go-test-table-driven（去测试表驱动）

Go Table-Driven Tests

Table-driven tests are a powerful Go idiom — when used correctly. The problem is that most codebases either underuse them (writing 10 copy-paste tests) or overuse them (jamming complex branching logic into a 200-line struct). This skill covers the sweet spot.

1. When Table-Driven Tests Shine

Use table tests when ALL of these are true:

• Same function under test across all cases
• Same assertion pattern — input goes in, output comes out, compare
• Cases differ only in data, not in setup or verification logic
• 3+ cases — fewer than 3, explicit subtests are clearer

The canonical use case: pure functions, parsers, validators, formatters.

func TestFormatCurrency(t *testing.T) {
    tests := []struct {
        name     string
        cents    int64
        currency string
        want     string
    }{
        {
            name:     "USD whole dollars",
            cents:    1000,
            currency: "USD",
            want:     "$10.00",
        },
        {
            name:     "USD with cents",
            cents:    1050,
            currency: "USD",
            want:     "$10.50",
        },
        {
            name:     "EUR formatting",
            cents:    999,
            currency: "EUR",
            want:     "€9.99",
        },
        {
            name:     "zero amount",
            cents:    0,
            currency: "USD",
            want:     "$0.00",
        },
        {
            name:     "negative amount",
            cents:    -500,
            currency: "USD",
            want:     "-$5.00",
        },
    }

    for _, tt := range tests {
        t.Run(tt.name, func(t *testing.T) {
            got := FormatCurrency(tt.cents, tt.currency)
            assert.Equal(t, tt.want, got)
        })
    }
}

Why this works: every case has the same shape, the loop body is 2 lines, and adding a new case is one struct literal. No branching, no conditionals.

2. When NOT to Use Table-Driven Tests

Complex per-case setup

If each case needs different mocks, different state, or different dependencies:

// ❌ Bad — table test with branching setup
tests := []struct {
    name        string
    setupMock   func(*mockStore)     // each case wires differently
    setupAuth   func(*mockAuth)      // more per-case wiring
    input       Request
    wantStatus  int
    shouldNotify bool                // branching assertion
}{...}

for _, tt := range tests {
    t.Run(tt.name, func(t *testing.T) {
        store := &mockStore{}
        tt.setupMock(store)          // hiding logic inside functions
        auth := &mockAuth{}
        tt.setupAuth(auth)
        // ... 20 lines of conditional assertions
    })
}

This is a code smell. The table is just hiding complexity behind function fields. Write separate subtests instead — they're longer but honest:

// ✅ Good — explicit subtests for different scenarios
func TestOrderHandler_Create(t *testing.T) {
    t.Run("succeeds with valid order", func(t *testing.T) {
        store := &mockStore{createFunc: func(...) (*Order, error) {
            return &Order{ID: "1"}, nil
        }}
        handler := NewHandler(store)
        // ... clear, readable, self-contained
    })

    t.Run("returns 401 when unauthenticated", func(t *testing.T) {
        handler := NewHandler(&mockStore{})
        // ... different setup, different assertions
    })
}

Fewer than 3 cases

Two cases don't need a table. The overhead of defining the struct is more code than just writing two tests:

// ❌ Overkill for 2 cases
tests := []struct {
    name    string
    input   string
    wantErr bool
}{
    {"valid", "hello", false},
    {"empty", "", true},
}

// ✅ Just write them
func TestValidate_AcceptsNonEmptyString(t *testing.T) {
    require.NoError(t, Validate("hello"))
}

func TestValidate_RejectsEmptyString(t *testing.T) {
    require.Error(t, Validate(""))
}

Multiple branching paths

If your loop body has if tt.shouldError / if tt.expectNotification / if tt.wantRedirect — you've outgrown the table. Each branch is a different test pretending to share a structure.

3. Struct Design

Keep fields minimal

Every field should change between at least 2 cases. If a field has the same value in all cases, it's not a variable — it's setup:

// ❌ Bad — userRole is "admin" in every case
tests := []struct {
    name     string
    userRole string  // always "admin"
    input    string
    want     string
}{
    {"case1", "admin", "a", "A"},
    {"case2", "admin", "b", "B"},
}

// ✅ Good — remove constants from the struct
func TestAdminFormatter(t *testing.T) {
    ctx := contextWithRole("admin") // shared setup, outside table

    tests := []struct {
        name  string
        input string
        want  string
    }{
        {"case1", "a", "A"},
        {"case2", "b", "B"},
    }
    // ...
}

Name the name field well

The name field appears in test output. Make it a short sentence that explains the scenario, not a label:

// ✅ Good names
{name: "trims leading whitespace"},
{name: "returns error for negative amount"},
{name: "handles unicode characters"},

// ❌ Bad names
{name: "case1"},
{name: "success"},
{name: "test with special chars"},

Use wantErr correctly

// ✅ Good — simple boolean for "should it error?"
tests := []struct {
    name    string
    input   string
    want    int
    wantErr bool
}{
    {name: "valid number", input: "42", want: 42},
    {name: "empty string", input: "", wantErr: true},
    {name: "not a number", input: "abc", wantErr: true},
}

for _, tt := range tests {
    t.Run(tt.name, func(t *testing.T) {
        got, err := ParseInt(tt.input)
        if tt.wantErr {
            require.Error(t, err)
            return
        }
        require.NoError(t, err)
        assert.Equal(t, tt.want, got)
    })
}

When you need to check specific errors

Use a wantErrIs field with a sentinel error, not just a boolean:

tests := []struct {
    name      string
    id        string
    wantErrIs error // nil means no error expected
}{
    {name: "valid id", id: "123"},
    {name: "empty id", id: "", wantErrIs: ErrInvalidID},
    {name: "not found", id: "999", wantErrIs: ErrNotFound},
}

for _, tt := range tests {
    t.Run(tt.name, func(t *testing.T) {
        _, err := store.GetByID(ctx, tt.id)
        if tt.wantErrIs != nil {
            require.ErrorIs(t, err, tt.wantErrIs)
            return
        }
        require.NoError(t, err)
    })
}

4. The Loop Body Must Be Trivial

The entire point of a table test is that the execution logic is identical for every case. If your loop body exceeds ~10 lines, something is wrong.

// ✅ Good — loop body is 5 lines
for _, tt := range tests {
    t.Run(tt.name, func(t *testing.T) {
        got, err := Process(tt.input)
        require.NoError(t, err)
        assert.Equal(t, tt.want, got)
    })
}

// ❌ Bad — loop body has become a mini-program
for _, tt := range tests {
    t.Run(tt.name, func(t *testing.T) {
        if tt.setupDB {
            db := setupDB(t)
            defer db.Close()
        }
        svc := NewService()
        if tt.withCache { svc.EnableCache() }
        got, err := svc.Process(tt.input)
        if tt.wantErr {
            require.Error(t, err)
            if tt.wantErrMsg != "" { assert.Contains(t, err.Error(), tt.wantErrMsg) }
            return
        }
        // ... more conditionals, more branches
    })
}

If you see this, split into separate subtests or separate test functions.

5. Parallel Table Tests

for _, tt := range tests {
    t.Run(tt.name, func(t *testing.T) {
        t.Parallel()
        got := Transform(tt.input)
        assert.Equal(t, tt.want, got)
    })
}

In Go 1.22+, the loop variable is scoped per iteration, so the old tt:= tt capture is unnecessary. For Go <1.22, you still need it:

// Go <1.22 only
for _, tt := range tests {
    tt := tt // capture range variable
    t.Run(tt.name, func(t *testing.T) {
        t.Parallel()
        // ...
    })
}

Only use t.Parallel() in table tests when the function under test has no side effects and no shared mutable state.

6. Readability Tricks

Align struct literals for scanning

tests := []struct {
    name  string
    input string
    want  string
}{
    {"lowercase",           "hello",       "hello"},
    {"uppercase",           "HELLO",       "hello"},
    {"mixed case",          "HeLLo",       "hello"},
    {"with spaces",         "Hello World", "hello world"},
    {"already lowercase",   "test",        "test"},
}

This works for simple cases. For complex structs, use the multi-line format:

tests := []struct {
    name   string
    config Config
    want   string
}{
    {
        name: "default timeout",
        config: Config{
            Host:    "localhost",
            Timeout: 0, // should get default
        },
        want: "localhost:8080",
    },
    {
        name: "custom port",
        config: Config{
            Host: "localhost",
            Port: 9090,
        },
        want: "localhost:9090",
    },
}

Map-based tables for ultra-simple cases

When the struct would just be {name, input, want}:

func TestStatusText(t *testing.T) {
    cases := map[string]struct {
        code int
        want string
    }{
        "ok":          {200, "OK"},
        "not found":   {404, "Not Found"},
        "server error": {500, "Internal Server Error"},
    }

    for name, tc := range cases {
        t.Run(name, func(t *testing.T) {
            assert.Equal(t, tc.want, StatusText(tc.code))
        })
    }
}

Note: map iteration order is random, so this also stress-tests that your cases are truly independent.

7. Error-Only Tables

When you're testing a validator and only care about which inputs fail:

func TestValidateEmail(t *testing.T) {
    valid := []string{
        "user@example.com",
        "user+tag@example.com",
        "user@sub.domain.com",
    }
    for _, email := range valid {
        t.Run("valid/"+email, func(t *testing.T) {
            require.NoError(t, ValidateEmail(email))
        })
    }

    invalid := []string{
        "",
        "@",
        "user@",
        "@domain.com",
        "user space@example.com",
    }
    for _, email := range invalid {
        t.Run("invalid/"+email, func(t *testing.T) {
            require.Error(t, ValidateEmail(email))
        })
    }
}

Two simple slices. No struct needed. The test name includes the input value, so failures are self-documenting.

8. Refactoring Bloated Tables

Signs your table test needs refactoring:

Decision Flowchart

1. Is the function pure (input → output, no side effects)? Yes → table test is probably ideal. Go to 2. No → consider explicit subtests first.
2. Do all cases share the exact same assertion pattern? Yes → table test. Go to 3. No → explicit subtests.
3. Can each case be expressed in ≤5 struct fields? Yes → table test. No → split by scenario into separate test functions.
4. Is the loop body ≤10 lines? Yes → you're golden. No → the table is hiding complexity. Refactor.

Verification Checklist

1. Table struct has only fields that vary between cases
2. Every case has a descriptive name field
3. Loop body is ≤10 lines with no branching
4. No setupFunc or mockFunc fields in the struct
5. wantErr is a simple bool or sentinel, not a string match
6. Cases cover: happy path, error path, edge cases (empty, nil, zero, max)
7. t.Run wraps each case for named subtests
8. t.Parallel() used only when function is side-effect-free