These points do not apply to prototype / throwaway phase.
To run testing, perform code quality analysis and build artifacts a Continuous Integration server can be used. The build will be performed every git push and pull request. Using a CI server will help with
it works for me problems.
The Netherlands eScience Center uses continuous integration services as much as possible when creating code.
CI meaning: compile, unit test, integration test, quality analysis etc. Once there is some build process established and tests set up, CI should be configured too. It will save you a lot of time on debugging and allow for much quicker problem diagnosis.
The Netherlands eScience Center public repositories should be built with Travis-CI. Travis-CI is free for Open Source projects. A Github repository can be added to Travis-CI by a Github user with admin right on the repository. At the moment Travis-CI performs builds in Ubuntu and OS X operating systems.
PS. If you want to get mails from Travis-CI then you have to login at https://travis-ci.org
To build repositories inside the Microsoft Windows operation system use AppVeyor. AppVeyor is free for Open Source projects.
Most CI builds are triggered by a git push, but sometimes the repository must be build every night. Possible reasons for nightly builds:
- Make sure the repository stays working even if there are no changes pushed to the repository, but it's dependencies are changing possibly breaking the code in the repository.
- The build performs an action that needs to be performed daily like updating a cache.
For triggering nightly builds in Travis-CI Cron jobs can be used.
All major CI services support some form of cctray.xml feed. This feed can be read by polling tools to automatically keep an eye on your project builds. For instance, BuildNotify, CCMenu and CCTray give you a tray icon that turns red when a build fails.
Code coverage is a measure which describes how much of the source code is exercised by the test suite. At the Netherlands eScience Center we require minimum of 70% coverage.
Setting up code coverage for a repository depends on the programming language, see the language specific guides for setup instructions.
The code coverage should be performed when a test suite is run as part of Continuous Integration build job. The code coverage results can be published on code coverage and/or code quality services.
Code coverage services
The publishing of the code coverage can be performed during a Continuous Integration build job. The code coverage service offers a visualization of the coverage and a metric which can be displayed as a badge/shield icon on the repository website. See the language specific guides which code coverage services are available and preferred for that language.
Code coverage services support many languages and a usually free for Open Source projects. Below is a short list of services and their strengths.
Shows unified coverage and separate coverage for build matrix e.g. different Python versions. For example project see https://codecov.io/gh/NLeSC/Xenon, with a Java 7/8 and Linux/Windows/OSX OS build matrix.
More popular then Codecov. For example project see https://coveralls.io/r/NLeSC/MAGMa
For (web) user interfaces. example with protractor and angular
Once the web page has any interface, e2e tests should be implemented.
Checking for dependency updates should be done regularly. It can save a lot of time, avoiding code dependent on deprecated functionality.