Skip to content

Latest commit

 

History

History
154 lines (104 loc) · 7.08 KB

README.md

File metadata and controls

154 lines (104 loc) · 7.08 KB
Raw

Build Status

Usage:

Running test-mate

either

lein run <parameters> from the git repository with leiningen installed

or

test-mate <parameters> where test-mate is an alias for java -jar <path-to-test-mate-jar> that is used throughout this documentation.

Configuration

test-mate expects a config file in ~/.test-mate-config. You can provide the following data in the config file:

{:statistic-server-url "https://localhost:8443"
 :auth-tokens {
   :publish "test"
 }
 :default-project {
   :project "test"
   :subproject "test-sub"
   :language "java"
 }
 :allow-decreasing-coverage false
 :coverage-threshold 85.0
}

The option :allow-decreasing-coverage defaults to true if not configured. If set to false the push will fail with a non-zero exit value if the new coverage values is below the current coverage value (compared in percentage with an epsilon of 0.001).

Optionally, you can set the Java system property fdc.test-mate.config.file with a file-path to the config file. Setting the system property overrides the configuration in the home directory.

This config file is only needed if you want do interface with statistic server.

Interfacing with statistic server

pushing coverage data

test-mate statistic-server publish-coverage <path/to/jacoco_coverage.xml>

Aggregates the coverage file with "/" and submits the result of this aggregation to the configured statistic-server.

You can overwrite the default project data with an extra argument to the publish command:

test-mate statistic-server publish-coverage <path/to/jacoco_coverage.xml> ''{:project "<project-name>" :subproject "<subproject-name>" :language "<language>"}'

The provided data are merged with the default-project data and you can omit any key as long as they are defined in the config file.

checking coverage difference

test-mate statistic-server check-coverage <path/to/jacoco_coverage.xml>

The command check-coverage will calculate the aggregated coverage of the provided coverage file and compare it to the value currently pushed to the server, failing if the provided file's coverage is lower. This allows easily failing the build if coverage sinks compared to the current master.

adding projects

test-mate statistic-server add-project <path/to/project-def-file>

The project-def-file expects this format:

[
  {:skip true :project "project-name :subproject "subproject-name" :language "java"}
  {:project "project-name :subproject "subproject-name :language "clojure"}
  {:project "test" :subproject "test-sub3" :language "java"}
]

test-mate tries to generate the projects from the file onto the statistic-server. The result is printed to stdout.

To ignore a project in the creation mark it with :skip true (this allows to keep a file with all projects in the statistic server and only append new projects).

Coverage data Aggregation

test-mate aggregate <path/to/coverage_file> package-regex-1 package-regex-2 ...

This command aggregates coverage data for each package-regex (Java regex syntax) and prints the result in EDN-format (https://github.com/edn-format/edn) to the console.

Note: You can supply the special package / (or use the regex .*) to aggregate the overall coverage result. Sub-package aggregation is not supported by all coverage formats.

test-mate automatically infers the coverage format from coverage_file. Currently the following formats are supported:

  • Emma/Jacoco (also supports the old coverage format, as written by cloverage)
  • Cobertura (does not yet support the subpackage aggregation)
  • Go (does not yet support the subpackage aggregation)

Analysis

test-need

The test-need analysis helps to find classes in Java-based projects that probably need tests added the most (hence the name test-need).

It outputs a csv file with some metrics for each java file in the repository.

To run the analysis you will need:

  • a git repository
  • an Emma/Jacoco coverage xml report file

executing test-mate analysis test-need will print the options for the analysis, currently:

test-need analysis options:
-c, --coverage-file FILE                       coverage file (emma/jacoco format)
-r, --git-repo REPO                            git repository
-o, --output FILE         ./test_need_out.csv  output file (csv format)
-n, --num-commits NUM     1000                 number of commits to consider (this affects runtime)
-p, --prefix PREFIX       src/main/java/       prefix used to put before class names in the emma report to get a valid file in the git repo

Only the options --coverage-file and --git-repo are mandatory. With --output you can select a different output file location if the default file location is not your first choice.

The --num-commits options controls the considered set of files, for which the repository is scanned. Supplying a high number affects runtime of the analysis run, since the git operations are not fast enough.

The first ranking is done by absolute uncovered lines for each file. This information is extracted from the supplied coverage-file. The first --num-commits files in this ranked list are considered for further analysis. (The idea behind this filter: files with good coverage probably do not need tests badly. This assumption can, of course, be completely wrong).

In order to match files from the coverage-file to files in the git-repository the --prefix option can be used. This prefix is prepended to the file name in the coverage-file and should provide the file location of the java source file in the git-repository. The analysis prints warnings if it cannot find the file in the git-repository and ultimately ignores the file.

After analysis execution outputs a csv file with this information:

class,commits,bugfixes,uncovered,lines,last-changed,'=>,coverage,bugfix/uncovered,bugfix/commit,bugfix/lines,days-last-update

with:

  • class: the java source file
  • commits: the number of commits for this source file
  • bugfixes: the number of bugfixes for the source file. A commit is considered a bugfix if the commit message contains the string fix (case-insensitive) in it
  • uncovered: absolute number of uncovered lines
  • last-changed: unix-timestamp of last commit for this file

separated by a => follow the derived data, which are:

  • coverage: in percentage
  • bugfix/uncovered: number of bugfixes per uncovered line
  • bugfix/commit: ratio of bugfixes on total commits of this file
  • bugfix/lines: ratio of bugfixes on total lines of this file
  • days-last-update: 'freshness' of the file. A file that was not accessed in a long time, probably does not need new tests added to it.

Finding an interesting case for a class that needs tests added to it is now a manual task. The idea is to open the csv file with an editor that supports sorting by different columns and see which class bubbles up with different metrics.