/piggieback

nREPL support for ClojureScript REPLs

Primary LanguageClojure

CircleCI Clojars Project

Piggieback

nREPL middleware that enables the use of a ClojureScript REPL on top of an nREPL session.

Why?

Two reasons:

  • The default ClojureScript REPL (as described in the "quick start" tutorial) assumes that it is running in a teletype environment. This works fine with nREPL tools in that environment (e.g. lein repl in Terminal.app or gnome-terminal, etc), but isn't suitable for development environments that have richer interaction models (including editors like vim (vim-fireplace) and Emacs (CIDER), and IDEs like Intellij (Cursive) and Eclipse (Counterclockwise)).

  • Most of the more advanced tool support for Clojure and ClojureScript (code completion, introspection and inspector utilities, refactoring tools, etc) is packaged and delivered as nREPL extensions (e.g. cider-nrepl and refactor-nrepl).

Piggieback provides an alternative ClojureScript REPL entry point (cider.piggieback/cljs-repl) that changes an nREPL session into a ClojureScript REPL for eval and load-file operations, while accepting all the same options as cljs.repl/repl. When the ClojureScript REPL is terminated (by sending :cljs/quit for evaluation), the nREPL session is restored to it original state.

Installation

Piggieback is compatible with Clojure 1.8.0+, and requires ClojureScript 1.9 or later and nREPL 0.6.0 or later.

To use the default Node.js REPL (cljs.repl.node) you'll also need to install a recent version of Node.js.

Leiningen

These instructions are for Leiningen. Translating them for use in Boot should be straightforward.

Modify your project.clj to include the following :dependencies and :repl-options:

:profiles {:dev {:dependencies [[cider/piggieback "0.5.3"]]
                 :repl-options {:nrepl-middleware [cider.piggieback/wrap-cljs-repl]}}}

The :repl-options bit causes lein repl to automagically mix the Piggieback nREPL middleware into its default stack.

If you're using Leiningen directly, or as the basis for the REPLs in your local development environment (e.g. CIDER, fireplace, counterclockwise, etc), you're done. Skip to starting a ClojureScript REPL.

Boot

Contributions welcome!

Clojure CLI (aka tools.deps)

The instructions below require nREPL 0.6.0 or newer

Add this alias to ~/.clojure/deps.edn:

{
;; ...
:aliases {:nrepl
          {:extra-deps
            {nrepl/nrepl {:mvn/version "0.8.3"}
             cider/piggieback {:mvn/version "0.5.3"}}}}
}

Then you can simply run a ClojureScript-capable nREPL server like this:

clj -R:nrepl -m nrepl.cmdline --middleware "[cider.piggieback/wrap-cljs-repl]"

When you connect to the running server with your favourite nREPL client (e.g. CIDER), you will be greeted by a Clojure REPL. Within this Clojure REPL, you can now start a ClojureScript REPL.

Embedded

If you're not starting nREPL through a build tool (e.g. maybe you're starting up an nREPL server from within an application), you can achieve the same thing by specifying that the wrap-cljs-repl middleware be mixed into nREPL's default handler:

(require '[nrepl.server :as server]
         '[cider.piggieback :as pback])

(server/start-server
  :handler (server/default-handler #'pback/wrap-cljs-repl)
  ; ...additional `start-server` options as desired
  )

Alternatively, you can add wrap-cljs-repl to your application's hand-tweaked nREPL handler. Keep two things in mind when doing so:

  • Piggieback needs to be "above" nREPL's nrepl.middleware.interruptible-eval/interruptible-eval; it doesn't use interruptible-eval's evaluation machinery, but it does reuse its execution queue and thus inherits its interrupt capability.
  • Piggieback depends upon persistent REPL sessions, like those provided by nrepl.middleware.session/session.)

Usage

Before you run the following, you must have gone through the setup steps. Instead of using lein repl, you might also connect to a headless nREPL using your development environment.

$ lein repl
....
user=> (require 'cljs.repl.node)
nil
user=> (cider.piggieback/cljs-repl (cljs.repl.node/repl-env))
To quit, type: :cljs/quit
nil
cljs.user=> (defn <3 [a b] (str a " <3 " b "!"))
#<
function cljs$user$_LT_3(a, b) {
    return [cljs.core.str(a), cljs.core.str(" <3 "), cljs.core.str(b), cljs.core.str("!")].join("");
}
>
cljs.user=> (<3 "nREPL" "ClojureScript")
"nREPL <3 ClojureScript!"

See how the REPL prompt changed after invoking cider.piggieback/cljs-repl? After that point, all expressions sent to the REPL are evaluated within the ClojureScript environment. cider.piggieback/cljs-repl's passes along all of its options to cljs.repl/repl, so all of the tutorials and documentation related to it hold.

Important Notes

  1. When using Piggieback to enable a browser REPL: the ClojureScript compiler defaults to putting compilation output in out, which is probably not where your ring app is serving resources from (resources, target/classes/public, etc). Either configure your ring app to serve resources from out, or pass a cljs-repl :output-dir option so that a reasonable correspondence is established.
  2. The load-file nREPL operation will only load the state of files from disk. This is in contrast to "regular" Clojure nREPL operation, where the current state of a file's buffer is loaded without regard to its saved state on disk.

Of course, you can concurrently take advantage of all of nREPL's other facilities, including connecting to the same nREPL server with other clients (so as to easily modify Clojure and ClojureScript code via the same JVM), and interrupting hung ClojureScript invocations:

cljs.user=> (iterate inc 0)
^C
cljs.user=> "Error evaluating:" (iterate inc 0) :as "cljs.core.iterate.call(null,cljs.core.inc,0);\n"
java.lang.ThreadDeath
        java.lang.Thread.stop(Thread.java:776)
		....
cljs.user=> (<3 "nREPL still" "ClojureScript")
"nREPL still <3 ClojureScript!"

(The ugly ThreadDeath exception will be eliminated eventually.)

Piggieback works well with all known ClojureScript REPL environments, including Node and browser REPLs.

Support for Rhino was dropped in version 0.3, and Nashorn support was dropped from ClojureScript in 1.10.741.

Design

This section documents some of the main design decisions in Piggieback and the differences between similar functionality in nREPL and Piggieback.

Perhaps the most important thing to remember is that Piggieback is written in Clojure and runs on Clojure. It drives ClojureScript evaluation by using ClojureScript's Clojure API (cljs.repl/IJavaScriptEnv). This allows you to host both Clojure and ClojureScript evaluation sessions on the same nREPL server, which is pretty cool. On the other hand it also means that you can't use Piggieback with self-hosted ClojureScript REPLs (e.g. Lumo).

Note: For self-hosted ClojureScript you'll need an nREPL implementation that can run natively on it (e.g. nrepl-cljs).

No hard dependency on ClojureScript

Piggieback doesn't have a hard dependency on ClojureScript, as users are expected to provide the necessary ClojureScript dependency themselves. If ClojureScript is not present, Piggieback simply won't do anything (see piggieback_shim.clj for details).

This allows tools to safely load Piggieback without having to consider whether something would blow up.

Session type based dispatch

Clients don't have to specify explicitly whether they are doing a ClojureScript eval operation (e.g. by passing some :env :cljs request params). As Piggieback operates at the nREPL session level all clients need to do is to pass a Piggieback session to ops like eval and that would trigger the Piggieback version of those ops.

Evaluation

As noted above Piggieback provides alternative versions of the standard nREPL ops eval and load-file for ClojureScript evaluation. Due to some differences between Clojure and ClojureScript they don't behave exactly the same.

Most notably - for performance reasons we don't spin separate instances of cljs.repl for each evaluation, as nREPL does for Clojure. In practice this means that if you try to evaluate multiple forms together only the first of them would be evaluated:

;; standard ClojureScript REPL behaviour
cljs.user>
(declare is-odd?)
(defn is-even? [n] (if (= n 0) true (is-odd? (dec n))))
(defn is-odd? [n] (if (= n 0) false (is-even? (dec n))))
#'cljs.user/is-odd?
#'cljs.user/is-even?
#'cljs.user/is-odd?
cljs.user> (is-even? 4)
true

Let's compare this to a REPL powered by Piggieback:

cljs.user>
(declare is-odd?)
(defn is-even? [n] (if (= n 0) true (is-odd? (dec n))))
(defn is-odd? [n] (if (= n 0) false (is-even? (dec n))))
#'cljs.user/is-odd?
cljs.user> (is-even? 4)
Compile Warning   <cljs repl>   line:1  column:2

  Use of undeclared Var cljs.user/is-even?

  1  (is-even? 4)
      ^---

#object[TypeError TypeError: Cannot read property 'call' of undefined]
	 (<NO_SOURCE_FILE>)
cljs.user>

Normally that's not a big deal in practice, as you'd rarely want to evaluate multiple expressions together, but it's something to be kept in mind.

Note: Check out this discussion for more details on the subject.

Pretty-printing

Note: Piggieback introduced support for nREPL's pretty-printing interface in version 0.5.

Support for pretty printing ClojureScript evaluation results is not entirely straightforward. This is because Piggieback mostly relies on the underlying nREPL server implementation to support the features of the nREPL protocol and on the cljs.repl/IJavaScriptEnv interface for ClojureScript evaluation.

nREPL 0.6 introduced nrepl.middleware.print to facilitate printing evaluation results in a configurable way. Since nREPL is implemented in Clojure and runs on the JVM, the middleware relies on receiving Clojure values for printing them. Conversely when evaluating a ClojureScript expression in a JavaScript environment, the resulting Clojure value of the evaluation is always a string. If this value would simply be passed on as is to the middleware, only the string itself could be printed by it instead of the evaluation result within the string.

There are multiple approaches for working around this issue with various trade-offs. The current implementation has the following main considerations:

  1. nrepl.middleware.print is used to print ClojureScript evaluation results whenever possible, so that the same nREPL (pretty) printing configuration is applied to both Clojure and ClojureScript.

  2. For cases where the above is not possible (see below), there is a fallback to support basic pretty printing.

In order to support nrepl.middleware.print for ClojureScript evaluation results, they first need to be read. The resulting Clojure values can then be normally printed by the middleware. However there are various cases where ClojureScript evaluation results can not be read by the default Clojure reader. Some examples:

  • Functions: #object[Function]
  • Objects: #object[cljs.user.Cheese], #object[Window [object Window]]
  • #js literals: #js {:foo 1, :bar 2}
  • #queue literals: #queue [1 2 3]
  • Custom tagged literals: #user/cheese "Pálpusztai"
  • Types implementing IPrintWithWriter in a way that is incompatible with the Clojure reader

To work around some of these cases Piggieback provides its own UnknownTaggedLiteral type. It is used as the default tag reader when reading ClojureScript evaluation results. It doesn't parse the contents of the literal and has print-method defined to simply print the original.

Note: When a pretty-printer which doesn't rely on print-method to serialize values (such as fipp, puget, etc.) is used, UnknownTaggedLiteral will be serialized in the output instead of the original literal.

There are still cases left which can prevent the Clojure reader from successfully reading ClojureScript evaluation results (mostly custom IPrintWithWriter implementations). In order to support pretty printing these results as well, the ClojureScript expression to be evaluated is always wrapped with cljs.pprint/pprint (unless :nrepl.middleware.print/print is set to nrepl.util.print/pr or cider.nrepl.pprint/pr, in which case cljs.core/pr is used instead). This means that whenever the Clojure reader fails to read the value for any reason, we can safely fall back on an already (pretty) printed string, albeit disabling nrepl.middleware.print and hence effectively ignoring the nrepl.middleware.print configuration. Special care is taken that output written to *out* during evaluation is not affected by the wrapping.

For the cases where the (pretty) printing configuration is not being applied, the reader probably failed to read the evaluation results and the above fallbacks are being used instead.

Note: See this pull request for more background and discussion on the current solution.

FAQ

Why "piggieback" instead of "piggyback"?

That's one of life's greatest mysteries. Only Chas can answer that one.

Why is the artifact group id "cider" instead of "nrepl"?

Bozhidar took over the maintenance of Piggieback before taking over the maintenance of nREPL. That's why for a period of time Piggieback lived under CIDER's GitHub org and back then it made sense to use CIDER's group id. Eventually, it got reunited with nREPL, but we've opted to preserve the CIDER group id to avoid further breakages.

For the same reason the main namespace is cider.piggieback instead of nrepl.piggieback.

Does Piggieback work with self-hosted ClojureScript REPLs (e.g. Lumo)?

No, it doesn't. Piggieback is implemented in Clojure and relies on Clojure's ClojureScript evaluation API (cljs.repl/IJavaScriptEnv).

For self-hosted ClojureScript you'll need a native ClojureScript nREPL implementation like nrepl-cljs.

Does shadow-cljs use Piggieback?

No, it doesn't use it. It's most recommended for shadow-cljs users to avoid including the cider.piggieback/wrap-cljs-repl middleware.

Unlike figwheel, which relies on Piggieback, shadow-cljs provides its own nREPL middleware. That's why some features of Piggieback (e.g. pretty-printing) might not be available with shadow-cljs.

You can find shadow-cljs's middleware here.

Need Help?

Feel free to create a Github issue or ask on #cider on Clojurians Slack if you have questions or would like to contribute patches.

Acknowledgements

Nelson Morris was instrumental in the initial development of piggieback.

License

Copyright © 2012-2023 Chas Emerick, Bruce Hauman, Bozhidar Batsov and other contributors.

Distributed under the Eclipse Public License, the same as Clojure.