Courtney makes your code coverage more meaningful, by excluding some of the less important parts.
- Packages are tested with coverage.
- Coverage files are merged.
- Some code is less important to test. This is excluded from the coverage file.
- Optionally we enforce that all remaining code is covered.
What do we exclude from the coverage report?
If you need to test that your code panics correctly, it should probably be an error rather than a panic.
Blocks or files with a // notest
comment are excluded.
We only exclude blocks where the error being returned has been tested to be non-nil, so:
err := foo()
if err != nil {
return err // excluded
}
... however:
if i == 0 {
return errors.New("...") // not excluded
}
All errors are originally created with code similar to errors.New
, which is
not excluded from the coverage report - it's important your tests hit these.
It's less important your tests cover all the points that an existing non-nil error is passed back, so these are excluded.
A few more rules:
- If multiple return values are returned, error must be the last, and all others must be nil or zero values.
- We also exclude blocks returning an error which is the result of a function
taking a non-nil error as a parameter, e.g.
errors.Wrap(err, "...")
. - We also exclude blocks containing a bare return statement, where the function has named result parameters, and the last result is an error that has been tested non-nil. Be aware that in this scenario no attempt is made to verify that the other result parameters are zero values.
- Having test coverage doesn't mean your code is well tested.
- It's up to you to make sure that your tests explore the appropriate edge cases.
go get -u github.com/dave/courtney
Run the courtney command followed by a list of packages. Use .
for the
package in the current directory, and adding /...
tests all sub-packages
recursively. If no packages are provided, the default is ./...
.
To test the current package, and all sub-packages recursively:
courtney
To test just the current package:
courtney .
To test the a
package, it's sub-packages and the b
package:
courtney github.com/dave/a/... github.com/dave/b
Enforce 100% code coverage.
The command will exit with an error if any code remains uncovered. Combining a
CI system with a fully tested package and the -e
flag is extremely useful. It
ensures any pull request has tests that cover all new code. For example, here
is a PR for this project that lacks
tests. As you can see the Travis build failed with a descriptive error.
Override coverage file location.
Provide a custom location for the coverage file. The default is ./coverage.out
.
Argument to pass to the 'go test' command.
If you have special arguments to pass to the go test
command, add them here.
Add one -t
flag per argument e.g.
courtney -t="-count=2" -t="-parallel=4"
Verbose output
All the output from the go test -v
command is shown.
Courtney will fail if the tests fail. If the tests succeed, it will create or
overwrite a coverage.out
file in the current directory.
To upload your coverage to codecov.io via
travis, use a .travis.yml
file something like this:
language: go
go:
- 1.x
notifications:
email:
recipients: <your-email>
on_failure: always
install:
- go get -u github.com/dave/courtney
- go get -t -v ./...
script:
- courtney
after_success:
- bash <(curl -s https://codecov.io/bash)
For coveralls.io, use something like this:
language: go
go:
- 1.x
notifications:
email:
recipients: <your-email>
on_failure: always
install:
- go get -u github.com/mattn/goveralls
- go get -u github.com/dave/courtney
- go get -t -v ./...
script:
- courtney
after_success:
- goveralls -coverprofile=coverage.out -service=travis-ci