Skip to content

Latest commit

 

History

History
149 lines (109 loc) · 5.28 KB

BUILD_DETAIL.md

File metadata and controls

149 lines (109 loc) · 5.28 KB

The CI build, in detail

The Travis CI build runs many verification steps to prevent regressions and ensure high-quality code. To run the Travis build locally, run:

$ script/run_build

It can be useful to run the build steps individually to repro a failing part of a Travis build. Let's break the build down into the individual steps.

Specs

RSpec dogfoods itself. Its primary defense against regressions is its spec suite. Run with:

$ bundle exec rspec

# or, if you installed your bundle with `--standalone --binstubs`:

$ bin/rspec

The spec suite performs a couple extra checks that are worth noting:

  • That all the code is warning-free. Any individual example that produces output to stderr will fail. We also have a spec that loads all the lib and spec files in a newly spawned process to detect load-time warnings and fail if there are any. RSpec must be warning-free so that users who enable Ruby warnings will not get warnings from our code.
  • That only a minimal set of stdlibs are loaded. Since Ruby makes loaded libraries available for use in any context, we want to minimize how many bits of the standard library we load and use. Otherwise, RSpec's use of part of the standard library could mask a problem where a gem author forgets to load a part of the standard library they rely on. The spec suite contains a spec that defines a list of allowed loaded stdlibs.

In addition, we use SimpleCov to measure and enforce test coverage. If the coverage falls below a project-specific threshold, the build will fail.

Cukes

RSpec uses cucumber for both acceptance testing and documentation. Since we publish our cukes as documentation, please limit new cucumber scenarios to user-facing examples that help demonstrate usage. Any tests that exist purely to prevent regressions should be written as specs, even if they are written in an acceptance style. Duplication between our YARD API docs and the cucumber documentation is fine.

Run with:

$ bundle exec cucumber

# or, if you installed your bundle with `--standalone --binstubs`:

$ bin/cucumber

YARD documentation

RSpec uses YARD for API documentation on the rspec.info site. Our commitment to SemVer requires that we explicitly declare our public API, and our build uses YARD to ensure that every class, module and method has either been labeled @private or has at least some level of documentation. For new APIs, this forces us to make an intentional decision about whether or not it should be part of RSpec's public API or not.

To run the YARD documentation coverage check, run:

$ bundle exec yard stats --list-undoc

# or, if you installed your bundle with `--standalone --binstubs`:

$ bin/yard stats --list-undoc

We also want to prevent YARD errors or warnings when actually generating the docs. To check for those, run:

$ bundle exec yard doc --no-cache

# or, if you installed your bundle with `--standalone --binstubs`:

$ bin/yard doc --no-cache

RuboCop

We use RuboCop to enforce style conventions on the project so that the code has stylistic consistency throughout. Run with:

$ bundle exec rubocop lib

# or, if you installed your bundle with `--standalone --binstubs`:

$ bin/rubocop lib

Our RuboCop configuration is a work-in-progress, so if you get a failure due to a RuboCop default, feel free to ask about changing the configuration. Otherwise, you'll need to address the RuboCop failure, or, as a measure of last resort, by wrapping the offending code in comments like # rubocop:disable SomeCheck and # rubocop:enable SomeCheck.

Run spec files one-by-one

A fast TDD cycle depends upon being able to run a single spec file, without the rest of the test suite. While rare, it's fairly easy to create a situation where a spec passes when the entire suite runs but fails when its individual file is run. To guard against this, our CI build runs each spec file individually, using a bit of bash like:

for file in `find spec -iname '*_spec.rb'`; do
  echo "Running $file"
  bin/rspec $file -b --format progress
done

Since this step boots RSpec so many times, it runs much, much faster when we can avoid the overhead of bundler. This is a main reason our CI build installs the bundle with --standalone --binstubs and runs RSpec via bin/rspec rather than bundle exec rspec.

Running the spec suite for each of the other repos

While each of the RSpec repos is an independent gem (generally designed to be usable on its own), there are interdependencies between the gems, and the specs for each tend to use features from the other gems. We don't want to merge a pull request for one repo that might break the build for another repo, so our CI build includes a spec that runs the spec suite of each of the other project repos. Note that we only run the spec suite, not the full build, of the other projects, as the spec suite runs very quickly compared to the full build.