Skip to content

Latest commit

 

History

History
269 lines (193 loc) · 9.63 KB

hacking.md

File metadata and controls

269 lines (193 loc) · 9.63 KB

Hacking on JX

This guide is for developers who want to improve the jenkins-x jx CLI. These instructions will help you set up a development environment for working on the jx source code.

Prerequisites

To compile and test jx binaries you will need:

  • git
  • Go 1.9 or later, with support for compiling to linux/amd64
  • dep

In most cases, install the prerequisite according to its instructions. See the next section for a note about Go cross-compiling support.

Configuring Go

The jx's binary eCLI is built on your machine in your GO Path.

On macOS, Go can be installed with Homebrew:

$ brew install go 

It is also straightforward to build Go from source:

$ sudo su
$ curl -sSL https://storage.googleapis.com/golang/go1.7.5.src.tar.gz | tar -C /usr/local -xz
$ cd /usr/local/go/src
$ # compile Go for the default platform first, then add cross-compile support
$ ./make.bash --no-clean
$ GOOS=linux GOARCH=amd64 ./make.bash --no-clean

Fork the Repository

Begin at Github by forking jx, then clone your fork locally. Since jx is a Go package, it should be located at $GOPATH/src/github.com/jenkins-x/jx.

$ mkdir -p $GOPATH/src/github.com/jenkins-x
$ cd $GOPATH/src/github.com/jenkins-x
$ git clone [email protected]:<username>/jx.git
$ cd jx

Add the conventional upstream git remote in order to fetch changes from jx's main master branch and to create pull requests:

$ git remote add upstream https://github.com/jenkins-x/jx.git

Build Your Changes

With the prerequisites installed and your fork of jx cloned, you can make changes to local jx source code.

Run make to build the jx binaries:

$ make build      # runs dep and builds `jx`  inside the build/

Testing

The jx test suite is divided into three sections:

  • The standard unit test suite
  • Slow unit tests
  • Integration tests

To run the standard test suite: make test

To run the standard test suite including slow running tests: make test-slow

To run all tests including integration tests (NOTE These tests are not encapsulated): make test-slow-integration

To get a nice HTML report on the tests: make test-report-html

Writing tests

Unit Tests

Unit tests should be issolated (see what is an unencapsulated test), and should contain the t.Parallel() directive in order to keep things nice and speedy.

If you add a slow running (more than a couple of seconds) test, it needs to be wrapped like so:

if testing.Short() {
	t.Skip("skipping a_long_running_test")
} else {
	// Slow test goes here...
}

Slows tests can (and should) still include t.Parallel()

Best practice for unit tests is to define the testing package appending _test to the name of your package, e.g. mypackage_test and then import mypackage inside your tests. This encourages good package design and will enable you to define the exported package API in a composable way.

Integration Tests

To add an integration test, create a separate file for your integration tests using the naming convention mypackage_integration_test.go Use the same package declaaration as your unit tests: mypackage_test. At the very top of the file before the package declaration add this custom build directive:

// +build integration

Note that there needs to be a blank line before you declare the package name.

This directive will ensure that integration tests are automatically separated from unit tests, and will not be run as part of the normal test suite. You should NOT add t.Parallel() to an unencapsulated test as it may cause intermittent failures.

What is an unencapsulated test?

A test is unencapsulated (not issolated) if it cannot be run (with repeatable success) without a certain surrounding state. Relying on external binaries that may not be present, writing or reading from the filesystem without care to specifically avoid collisions, or relying on other tests to run in a specific sequence for your test to pass are all examples of a test that you should carefully consider before committing. If you would like to easily check that your test is issolated before committing simply run: make docker-test, or if your test is marked as slow: make docker-test-slow. This will mount the jx project folder into a golang docker container that does not include any of your host machines environment. If your test passes here, then you can be happy that the test is encapsulated.

Mocking / Stubbing

Mocking or stubbing methods in your unit tests will get you a long way towards test isolation. Coupled with the use of interface based APIs you should be able to make your methods easily testable and useful to other packages that may need to import them. https://github.com/petergtz/pegomock Is our current mocking library of choice, mainly because it is very easy to use and doesn't require you to write your own mocks (Yay!) We place all interfaces for each package in a file called interface.go in the relevant folder. So you can find all interfaces for github.com/jenkins-x/jx/pkg/util in github.com/jenkins-x/jx/pkg/util/interface.go Generating/Regenerating a mock for a given interface is easy, just go to the interface.go file that corresponds with the interface you would like to mock and add a comment directly above your interface definition that will look something like this:

// CommandInterface defines the interface for a Command
//go:generate pegomock generate github.com/jenkins-x/jx/pkg/util CommandInterface -o mocks/command_interface.go
type CommandInterface interface {
	DidError() bool
	DidFail() bool
	Error() error
	Run() (string, error)
	RunWithoutRetry() (string, error)
	SetName(string)
	SetDir(string)
	SetArgs([]string)
	SetTimeout(time.Duration)
	SetExponentialBackOff(*backoff.ExponentialBackOff)
}

In the example you can see that we pass the generator to use: pegomock generate the package path name: github.com/jenkins-x/jx/pkg/util the name of the interface: CommandInterface and finally an output directive to write the generated file to a mock subfolder. To keep things nice and tidy it's best to write each mocked interface to a separate file in this folder. So in this case: -o mocks/command_interface.go

Now simply run:

go generate ./...

or

make generate

You now have a mock to test your new interface! The new mock can now be imported into your test file and used for easy mocking/stubbing. Here's an example:

package util_test

import (
	"errors"
	"testing"

	"github.com/jenkins-x/jx/pkg/util"
	mocks "github.com/jenkins-x/jx/pkg/util/mocks"
	. "github.com/petergtz/pegomock"
	"github.com/stretchr/testify/assert"
)

func TestJXBinaryLocationSuccess(t *testing.T) {
	t.Parallel()
	commandInterface := mocks.NewMockCommandInterface()
	When(commandInterface.RunWithoutRetry()).ThenReturn("/test/something/bin/jx", nil)

	res, err := util.JXBinaryLocation(commandInterface)
	assert.Equal(t, "/test/something/bin", res)
	assert.NoError(t, err, "Should not error")
}

Here we're importing the mock we need in our import declaration:

mocks "github.com/jenkins-x/jx/pkg/util/mocks"

Then inside the test we're instantiating NewMockCommandInterface which was automatically generated for us by pegomock.

Next we're stubbing something that we don't actually want to run when we execute our test. In this case we don't want to make a call to an external binary as that could break our tests isolation. We're using some handy matchers which are provided by pegomock, and importing using a . import to keep the syntax neat (You probably shouldn't do this outside of tests):

When(commandInterface.RunWithoutRetry()).ThenReturn("/test/something/bin/jx", nil)

Now when we can setup our test using the mock interface and make assertions as normal.

Debug logging

Lots of the test have debug output to try figure out when things fail. You can enable verbose debug logging for tests via

export JX_TEST_DEBUG=true

Debugging

First you need to install Delve

Then you should be able to run a debug version of a jx command:

dlv --listen=:2345 --headless=true --api-version=2 exec ./build/jx -- some arguments

Then in you IDE you should be able to then set a breakpoint and connect to 2345.

e.g. in IntellJ you create a new Go Remote execution and then hit Debug

Debugging jx with stdin

If you want to debug using jx with stdin to test out terminal interaction, you can start jx as usual from the command line then:

  • find the pid of the jx command via something like ps -elaf | grep jx
  • start Delve attaching to the pid:
dlv --listen=:2345 --headless=true --api-version=2 attach SomePID

Debugging a unit test

You can run a single unit test via

export TEST="TestSomething"
make test1

You can then start a Delve debug session on a unit test via:

export TEST="TestSomething"
make debugtest1

Then set breakpoints and debug in your IDE like in the above debugging.

Using a helper script

If you create a bash file called jxDebug as the following (replacing SomePid with the actual pid):

#!/bin/sh
echo "Debugging jx"
dlv --listen=:2345 --headless=true --api-version=2 exec `which jx` -- $*

Then you can change your jx someArgs CLI to jxDebug someArgs then debug it!