Crytic, pronounced /ˈkrɪtɪk/, is a mutation testing framework for the crystal programming language. Mutation testing is a type of software testing where specific statements in the code are changed to determine if test cases find this defect.
Crytic is in a very early state of development. It is not very clever, making it slow as well.
- Introducing crytic - mutation testing in crystal-lang
- Building a mutation test framework in crystal - Part 1
Add this to your application's
development_dependencies: crytic: github: hanneskaeufler/crytic version: ~> 8
shards install, this will place the
crytic executable into the
bin/ folder inside your project.
Crytic understands two commands:
crytic test without further arguments will mutate all of the source files found by
src/**/*.cr and use the test-suite containing
spec/**/*_spec.cr. Depending on the size of your project and the duration of a full
crystal spec, this might take quite a bit of time.
Crytic can also be run to only mutate statements in one file, let's call that our subject, or
--subject in the command line interface. You can also provide a list of test files to be executed in order to find the defects. This might be helpful to exclude certain long-running integration specs in order to speed up the test suite.
./bin/crytic test --subject src/blog/pages/archive.cr spec/blog_spec.cr spec/blog/pages/archive_spec.cr
The above command determines a list of mutations that can be performed on the source code of
archive.cr and joins the
archive_spec.cr as a test-suite to find suriving mutants.
./bin/crytic noop will output the source code as if crytic applied a no-operation mutation. This can help with debugging crytic bugs by
./bin/crytic noop | crystal eval.
For "test" command
-s specifies a relative filepath to the sourcecode being mutated.
-m specifies a threshold as to when to exit the program with 0 even when mutants survived. MSI is the Mutation Score Indicator.
-p specifies some source code to prepended to the combination of mutated source and specs. By default this will inject a bit of code to enable the "fail fast" mode of crystal spec. This can be used to disable the fail fast behaviour or avoid errors if you don't use crystal spec.
--reporters/-r specifies which reporters to enable via a comma separated list. Reporters
Console is enabled by default if the option is omitted.
The rest of the unnamed positional arguments are relative filepaths to the specs to be run.
For "noop" command
All unnamed positional arguments are relative filepaths to the specs source code to be printed.
How to read the output
✅ Original test suite passed. Running 138 mutations. ❌ AndOrSwap in source.cr:26:7 The following change didn't fail the test-suite: @@ -26,7 +26,7 @@ end end def ==(other : Chunk) - ((type == other.type) && (range_a == other.range_a)) && (range_b == other.range_b) + ((type == other.type) && (range_a == other.range_a)) || (range_b == other.range_b) end end enum Type ✅ AndOrSwap in source.cr:109:13 Finished in 14:02 minutes: 138 mutations, 85 covered, 36 uncovered, 0 errored, 17 timeout. Mutation Score Indicator (MSI): 73.91%
The first good message here is that the
Original test suite passed. Crytic ran
crystal spec [with all spec files] and that exited with exit code
0. Any other result on your inital test suite and it would not have made sense to continue. Intentionally breaking source code which is already broken is of no use.
Each occurance of
✅ shows that a mutant has been killed, ergo that the change in the source code was detected by the test suite. The line and column numbers are printed to follow the progress through the subject file.
❌ AndOrSwap is signaling that indeed a mutant, an intentional change in the subject, was not detected. The diff below shows the change that was made which was not caught by the test suite.
To show a badge about your mutation testing efforts like at the top of this readme you can make use of the dashboard of stryker by letting crytic post the msi score to the stryker API. To do that, make sure to have the following env vars set:
CIRCLE_BRANCH => "master", CIRCLE_PROJECT_REPONAME => "crytic", CIRCLE_PROJECT_USERNAME => "hanneskaeufler", STRYKER_DASHBOARD_API_KEY => "apikey",
It is currently limited to work with Circle CI and assumes your project is hosted on GitHub.
There are many ways a code-base can be modified to introduce arbitrary failures. Crytic only provides mutators which keep the code compiling (at least in theory). Currently, available mutators are:
This mutant replaces the
&& operator by the
|| operator and vice-versa. A typical mutation is:
- if cool && nice + if cool || nice
This mutant flips literal occurances of
false. A typical mutation is:
def valid - return true + return false end
This mutant flips the
else branch in conditions. It will create an
else branch even if there is none. A typical mutation is:
if true + else doSomething() end
This mutant changes literal occurances of numbers by replacing them with "0". "0" gets replaces by "1". A typical mutation is:
- 0 + 1
This mutant changes the sign of literal numbers. It ignores literal "0". A typical mutation is:
- 5 + -5
This mutant changes literal occurances of string by replacing empty strings with
__crytic__ and all other strings with the empty string. Typical mutations are:
- "Welcome" + ""
- [false].all? + [false].any?
This mutant modifies any regular expression literal to never match anything. A typical mutation is:
- /\d+/ + /a^/
- .select(&.nil?) + .reject(&.nil?)
This mutant replaces method calls with
nil. The mutation is only performed inside methods that are of void return type.
def some_method : Nil - perform_side_effect + nil end
This mutant modifies symbol literals by prefixing them with
- :hello + :__crytic__hello
Credits & inspiration
I have to credit the crystal code-coverage shard which finally helped me create a working mutation testing tool after one or two failed attempts. I took heavy inspirations from its SourceFile class and actually lifted nearly all the code.
One of the more difficult parts of crytic was the resolving of
require statements. In order to work for most projects, crytic has to resolve those statements identical to the way crystal itself does. I achieved this (for now) by copying a bunch of methods from crystal-lang itself.
Obviously I didn't invent mutation testing. While I cannot remember where I have read about it initially, my first recollection is the mutant gem for ruby. Markus Shirp, author of mutant is also the one who explained the notion of a "neutral" mutant to me in private chat. I took his idea and implemented it in crytic as well. Thanks!
The logo above is free from icons8.com.
Although not having tested it myself yet, the mull libray is supposed to work for any llvm based language, which I believe crystal is.
The following diagram shows a rough sequence of how crytic works. It was generated in mermaidjs.github.io.
- Fork it (https://github.com/hanneskaeufler/crytic/fork)
- Create your feature branch (
git checkout -b my-new-feature)
- Run tests locally with
- Commit your changes (
git commit -am 'Add some feature')
- Push to the branch (
git push origin my-new-feature)
- Create a new Pull Request