nREPL middleware to support refactorings in an editor agnostic way
Clojure Other
Latest commit 10089ca Jan 22, 2017 @plexus plexus committed with expez Handle NoClassDefFoundError when determining class type
When `resolve-missing` gets a number of candidate classes, it will load them to
determine the type (Clojure type/record vs regular Java class). However a class
on the classpath may be compiled with dependencies that are not currently
present, causing the loading to fail.

This would cause the whole operation to fail, instead this change handles the
exception, takes it as a sign that it's a regular Java class, and continues.

For example, I do a `resolve-missing` on the symbol URL, this yields two
candidates: ``, and
``. The latter depends on
`org.w3c.dom.ElementTraversal`, which is not currently present, causing a

Build Status Gitter

Refactor nREPL

nREPL middleware to support refactorings in an editor agnostic way.

The role of this nREPL middleware is to provide refactoring support for clients such as clj-refactor.el.


With CIDER and clj-refactor

If you're using CIDER and clj-refator you don't have to do anything except call cider-jack-in. The dependencies are injected automagically.

Be aware that this isn't the case if you connect to an already running REPL process. See the cider documentation for more details.

Adding the middleware via Leiningen

Add the following, either in your project's project.clj, or in the :user profile found at ~/.lein/profiles.clj:

:plugins [[refactor-nrepl "2.2.0"]
          [cider/cider-nrepl "0.13.0"]]

Adding the middleware via Boot

Add the following in ~/.boot/profile.boot:

(require 'boot.repl)

(swap! boot.repl/*default-dependencies* conj
       '[refactor-nrepl "2.2.0"]
       '[cider/cider-nrepl "0.11.0"])

(swap! boot.repl/*default-middleware* conj

Passing messages to and from refactor-nrepl

We've already called this a middleware, but we haven't really talked about what that means. refactor-nrepl is middleware for a REPL. Specifically it's middleware for a networked REPL, which is managed by refactor-nrepl uses the running REPL to to gain insight about your project, in order to offer various refactorings.

Most likely you're already in an environment with a nREPL client available, so you don't have to worry about anything except sending an receiving messages:

=> (require '[ :as repl])
=> (with-open [conn (repl/connect :port 59258)]
     (-> (repl/client conn 1000)    ; message receive timeout required
       (repl/message {:op "eval" :code "(+ 2 3)"})
;;=> [5]

In the example above we're talking to one of the built-in nREPL ops, eval, passing it the data :code "(+ 2 3)". The rest of the readme details or own nREPL ops which provide various refactoring support.

Available features


Configuration settings are passed along with each msg, currently the recognized options are as follows:

 ;; Verbose setting for debugging.  The biggest effect this has is
 ;; to not catch any exceptions to provide meaningful error
 ;; messages for the client.

 :debug false

 ;; When true `clean-ns` will remove unused symbols, otherwise just
 ;; sort etc
 :prune-ns-form true

 ;; Should `clean-ns` favor prefix forms in the ns macro?
 :prefix-rewriting true

 ;; Some libspecs are side-effecting and shouldn't be pruned by `clean-ns`
 ;; even if they're otherwise unused.
 ;; This seq of strings will be used as regexp patterns to match
 ;; against the libspec name.
 :libspec-whitelist ["^cljsjs"]

Any configuration settings passed along with the message will replace the defaults above.

Artifact lookup

This middleware provides operations for obtaining information about artifacts from clojars, or mvn central.

Two ops are available:


Takes no arguments and returns a list of all available artifacts.


Takes one required argument, artifact which is the full name of the artifact e.g. org.clojure/clojure, and one optional argument force which optionally triggers a forced update of the cached artifacts.

The return value is a sorted list, in decreasing order of relevance, with all the available versions.


This op finds occurrences of a single symbol.

find-symbol requires:

file The absolute path to the file containing the symbol to lookup.

dir Only files below this dir will be searched.

ns The ns where the symbol is defined.

name The name of the symbol.

line The line number where the symbol occurrs, counting from 1.

column The column number where the symbol occurs, counting from 1.

ignore-errors [optional] if set find symbol carries on even if there is broken namespace which we can not build AST for

The return value is a stream of occurrences under the key occurrence which is an list of maps like this:

{:line-beg 5 :line-end 5 :col-beg 19 :col-end 26 :name a-name :file \"/aboslute/path/to/file.clj\" :match (fn-name some args)}

When the final occurrence has been sent a final message is sent with count, indicating the total number of matches, and status done.

Clients are advised to set ignore-errors on only for find usages as the rest of the operations built on find-symbol supposed to modify the project as well therefore can be destructive if some namespaces can not be analyzed.


The clean-ns op will perform the following cleanups on an ns form:

  • Eliminate :use clauses
  • Sort required libraries, imports and vectors of referred symbols
  • Rewrite to favor prefix form, e.g. [clojure [string test]] instead of two separate libspecs
  • Raise errors if any inconsistencies are found (e.g. a libspec with more than one alias.
  • Remove any unused namespaces, referred symbols or imported classes.
  • Remove any duplication in the :require and :import form.
  • Prune or remove any :rename clause.

The clean-ns requires a path which is the path to the file containing the ns to be operated upon.

The return value, ns is the entire (ns ..) form in prestine condition, or nil if nothing was done (so the client doesn't update the timestamp on files when nothing actually needs doing).

Pretty-printing the (ns ..) form is surprisingly difficult. The current implementation just puts stuff on the right line and delegates the actual indentation to the client.

In the event of an error clean-ns will return error which is an error message intended for display to the user.

Warning: The clean-ns op dependes on tools.analyzer to determine which vars in a file are actually being used. This means the code is evaluated and any top-level occurrences of (launch-missiles) should be avoided.

This op can be configured.


The goal of the op is to provide intelligent suggestions when the user wants to import or require the unresolvable symbol at point.

The op requires symbol which represents a name to look up on the classpath. This symbol can be qualified, e.g. walk/postwalk or Pattern/quote will yield the correct result, even though the first is a qualified reference to a clojure var and the second a reference to a static java method.

The return value candidates is a list of ({:name candidate.ns :type :ns} {:name candidate.package :type :type} ...) where type is in #{:type :class :ns :macro} so we can branch on the various ways to make the symbol available. :type means the symbol resolved to a var created by defrecord or deftype, :class is for Java classes but also includes interfaces. :macro is only used if the op is called in a cljs context and means the var resolved to macro.


Loads a new project dependency into the currently active repl.

The op requires coordinates which is a leiningen style dependency.

The return value is a status of done and dependency which is the coordinate vector that was hotloaded, or error when something went wrong.


This op finds available and used local vars in a selected s-expression in a ns on the classpath. In clj-refactor we use this as the underlying op for the extract-function refactoring.

This op requires file which is the path of the file to work on as well as line and column. The enclosing s-expression will be used to determine the available and used locals.

Both line and column start counting at 1.

Return values status of done and used-locals which is a list of unbound vars, or error when something went wrong.

The returned symbols' order is based on the order of their occurrence in the macro expanded s-expression (that means reversed order for threading macros naturally -- compared to what you actually see).


stubs-for-interface takes a single input interface which is a fully qualified symbol which resolves to either an interface or protocol.

The return value is edn and looks like this:

user> (stubs-for-interface {:interface "java.lang.Iterable"})
({:parameter-list "[^java.util.function.Consumer arg0]", :name "forEach"}
 {:parameter-list "[]", :name "iterator"}
 {:parameter-list "[]", :name "spliterator"})

The intended use-case for stubs-for-interface is to provide enough info to create skeleton implementations when implementing e.g. an interface in a defrecord.


extract-definition is based on find-symbol so it takes the same input values. The return value, definition is a string of edn which looks like this:

{:definition {:line-beg 4
              :line-end 4
              :col-beg 9
              :col-end 21
              :name \"another-val\"
              :file \"core.clj\"
              :match \"(let [another-val 321]\"
              :definition \"321\"}
 :occurrences ({:match \"(println my-constant my-constant another-val)))\"
                :file \"core.clj\"
                :name \"another-val\"
                :col-end 50
                :col-beg 38
                :line-end 5
                :line-beg 5})}

The key :definition contains information about the defining form, so the client can delete it.

The key :occurrences is a seq of all occurrences of the symbol which need to be inlined. This means the definition itself is excluded to avoid any special handling by the client.


This op returns, version, which is the current version of this project.


Eagerly builds, and caches ASTs for all clojure files in the project. Returns status done on success and stats for the ASTs built: a list of namespace names as the odd members of the list and either 'OK' as the even member or the error message generated when the given namespace was analyzed. For example

'( "OK" "OK" com.baz '("error" "Could not resolve var: keyw"))


The rename-file-or-dir op takes an old-path and a new-path which are absolute paths to a file or directory.

If old-path is a directory, all files, including any non-clj files, are moved to new-path.

The op returns touched which is a list of all files that were affected by the move, and needs to be visited by the client to indent the updated ns form while we await proper pretty printing support in the middleware.

This op can cause serious havoc if it crashes midway through the refactoring. I recommend not running it without first creating a restore point in your version control system.


Returns namespace-aliases which is a list of all the namespace aliases that are in use in the project. The reply looks like this:

 {t (clojure.test),
  set (clojure.set),
  tracker (refactor-nrepl.ns.tracker},
 :cljs {set (clojure.set), pprint (cljs.pprint)}}

The list of suggestions is sorted by frequency in decreasing order, so the first element is always the best suggestion.


In case namespace B depends on namespace A this operation finds occurrences of symbols in namespace B defined in namespace A.

file The absolute path to the file being analyzed (namespace B).

used-ns The namespace that defines symbols we are searching for (namespace A).

Possible application of this operation to refactor a :refer :all style require into a refer or aliased style require.


The middleware returns errors under one of two keys: :error or :err. The key :error contains an error string which is intended for the end user. The key :err is used for unexpected failures and contains among other things a full stacktrace.

Development with mranderson

mranderson is used to avoid classpath collisions.

To work with mranderson the first thing to do is:

lein do clean, source-deps :prefix-exclusions "[\"classlojure\"]"

this creates the munged local dependencies in target/srcdeps directory

after that you can run your tests or your repl with:

lein with-profile +plugin.mranderson/config repl

lein with-profile +plugin.mranderson/config test

note the plus sign before the leiningen profile.

If you want to use mranderson while developing locally with the repl the source has to be modified in the target/srcdeps directory.

When you want to release locally:

lein with-profile plugin.mranderson/config install

to clojars:

lein with-profile plugin.mranderson/config deploy clojars

Or alternatively run

./ install

./ deploy clojars cleans, runs source-deps with the right parameters, runs the tests and then runs the provided lein target.


An extensive changelog is available here.


Copyright © 2013-2016 Benedek Fazekas, Magnar Sveen, Alex Baranosky, Lars Andersen

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