Automated code reviews via mutation testing - semantic code coverage (fork of the latest free version)
Find a file
2016-08-14 15:01:21 -07:00
bin Reduce direct reference of globals 2015-11-21 22:02:51 +00:00
config Extract meta duplication to regexp_mutations 2016-08-14 15:01:21 -07:00
lib Extract meta duplication to regexp_mutations 2016-08-14 15:01:21 -07:00
meta Extract meta duplication to regexp_mutations 2016-08-14 15:01:21 -07:00
spec Extract meta duplication to regexp_mutations 2016-08-14 15:01:21 -07:00
test_app Drop support for rspec-3.3 2016-07-02 21:09:15 -07:00
.gitignore Test both rspec 2 and 3 2014-03-03 20:00:57 +02:00
.rspec Use RSpec as receiver for rspec DSL methods 2014-08-10 21:04:07 +00:00
.rubocop.yml Remove stale entries in rubocop Exclude 2016-03-20 18:43:26 -04:00
.ruby-gemset
Changelog.md Fix changelog date, again 2016-08-02 05:29:34 +00:00
circle.yml Bump ruby version on circle to 2.3 2016-01-24 03:55:23 +00:00
Gemfile Use devtools from rubygems 2015-09-10 13:02:04 +00:00
Gemfile.shared Fix shared gems to be deduplicated 2015-09-04 22:31:04 +00:00
LICENSE
mutant-rspec.gemspec Remove TODO files from gemspecs 2016-08-01 19:19:01 +00:00
mutant.gemspec Remove TODO files from gemspecs 2016-08-01 19:19:01 +00:00
Rakefile Fix coverage of Mutant::Mutator namespace 2016-03-22 01:01:21 +00:00
README.md Add Mutant deep dive blog post to README. 2016-08-10 13:32:05 +02:00

mutant

Build Status Dependency Status Code Climate Inline docs Gem Version Flattr Slack Status

Mutant is a mutation testing tool for Ruby.

The idea is that if code can be changed and your tests do not notice, then either that code isn't being covered or it does not have a speced side effect.

Mutant supports ruby >= 2.1, while support for JRuby is planned. It should also work under any Ruby engine that supports POSIX-fork(2) semantics.

Mutant uses a pure Ruby parser and an unparser to do its magic.

Mutant does not have really good "getting started" documentation currently so please refer to presentations and blog posts below.

Installation

As mutant right now only supports rspec, install the gem mutant-rspec via your preferred method. It'll pull the mutant gem (in the correct version), that contains the main engine.

gem install mutant-rspec

The minitest integration is still in the works.

Examples

cd virtus
# Run mutant on virtus namespace
mutant --include lib --require virtus --use rspec Virtus*
# Run mutant on specific virtus class
mutant --include lib --require virtus --use rspec Virtus::Attribute
# Run mutant on specific virtus class method
mutant --include lib --require virtus --use rspec Virtus::Attribute.build
# Run mutant on specific virtus instance method
mutant --include lib --require virtus --use rspec Virtus::Attribute#type

Rails

To mutation test Rails models with rspec, comment out require 'rspec/autorun' from your spec_helper.rb file. Having done so you should be able to use commands like the following:

RAILS_ENV=test bundle exec mutant -r ./config/environment --use rspec User

Passing in RSpec Options

NOTE: Experimental

You can control some aspects of RSpec using the SPEC_OPTS environment variable as usual. If you want mutant to only pay attention to specs in a certain directory, you can run

SPEC_OPTS="--pattern spec/subdir_only/**/*_spec.rb" mutant --use rspec SomeClass

Limitations

Mutant cannot emit mutations for...

  • methods defined within a closure. For example, methods defined using module_eval, class_eval, define_method, or define_singleton_method:

    class Example
      class_eval do
        def example1
        end
      end
    
      module_eval do
        def example2
        end
      end
    
      define_method(:example3) do
      end
    
      define_singleton_method(:example4) do
      end
    end
    
  • singleton methods not defined on a constant or self

    class Foo
      def self.bar; end   # ok
      def Foo.baz; end    # ok
    
      myself = self
      def myself.qux; end # cannot mutate
    end
    
  • methods defined with eval:

    class Foo
      class_eval('def bar; end') # cannot mutate
    end
    

Mutation-Operators

Mutant supports a wide range of mutation operators. An exhaustive list can be found in the mutant-meta. The mutant-meta is arranged to the AST-Node-Types of parser. Refer to parsers AST documentation in doubt.

There is no easy and universal way to count the number of mutation operators a tool supports.

Subjects

Mutant currently mutates code in instance and singleton methods. It is planned to support mutation of constant definitions and domain specific languages, DSL probably as plugins.

Test-Selection

Mutation testing is slow. The key to making it fast is selecting the correct set of tests to run. Mutant currently supports the following built-in strategy for selecting tests/specs:

Mutant uses the "longest rspec example group descriptions prefix match" to select the tests to run.

Example for a subject like Foo::Bar#baz it will run all example groups with description prefixes in Foo::Bar#baz, Foo::Bar and Foo. The order is important, so if mutant finds example groups in the current prefix level, these example groups must kill the mutation.

Reading Reports

Mutation output is grouped by selection groups. Each group contains three sections:

  1. An identifier for the current group.

    Format:

    [SUBJECT EXPRESSION]:[SOURCE LOCATION]:[LINENO]
    

    Example:

    Book#add_page:Book#add_page:/home/dev/mutant-examples/lib/book.rb:18
    
  2. A list of specs that mutant ran to try to kill mutations for the current group.

    Format:

    - [INTEGRATION]:0:[SPEC LOCATION]:[SPEC DESCRIPTION]
    - [INTEGRATION]:1:[SPEC LOCATION]:[SPEC DESCRIPTION]
    

    Example:

    - rspec:0:./spec/unit/book_spec.rb:9/Book#add_page should return self
    - rspec:1:./spec/unit/book_spec.rb:13/Book#add_page should add page to book
    
  3. A list of unkilled mutations diffed against the original unparsed source

    Format:

    [MUTATION TYPE]:[SUBJECT EXPRESSION]:[SOURCE LOCATION]:[SOURCE LINENO]:[IDENTIFIER]
    [DIFF]
    -----------------------
    
    • [MUTATION TYPE] will be one of the following:
      • evil - a mutation of your source was not killed by your tests
      • neutral your original source was injected and one or more tests failed
    • [IDENTIFIER] - Unique identifier for this mutation

    Example:

    evil:Book#add_page:Book#add_page:/home/dev/mutant-examples/lib/book.rb:18:01f69
    @@ -1,6 +1,6 @@
     def add_page(page)
    -  @pages << page
    +  @pages
       @index[page.number] = page
       self
     end
    -----------------------
    evil:Book#add_page:Book#add_page:/home/dev/mutant-examples/lib/book.rb:18:b1ff2
    @@ -1,6 +1,6 @@
     def add_page(page)
    -  @pages << page
    +  self
       @index[page.number] = page
       self
     end
    -----------------------
    

Concurrency

By default, mutant will test mutations in parallel by running up to one process for each core on your system. You can control the number of processes created using the --jobs argument.

Mutant forks a new process for each mutation to be tested to prevent side affects in your specs and the lack of thread safety in rspec from impacting the results.

If the code under test relies on a database, you may experience problems when running mutant because of conflicting data in the database. For example, if you have a test like this:

m = MyModel.create!(...)
expect(MyModel.first.name).to eql(m.name)

It might fail if some other test wrote a record to the MyModel table at the same time as this test was executed. (It would find the MyModel record created by the other test.) Most of these issues can be fixed by writing more specific tests. Here is a concurrent safe version of the same test:

m = MyModel.create!(...)
expect(MyModel.find_by_id(m.id).name).to eql(m.name)

You may also try wrapping your test runs in transactions.

Note that some databases, SQLite in particular, are not designed for concurrent access and will fail if used in this manner. If you are using SQLite, you should set the --jobs to 1.

Neutral (noop) Tests

Mutant will also test the original, unmutated, version your code. This ensures that mutant is able to properly setup and run your tests. If an error occurs while mutant/rspec is running testing the original code, you will receive an error like the following:

--- Neutral failure ---
Original code was inserted unmutated. And the test did NOT PASS.
Your tests do not pass initially or you found a bug in mutant / unparser.
...
Test Output:
marshal data too short

Currently, troubleshooting these errors requires using a debugger and/or modyifying mutant to print out the error. You will want to rescue and inspect exceptions raised in this method: lib/mutant/integration/rspec.rb:call

Only Mutating Changed Code

Running mutant for the first time on an existing codebase can be a rather disheartening experience due to the large number of alive mutations found! Mutant has a setting that can help. Using the --since argument, mutant will only mutate code that has been modified. This allows you to introduce mutant into an existing code base without drowning in errors. Example usage that will mutate all code changed between master and the current branch:

mutant --include lib --require virtus --since master --use rspec Virtus::Attribute#type

Known Problems

Mutations with Infinite Runtimes

Occasionally mutant will produce a mutation with an infinite runtime. When this happens mutant will look like it is running indefinitely without killing a remaining mutation. To avoid mutations like this, consider adding a timeout around your tests. For example, in RSpec you can add the following to your spec_helper:

config.around(:each) do |example|
  Timeout.timeout(5, &example)
end

which will fail specs which run for longer than 5 seconds.

The Crash / Stuck Problem (MRI)

Mutations generated by mutant can cause MRI to enter VM states its not prepared for. All MRI versions > 1.9 and < 2.2.1 are affected by this depending on your compiler flags, compiler version, and OS scheduling behavior.

This can have the following unintended effects:

  • MRI crashes with a segfault. Mutant kills each mutation in a dedicated fork to isolate the mutations side effects when this fork terminates abnormally (segfault) mutant counts the mutation as killed.

  • MRI crashes with a segfault and gets stuck when handling the segfault. Depending on the number of active kill jobs mutant might appear to continue normally until all workers are stuck into this state when it begins to hang. Currently mutant must assume that your test suite simply not terminated yet as from the outside (parent process) the difference between a long running test and a stuck MRI is not observable. Its planned to implement a timeout enforced from the parent process, but ideally MRI simply gets fixed.

References:

Presentations

There are some presentations about mutant in the wild:

Planning a presentation?

Mutation testing lately (not only mutant) seems to attract some attention. So naturally people do talks about it at conferences, user groups or other chances. Thanks for that!

As I (the author @mbj) am not too happy with some of the facts being presented about mutant the last month.

So if you plan to do a presentation: I offer to review your slides / talk - for free of course. My intention is NOT to change your bias pro / against this tool. Just to help to fix invalid statements about the tool.

Also in many cases a conversation to the author should help you to improve the talk significantly. One of mutants biggest weaknesses is the bad documentation, but instead of assumptions based on the absence of docs, use the tool authors brain to fill the gaps.

Hint, same applies to papers.

Blog posts

Sorted by recency:

Support

I'm very happy to receive/answer feedback/questions and criticism.

Your options:

There is also a mutation testing slack chat. Get an invite here. For discussing this project, join #mutant.

Other Channels:

  • #cosmic-ray: for discussing cosmic-ray, the python mutation testing tool.
  • #devtools: for discussing the devtools metagem.
  • #general: for general discussions about mutation testing.
  • #mutagen: for discussing mutagen, the javascript mutation testing tool.
  • #random: for misc. off topic discussion.
  • #stryker: for discussing stryker, the javascript mutation testing tool.
  • #wtf-dev: for sharing software development wtfs.

Credits

Contributing

  • Fork the project.
  • Make your feature addition or bug fix.
  • Add tests for it. This is important so I don't break it in a future version unintentionally.
  • Commit, do not mess with Rakefile or version (if you want to have your own version, that is fine but bump version in a commit by itself I can ignore when I pull)
  • Send me a pull request. Bonus points for topic branches.

License

See LICENSE file.