Simple, intuitive snapshot testing with Go 📸
Warning
snapshot is in early development and is not yet stable
Snapshot testing is where you assert the result of your code is identical to a specific reference value... which is basically all testing. If you've ever written:
if got != want {
t.Errorf("got %v, wanted %v", got, want)
}Then congratulations, you've done snapshot testing 🎉 In this case want is the snapshot.
The trick is, when these values get large or complicated (imagine a complicated JSON document), it's difficult to manually create and maintain the snapshot every time.
The next jump up is what's typically called "golden files".
These are files (typically manually created) that contain the expected output, any difference in what your code produces to what's in the file is an error.
Enter snapshot testing 📸
Think of snapshot testing as an automated, configurable, and simple way of managing golden files. All you need to do is call Snap and everything is handled for you!
go get github.com/FollowTheProcess/snapshot@latestimport (
"testing"
"github.com/FollowTheProcess/snapshot"
)
func TestSnapshot(t *testing.T) {
snap := snapshot.New(t)
snap.Snap([]string{"hello", "there", "this", "is", "a", "snapshot"})
// This will store the above slice in testdata/snapshots/TestSnapshot.snap.txt
// then all future checks will compare against this snapshot
}A few other libraries are out there for snapshot testing in Go, but they typically control the serialisation for you, using a generic object dumping library. This means you get what you get and there's not much option to change it.
Not very helpful if you want your snapshots to be as readable as possible!
With snapshot, you have full control over how your type is serialised to the snapshot file (if you need it). You can either:
- Let
snapshottake a best guess at how to serialise your type- With or without "filters" to tweak the result, see Filters below for more info
- Implement one of
snapshot.Snapper, json.Marshaler, encoding.TextMarshaler, or fmt.Stringer to override how it's serialised
See Serialisation Rules 👇🏻 for more info on how snapshot decides how to snap your value
Let's say you've got a bunch of snapshots saved already, and you change your implementation. All those snapshots will now likely need to change (after you've carefully reviewed the changes and decided they are okay!)
snapshot lets you do this with one line of configuration, which you can set with a test flag or environment variable, or however you like:
// something_test.go
import (
"testing"
"github.com/FollowTheProcess/snapshot"
)
var update = flag.Bool("update", false, "Update snapshots automatically")
func TestSomething(t *testing.T) {
// Tell snapshot to update everything if the -update flag was used
snap := snapshot.New(t, snapshot.Update(*update))
// .... rest of the test
}Tip
If you declare top level flags in a test file, you can pass them to go test. So in this case, go test -update would store true in the update var. You can also use environments variables and test them with os.Getenv e.g. UPDATE_SNAPSHOTS=true go test. Whatever works for you.
Warning
This will update all snapshots in one go, so make sure you run the tests normally first and check the diffs to make sure the changes are as expected
One criticism of snapshot testing is that if you restructure or rename your tests, you could end up with duplicated snapshots and/or messy unused ones cluttering up your repo. This is where the Clean option comes in:
// something_test.go
import (
"testing"
"github.com/FollowTheProcess/snapshot"
)
var clean = flag.Bool("clean", false, "Clean up unused snapshots")
func TestSomething(t *testing.T) {
// Tell snapshot to prune the snapshots directory of unused snapshots
snap := snapshot.New(t, snapshot.Clean(*clean))
// .... rest of the test
}This will erase all the snapshots currently managed by snapshot, and then run the tests as normal, creating the snapshots for all the new or renamed tests for the first time. The net result is a tidy snapshots directory with only what's needed
Snapshots are stored in a snapshots directory in the current package under testdata which is the canonical place to store test fixtures and other files of this kind, the go tool completely ignores testdata so you know these files will never impact your binary!
See go help test...
The go tool will ignore a directory named "testdata", making it available
to hold ancillary data needed by the tests.
The files will be named automatically after the test:
- Single tests will be given the name of the test e.g.
func TestMyThing(t *testing.T)will produce a snapshot file oftestdata/snapshots/TestMyThing.snap.txt - Sub tests (including table driven tests) will use the sub test name e.g.
testdata/snapshots/TestAdd/positive_numbers.snap.txt
Tip
If you want to split your snapshots with more granularity, you can name your table driven cases with a / in them (e.g. "Group/subtest name") and the directory hierarchy will be created automatically for you, completely cross platform!
See an example of this here
snapshot deals with plain text files as snapshots, this keeps them easy to read/write for both computers and humans. But crucially, easy to diff in pull request reviews!
Because of this, it needs to know how to serialise your value (which could be basically any valid construct in Go) to plain text, so we follow a few basic rules in priority order:
snapshot.Snapper: If your type implements theSnapperinterface, this is preferred over all other potential serialisation, this allows you to have total control over how your type is snapshotted, do whatever you like in theSnapmethod, just return a[]bytethat you'd like to look at in the snapshot and thats it!- json.Marshaler: If your type implements json.Marshaler, this will be used and the snapshot will be a valid JSON file (using
MarshalIndentfor readability) - encoding.TextMarshaler: If your type implements encoding.TextMarshaler, this will be used to render your value to the snapshot
- fmt.Stringer: If your type implements the fmt.Stringer interface, this is then used instead
- Primitive Types: Any primitive type in Go (
bool,int,stringetc.) is serialised according to the%vverb in the fmt package - Fallback: If your type hasn't been caught by any of the above rules, we will snap it using the
GoStringmechanism (the%#vprint verb) so e.g. a[]stringwould serialise as[]string{"one", "two", "three"}
Tip
snapshot effectively goes through this list top to bottom to discover how to serialise your type, so mechanisms at the top are preferentially chosen over mechanisms lower down. If your snapshot doesn't look quite right, consider implementing a method higher up the list to get the behaviour you need
Sometimes, your snapshots might contain data that is randomly generated like UUIDs, or constantly changing like timestamps, or that might change on different platforms like filepaths, temp directory names etc.
These things make snapshot testing annoying because your snapshot changes every time or passes locally but fails in CI on a windows GitHub Actions runner (my personal bane!). snapshot lets you add "filters" to your configuration to solve this.
Filters are simply regex replacements that you specify ahead of time, and if any of them appear in your snapshot, they can be replaced by whatever you want!
For example:
func TestWithFilters(t *testing.T) {
// Some common ones to give you inspiration
snap := snapshot.New(
t,
snapshot.Filter("(?i)[0-9a-f]{8}-[0-9a-f]{4}-[0-9a-f]{4}-[0-9a-f]{4}-[0-9a-f]{12}", "[UUID]"), // Replace uuids with the literal string "[UUID]"
snapshot.Filter(`\\([\w\d]|\.)`, "/$1"), // Replace windows file paths with unix equivalents
snapshot.Filter(`/var/folders/\S+?/T/\S+`, "[TEMP_DIR]"), // Replace a macos temp dir with the literal string "[TEMP_DIR]"
)
}But you can imagine more:
- Unix timestamps
- Go
time.Durationvalues - Other types of uuids, ulids etc.
If you can find a regex for it, you can filter it out!
This package was created with copier and the FollowTheProcess/go_copier project template.