if @server
    (println "Warning: already initialized")
    (let [port 8001]
      (println (format "Starting http://localhost:%s/" port))
      (swap! server (fn [_] (start-http-server
                             (wrap-ring-handler app-routes)
                             {:port port}))))))

(defn shutdown []
  (when @server
    (do
      (println "Shutting down web server")
      (@server)
      (swap! server (fn [_] nil)))))

(defn reinitialize []
  "Run this on the REPL to reload web.clj and restart the web server"
  (myproject.web/shutdown)
  (use :reload-all 'myproject.web)
  (myproject.web/initialize))

The server instance is stored in a Clojure atom , so that it can be stopped later.

I use Emacs and Swank to directly launch the server on the REPL like this (after compiling web.clj with Cc Ck ):

user> (myproject.web/initialize)

Whenever web.clj or other dependent modules are edited, I must

1. remember NOT to recompile web.clj using Cc Ck because the atom holding the running instance would vanish (due to the atom from a newly compiled module) from the REPL.

2. Run (myproject.web/reinitialize) which stops the server and then reloads the module before starting it again.

There are two problems with this:

• Often I forget point #1 and press Cc Ck anyway. This causes the loss of server atom in REPL, leading to having to kill swank (or restart emacs) so that I can start the server at the same port number.

• :reload-all doesn't report compile errors as friendly as Cc Ck (ugly traceback vs concise clickable errors).

How can I best address these two problems in this edit-compile-restart workflow?

Answer 1

You can replace

(def server (atom nil))

with

(defonce server (atom nil))

that way when you evaluate the buffer it will not redefine server.

Answer 2

for your first problem you can store the atom in a different namespace and on load only overwrite it if it is not yet defined. putting it in it's own namepspace will prevent it from being erased by the reload-all