visionA/local-agent/visiona-agent/startup_pipeline_test.go

package main

// startup_pipeline_test.go — M8-4b StartupPipeline 單元測試
//
// 涵蓋情境：
//  1. CompleteStage(n) 正確切換內部 state 並進入下一階段
//  2. SkipStage(n) 把 stage 標記為 skipped 並進入下一階段
//  3. FailStage 把 pipeline 切到 failed
//  4. Soft timeout（單一階段 > 20 s）→ softTimeoutEmitted flag 被設
//  5. Hard timeout（總時 > 60 s）→ pipeline 進 failed
//  6. AutoOpenBrowser=false 時階段 5 skip + 階段 6 不檢查 timeout
//  7. RestartStartupSequence 5 步驟（不實際呼叫 ctrl.Start，因為會跑真的 server）
//  8. removeSentinelFile / sentinel file 偵測機制
//
// 測試使用 newTestApp()（見 server_control_test.go），所有 ctx == nil，
// emit 路徑會 short-circuit 但內部 state 仍會更新。

import (
	"context"
	"errors"
	"os"
	"path/filepath"
	"testing"
	"time"
)

// newPipelineTestApp 建一個 App 並設好 dataDir + prefs（測試用）。
// 不啟動 watcher（除非測試自己呼叫 Start）。
func newPipelineTestApp(t *testing.T) (*App, string) {
	t.Helper()
	a := newTestApp()
	dir, err := os.MkdirTemp("", "visiona-agent-pipeline-test-*")
	if err != nil {
		t.Fatalf("mkdtemp: %v", err)
	}
	t.Cleanup(func() { _ = os.RemoveAll(dir) })
	a.dataDir = dir
	a.prefs = DefaultPreferences() // macOS/Windows: AutoOpenBrowser=true
	return a, dir
}

// -----------------------------------------------------------------------
// CompleteStage / SkipStage / FailStage 基本行為
// -----------------------------------------------------------------------

func TestStartupPipeline_CompleteStage_AdvancesToNext(t *testing.T) {
	a, _ := newPipelineTestApp(t)
	p := NewStartupPipeline(a)
	// 不啟動 watcher，手動設 current=1 running 模擬 Start()
	p.startedAt = time.Now()
	p.current = 1
	p.stages[1].status = "running"
	p.stages[1].startedAt = time.Now()

	p.CompleteStage(1)

	if p.stages[1].status != "completed" {
		t.Fatalf("stage 1 status=%q, want completed", p.stages[1].status)
	}
	if p.current != 2 {
		t.Fatalf("after CompleteStage(1), current=%d, want 2", p.current)
	}
	if p.stages[2].status != "running" {
		t.Fatalf("stage 2 status=%q, want running", p.stages[2].status)
	}
}

func TestStartupPipeline_CompleteStage_OutOfOrder_Ignored(t *testing.T) {
	a, _ := newPipelineTestApp(t)
	p := NewStartupPipeline(a)
	p.current = 1
	p.stages[1].status = "running"

	// 嘗試 complete 階段 3（current 是 1）→ 應該被 ignore
	p.CompleteStage(3)

	if p.current != 1 {
		t.Fatalf("after out-of-order CompleteStage(3), current=%d, want 1", p.current)
	}
	if p.stages[3].status != "" {
		t.Fatalf("stage 3 status=%q, want empty (untouched)", p.stages[3].status)
	}
}

func TestStartupPipeline_CompleteStage_LastStageMarksReady(t *testing.T) {
	a, _ := newPipelineTestApp(t)
	p := NewStartupPipeline(a)
	p.current = startupTotalStages
	p.stages[startupTotalStages].status = "running"

	p.CompleteStage(startupTotalStages)

	if p.current != startupTotalStages+1 {
		t.Fatalf("after final CompleteStage, current=%d, want %d (ready)", p.current, startupTotalStages+1)
	}
	if p.stages[startupTotalStages].status != "completed" {
		t.Fatalf("final stage status=%q, want completed", p.stages[startupTotalStages].status)
	}
}

func TestStartupPipeline_SkipStage_AdvancesAndMarksSkipped(t *testing.T) {
	a, _ := newPipelineTestApp(t)
	p := NewStartupPipeline(a)
	p.current = 5
	p.stages[5].status = "running"

	p.SkipStage(5)

	if p.stages[5].status != "skipped" {
		t.Fatalf("stage 5 status=%q, want skipped", p.stages[5].status)
	}
	if p.current != 6 {
		t.Fatalf("after SkipStage(5), current=%d, want 6", p.current)
	}
	if p.stages[6].status != "running" {
		t.Fatalf("stage 6 status=%q, want running", p.stages[6].status)
	}
}

func TestStartupPipeline_FailStage_StopsPipeline(t *testing.T) {
	a, _ := newPipelineTestApp(t)
	p := NewStartupPipeline(a)
	p.current = 3
	p.stages[3].status = "running"

	p.FailStage(3, errors.New("boom"))

	if p.stages[3].status != "failed" {
		t.Fatalf("stage 3 status=%q, want failed", p.stages[3].status)
	}
	if p.current != -1 {
		t.Fatalf("after FailStage, current=%d, want -1", p.current)
	}
	// FailStage 觸發 emitError → setState(Error)
	if got := a.ctrl.State(); got != ServerStateError {
		t.Fatalf("ctrl state after FailStage=%s, want error", got)
	}
}

// -----------------------------------------------------------------------
// Watcher：soft timeout
// -----------------------------------------------------------------------

func TestStartupPipeline_Watcher_SoftTimeout(t *testing.T) {
	a, _ := newPipelineTestApp(t)
	p := NewStartupPipeline(a)
	// 模擬「階段 2 已經跑了 25 秒」
	now := time.Now()
	p.startedAt = now.Add(-25 * time.Second) // 總時 25s（< 60s hard）
	p.current = 2
	p.stages[2].status = "running"
	p.stages[2].startedAt = now.Add(-25 * time.Second) // 階段時間 25s（> 20s soft）

	// 開 watcher，等 ~1.2 秒讓它跑一次 tick
	ctx, cancel := context.WithCancel(context.Background())
	defer cancel()
	p.watcherDone = make(chan struct{})
	go p.watcher(ctx)
	time.Sleep(1300 * time.Millisecond)

	p.mu.Lock()
	emitted := p.stages[2].softTimeoutEmitted
	cur := p.current
	p.mu.Unlock()

	if !emitted {
		t.Fatalf("expected softTimeoutEmitted=true after watcher tick")
	}
	if cur != 2 {
		t.Fatalf("current=%d, want 2 (soft timeout 不中斷流程)", cur)
	}
	cancel()
	<-p.watcherDone
}

// -----------------------------------------------------------------------
// Watcher：hard timeout
// -----------------------------------------------------------------------

func TestStartupPipeline_Watcher_HardTimeout(t *testing.T) {
	a, _ := newPipelineTestApp(t)
	p := NewStartupPipeline(a)
	// 模擬「總時已經 305 秒（超過 300 秒 hard timeout），當前在階段 3」
	now := time.Now()
	p.startedAt = now.Add(-305 * time.Second)
	p.current = 3
	p.stages[3].status = "running"
	p.stages[3].startedAt = now.Add(-30 * time.Second)

	ctx, cancel := context.WithCancel(context.Background())
	defer cancel()
	p.watcherDone = make(chan struct{})
	p.watcherCancel = cancel
	go p.watcher(ctx)
	time.Sleep(1300 * time.Millisecond)

	p.mu.Lock()
	cur := p.current
	status := p.stages[3].status
	p.mu.Unlock()

	if cur != -1 {
		t.Fatalf("after hard timeout, current=%d, want -1", cur)
	}
	if status != "failed" {
		t.Fatalf("stage 3 status after hard timeout=%q, want failed", status)
	}
	if got := a.ctrl.State(); got != ServerStateError {
		t.Fatalf("ctrl state after hard timeout=%s, want error", got)
	}
}

// -----------------------------------------------------------------------
// PauseHardTimeout / ResumeHardTimeout：首次 Python bootstrap 豁免
// -----------------------------------------------------------------------

// Pause 期間 sinceTotal 不會累積，Resume 後從暫停時點繼續算。
func TestStartupPipeline_PauseHardTimeout_ExcludesPausedDuration(t *testing.T) {
	a, _ := newPipelineTestApp(t)
	p := NewStartupPipeline(a)

	now := time.Now()
	p.startedAt = now.Add(-10 * time.Second)
	p.current = 2
	p.stages[2].status = "running"
	p.stages[2].startedAt = now.Add(-10 * time.Second)

	// 未暫停：effective = 10s
	p.mu.Lock()
	eff := p.effectiveSinceTotalLocked()
	p.mu.Unlock()
	if eff < 9*time.Second || eff > 11*time.Second {
		t.Fatalf("effective before pause=%s, want ~10s", eff)
	}

	// 暫停 500ms 再看
	p.PauseHardTimeout()
	time.Sleep(500 * time.Millisecond)
	p.mu.Lock()
	effDuringPause := p.effectiveSinceTotalLocked()
	p.mu.Unlock()
	// 暫停期間 effective 不該跟 wall clock 一起漲 —— 應該還是 ~10s
	if effDuringPause > 10500*time.Millisecond {
		t.Fatalf("effective during pause=%s, should stay ~10s (paused window excluded)", effDuringPause)
	}

	// Resume 後再等 200ms，effective 繼續前進
	p.ResumeHardTimeout()
	time.Sleep(200 * time.Millisecond)
	p.mu.Lock()
	effAfterResume := p.effectiveSinceTotalLocked()
	p.mu.Unlock()
	// 預期 ~10.2s，誤差允許 300ms
	if effAfterResume < 10*time.Second || effAfterResume > 10500*time.Millisecond {
		t.Fatalf("effective after resume=%s, want ~10.2s", effAfterResume)
	}
}

// 首次 bootstrap 情境：即使 wall clock 已超過 60s，只要暫停時間夠多，
// hard timeout 就不該觸發。
func TestStartupPipeline_PauseHardTimeout_PreventsHardTimeout(t *testing.T) {
	a, _ := newPipelineTestApp(t)
	p := NewStartupPipeline(a)

	// 模擬 wall clock 已過 300 秒，但其中 250 秒是「首次 bootstrap」暫停
	// effective = 50s < 300s hard timeout，pipeline 不該 fail
	now := time.Now()
	p.startedAt = now.Add(-300 * time.Second)
	p.pausedDuration = 250 * time.Second
	p.current = 2
	p.stages[2].status = "running"
	p.stages[2].startedAt = now.Add(-300 * time.Second)

	ctx, cancel := context.WithCancel(context.Background())
	defer cancel()
	p.watcherDone = make(chan struct{})
	p.watcherCancel = cancel
	go p.watcher(ctx)
	time.Sleep(1300 * time.Millisecond)

	p.mu.Lock()
	cur := p.current
	status := p.stages[2].status
	p.mu.Unlock()

	if cur == -1 {
		t.Fatal("pipeline failed due to hard timeout, but effective=50s should be under the 300s limit")
	}
	if status == "failed" {
		t.Fatalf("stage 2 failed, want still running (effective time under limit)")
	}
	// 不 Fatal ctrl state—— watcher 在 test 裡不會被真的 Stop，只檢查 pipeline 不進 failed 就好
}

// Resume 未暫停時為 no-op。
func TestStartupPipeline_ResumeHardTimeout_NoopWhenNotPaused(t *testing.T) {
	a, _ := newPipelineTestApp(t)
	p := NewStartupPipeline(a)
	p.startedAt = time.Now()
	// 不呼叫 Pause，直接 Resume 應不 panic 且 pausedDuration 保持為 0
	p.ResumeHardTimeout()
	p.mu.Lock()
	dur := p.pausedDuration
	p.mu.Unlock()
	if dur != 0 {
		t.Fatalf("pausedDuration after noop resume=%s, want 0", dur)
	}
}

// -----------------------------------------------------------------------
// Skip 規則：階段 5/6 + AutoOpenBrowser=false
// -----------------------------------------------------------------------

func TestStartupPipeline_Watcher_SkippedStageNoTimeout(t *testing.T) {
	a, _ := newPipelineTestApp(t)
	a.prefs.AutoOpenBrowser = false
	p := NewStartupPipeline(a)

	// 階段 6 + AutoOpenBrowser=false：總時 320s（已超 300s hard timeout）也不該觸發
	now := time.Now()
	p.startedAt = now.Add(-320 * time.Second)
	p.current = 6
	p.stages[6].status = "running"
	p.stages[6].startedAt = now.Add(-320 * time.Second)

	ctx, cancel := context.WithCancel(context.Background())
	defer cancel()
	p.watcherDone = make(chan struct{})
	p.watcherCancel = cancel
	go p.watcher(ctx)
	time.Sleep(1300 * time.Millisecond)

	p.mu.Lock()
	cur := p.current
	status := p.stages[6].status
	p.mu.Unlock()

	if cur != 6 {
		t.Fatalf("stage 6 + AutoOpenBrowser=false should not timeout, current=%d", cur)
	}
	if status != "running" {
		t.Fatalf("stage 6 status=%q, want running (no timeout fired)", status)
	}
	cancel()
	<-p.watcherDone
}

func TestStartupPipeline_Watcher_SkippedStatusBypassesTimeout(t *testing.T) {
	a, _ := newPipelineTestApp(t)
	p := NewStartupPipeline(a)

	// 階段 5 已 skipped，總時 320s 不該觸發 hard timeout（skipped 跳過所有檢查）
	// 注意：skip 之後實際上 current 會是 6，但這裡測試的是 skip 狀態本身的 bypass 行為
	now := time.Now()
	p.startedAt = now.Add(-320 * time.Second)
	p.current = 5
	p.stages[5].status = "skipped"
	p.stages[5].startedAt = now.Add(-30 * time.Second)

	ctx, cancel := context.WithCancel(context.Background())
	defer cancel()
	p.watcherDone = make(chan struct{})
	p.watcherCancel = cancel
	go p.watcher(ctx)
	time.Sleep(1300 * time.Millisecond)

	p.mu.Lock()
	cur := p.current
	status := p.stages[5].status
	p.mu.Unlock()

	if cur != 5 {
		t.Fatalf("skipped stage 5 should not be failed, current=%d", cur)
	}
	if status != "skipped" {
		t.Fatalf("stage 5 status=%q, want skipped", status)
	}
	cancel()
	<-p.watcherDone
}

// -----------------------------------------------------------------------
// Sentinel file
// -----------------------------------------------------------------------

func TestStartupPipeline_CheckSentinelFile(t *testing.T) {
	a, dir := newPipelineTestApp(t)
	p := NewStartupPipeline(a)

	if p.checkSentinelFile() {
		t.Fatalf("sentinel file 不存在時 checkSentinelFile 應回 false")
	}

	// 寫一個 sentinel
	path := filepath.Join(dir, ".first-ws-connected")
	if err := os.WriteFile(path, []byte("bootId=test\n"), 0o644); err != nil {
		t.Fatalf("write sentinel: %v", err)
	}

	if !p.checkSentinelFile() {
		t.Fatalf("sentinel file 存在時 checkSentinelFile 應回 true")
	}
}

func TestStartupPipeline_RemoveSentinelFile(t *testing.T) {
	_, dir := newPipelineTestApp(t)
	path := filepath.Join(dir, ".first-ws-connected")
	if err := os.WriteFile(path, []byte("x"), 0o644); err != nil {
		t.Fatalf("write: %v", err)
	}

	removeSentinelFile(dir)

	if _, err := os.Stat(path); !os.IsNotExist(err) {
		t.Fatalf("sentinel file 應該被刪除，stat err=%v", err)
	}

	// 重複呼叫不會出錯
	removeSentinelFile(dir)

	// 空 dataDir 也不會 panic
	removeSentinelFile("")
}

func TestStartupPipeline_Watcher_Stage6CompletesOnSentinel(t *testing.T) {
	a, dir := newPipelineTestApp(t)
	p := NewStartupPipeline(a)

	// 模擬「階段 6 剛 running 1 秒」
	now := time.Now()
	p.startedAt = now.Add(-30 * time.Second) // 總時 30s（< 60s）
	p.current = 6
	p.stages[6].status = "running"
	p.stages[6].startedAt = now.Add(-1 * time.Second)

	ctx, cancel := context.WithCancel(context.Background())
	defer cancel()
	p.watcherDone = make(chan struct{})
	p.watcherCancel = cancel
	go p.watcher(ctx)

	// 等 watcher 跑一次 tick 確認沒誤判
	time.Sleep(300 * time.Millisecond)
	p.mu.Lock()
	if p.current != 6 {
		p.mu.Unlock()
		t.Fatalf("watcher 在 sentinel 不存在時就把 current 改了 (=%d)", p.current)
	}
	p.mu.Unlock()

	// 寫 sentinel file，等下一次 tick
	if err := os.WriteFile(filepath.Join(dir, ".first-ws-connected"), []byte("ok"), 0o644); err != nil {
		t.Fatalf("write sentinel: %v", err)
	}
	time.Sleep(1300 * time.Millisecond)

	p.mu.Lock()
	cur := p.current
	status := p.stages[6].status
	p.mu.Unlock()

	if cur != startupTotalStages+1 {
		t.Fatalf("watcher 看到 sentinel 後 current=%d, want %d (ready)", cur, startupTotalStages+1)
	}
	if status != "completed" {
		t.Fatalf("stage 6 status=%q, want completed", status)
	}
}

// -----------------------------------------------------------------------
// NOTE: 以下 2 個原本針對 RestartStartupSequence Wails binding 的測試，於
// AB2+AB3 連同 binding 本身一併刪除：
//   - TestStartupPipeline_RestartStartupSequence_StepsExecution
//   - TestRestartStartupSequence_ColdStartOpenBrowser_OnlyOnce
//
// 理由：visionA Agent 的 3 個配置頁不提供 Retry 按鈕；使用者不直接操作 server
// 生命週期。rebuildStartupPipeline 的內部邏輯仍保留供 startInternal 防呆路徑
// 使用，但不對前端暴露。若未來要加回類似功能，測試也要重新設計。
// -----------------------------------------------------------------------

// -----------------------------------------------------------------------
// M8-4b 補丁 M-1 regression：HasFailedStage + 不重複通知
// -----------------------------------------------------------------------
//
// 驗證 HasFailedStage 的語義：
//   - 新 pipeline（current=0）→ false
//   - running（current=2, stage[2]=running）→ false
//   - 整個 pipeline 已 fail（current=-1）→ true
//   - 某 stage status=failed 但 current 還沒切 → 也要回 true（defensive）
//
// 這個 helper 是 startInternal 判斷「pipeline 已發過錯誤通知」的依據，
// 必須覆蓋所有 fail 路徑。

func TestStartupPipeline_HasFailedStage(t *testing.T) {
	a, _ := newPipelineTestApp(t)
	p := NewStartupPipeline(a)

	// 1. 新 pipeline → false
	if p.HasFailedStage() {
		t.Fatalf("new pipeline HasFailedStage should be false")
	}

	// 2. 正在 running → false
	p.current = 2
	p.stages[2].status = "running"
	if p.HasFailedStage() {
		t.Fatalf("running pipeline HasFailedStage should be false")
	}

	// 3. FailStage 後（current=-1）→ true
	p.FailStage(2, errors.New("test failure"))
	if !p.HasFailedStage() {
		t.Fatalf("after FailStage HasFailedStage should be true, current=%d", p.current)
	}

	// 4. 另一種 case：某 stage 被手動寫成 failed 但 current 還沒同步（defensive）
	p2 := NewStartupPipeline(a)
	p2.current = 3
	p2.stages[2].status = "failed" // 不該發生但 defensive
	if !p2.HasFailedStage() {
		t.Fatalf("any stage failed should make HasFailedStage true")
	}
}

// -----------------------------------------------------------------------
// M8-4b 補丁 M-2 regression：IsInColdStart + openBrowser 不被呼叫兩次
// -----------------------------------------------------------------------
//
// 驗證 IsInColdStart 的語義：
//   - current = 0（未 Start）→ false（非冷啟動中）
//   - current = 1..6（running）→ true（冷啟動進行中）
//   - current = 7（ready）→ false（已完成，屬於 RestartServer 等場景）
//   - current = -1（failed）→ false

func TestStartupPipeline_IsInColdStart(t *testing.T) {
	a, _ := newPipelineTestApp(t)
	p := NewStartupPipeline(a)

	// current = 0
	if p.IsInColdStart() {
		t.Fatalf("current=0 IsInColdStart should be false")
	}

	// current = 1..6
	for i := 1; i <= startupTotalStages; i++ {
		p.current = i
		if !p.IsInColdStart() {
			t.Fatalf("current=%d IsInColdStart should be true", i)
		}
	}

	// current = 7 (ready)
	p.current = startupTotalStages + 1
	if p.IsInColdStart() {
		t.Fatalf("current=7 IsInColdStart should be false (ready)")
	}

	// current = -1 (failed)
	p.current = -1
	if p.IsInColdStart() {
		t.Fatalf("current=-1 IsInColdStart should be false (failed)")
	}
}

// NOTE: TestRestartStartupSequence_ColdStartOpenBrowser_OnlyOnce 已於 AB2+AB3
// 隨 RestartStartupSequence Wails binding 一併刪除（visionA Agent 的 UI 不提供
// Retry）。openBrowser 變數本身仍在（未來 stub 可能用），保留其他冷啟動/開瀏覽器
// 測試覆蓋此路徑。

// -----------------------------------------------------------------------
// stopWatcher 的 idempotent 行為
// -----------------------------------------------------------------------

func TestStartupPipeline_StopWatcher_Idempotent(t *testing.T) {
	a, _ := newPipelineTestApp(t)
	p := NewStartupPipeline(a)
	// 沒 watcherCancel → 不該 panic
	p.stopWatcher()

	// 設一個 cancel func，呼叫兩次也不該 panic（context.CancelFunc 文件保證可重複呼叫）
	_, cancel := context.WithCancel(context.Background())
	p.watcherCancel = cancel
	p.stopWatcher()
	p.stopWatcher()
}