autogold - automatically update your Go tests

autogold makes go test -update automatically update your Go tests (golden files and Go values in e.g. foo_test.go).

~5m introduction available on YouTube:

"It's 2021: you shouldn't have to update Go tests manually"

Installation

go get -u github.com/hexops/autogold/v2

Automatic golden files

Write in a Go test:

import "github.com/hexops/autogold/v2"
...
autogold.ExpectFile(t, got)

go test -update will now create/update a testdata/<test name>.golden file for you automatically. If your tests change over time you can use go test -update -clean to also have it remove unused golden files.

Automatic inline test updating

In a Go test, simply call autogold.Expect(want).Equal(t, got), passing nil as the value you want initially:

func TestFoo(t *testing.T) {
	...
	autogold.Expect(nil).Equal(t, got)
}

Run go test -update and autogold will automatically update the autogold.Expect(want) Go syntax with the actual value your test got. It works with complex Go structs, slices, strings, etc.

Diffs

Anytime your test produces a result that is unexpected, you'll get a nice diff showing exactly what changed. It does this by converting values at runtime directly to a formatted Go AST, and using the same diffing library the Go language server uses:

--- FAIL: TestFoo (0.08s)
    autogold.go:91: mismatch (-want +got):
        --- want
        +++ got
        @@ -1 +1 @@
        +&example.Baz{Name: "Jane", Age: 31}

Subtesting

Table-driven Go subtests are supported nicely as you can call .Equal(got) later, so that go test -update will update your table-driven test values defined earlier for you:

func TestTime(t *testing.T) {
	testCases := []struct {
		gmt    string
		loc    string
		expect autogold.Value // autogold: the value we expect
	}{
		{"12:31", "Europe/Zuri", autogold.Expect(nil)},
		{"12:31", "America/New_York", autogold.Expect(nil)},
		{"08:08", "Australia/Sydney", autogold.Expect(nil)},
	}
	for _, tc := range testCases {
		t.Run(tc.loc, func(t *testing.T) {
			loc, err := time.LoadLocation(tc.loc)
			if err != nil {
				t.Fatal("could not load location")
			}
			gmt, _ := time.Parse("15:04", tc.gmt)
			got := gmt.In(loc).Format("15:04")

			tc.expect.Equal(t, got) // autogold: tell it the value our test produced
		})
	}
}

It works by finding the relevant autogold.Expect(want) call for you based on callstack information / matching line number in the file, and then rewrites the nil parameter (or any other value that was there.)

What are golden files, when should they be used?

Golden files are used by the Go authors for testing the standard library, the gofmt tool, etc. and are a common pattern in the Go community for snapshot testing. See also "Testing with golden files in Go" - Chris Reeves

Golden files make the most sense when you'd otherwise have to write a complex multi-line string or large Go structure inline in your test, making it hard to read.

In most cases, you should prefer inline snapshots, subtest golden values, or traditional Go tests.

Command line syntax: put `-update` at the end

-update should go at the end of your go test command, otherwise for some reason stdout will be considered a terminal and color will be turned on for libraries like fatih/color. Example:

go test -count=1 -run TestSomething . -update

Custom formatting

valast is used to produce Go syntax at runtime for the Go value you provide. If the default output is not to your liking, you have options:

Pass a string to autogold: It will be formatted as a Go string for you in the resulting .golden file / in Go tests.
Use your own formatting (JSON, etc.): Make your got value of type autogold.Raw("foobar"), and it will be used as-is for .golden files (not allowed with inline tests.)
Exclude unexported fields: autogold.ExpectFile(t, got, autogold.ExportedOnly())

Backwards compatibility

As is the case with gofmt, different Go versions may produce different formattings (although rare.)
Minor versions of autogold (e.g. v1.0, v1.1) may alter the formatting of .golden files, although we will be mindful of such changes.
Major versions of autogold (e.g. v1, v2) will be used for any major changes in output that would be difficult to review (we expect this will be rare in practice.)

Alternatives comparison

The following are alternatives to autogold, making note of the differences we found that let us to create autogold:

github.com/xorcare/golden
- Supports []byte inputs only, defers formatting to users.
- Does not support inline snapshots / code updating.
github.com/sebdah/goldie
- Supports []byte inputs only, provides helpers for JSON, XML, etc.
- Does not support inline snapshots / code updating.
github.com/bradleyjkemp/cupaloy
- Works on interface{} inputs.
- Uses inactive go-spew project to format Go structs.
- Does not support inline snapshots / code updating.

Changelog

v2.2.1

Updated to valast v1.4.4:

valast.Addr is replaced by valast.Ptr, which uses Go generics and looks cleaner.
time.Time values are now supported.

v2.2.0

If autogold is used in packages with an -update flag already defined, now no conflict occurs. This enables autogold to be used with other 'golden' packages without conflict.
Fixed an issue where _test packages using types from non-test packages would sometimes result in the wrong package name qualifier.

v2.1.0

Added support for building in Bazel / working around a bug in Bazel / Go's packages.Load functionality. This feature can be enabled using ENABLE_BAZEL_PACKAGES_LOAD_HACK=true. For more details see #40 and golang/go#57304

v2.0.3

Fixed an issue where updating inline tests could cause a deadlock.

v2.0.2

Writing a unique name with inline tests is no longer required. Previously you must write a unique name as the first parameter to Want and it must have been inside a TestFoo function for autogold to find it:

func TestFoo(t *testing.T) {
	...
	autogold.Want("unique name inside TestFoo", want).Equal(t, nil)
}

This can be rewritten as autogold.Expect(want).Equal(t, got) and no unique name is required, the function call can be placed anywhere inside your Go test file as autogold will now update the invocation based on callstack information:

func TestFoo(t *testing.T) {
	...
	autogold.Expect(want).Equal(t, nil)
}

Additionally, CLI flag behavior has been improved substantially based on experience working in very large enterprise Go codebases:

-update now behaves like -update-only, it no longer removes unused golden files which is faster in very large codebases. Instead, you may use -update -clean to remove unused golden files. -update-only is removed.
Previously autogold would fail tests when running -update, meaning you may need to run go test -update many times to get to your desired end-state if updating a lot of test cases. Now we match the behavior of OCaml expect tests in not failing tests by default (you can now specify -fail-on-update)
-fail-on-update now uses t.FailNow() instead of t.Fatal() to allow as many tests as possible to succeed when performing an update operation.
autogold.Want has been deprecated in favor of autogold.Expect
Fixed invalid cross-device link errors on some systems.

Finally, please note the renaming of functions before and after:

Inline tests: autogold.Want -> autogold.Expect
File tests: autogold.Equal -> autogold.ExpectFile

Automating the migration with Comby

You can automatically migrate from v1 to v2 using the following Comby configuration:

autogold.comby.toml

# autogold.comby.toml
[update-imports]

match="\"github.com/hexops/autogold\""
rewrite="\"github.com/hexops/autogold/v2\""

[update-api-want]

match="autogold.Want(:[desc], :[v])"
rewrite="autogold.Expect(:[v])"

[update-api-equal]

match="autogold.Equal(:[v])"
rewrite="autogold.ExpectFile(:[v])"

Assuming Comby is available on your system, you can run the following command to apply the changes:

$ go get -u github.com/hexops/autogold/v2
$ comby -config autogold.comby.toml -matcher .go -exclude-dir vendor,node_modules -in-place
$ go mod tidy
$ git diff # show the changes applied by comby

v1.3.1

Improved Go code formatting (updated valast and gofumpt versions)
Added usage of t.Helper to improve line:column information of test failures.
Fixed an issue where -update subtest names could collide and incorrectly fail tests.
Fixed a data race when -update is used
Diffs are now printed with ANSII color codes

hexops/autogold