| name | golang-testing |
| description | Write idiomatic Go tests. Use when this capability is needed. |
| metadata | {"author":"brpaz"} |
Golang Testing
Use this skill when writing, reviewing, or refactoring Go tests.
When to Use
- Writing new unit, integration, or package-level tests for Go code
- Refactoring tests toward idiomatic Go structure and stronger signal
- Reviewing Go test suites for maintainability, determinism, and coverage gaps
- Designing fakes, mocks, or test seams around external dependencies
Default Testing Stance
Unless there is a clear reason not to, prefer these defaults:
- Use black-box tests with an external test package:
package mypkg_test
- Test public behaviour and interfaces, not internal implementation details
- Use one top-level test function per public function/method/behaviour, with subtests for scenarios
- Call
t.Parallel() by default in top-level tests and subtests unless shared state, process-wide mutation, timing, or external resources make it unsafe
- Use
t.Run subtests by default, even when not using table-driven tests
- Use table-driven tests only when there is a finite, related set of input/output scenarios that benefits from shared structure
- Use
testify/require for preconditions and fatal assertions, and testify/assert for non-fatal assertions
- Use
testify/mock only for true external-system boundaries or very small/simple seams; otherwise prefer fakes or in-memory implementations
- Keep tests deterministic, hermetic, and readable
Core Principles
- Behaviour over implementation - Assert externally visible outcomes
- Fast feedback - Keep most tests cheap enough to run constantly
- Deterministic - Avoid sleeps, real clocks, network access, and ambient environment dependence
- Minimal setup - Prefer small helpers over deep fixture stacks
- High signal - Each failing test should point to one clear behaviour regression
- Coverage with intent - Cover important branches, edge cases, and error paths, not just lines
Package and File Conventions
Black-Box Test Packages
Prefer:
package user_test
import (
"testing"
"github.com/stretchr/testify/require"
"example.com/project/user"
)
Avoid:
package user
Use same-package tests only when you intentionally need access to unexported helpers and there is no better public seam. Default to _test packages.
File Organisation
- Group tests by public API surface or behaviour
- Keep helper functions near the tests that use them unless broadly reused
- Prefer
_test.go files named after the subject under test, such as service_test.go, handler_test.go, client_test.go
- Avoid giant catch-all files once a package grows
Preferred Test Structure
One Top-Level Test per Behaviour, Subtests per Scenario
For a public function, prefer one top-level test and scenario subtests instead of many separate TestXxx_Yyy functions.
Use t.Run even without a table when the scenarios are clearer as explicitly written examples.
func TestCreateUser(t *testing.T) {
t.Parallel()
t.Run("success", func(t *testing.T) {
t.Parallel()
got, err := user.Create("alice@example.com")
require.NoError(t, err)
assert.Equal(t, "alice@example.com", got.Email)
})
t.Run("rejects invalid email", func(t *testing.T) {
t.Parallel()
_, err := user.Create("not-an-email")
require.Error(t, err)
require.ErrorIs(t, err, user.ErrInvalidEmail)
})
}
func TestParseUserID(t *testing.T) {
t.Parallel()
tests := []struct {
name string
input string
want int
wantErr string
}{
{name: "valid", input: "42", want: 42},
{name: "empty", input: "", wantErr: "empty user id"},
{name: "invalid", input: "abc", wantErr: "invalid user id"},
}
for _, tt := range tests {
t.Run(tt.name, func(t *testing.T) {
t.Parallel()
got, err := user.ParseUserID(tt.input)
if tt.wantErr != "" {
require.Error(t, err)
assert.Contains(t, err.Error(), tt.wantErr)
return
}
require.NoError(t, err)
assert.Equal(t, tt.want, got)
})
}
}
Why This Pattern
- Keeps all scenarios for one behaviour together
- Makes it easy to add edge cases without duplicating setup
- Produces better output in
go test -run and CI logs
- Works naturally with table-driven tests and
t.Parallel()
Parallel Test Execution
Default Rule
Call t.Parallel() unless there is a concrete reason not to.
Good candidates:
- Pure functions
- Tests using isolated in-memory state
- Tests using
t.TempDir()
- Tests with per-test
httptest.Server
- Table-driven subtests with isolated fixtures
Avoid or carefully isolate parallelism when tests:
- Mutate global variables, singleton state, env vars, or process working directory
- Depend on wall-clock timing or
time.Sleep
- Reuse shared mocks/fakes with non-thread-safe state
- Use fixed ports, shared files, or shared databases without isolation
Parallel Subtest Safety
In Go 1.22+ modules, loop variables declared in the for statement are created per iteration, so rebinding like tt := tt is usually not needed.
Still rebind when:
- the module/package targets pre-Go-1.22 semantics
- the variable is declared outside the loop and reused
- you intentionally want compatibility with older Go module targets
For modern Go:
for _, tt := range tests {
t.Run(tt.name, func(t *testing.T) {
t.Parallel()
})
}
For older module targets:
for _, tt := range tests {
tt := tt
t.Run(tt.name, func(t *testing.T) {
t.Parallel()
})
}
Testify Usage
Prefer Testify for clearer assertions.
import (
"testing"
"github.com/stretchr/testify/assert"
"github.com/stretchr/testify/require"
)
require vs assert
- Use
require when the test cannot continue after failure
- Use
assert when you want multiple related checks in one scenario
func TestUserName(t *testing.T) {
t.Parallel()
u, err := user.New("alice")
require.NoError(t, err)
assert.Equal(t, "alice", u.Name())
assert.True(t, u.IsActive())
}
Prefer semantic assertions when available:
require.ErrorIs
assert.Equal
assert.ElementsMatch
assert.Len
assert.Empty / assert.NotEmpty
assert.WithinDuration
assert.JSONEq
Error Testing
- Prefer asserting exported sentinel/type behaviour with
ErrorIs / ErrorAs
- Avoid brittle full-string equality unless the exact message is part of the contract
- Assert both success and failure paths for public APIs
require.ErrorIs(t, err, user.ErrNotFound)
Good Coverage Practices
Aim for coverage that reflects risk and behaviour, not vanity percentages.
Always Cover
- Happy path
- Input validation failures
- Boundary values
- Zero values and empty collections
- Domain-specific edge cases
- External dependency failures
- Serialization/parsing failures
- Context cancellation / timeout paths when relevant
- Idempotency or retry semantics when relevant
Coverage Heuristics
- Add a test for every bug fix before or with the fix
- Cover each exported public method/function with at least one success and one failure/edge scenario where meaningful
- Prefer focused unit tests for combinatorial logic and a smaller number of integration tests for end-to-end confidence
- If a branch matters to production behaviour, it deserves an assertion
- Remove duplicate tests that do not add behavioural signal
Coverage Commands
go test ./...
go test ./... -cover
go test ./... -coverprofile=coverage.out
go tool cover -func=coverage.out
go tool cover -html=coverage.out
Integration vs Unit Tests
Use the lightest test that proves the behaviour.
Prefer Unit Tests For
- Pure transformations and validation
- Orchestration logic with fake dependencies
- Error mapping and branching logic
Prefer Integration Tests For
- SQL queries and repository behaviour
- HTTP handlers, middleware, and routing
- Serialization boundaries
- Interactions with real adapters that are cheap and deterministic to run locally/CI
When an integration test gives stronger confidence with similar complexity, prefer it over elaborate mocking.
Mocking Guidance
Default Rule
Mock only external systems or very thin boundaries. Prefer fakes and in-memory implementations for richer domain behaviour.
Prefer:
- In-memory repositories
httptest.Server
- Temporary directories
fstest.MapFS
- Fixed clocks / injectable time sources
Reach for testify/mock when:
- The dependency is an external system boundary
- A fake would be more complex than the behaviour being tested
- You need precise call assertions on a thin collaborator
Avoid mocks for:
- Your own core domain models
- Simple data containers
- Behaviour that is easier to express with an in-memory fake
- Deep implementation-driven interaction testing
Testify Mock Pattern
type MockMailer struct {
mock.Mock
}
func (m *MockMailer) Send(ctx context.Context, msg mail.Message) error {
args := m.Called(ctx, msg)
return args.Error(0)
}
func TestService_SendWelcomeEmail(t *testing.T) {
t.Parallel()
mailer := new(MockMailer)
mailer.
On("Send", mock.Anything, mail.Message{To: "alice@example.com"}).
Return(nil).
Once()
svc := user.NewService(mailer)
err := svc.SendWelcomeEmail(context.Background(), "alice@example.com")
require.NoError(t, err)
mailer.AssertExpectations(t)
}
Rules:
- One mock per external dependency seam, not everywhere
- Keep expectations minimal and behaviour-focused
- Avoid over-asserting call order unless it is part of the contract
- Do not share mock instances across parallel tests
Filesystem Testing
Prefer standard-library-friendly seams and hermetic storage.
Recommended Patterns
- Inject an
fs.FS for read-only filesystem behaviour
- Use
fstest.MapFS for small read scenarios
- Use
t.TempDir() for realistic file creation/update flows
- Keep filesystem access behind a small adapter if the production code is otherwise hard to isolate
fstest.MapFS Example
func TestLoadConfig(t *testing.T) {
t.Parallel()
files := fstest.MapFS{
"config.json": {Data: []byte(`{"env":"test"}`)},
}
cfg, err := config.Load(files, "config.json")
require.NoError(t, err)
assert.Equal(t, "test", cfg.Env)
}
t.TempDir() Example
func TestWriteReport(t *testing.T) {
t.Parallel()
dir := t.TempDir()
path := filepath.Join(dir, "report.txt")
err := report.Write(path, "hello")
require.NoError(t, err)
data, err := os.ReadFile(path)
require.NoError(t, err)
assert.Equal(t, "hello", string(data))
}
Avoid mocking os calls directly when a temp dir or fs.FS seam would be simpler.
HTTP Client Testing
Prefer real HTTP semantics without real network dependencies.
Recommended Patterns
- Use
httptest.Server to test client behaviour against realistic responses
- Inject
*http.Client into your code
- For very small seams, a custom
http.RoundTripper fake can be enough
- Mock higher-level gateways only when HTTP itself is not the thing under test
httptest.Server Example
func TestClient_GetUser(t *testing.T) {
t.Parallel()
srv := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
assert.Equal(t, http.MethodGet, r.Method)
assert.Equal(t, "/users/42", r.URL.Path)
w.Header().Set("Content-Type", "application/json")
_, _ = w.Write([]byte(`{"id":42,"name":"alice"}`))
}))
t.Cleanup(srv.Close)
client := api.NewClient(srv.URL, srv.Client())
got, err := client.GetUser(context.Background(), 42)
require.NoError(t, err)
assert.Equal(t, "alice", got.Name)
}
Custom Transport Fake
For narrow client logic, prefer a tiny fake over a heavyweight mock:
type roundTripFunc func(*http.Request) (*http.Response, error)
func (f roundTripFunc) RoundTrip(r *http.Request) (*http.Response, error) {
return f(r)
}
Use this when you only need to simulate one request/response pair and httptest.Server would be unnecessary ceremony.
Time Testing
Do not depend on real time in tests.
Recommended Patterns
- Inject time via a function like
now func() time.Time
- Or define a tiny clock interface at the boundary that needs it
- Use fixed timestamps in tests
- Avoid
time.Sleep as a synchronisation mechanism
- Prefer contexts, channels, and explicit signals over waiting for elapsed time
Function Injection Example
type Service struct {
now func() time.Time
}
func NewService() *Service {
return &Service{now: time.Now}
}
func TestTokenExpired(t *testing.T) {
t.Parallel()
fixed := time.Date(2025, 1, 1, 12, 0, 0, 0, time.UTC)
svc := token.NewService(func() time.Time { return fixed })
assert.True(t, svc.IsExpired(fixed.Add(-time.Second)))
assert.False(t, svc.IsExpired(fixed.Add(time.Second)))
}
If production code currently calls time.Now() directly in many places, first introduce a small seam at the package boundary rather than mocking time per call site.
Table-Driven Testing Guidance
- Prefer plain
t.Run subtests first without a table when scenarios are clearer as explicitly written examples
- Use tables when the setup is mostly the same and only inputs/expectations vary
- Name cases for behaviour, not raw input values
- Keep test case structs focused; avoid giant anonymous structs with dozens of fields
- Split unrelated behaviours into separate tests instead of massive tables
Good fit for table-driven tests:
- Parsing and validation matrices
- Boundary-value checks on the same function
- Multiple related input/output combinations for deterministic logic
Poor fit for table-driven tests:
- Scenarios with materially different setup or assertions
- Behaviour tests that read more clearly as individually named
t.Run blocks
- Large tables that hide intent behind many mostly-empty fields
Test Helpers
- Prefer helper functions that build valid defaults and allow explicit overrides
- Mark helpers with
t.Helper()
- Return concrete values rather than hidden global state
- Keep helpers local to the package unless broad reuse justifies promotion
func newTestUser(t *testing.T, opts ...UserOption) user.User {
t.Helper()
u, err := user.New("alice", opts...)
require.NoError(t, err)
return u
}
Anti-Patterns
Avoid these unless there is a strong reason:
- Same-package white-box tests by default
- Multiple top-level tests for each small scenario of one function
- Reaching for table-driven tests when explicit
t.Run scenarios are clearer
- No
t.Parallel() without reason
- Real sleeps, real network calls, or dependence on current wall-clock time
- Assertion on internal private fields in black-box tests
- Brittle exact error string checks for wrapped/domain errors
- Massive shared fixtures that hide what the test actually needs
- Mock-heavy tests that mirror implementation rather than behaviour
- Coverage chasing without meaningful assertions
Review Checklist
When reviewing Go tests, check for:
- Black-box
_test package usage by default
t.Parallel() in top-level tests and subtests unless unsafe
- Subtests for multiple scenarios of one behaviour
testify/assert and testify/require used appropriately
- Fakes preferred over mocks when practical
testify/mock limited to external boundaries or simple seams
- No hidden global state or cross-test coupling
- Deterministic handling of filesystem, HTTP, and time
- Meaningful edge-case and error-path coverage
- Tests that read like executable specifications of public behaviour
Source: brpaz/agent-skills — distributed by TomeVault.