test(runner): retire the scaffolding the package grew by copying

Four consolidations, no assertion changed and the same 217 tests passing:

harness gains run and tryRun, retiring 68 copies of context.WithTimeout plus
Run(ctx, Options{}) plus an error check. The seven sites whose error check said
something specific keep their own wording.

mock.FailurePlan retires six wrapper drivers that were the same seven-line
method with a different error constant. The three that need more than an error
on call N stay.

traceStepLine absorbs the 15 local trace-line declarations and readTraceLines is
now the only reader. No local copy disagreed with internal/trace about a JSON
tag. Three raw readers stay on purpose: one decodes a real trace.Step, one
snapshots bytes, one asserts on raw JSON text.

One embeddable base carries the five V8-path methods the five web fakes each
redeclared.
This commit is contained in:
pj committed 2026-08-22 21:39:22 +05:30
1 parent d9f96602d9
commit 577993b540
14 files changed
+233 -1282

No files matched your search

+6 -33
View File
@@ -493,7 +493,7 @@ func newSeededVerifier(t *testing.T) *verifier.Verifier {
func TestPickSourcesRejectsLLMWithoutSpecConfig(t *testing.T) {
for name, activeDriver := range map[string]driver.DeviceDriver{
"native": nil,
"web": &webMockDriver{Driver: mockdriver.New()},
"web": &webMockDriver{webDriverBase: webDriverBase{Driver: mockdriver.New()}},
} {
t.Run(name, func(t *testing.T) {
action, extractor, err := pickSources(Options{
@@ -523,7 +523,7 @@ func TestPickSourcesOnWebComposesLLMWithWebExtractors(t *testing.T) {
fake := newFakeOpenRouter(t)
_, verifierInstance := newLLMSource(t, fake)
action, extractor, err := pickSources(Options{
Driver: &webMockDriver{Driver: mockdriver.New()},
Driver: &webMockDriver{webDriverBase: webDriverBase{Driver: mockdriver.New()}},
Verifier: verifierInstance,
Generator: "llm",
Logger: slog.New(slog.NewTextHandler(io.Discard, nil)),
@@ -541,7 +541,7 @@ func TestPickSourcesOnWebComposesLLMWithWebExtractors(t *testing.T) {
func TestPickSourcesOnWebSeededKeepsBothOnV8(t *testing.T) {
action, extractor, err := pickSources(Options{
Driver: &webMockDriver{Driver: mockdriver.New()},
Driver: &webMockDriver{webDriverBase: webDriverBase{Driver: mockdriver.New()}},
Verifier: newSeededVerifier(t),
Generator: "seeded",
Logger: slog.New(slog.NewTextHandler(io.Discard, nil)),
@@ -768,22 +768,13 @@ func TestRunner_EveryModelCallFailingIsNotACleanRun(t *testing.T) {
state := newHarnessWithSpec(t, llmFixtureSpec)
state.mock.HierarchyJSON = llmTreeJSON
ctx, cancel := context.WithTimeout(context.Background(), 30*time.Second)
defer cancel()
summary, err := Run(ctx, Options{
summary := state.run(t, Options{
Duration: 30 * time.Second,
IdleTimeout: 20 * time.Millisecond,
MaxSteps: 3,
Driver: state.mock,
Verifier: state.verifier,
TraceWriter: state.writer,
Generator: "llm",
LabelSource: verifier.LabelSourceVisibleText,
Logger: slog.New(slog.NewTextHandler(io.Discard, nil)),
})
if err != nil {
t.Fatalf("Run: %v", err)
}
calls := readLLMCalls(t, state.writer.Directory())
if len(calls) != 3 {
@@ -845,22 +836,13 @@ func TestRunner_SetupActionsAreNotTheGeneratorDrivingTheApp(t *testing.T) {
state := newHarnessWithSpec(t, llmLoginSetupSpec)
state.mock.HierarchyJSON = llmTreeJSON
ctx, cancel := context.WithTimeout(context.Background(), 30*time.Second)
defer cancel()
summary, err := Run(ctx, Options{
summary := state.run(t, Options{
Duration: 30 * time.Second,
IdleTimeout: 20 * time.Millisecond,
MaxSteps: 4,
Driver: state.mock,
Verifier: state.verifier,
TraceWriter: state.writer,
Generator: "llm",
LabelSource: verifier.LabelSourceVisibleText,
Logger: slog.New(slog.NewTextHandler(io.Discard, nil)),
})
if err != nil {
t.Fatalf("Run: %v", err)
}
wantOutcomes := []string{
trace.LLMOutcomeSetupAction, trace.LLMOutcomeSetupAction,
@@ -933,22 +915,13 @@ func TestRunner_SetupAndGeneratorBothDrivingIsAHealthyRun(t *testing.T) {
fake.choice = tap.Index
fake.chosenAction = tap.Description
ctx, cancel := context.WithTimeout(context.Background(), 30*time.Second)
defer cancel()
summary, err := Run(ctx, Options{
summary := state.run(t, Options{
Duration: 30 * time.Second,
IdleTimeout: 20 * time.Millisecond,
MaxSteps: 4,
Driver: state.mock,
Verifier: state.verifier,
TraceWriter: state.writer,
Generator: "llm",
LabelSource: verifier.LabelSourceVisibleText,
Logger: slog.New(slog.NewTextHandler(io.Discard, nil)),
})
if err != nil {
t.Fatalf("Run: %v", err)
}
if summary.DispatchedActions != 4 {
t.Errorf("DispatchedActions = %d, want 4: every step drove the app",