Files
gh-action-readme/testutil/testutil_test.go

931 lines
26 KiB
Go

package testutil
import (
"context"
"fmt"
"io"
"net/http"
"os"
"path/filepath"
"strings"
"testing"
"time"
)
// TestMockHTTPClient tests the MockHTTPClient implementation.
func TestMockHTTPClient(t *testing.T) {
t.Parallel()
t.Run("returns configured response", func(t *testing.T) {
t.Parallel()
testMockHTTPClientConfiguredResponse(t)
})
t.Run("returns 404 for unconfigured endpoints", func(t *testing.T) {
t.Parallel()
testMockHTTPClientUnconfiguredEndpoints(t)
})
t.Run("tracks requests", func(t *testing.T) {
t.Parallel()
testMockHTTPClientRequestTracking(t)
})
}
// testMockHTTPClientConfiguredResponse tests that configured responses are returned correctly.
func testMockHTTPClientConfiguredResponse(t *testing.T) {
t.Helper()
client := createMockHTTPClientWithResponse("GET https://api.github.com/test", 200, `{"test": "response"}`)
req := createTestRequest(t, "GET", ""+TestURLGitHubAPI+"test")
resp := executeRequest(t, client, req)
defer func() { _ = resp.Body.Close() }()
validateResponseStatus(t, resp, 200)
validateResponseBody(t, resp, `{"test": "response"}`)
}
// testMockHTTPClientUnconfiguredEndpoints tests that unconfigured endpoints return 404.
func testMockHTTPClientUnconfiguredEndpoints(t *testing.T) {
t.Helper()
client := &MockHTTPClient{
Responses: make(map[string]*http.Response),
}
req := createTestRequest(t, "GET", ""+TestURLGitHubAPI+"nonexistent")
resp := executeRequest(t, client, req)
defer func() { _ = resp.Body.Close() }()
validateResponseStatus(t, resp, 404)
}
// testMockHTTPClientRequestTracking tests that requests are tracked correctly.
func testMockHTTPClientRequestTracking(t *testing.T) {
t.Helper()
client := &MockHTTPClient{
Responses: make(map[string]*http.Response),
}
req1 := createTestRequest(t, "GET", ""+TestURLGitHubAPI+"test1")
req2 := createTestRequest(t, "POST", ""+TestURLGitHubAPI+"test2")
executeAndCloseResponse(client, req1)
executeAndCloseResponse(client, req2)
validateRequestTracking(t, client, 2, ""+TestURLGitHubAPI+"test1", "POST")
}
// createMockHTTPClientWithResponse creates a mock HTTP client with a single configured response.
func createMockHTTPClientWithResponse(key string, statusCode int, body string) *MockHTTPClient {
return &MockHTTPClient{
Responses: map[string]*http.Response{
key: {
StatusCode: statusCode,
Body: io.NopCloser(strings.NewReader(body)),
},
},
}
}
// createTestRequest creates an HTTP request for testing purposes.
func createTestRequest(t *testing.T, method, url string) *http.Request {
t.Helper()
req, err := http.NewRequest(method, url, nil)
if err != nil {
t.Fatalf("failed to create request: %v", err)
}
return req
}
// executeRequest executes an HTTP request and returns the response.
func executeRequest(t *testing.T, client *MockHTTPClient, req *http.Request) *http.Response {
t.Helper()
resp, err := client.Do(req)
if err != nil {
t.Fatalf(TestErrUnexpected, err)
}
return resp
}
// executeAndCloseResponse executes a request and closes the response body.
func executeAndCloseResponse(client *MockHTTPClient, req *http.Request) {
if resp, _ := client.Do(req); resp != nil {
_ = resp.Body.Close()
}
}
// validateResponseStatus validates that the response has the expected status code.
func validateResponseStatus(t *testing.T, resp *http.Response, expectedStatus int) {
t.Helper()
if resp.StatusCode != expectedStatus {
t.Errorf("expected status %d, got %d", expectedStatus, resp.StatusCode)
}
}
// validateResponseBody validates that the response body matches the expected content.
func validateResponseBody(t *testing.T, resp *http.Response, expected string) {
t.Helper()
body, err := io.ReadAll(resp.Body)
if err != nil {
t.Fatalf("failed to read response body: %v", err)
}
if string(body) != expected {
t.Errorf("expected body %s, got %s", expected, string(body))
}
}
// validateRequestTracking validates that requests are tracked correctly.
func validateRequestTracking(
t *testing.T,
client *MockHTTPClient,
expectedCount int,
expectedURL, expectedMethod string,
) {
t.Helper()
if len(client.Requests) != expectedCount {
t.Errorf("expected %d tracked requests, got %d", expectedCount, len(client.Requests))
return
}
if client.Requests[0].URL.String() != expectedURL {
t.Errorf("unexpected first request URL: %s", client.Requests[0].URL.String())
}
if len(client.Requests) > 1 && client.Requests[1].Method != expectedMethod {
t.Errorf("unexpected second request method: %s", client.Requests[1].Method)
}
}
func TestMockGitHubClient(t *testing.T) {
t.Parallel()
t.Run("creates client with mocked responses", func(t *testing.T) {
t.Parallel()
responses := map[string]string{
"GET https://api.github.com/repos/test/repo": `{"name": "repo", "full_name": "test/repo"}`,
}
client := MockGitHubClient(responses)
if client == nil {
t.Fatal("expected client to be created")
}
// Test that we can make a request (this would normally hit the API)
// The mock transport should handle this
ctx := context.Background()
_, resp, err := client.Repositories.Get(ctx, "test", "repo")
if err != nil {
t.Fatalf(TestErrUnexpected, err)
}
if resp.StatusCode != http.StatusOK {
t.Errorf(TestErrStatusCode, resp.StatusCode)
}
})
t.Run("uses MockGitHubResponses", func(t *testing.T) {
t.Parallel()
responses := MockGitHubResponses()
client := MockGitHubClient(responses)
// Test a specific endpoint that we know is mocked
ctx := context.Background()
_, resp, err := client.Repositories.Get(ctx, "actions", "checkout")
if err != nil {
t.Fatalf(TestErrUnexpected, err)
}
if resp.StatusCode != http.StatusOK {
t.Errorf(TestErrStatusCode, resp.StatusCode)
}
})
}
func TestMockTransport(t *testing.T) {
t.Parallel()
client := &MockHTTPClient{
Responses: map[string]*http.Response{
"GET https://api.github.com/test": {
StatusCode: 200,
Body: io.NopCloser(strings.NewReader(`{"success": true}`)),
},
},
}
transport := &MockTransport{Client: client}
req, err := http.NewRequest(http.MethodGet, ""+TestURLGitHubAPI+"test", nil)
if err != nil {
t.Fatalf("failed to create request: %v", err)
}
resp, err := transport.RoundTrip(req)
if err != nil {
t.Fatalf(TestErrUnexpected, err)
}
defer func() { _ = resp.Body.Close() }()
if resp.StatusCode != http.StatusOK {
t.Errorf(TestErrStatusCode, resp.StatusCode)
}
}
func TestTempDir(t *testing.T) {
t.Parallel()
t.Run("creates temporary directory", func(t *testing.T) {
t.Parallel()
dir, cleanup := TempDir(t)
defer cleanup()
// Verify directory exists
if _, err := os.Stat(dir); os.IsNotExist(err) {
t.Error("temporary directory was not created")
}
// Verify it's in temp location
if !strings.Contains(dir, os.TempDir()) && !strings.Contains(dir, "/tmp") {
t.Errorf("directory not in temp location: %s", dir)
}
// Verify directory name pattern (t.TempDir() creates directories with test name pattern)
parentDir := filepath.Base(filepath.Dir(dir))
if !strings.Contains(parentDir, "TestTempDir") {
t.Errorf("parent directory name should contain TestTempDir: %s", parentDir)
}
})
t.Run("cleanup removes directory", func(t *testing.T) {
t.Parallel()
dir, cleanup := TempDir(t)
// Verify directory exists
if _, err := os.Stat(dir); os.IsNotExist(err) {
t.Error("temporary directory was not created")
}
// Clean up - this is now a no-op since t.TempDir() handles cleanup automatically
cleanup()
// Note: We can't verify directory removal here because t.TempDir() only
// cleans up at the end of the test, not when cleanup() is called.
// The directory will be automatically cleaned up when the test ends.
})
}
func TestWriteTestFile(t *testing.T) {
t.Parallel()
tmpDir, cleanup := TempDir(t)
defer cleanup()
t.Run("writes file with content", func(t *testing.T) {
t.Parallel()
testPath := filepath.Join(tmpDir, "test.txt")
testContent := "Hello, World!"
WriteTestFile(t, testPath, testContent)
// Verify file exists
if _, err := os.Stat(testPath); os.IsNotExist(err) {
t.Error("file was not created")
}
// Verify content
content, err := os.ReadFile(testPath) // #nosec G304 -- test file path
if err != nil {
t.Fatalf("failed to read file: %v", err)
}
if string(content) != testContent {
t.Errorf("expected content %s, got %s", testContent, string(content))
}
})
t.Run("creates nested directories", func(t *testing.T) {
t.Parallel()
nestedPath := filepath.Join(tmpDir, "nested", "deep", "file.txt")
testContent := "nested content"
WriteTestFile(t, nestedPath, testContent)
// Verify file exists
if _, err := os.Stat(nestedPath); os.IsNotExist(err) {
t.Error("nested file was not created")
}
// Verify parent directories exist
parentDir := filepath.Dir(nestedPath)
if _, err := os.Stat(parentDir); os.IsNotExist(err) {
t.Error("parent directories were not created")
}
})
t.Run("sets correct permissions", func(t *testing.T) {
t.Parallel()
testPath := filepath.Join(tmpDir, "perm-test.txt")
WriteTestFile(t, testPath, "test")
info, err := os.Stat(testPath)
if err != nil {
t.Fatalf("failed to stat file: %v", err)
}
// File should have 0600 permissions
expectedPerm := os.FileMode(0600)
if info.Mode().Perm() != expectedPerm {
t.Errorf("expected permissions %v, got %v", expectedPerm, info.Mode().Perm())
}
})
}
func TestSetupTestTemplates(t *testing.T) {
t.Parallel()
tmpDir, cleanup := TempDir(t)
defer cleanup()
SetupTestTemplates(t, tmpDir)
// Verify template directories exist
templatesDir := filepath.Join(tmpDir, "templates")
if _, err := os.Stat(templatesDir); os.IsNotExist(err) {
t.Error("templates directory was not created")
}
// Verify theme directories exist
themes := []string{TestThemeGitHub, TestThemeGitLab, TestThemeMinimal, TestThemeProfessional}
for _, theme := range themes {
themeDir := filepath.Join(templatesDir, "themes", theme)
if _, err := os.Stat(themeDir); os.IsNotExist(err) {
t.Errorf("theme directory %s was not created", theme)
}
// Verify theme template file exists
templateFile := filepath.Join(themeDir, TestTemplateReadme)
if _, err := os.Stat(templateFile); os.IsNotExist(err) {
t.Errorf("template file for theme %s was not created", theme)
}
// Verify template content
content, err := os.ReadFile(templateFile) // #nosec G304 -- test file path
if err != nil {
t.Errorf("failed to read template file for theme %s: %v", theme, err)
}
if string(content) != SimpleTemplate {
t.Errorf("template content for theme %s doesn't match SimpleTemplate", theme)
}
}
// Verify default template exists
defaultTemplate := filepath.Join(templatesDir, TestTemplateReadme)
if _, err := os.Stat(defaultTemplate); os.IsNotExist(err) {
t.Error("default template was not created")
}
}
func TestCreateTestAction(t *testing.T) {
t.Parallel()
t.Run("creates basic action", func(t *testing.T) {
t.Parallel()
testCreateBasicAction(t)
})
t.Run("creates action with no inputs", func(t *testing.T) {
t.Parallel()
testCreateActionNoInputs(t)
})
}
// testCreateBasicAction tests creating an action with name, description, and inputs.
func testCreateBasicAction(t *testing.T) {
t.Helper()
name := "Test Action"
description := "A test action for testing"
inputs := map[string]string{
"input1": "First input",
"input2": "Second input",
}
action := CreateTestAction(name, description, inputs)
validateActionNonEmpty(t, action)
validateActionContainsNameAndDescription(t, action, name, description)
validateActionContainsInputs(t, action, inputs)
}
// testCreateActionNoInputs tests creating an action without inputs.
func testCreateActionNoInputs(t *testing.T) {
t.Helper()
action := CreateTestAction("Simple Action", "No inputs", nil)
validateActionNonEmpty(t, action)
if !strings.Contains(action, "Simple Action") {
t.Error("action should contain the name")
}
}
// validateActionNonEmpty checks that the action is not empty.
func validateActionNonEmpty(t *testing.T, action string) {
t.Helper()
if action == "" {
t.Fatal(TestErrNonEmptyAction)
}
}
// validateActionContainsNameAndDescription validates action contains name and description.
func validateActionContainsNameAndDescription(t *testing.T, action, name, description string) {
t.Helper()
if !strings.Contains(action, name) {
t.Errorf("action should contain name: %s", name)
}
if !strings.Contains(action, description) {
t.Errorf("action should contain description: %s", description)
}
}
// validateActionContainsInputs validates action contains all expected inputs.
func validateActionContainsInputs(t *testing.T, action string, inputs map[string]string) {
t.Helper()
for inputName, inputDesc := range inputs {
if !strings.Contains(action, inputName) {
t.Errorf("action should contain input name: %s", inputName)
}
if !strings.Contains(action, inputDesc) {
t.Errorf("action should contain input description: %s", inputDesc)
}
}
}
func TestCreateCompositeAction(t *testing.T) {
t.Parallel()
t.Run("creates composite action with steps", func(t *testing.T) {
t.Parallel()
name := "Composite Test"
description := "A composite action"
steps := []string{
TestActionCheckoutV4,
"actions/setup-node@v4",
}
action := CreateCompositeAction(name, description, steps)
if action == "" {
t.Fatal(TestErrNonEmptyAction)
}
// Verify the action contains our values
if !strings.Contains(action, name) {
t.Errorf("action should contain name: %s", name)
}
if !strings.Contains(action, description) {
t.Errorf("action should contain description: %s", description)
}
for _, step := range steps {
if !strings.Contains(action, step) {
t.Errorf("action should contain step: %s", step)
}
}
})
t.Run("creates composite action with no steps", func(t *testing.T) {
t.Parallel()
action := CreateCompositeAction("Empty Composite", "No steps", nil)
if action == "" {
t.Fatal(TestErrNonEmptyAction)
}
if !strings.Contains(action, "Empty Composite") {
t.Error("action should contain the name")
}
})
}
func TestMockAppConfig(t *testing.T) {
t.Parallel()
t.Run("creates default config", func(t *testing.T) {
t.Parallel()
testMockAppConfigDefaults(t)
})
t.Run("applies overrides", func(t *testing.T) {
t.Parallel()
testMockAppConfigOverrides(t)
})
t.Run("partial overrides keep defaults", func(t *testing.T) {
t.Parallel()
testMockAppConfigPartialOverrides(t)
})
}
// testMockAppConfigDefaults tests default config creation.
func testMockAppConfigDefaults(t *testing.T) {
t.Helper()
config := MockAppConfig(nil)
validateConfigCreated(t, config)
validateConfigDefaults(t, config)
}
// testMockAppConfigOverrides tests full override application.
func testMockAppConfigOverrides(t *testing.T) {
t.Helper()
overrides := createFullOverrides()
config := MockAppConfig(overrides)
validateOverriddenValues(t, config)
}
// testMockAppConfigPartialOverrides tests partial override application.
func testMockAppConfigPartialOverrides(t *testing.T) {
t.Helper()
overrides := createPartialOverrides()
config := MockAppConfig(overrides)
validatePartialOverrides(t, config)
validateRemainingDefaults(t, config)
}
// createFullOverrides creates a complete set of test overrides.
func createFullOverrides() *TestAppConfig {
return &TestAppConfig{
Theme: "github",
OutputFormat: "html",
OutputDir: "docs",
Template: "custom.tmpl",
Schema: "custom.schema.json",
Verbose: true,
Quiet: true,
GitHubToken: "test-token",
}
}
// createPartialOverrides creates a partial set of test overrides.
func createPartialOverrides() *TestAppConfig {
return &TestAppConfig{
Theme: TestThemeProfessional,
Verbose: true,
}
}
// validateConfigCreated validates that config was created successfully.
func validateConfigCreated(t *testing.T, config *TestAppConfig) {
t.Helper()
if config == nil {
t.Fatal("expected config to be created")
}
}
// validateConfigDefaults validates all default configuration values.
func validateConfigDefaults(t *testing.T, config *TestAppConfig) {
t.Helper()
validateStringField(t, config.Theme, "default", "theme")
validateStringField(t, config.OutputFormat, "md", TestFieldOutputFormat)
validateStringField(t, config.OutputDir, ".", "output dir")
validateStringField(t, config.Schema, "schemas/action.schema.json", "schema")
validateBoolField(t, config.Verbose, false, "verbose")
validateBoolField(t, config.Quiet, false, "quiet")
validateStringField(t, config.GitHubToken, "", "GitHub token")
}
// validateOverriddenValues validates all overridden configuration values.
func validateOverriddenValues(t *testing.T, config *TestAppConfig) {
t.Helper()
validateStringField(t, config.Theme, "github", "theme")
validateStringField(t, config.OutputFormat, "html", TestFieldOutputFormat)
validateStringField(t, config.OutputDir, "docs", "output dir")
validateStringField(t, config.Template, "custom.tmpl", "template")
validateStringField(t, config.Schema, "custom.schema.json", "schema")
validateBoolField(t, config.Verbose, true, "verbose")
validateBoolField(t, config.Quiet, true, "quiet")
validateStringField(t, config.GitHubToken, "test-token", "GitHub token")
}
// validatePartialOverrides validates partially overridden values.
func validatePartialOverrides(t *testing.T, config *TestAppConfig) {
t.Helper()
validateStringField(t, config.Theme, TestThemeProfessional, "theme")
validateBoolField(t, config.Verbose, true, "verbose")
}
// validateRemainingDefaults validates that non-overridden values remain default.
func validateRemainingDefaults(t *testing.T, config *TestAppConfig) {
t.Helper()
validateStringField(t, config.OutputFormat, "md", TestFieldOutputFormat)
validateBoolField(t, config.Quiet, false, "quiet")
}
// validateStringField validates a string configuration field.
func validateStringField(t *testing.T, actual, expected, fieldName string) {
t.Helper()
if actual != expected {
t.Errorf("expected %s %s, got %s", fieldName, expected, actual)
}
}
// validateBoolField validates a boolean configuration field.
func validateBoolField(t *testing.T, actual, expected bool, fieldName string) {
t.Helper()
if actual != expected {
t.Errorf("expected %s to be %v, got %v", fieldName, expected, actual)
}
}
func TestSetEnv(t *testing.T) {
testKey := "TEST_TESTUTIL_VAR"
originalValue := "original"
newValue := "new"
// Ensure the test key is not set initially
_ = os.Unsetenv(testKey)
t.Run("sets new environment variable", func(t *testing.T) {
cleanup := SetEnv(t, testKey, newValue)
defer cleanup()
if os.Getenv(testKey) != newValue {
t.Errorf("expected env var to be %s, got %s", newValue, os.Getenv(testKey))
}
})
t.Run("cleanup unsets new variable", func(t *testing.T) {
cleanup := SetEnv(t, testKey, newValue)
cleanup()
// Note: We can't verify env var cleanup here because t.Setenv() only
// cleans up at the end of the test, not when cleanup() is called.
// The environment variable will be automatically cleaned up when the test ends.
})
t.Run("overrides existing variable", func(t *testing.T) {
// Set original value
t.Setenv(testKey, originalValue)
cleanup := SetEnv(t, testKey, newValue)
defer cleanup()
if os.Getenv(testKey) != newValue {
t.Errorf("expected env var to be %s, got %s", newValue, os.Getenv(testKey))
}
})
t.Run("cleanup restores original variable", func(t *testing.T) {
// Set original value
t.Setenv(testKey, originalValue)
cleanup := SetEnv(t, testKey, newValue)
cleanup()
// Note: We can't verify env var restoration here because t.Setenv() manages
// all environment variables automatically. The last call to t.Setenv() wins
// and cleanup is automatic at test end.
if os.Getenv(testKey) != newValue {
t.Errorf("expected env var to still be %s (last set value), got %s", newValue, os.Getenv(testKey))
}
})
// Clean up after test
_ = os.Unsetenv(testKey)
}
func TestWithContext(t *testing.T) {
t.Parallel()
t.Run("creates context with timeout", func(t *testing.T) {
t.Parallel()
timeout := 100 * time.Millisecond
ctx, cancel := WithContext(timeout)
defer cancel()
if ctx == nil {
t.Fatal("expected context to be created")
}
// Check that the context has a deadline
deadline, ok := ctx.Deadline()
if !ok {
t.Error("expected context to have a deadline")
}
// The deadline should be approximately now + timeout
expectedDeadline := time.Now().Add(timeout)
timeDiff := deadline.Sub(expectedDeadline)
if timeDiff < -time.Second || timeDiff > time.Second {
t.Errorf("deadline too far from expected: diff = %v", timeDiff)
}
})
t.Run("context eventually times out", func(t *testing.T) {
t.Parallel()
ctx, cancel := WithContext(1 * time.Millisecond)
defer cancel()
// Wait a bit longer than the timeout
time.Sleep(10 * time.Millisecond)
select {
case <-ctx.Done():
// Context should be done
if ctx.Err() != context.DeadlineExceeded {
t.Errorf("expected DeadlineExceeded error, got %v", ctx.Err())
}
default:
t.Error("expected context to be done after timeout")
}
})
}
func TestAssertNoError(t *testing.T) {
t.Parallel()
t.Run("passes with nil error", func(t *testing.T) {
t.Parallel()
// This should not fail
AssertNoError(t, nil)
})
// Testing the failure case is complex because AssertNoError calls t.Fatalf
// which causes the test to exit. We can't easily test this without
// complex mocking infrastructure, so we'll just test the success case
// The failure case is implicitly tested throughout the codebase where
// AssertNoError is used with actual errors.
}
func TestAssertError(t *testing.T) {
t.Parallel()
t.Run("passes with non-nil error", func(t *testing.T) {
t.Parallel()
// This should not fail
AssertError(t, io.EOF)
})
// Similar to AssertNoError, testing the failure case is complex
// The failure behavior is implicitly tested throughout the codebase
}
func TestAssertStringContains(t *testing.T) {
t.Parallel()
t.Run("passes when string contains substring", func(t *testing.T) {
t.Parallel()
AssertStringContains(t, "hello world", "world")
AssertStringContains(t, "test string", "test")
AssertStringContains(t, "exact match", "exact match")
})
// Failure case testing is complex due to t.Fatalf behavior
}
func TestAssertEqual(t *testing.T) {
t.Parallel()
t.Run("passes with equal basic types", func(t *testing.T) {
t.Parallel()
AssertEqual(t, 42, 42)
AssertEqual(t, "test", "test")
AssertEqual(t, true, true)
AssertEqual(t, 3.14, 3.14)
})
t.Run("passes with equal string maps", func(t *testing.T) {
t.Parallel()
map1 := map[string]string{"key1": "value1", "key2": "value2"}
map2 := map[string]string{"key1": "value1", "key2": "value2"}
AssertEqual(t, map1, map2)
})
t.Run("passes with empty string maps", func(t *testing.T) {
t.Parallel()
map1 := map[string]string{}
map2 := map[string]string{}
AssertEqual(t, map1, map2)
})
// Testing failure cases is complex due to t.Fatalf behavior
// The map comparison logic is tested implicitly throughout the codebase
}
func TestNewStringReader(t *testing.T) {
t.Parallel()
t.Run("creates reader from string", func(t *testing.T) {
t.Parallel()
testNewStringReaderBasic(t)
})
t.Run("creates reader from empty string", func(t *testing.T) {
t.Parallel()
testNewStringReaderEmpty(t)
})
t.Run("reader can be closed", func(t *testing.T) {
t.Parallel()
testNewStringReaderClose(t)
})
t.Run("handles large strings", func(t *testing.T) {
t.Parallel()
testNewStringReaderLarge(t)
})
}
// testNewStringReaderBasic tests basic string reader creation and reading.
func testNewStringReaderBasic(t *testing.T) {
t.Helper()
testString := "Hello, World!"
reader := NewStringReader(testString)
if reader == nil {
t.Fatal("expected reader to be created")
}
content, err := io.ReadAll(reader)
if err != nil {
t.Fatalf("failed to read from reader: %v", err)
}
if string(content) != testString {
t.Errorf("expected content %s, got %s", testString, string(content))
}
}
// testNewStringReaderEmpty tests string reader with empty string.
func testNewStringReaderEmpty(t *testing.T) {
t.Helper()
reader := NewStringReader("")
content, err := io.ReadAll(reader)
if err != nil {
t.Fatalf("failed to read from empty reader: %v", err)
}
if len(content) != 0 {
t.Errorf("expected empty content, got %d bytes", len(content))
}
}
// testNewStringReaderClose tests that the reader can be closed.
func testNewStringReaderClose(t *testing.T) {
t.Helper()
reader := NewStringReader("test")
err := reader.Close()
if err != nil {
t.Errorf("failed to close reader: %v", err)
}
}
// testNewStringReaderLarge tests reading large strings.
func testNewStringReaderLarge(t *testing.T) {
t.Helper()
largeString := strings.Repeat("test ", 10000)
reader := NewStringReader(largeString)
content, err := io.ReadAll(reader)
if err != nil {
t.Fatalf("failed to read large string: %v", err)
}
if string(content) != largeString {
t.Error("large string content mismatch")
}
}
func TestCaptureStdout(t *testing.T) {
// Note: Cannot run in parallel as it manipulates global os.Stdout
output := CaptureStdout(func() {
fmt.Print("test output")
})
if output != "test output" {
t.Errorf("expected 'test output', got %q", output)
}
}
func TestCaptureStderr(t *testing.T) {
// Note: Cannot run in parallel as it manipulates global os.Stderr
output := CaptureStderr(func() {
fmt.Fprint(os.Stderr, "test error")
})
if output != "test error" {
t.Errorf("expected 'test error', got %q", output)
}
}
func TestCaptureOutputStreams(t *testing.T) {
// Note: Cannot run in parallel as it manipulates global os.Stdout/Stderr
output := CaptureOutputStreams(func() {
fmt.Print("stdout message")
fmt.Fprint(os.Stderr, "stderr message")
})
if output.Stdout != "stdout message" {
t.Errorf("expected stdout 'stdout message', got %q", output.Stdout)
}
if output.Stderr != "stderr message" {
t.Errorf("expected stderr 'stderr message', got %q", output.Stderr)
}
}