A command line tool to identify unused code.
Unused has been rewritten and now lives at https://github.com/unused-code/unused
Issues, updates, and all future work will occur there.
Yes, literally anything.
It's probably best if you have a file generated from
ctags it can read from
(it looks in
tmp, and the root directory for a
tags file), but if
you have another way to pipe a bunch of
methods/functions/classes/modules/whatever in, that works too.
Right now, there are some special cases built in for Rails and Phoenix apps (specifically, assumptions about what's fine to only have one reference to, e.g. Controllers in Rails and Views in Phoenix), but it'll work on Rubygems, Elixir packages, or anything else.
brew tap joshuaclayton/formulae
brew install unused
This will install
unused and its corresponding dependencies.
To update, run:
brew update brew upgrade unused
Alternatively, you can install with Stack or by hand. Because it needs to compile, installation times may vary, but it's often several minutes.
If you already have Stack installed, ensure you have the latest list of packages:
Verify Stack is using at least
lts-6.0 when installing by checking the
global project settings in
Once that is complete, run:
stack install unused
This will install unused in the appropriate directory for Stack; you'll want
to ensure your
$PATH reflects this.
Once you have these tools installed and the project cloned locally:
stack setup stack install
This will generate a binary in
$HOME/.local/bin; ensure this directory is in
Once Docker is installed, create a binary within your
$PATH to run the
#!/usr/bin/env bash docker run --rm -it -v $(pwd):/code joshuaclayton/unused unused [email protected]
Note that, because Unused will be running inside of a virtual machine, it will take longer to generate output than were you to install via previously-mentioned methods.
unused attempts to read from common tags file locations (
In an application where the tags file exists, run:
If you don't have a tags file, you can generate one by running:
git ls-files | xargs ctags
If you want to specify a custom tags file, or load tokens from somewhere else, run:
cat .custom/tags | unused --stdin
To view more usage options, run:
Exuberant Ctags (or another tool that will generate a tags file, like
hasktags for Haskell projects) is required to use
however, the version of
ctags that ships with OS X (
/usr/bin/ctags) is an
older version won't work with many languages (that BSD version of
it "makes a tags file for ex(1) from the specified C, Pascal, Fortran, YACC,
lex, and lisp sources.")
Installation via Homebrew includes the
ctags dependency. You can also run
brew install ctags by hand. If you're not on OS X, use your favorite package
manager and refer to the Exuberant Ctags site for download instructions.
If you're using
ctags to generate a
tags file prior to running
don't have a workflow around automatically generating a
tags file, run:
git ls-files | xargs ctags -f tmp/tags
This will take your
.gitignore into account and write the tags file to
tmp/tags. Be sure to write this to a location that's ignored by
While this process allows a developer to get started, it requires remembering
to run this command before running
unused. Let's explore how to automate this
ctags installed, you'll likely want to configure your workflow such that
your tags file gets updated periodically without any action on your part. I
recommend following the instructions outlined by Tim Pope on this matter,
which discusses a workflow coupled to
git for managing the tags file. It
includes shell scripting that may not look "effortless"; however, the fact this
is automated helps to ensure
unused is running against new versions of the
code as you (and other teammates, if you have any) are committing.
As he suggests, you'll want to run
git init into the directories you want
this hook, and to manually run the hook:
unused is configured to look for a tags file in three different directories,
.git/ as the article suggests, so no further configuration will be
unused attempts to be intelligent at understanding if your codebase has
changed before running analysis (since it can be time-consuming on large
codebases). To do so, it calculates a "fingerprint" of the entire directory by
md5sum), along with
find and your
If you're checking in artifacts (e.g.
unused will likely take significantly longer to calculate the
--help documentation, you can disable caching with the
$ unused -C
If you're expecting to see results but
unused doesn't find anything, verify
that any artifacts
unused uses (e.g. the
tags file, wherever it's located)
or generates (e.g. in
What might be happening is, because unused searches for tokens with
.gitignore), it's running into checked-in versions of the
tokens from other files, resulting in duplicate occurrences that aren't
representative of the actual codebase. The most obvious might be the
file itself, although if you're using an IDE that runs any sort of analysis
and that's getting checked in somehow, that may cause it too.
One final piece to check is the number of tokens in the tags file itself; if
ctags is misconfigured and only a handful of tokens are being analyzed, they
all may have low removal likelihood and not display in the default results
In my experience, projects under 100,000LOC should have at most around 8,000 unique tokens found. This obviously depends on how you structure your classes/modules/functions, but it'll likely be close.
If you're seeing more than 15,000 terms matched (I've seen upwards of 70,000),
this is very likely due to misconfiguration of
ctags where it includes some
amount of build artifacts. In Ruby, this might be a
directory, or if you're using NPM,
ctags, be sure to include your
--exclude directives; you
can find an example here.
The first time you use
unused, you might see a handful of false positives.
unused will look in two additional locations in an attempt to load
additional custom configuration to help improve this.
# Language or framework name # e.g. Rails, Ruby, Go, Play - name: Framework or language # Collection of matches allowed to have one occurrence autoLowLikelihood: # Low likelihood match name - name: ActiveModel::Serializer # Flag to capture only capitalized names # e.g. would match `ApplicationController`, not `with_comments` classOrModule: true # Matcher for `.*Serializer$` # e.g. `UserSerializer`, `ProjectSerializer` termEndsWith: Serializer # Matcher for `^with_.*` # e.g. `with_comments`, `with_previous_payments` termStartsWith: with_ # Matcher for `^ApplicationController$` termEquals: ApplicationController # Matcher for `.*_factory.ex` # e.g. `lib/appname/user_factory.ex`, `lib/appname/project_factory.ex` pathEndsWith: _factory.ex # Matcher for `^app/policies.*` # e.g. `app/policies/user_policy.rb`, `app/policies/project_policy.rb` pathStartsWith: app/policies # list of termEquals # Matcher allowing any exact match from a list allowedTerms: - index? - edit? - create?
The first location is
~/.unused.yml. This should hold widely-used
configuration roughly applicable across projects. Here's an example of what
might be present:
- name: Rails autoLowLikelihood: - name: ActiveModel::Serializer termEndsWith: Serializer classOrModule: true - name: Pundit termEndsWith: Policy classOrModule: true pathEndsWith: .rb - name: Pundit Helpers allowedTerms: - Scope - index? - new? - create? - show? - edit? - destroy? - resolve - name: JSONAPI::Resources termEndsWith: Resource classOrModule: true pathStartsWith: app/resources - name: JSONAPI::Resources Helpers allowedTerms: - updatable_fields pathStartsWith: app/resources
I tend to work on different APIs, and the two libraries I most commonly use have a fairly similar pattern when it comes to class naming. They both also use that naming structure to identify serializers automatically, meaning they very well may only be referenced once in the entire application (when they're initially defined).
Similarly, with Pundit, an authorization library, naming conventions often mean only one reference to the class name.
This is a file that might grow, but is focused on widely-used patterns across codebases. You might even want to check it into your dotfiles.
The second location is
APP_ROOT/.unused.yml. This is where any
project-specific settings might live. If you're working on a library before
extracting to a gem or package, you might have this configuration take that
unused will attempt to parse both of these files, if it finds them. If
either is invalid either due to missing or mistyped keys, an error will be
Unused leverages Ag to
analyze the codebase; as such, you'll need to have
ag available in your
$PATH. This is set as an explicit dependency in Homebrew.
Alternatively, if you'd like to use
RipGrep, you can do so with the
--search rg flag. Be sure to have RipGrep installed first.
To run the test suite, run:
Copyright 2016-2018 Josh Clayton. See the LICENSE.