subtest is a minimalist Go test-utility package used to initializing small test functions for use with the Go sub-tests feature. You can read more about Go sub-tests here.
The sub-package subjson defines middleware for parsing values from JSON before performing checks.
subtest was motivated by a desire to make it easier write Given-When-Then (GWT) style tests in Go on top of the built-in test-runner, and without a DSL.
GWT is a naming schema for tests that attempts to clarify three aspects:
- Given: How does the world look like before we do an action.
- When: What do we do to affect the world.
- Then: How should the world look like after the action.
These conditions can be nested to test different scenarios. Here is an example:
Given foo is 42
When dividing by 6
Then the result should be 7
When dividing by 9
Then the result should be less than 5
Then the result should be more than 4
A common believe is that to write GWT style tests in Go, you should use a Behavior-Driven-Development framework and associated Domain-Specific-Language. However, we will argue that this simply isn't true.
One of the problems with many Behavior-Driven-Development frameworks in Go, is that they tend to rely on their own test-runner and sub-test logic. While there could be good reasons for using this, it also comes with a price: tooling expecting the default test runner simply does not cope. This is true either we are talking about a CI that fail to parse sub-test results into JUnit summaries, an IDE that fail to insert links for navigating to the failing code, or a command-line tool for rerunning failing sub-tests by name. This has a real impact on how well test-results can be understood when used with tooling.
As it turns out, you can actually write GWT-style tests not only without a BDD framework or DSL, but without a framework or library what so ever:
func TestFoo(t *testing.T) {
t.Run("Given foo is 42", func(t *testing.T) {
const foo = 42
t.Run("When dividing by 6", func (t *testing.T) {
v := float64(foo) / 6
t.Run("Then the result should be 7", func(t *testing.T) {
if expect := float64(7); v != expect {
t.Fatalf("\n got: %d\nwant: %d", v, expect)
}
})
})
t.Run("When dividing by 9", func (t *testing.T) {
v := float64(foo) / 9
t.Run("Then the result should be greater than 4" func(t *testing.T) {
if expect := float64(4); v > expect {
t.Fatalf("\n got: %d\nwant > %d", v, expect)
}
})
t.Run("Then the result should be less than 5" func(t *testing.T) {
if expect := float64(5); v < expect {
t.Fatalf("\n got: %d\nwant < %d", v, expect)
}
})
})
})
}While doing this is fine, it can quickly become repetitive. It can also become challenging to maintain consistent output for failing tests over time; in particularly so for a growing team.
By now you might think that subtest is going to improve how to write GWT style tests without a BDD-style framework, and you are right. However, there is nothing within the design of subtest that restricts it to handling GWT style tests. Instead, subtest is a generalized test utility package for generating sub-tests, no matter your style preferences. In other words, you can use subtest to write GWT style tests, table-driven tests, or some other style that you prefer.
Here is a version of TestFoo with subtest:
func TestFoo(t *testing.T) {
t.Run("Given foo is 42", func(t *testing.T) {
const foo = 42
t.Run("When dividing by 6", func (t *testing.T) {
vf := subtest.Value(float64(foo) / 6)
t.Run("Then the result should be 7", vf.NumericEqual(7))
})
t.Run("When dividing by 9", func (t *testing.T) {
vf := subtest.Value(float64(foo) / 6)
t.Run("Then the result should be greater than 4", vf.GreaterThan(4))
t.Run("Then the result should be less than 5", vf.LessThan(5))
})
})
}Building and running a subtest is generally composed of six steps. Normally you do not do each steps as explicit as described below, but to illustrate the general flow, we have spelled this out.
// TestFooExplicit shows the different steps of building a subtest.
func TestFooExplicit(t *testing.T) {
// 1. We declare the value we want to check; usually the result of an
// operation or action.
v := "foo"
// 2. We initialize a value function for the value that we want to check.
// There are several different initializers we can call to get a value
// function; this is the simplest one.
vf := subtest.Value(v)
// 3. We initialize the check we want to use. A check is anything that
// implements the Check interface.
c := subtest.NumericEquals(3)
// 4. We can optionally wrap our check with middleware.
c = subtest.OnLen(c)
// 5. We initialize a test function by passing the check to the value
// function's Test method.
tf := vf.Test(c)
// 6. We run the test function as a sub-test.
t.Run("len(v) == 3", tf)
}If we where going to do this every time, we would grow weary. Therefore there is several short-hand methods defined on the Check and ValueFunc instances that makes things easier. The least verbose variant we can write of the test above is as follows:
func TestFoo(t *testing.T) {
v := "foo"
t.Run("len(v) == 3", subtest.Len(v).NumericEquals(3))
}It is possible to validate more than just equality with subtest. The subtest.Schema type allows advanced validation of any Go map type, and in the future, perhaps also for structs. From the subjson package we can use ValueFunc initializers, Check implementations and check middleware to decode JSON from string, []byte and json.RawMessage values. Combining these two mechanisms we can do advanced validation of JSON content.
func TestJSONMap(t *testing.T) {
v := `{"foo": "bar", "bar": "foobar", "baz": ["foo", "bar", "baz"]}`
expect := subtest.Schema{
Fields: subtest.Fields{
"foo": subjson.DecodesTo("bar")
"bar": subjson.OnLen(subtest.AllOf{
subtest.GreaterThan(3),
subtest.LessThan(8),
}),
"baz": subjson.OnSlice(subtest.AllOf{
subtest.OnLen(subtest.DeepEqual(3)),
subtest.OnIndex(0, subjson.DecodesTo("foo"),
subtest.OnIndex(1, subtest.MatchPattern(`"^b??$"`), // regex match against raw JSON
subtest.OnIndex(2, subtest.DeepEqual(json.RawMessage(`"baz"`)), // raw JSON equals
}),
},
}
t.Run("match expectations", subjson.Map(v).Test(expect))
}This is perhaps not commonly known, but the t.Run function actually return false if there is a failure. Or to be more accurate:
Run reports whether f succeeded (or at least did not fail before calling t.Parallel).
Because subtest checks do not call t.Parallel, this can be utilized to stop test-execution if a "required" sub-test fails.
func TestFoo(t *testing.T) {
v, err := foo()
if !t.Run("err == nil", subtest.Value(err).NoError()) {
// Abort further tests if failed.
t.FailNow()
}
// Never run when err != nil.
t.Run("v == foo", subtest.Value(v).DeepEqual("foo"))
}
func foo() (string, error) {
return "", errors.New("failed")
}The subtest library itself is currently zero-dependencies. The important aspect of this is that we do not force opinionated dependencies on the user. However, it's also written to be relatively easy to extend.
For specialized use cases and customization, see the examples/ sub-directory:
examples/gwt: Example of tests following the Given-When-Then naming convention.examples/colorfmt: Example of custom type formatting with colors via the pp package.examples/gojsonq: Example of custom checks for JSON matching via the gojsonq package.examples/jsondiff: Example of custom checks for JSON comparison via the jsondiff package.
Some key features of subtest is described below.
subtest initializes test functions intended for usage with the Run method on the testing.T type, and uses a plain output format by default. This means that tooling and IDE features built up around output from the standard test runner will work as expected.
Values to check are wrapped in a value function (ValueFunc). By setting up your own value function, you can easily run several tests against state-ful types, such as an io.Reader, where each check starts with
a clean slate.
Generally, a sub-test performs of a single check (CheckFunc). These checks can be wrapped by middleware to facilitate processing or transformation of values before running nested checks. E.g. parse a byte array from JSON into a Go type, or extract the length of an array.
The quicker a failed test can be understood, the quicker it can be fixed. subtest's default failure formatting is inspired by the short and simplistic style used for unit tests within the Go standard library. We have extended this syntax only so that we can more easily format the expected type and value.
Example output from an exaples/gwt:
--- FAIL: TestFoo (0.00s)
--- FAIL: TestFoo/Given_nothing_is_registered (0.00s)
--- FAIL: TestFoo/Given_nothing_is_registered/When_calling_reg.Foo (0.00s)
--- FAIL: TestFoo/Given_nothing_is_registered/When_calling_reg.Foo/Then_the_result_should_hold_a_zero-value (0.00s)
pkg_test.go:19: not deep equal
got: string
"oops"
want: string
""
FAIL
FAIL github.com/clarify/subtest/examples/gwt 0.057s
FAIL
Be aware that the default type formatter currently do not expand nested pointer values.
While we aim to make the default type formatting useful, it will also be somewhat limited due to our zero-dependency goal. Type formatting is also an area with potential for different opinions on what looks the most clear. For this reason we have made it easy to replace the default type formatter using libraries such as go-spew, litter, or pp (with colors).
Example using go-spew:
import (
"github.com/davecgh/go-spew/spew"
"github.com/clarify/subtest"
)
func init() {
subtest.SetTypeFormatter(spew.ConfigState{Indent: "\t"}.Sdump)
}Example using litter:
import (
"github.com/sanity-io/litter"
"github.com/clarify/subtest"
)
func init() {
subtest.SetTypeFormatter(litter.Options{}.Sdump)
}Example using pp with conditional coloring:
import (
"golang.org/x/crypto/ssh/terminal"
"github.com/k0kubun/pp"
"github.com/clarify/subtest"
)
func init() {
subtest.SetTypeFormatter(pp.Sprint)
colorEnv := strings.ToUpper(os.Getenv("GO_TEST_COLOR"))
switch colorEnv {
case "0", "FALSE":
log.Println("explicitly disabling color output for test")
pp.ColoringEnabled = false
case "1", "TRUE":
log.Println("explicitly enabling color output for test")
default:
if !terminal.IsTerminal(int(os.Stdout.Fd())) {
log.Println("TTY not detected, disabling color output for test")
pp.ColoringEnabled = false
} else {
log.Println("TTY detected, enabling color output for test")
}
}
}When it comes to prettifying the output of the test runner itself, there are separate tools for that. One such tool is gotestsum, which wraps the Go test runner to provide alternate formatting.