// Ownership store 單元測試。
//
// 測試策略：
//   - Set/Get/Delete 用 race detector 驗 concurrent safety
//   - EnsureRebuilt 用 stub ConverterClient（atomic counter 紀錄 fetch 次數）
//     驗：first-call fetches / second-call noop / per-user 並行 / thundering herd 收斂
//   - 失敗路徑驗：error 不標 rebuilt → 下次再 fetch
//
// Phase 0.8 conversion (見 .autoflow/04-architecture/conversion.md §2.6.1)
package conversion

import (
	"context"
	"errors"
	"fmt"
	"io"
	"log/slog"
	"strconv"
	"sync"
	"sync/atomic"
	"testing"
	"time"

	"github.com/stretchr/testify/assert"
	"github.com/stretchr/testify/require"
)

// ==========================================================================
// stub ConverterClient — 只實作 ListInProgressJobs，其他 panic（測試不用）
// ==========================================================================

// stubConverterClient 是 test 用的 fake ConverterClient。
//
// 只實作 ListInProgressJobs（其他 method 測試不用，panic 防呆）；用 atomic counter
// 紀錄各 user 被呼叫次數。
type stubConverterClient struct {
	mu sync.Mutex

	// jobsByUser: user_id → 該 user 的 in_progress jobs（若 nil → 空 slice）
	jobsByUser map[string][]*ConverterJob

	// errByUser: user_id → 強制回傳的錯誤（用在失敗路徑測試）
	errByUser map[string]error

	// callCountByUser: user_id → ListInProgressJobs 被呼叫次數（atomic counter）
	callCountByUser sync.Map // map[string]*atomic.Int32

	// fetchDelay 模擬慢 fetch（讓併發測試有機會競態）
	fetchDelay time.Duration

	// blockSignal 若非 nil，每次 ListInProgressJobs 進入時發 signal（用在 timeout 測試）
	blockSignal chan struct{}
	// blockUntil 若非 nil，會 block 在 ctx.Done 或這個 channel 任一觸發
	blockUntil chan struct{}
}

func newStubConverterClient() *stubConverterClient {
	return &stubConverterClient{
		jobsByUser: make(map[string][]*ConverterJob),
		errByUser:  make(map[string]error),
	}
}

func (s *stubConverterClient) setJobs(userID string, jobs []*ConverterJob) {
	s.mu.Lock()
	defer s.mu.Unlock()
	s.jobsByUser[userID] = jobs
}

func (s *stubConverterClient) setError(userID string, err error) {
	s.mu.Lock()
	defer s.mu.Unlock()
	s.errByUser[userID] = err
}

// callCount 取某個 user 被呼叫的次數。
func (s *stubConverterClient) callCount(userID string) int32 {
	v, ok := s.callCountByUser.Load(userID)
	if !ok {
		return 0
	}
	return v.(*atomic.Int32).Load()
}

func (s *stubConverterClient) ListInProgressJobs(ctx context.Context, userID string) ([]*ConverterJob, error) {
	// atomic counter
	cnt, _ := s.callCountByUser.LoadOrStore(userID, &atomic.Int32{})
	cnt.(*atomic.Int32).Add(1)

	// 通知 caller 已進入（給 thundering herd 測試用）
	if s.blockSignal != nil {
		select {
		case s.blockSignal <- struct{}{}:
		default:
		}
	}

	// 若有 blockUntil，等到 signal 或 ctx.Done 才 return（模擬慢 / cancel）
	if s.blockUntil != nil {
		select {
		case <-s.blockUntil:
		case <-ctx.Done():
			return nil, ctx.Err()
		}
	}

	if s.fetchDelay > 0 {
		select {
		case <-time.After(s.fetchDelay):
		case <-ctx.Done():
			return nil, ctx.Err()
		}
	}

	s.mu.Lock()
	err := s.errByUser[userID]
	jobs := s.jobsByUser[userID]
	s.mu.Unlock()

	if err != nil {
		return nil, err
	}
	if jobs == nil {
		jobs = []*ConverterJob{}
	}
	return jobs, nil
}

// 其他 method panic（測試不會呼叫，撞到 panic 反而好 debug）。
func (s *stubConverterClient) InitJob(ctx context.Context, req InitConverterJobReq) (*ConverterJob, error) {
	panic("stubConverterClient.InitJob: not used in ownership_test")
}
func (s *stubConverterClient) GetJob(ctx context.Context, jobID string) (*ConverterJob, error) {
	panic("stubConverterClient.GetJob: not used in ownership_test")
}
func (s *stubConverterClient) Promote(ctx context.Context, jobID string, req PromoteReq) (*ConverterPromoteResult, error) {
	panic("stubConverterClient.Promote: not used in ownership_test")
}

// 確保 stubConverterClient 滿足 ConverterClient interface（編譯期驗）。
var _ ConverterClient = (*stubConverterClient)(nil)

// ==========================================================================
// helper：建立靜默 logger（避免測試 stdout 噪音）
// ==========================================================================

func newSilentLogger() *slog.Logger {
	return slog.New(slog.NewTextHandler(io.Discard, nil))
}

// ==========================================================================
// 基本 Set / Get / Delete
// ==========================================================================

// TestSet_Get_Delete_Basic：write / read / delete 標準操作。
func TestSet_Get_Delete_Basic(t *testing.T) {
	t.Parallel()

	stub := newStubConverterClient()
	o := NewOwnership(stub, newSilentLogger())

	// Set + Get
	o.Set("job-1", "alice")
	uid, ok := o.Get("job-1")
	assert.True(t, ok)
	assert.Equal(t, "alice", uid)

	// 覆寫
	o.Set("job-1", "bob")
	uid, _ = o.Get("job-1")
	assert.Equal(t, "bob", uid, "Set 同 jobID 應覆寫")

	// Delete
	o.Delete("job-1")
	_, ok = o.Get("job-1")
	assert.False(t, ok, "Delete 後 Get 應回 false")

	// 不存在的 jobID
	_, ok = o.Get("ghost")
	assert.False(t, ok)

	// 防呆：空字串不寫入
	o.Set("", "alice")
	o.Set("job-empty-uid", "")
	_, ok = o.Get("")
	assert.False(t, ok)
	_, ok = o.Get("job-empty-uid")
	assert.False(t, ok, "空 userID 不應寫入")
}

// TestDelete_RemovesFromCache：Delete 後 Get 回 false（規範必含）。
func TestDelete_RemovesFromCache(t *testing.T) {
	t.Parallel()

	stub := newStubConverterClient()
	o := NewOwnership(stub, newSilentLogger())

	o.Set("job-1", "alice")
	o.Delete("job-1")
	_, ok := o.Get("job-1")
	assert.False(t, ok)

	// 重複 Delete 不該 panic
	o.Delete("job-1")
	o.Delete("never-existed")
}

// TestSet_Concurrent：100 goroutine 同時 Set 不同 job → race detector 通過。
//
// 規範必含：跑 go test -race -count=3 必綠。
func TestSet_Concurrent(t *testing.T) {
	t.Parallel()

	stub := newStubConverterClient()
	o := NewOwnership(stub, newSilentLogger())

	const N = 100
	var wg sync.WaitGroup
	wg.Add(N)
	for i := 0; i < N; i++ {
		go func(idx int) {
			defer wg.Done()
			jobID := "job-" + strconv.Itoa(idx)
			userID := "user-" + strconv.Itoa(idx%10) // 10 種 user
			o.Set(jobID, userID)
			// 立即 Get 驗 not lost
			uid, ok := o.Get(jobID)
			assert.True(t, ok)
			assert.Equal(t, userID, uid)
		}(i)
	}
	wg.Wait()

	// 驗 100 個都進去了
	for i := 0; i < N; i++ {
		jobID := "job-" + strconv.Itoa(i)
		_, ok := o.Get(jobID)
		assert.True(t, ok)
	}
}

// TestSet_Get_Delete_Concurrent_Mixed：併發 mixed write/read/delete，race detector 驗。
func TestSet_Get_Delete_Concurrent_Mixed(t *testing.T) {
	t.Parallel()

	stub := newStubConverterClient()
	o := NewOwnership(stub, newSilentLogger())

	const N = 50
	var wg sync.WaitGroup
	wg.Add(N * 3)

	for i := 0; i < N; i++ {
		jobID := "job-" + strconv.Itoa(i)
		go func() { defer wg.Done(); o.Set(jobID, "alice") }()
		go func() { defer wg.Done(); _, _ = o.Get(jobID) }()
		go func() { defer wg.Done(); o.Delete(jobID) }()
	}
	wg.Wait()
	// 不驗結果（race 驗 deadlock / 共享 state corruption 即可）
}

// ==========================================================================
// EnsureRebuilt
// ==========================================================================

// TestEnsureRebuilt_FirstCall_Fetches：第一次該 user 真的打 converter（規範必含）。
func TestEnsureRebuilt_FirstCall_Fetches(t *testing.T) {
	t.Parallel()

	stub := newStubConverterClient()
	stub.setJobs("alice", []*ConverterJob{
		{JobID: "j-1", Status: "running"},
	})

	o := NewOwnership(stub, newSilentLogger())

	err := o.EnsureRebuilt(context.Background(), "alice")
	require.NoError(t, err)
	assert.Equal(t, int32(1), stub.callCount("alice"), "首次應打 converter 1 次")

	// 驗 jobToUser 已寫入
	uid, ok := o.Get("j-1")
	assert.True(t, ok)
	assert.Equal(t, "alice", uid)
}

// TestEnsureRebuilt_SecondCall_NoOp：第二次該 user noop（atomic counter 驗，規範必含）。
func TestEnsureRebuilt_SecondCall_NoOp(t *testing.T) {
	t.Parallel()

	stub := newStubConverterClient()
	stub.setJobs("alice", []*ConverterJob{{JobID: "j-1"}})

	o := NewOwnership(stub, newSilentLogger())

	require.NoError(t, o.EnsureRebuilt(context.Background(), "alice"))
	require.NoError(t, o.EnsureRebuilt(context.Background(), "alice"))
	require.NoError(t, o.EnsureRebuilt(context.Background(), "alice"))

	assert.Equal(t, int32(1), stub.callCount("alice"),
		"成功 rebuild 後同 user 後續呼叫應 noop")
}

// TestEnsureRebuilt_DifferentUsers_EachFetch：不同 user 各自 fetch 一次（規範必含）。
func TestEnsureRebuilt_DifferentUsers_EachFetch(t *testing.T) {
	t.Parallel()

	stub := newStubConverterClient()
	stub.setJobs("alice", []*ConverterJob{{JobID: "j-a"}})
	stub.setJobs("bob", []*ConverterJob{{JobID: "j-b"}})
	stub.setJobs("carol", []*ConverterJob{})

	o := NewOwnership(stub, newSilentLogger())

	require.NoError(t, o.EnsureRebuilt(context.Background(), "alice"))
	require.NoError(t, o.EnsureRebuilt(context.Background(), "bob"))
	require.NoError(t, o.EnsureRebuilt(context.Background(), "carol"))

	assert.Equal(t, int32(1), stub.callCount("alice"))
	assert.Equal(t, int32(1), stub.callCount("bob"))
	assert.Equal(t, int32(1), stub.callCount("carol"))

	// 二次呼叫 noop
	require.NoError(t, o.EnsureRebuilt(context.Background(), "alice"))
	require.NoError(t, o.EnsureRebuilt(context.Background(), "bob"))
	assert.Equal(t, int32(1), stub.callCount("alice"))
	assert.Equal(t, int32(1), stub.callCount("bob"))
}

// TestEnsureRebuilt_Concurrent_OnlyOneFetch：同 user 100 goroutine 同時 EnsureRebuilt
// → atomic counter 驗只 fetch 一次（規範必含 — thundering herd 收斂）。
func TestEnsureRebuilt_Concurrent_OnlyOneFetch(t *testing.T) {
	t.Parallel()

	stub := newStubConverterClient()
	stub.setJobs("alice", []*ConverterJob{{JobID: "j-1"}})
	stub.fetchDelay = 50 * time.Millisecond // 故意讓 fetch 慢，放大 race window

	o := NewOwnership(stub, newSilentLogger())

	const N = 100
	var wg sync.WaitGroup
	wg.Add(N)
	errs := make(chan error, N)
	for i := 0; i < N; i++ {
		go func() {
			defer wg.Done()
			if err := o.EnsureRebuilt(context.Background(), "alice"); err != nil {
				errs <- err
			}
		}()
	}
	wg.Wait()
	close(errs)

	for err := range errs {
		t.Errorf("EnsureRebuilt 失敗: %v", err)
	}

	assert.Equal(t, int32(1), stub.callCount("alice"),
		"同 user 100 個併發 caller 應只 fetch 1 次（DCL 收斂）")
}

// TestEnsureRebuilt_Concurrent_DifferentUsers_NotBlocked：不同 user 並行 rebuild
// 互不阻塞（per-user mutex 設計驗證）。
func TestEnsureRebuilt_Concurrent_DifferentUsers_NotBlocked(t *testing.T) {
	t.Parallel()

	stub := newStubConverterClient()
	stub.fetchDelay = 200 * time.Millisecond

	const N = 10
	for i := 0; i < N; i++ {
		stub.setJobs("u-"+strconv.Itoa(i), []*ConverterJob{})
	}

	o := NewOwnership(stub, newSilentLogger())

	start := time.Now()
	var wg sync.WaitGroup
	wg.Add(N)
	for i := 0; i < N; i++ {
		uid := "u-" + strconv.Itoa(i)
		go func() {
			defer wg.Done()
			_ = o.EnsureRebuilt(context.Background(), uid)
		}()
	}
	wg.Wait()
	elapsed := time.Since(start)

	// 若 per-user mutex 失效退化成全域鎖：N=10 * 200ms = 2s
	// 並行情況：應該接近單次 fetch 200ms（加上少量 schedule overhead）
	// 用 1s 當判斷線（給 CI 足夠寬裕）
	assert.Less(t, elapsed, time.Second,
		"不同 user rebuild 應並行（per-user mutex），elapsed=%v", elapsed)
}

// TestEnsureRebuilt_ConverterError_NotMarkedRebuilt：converter 5xx → 不標 rebuilt
// → 下次再 fetch（規範必含）。
func TestEnsureRebuilt_ConverterError_NotMarkedRebuilt(t *testing.T) {
	t.Parallel()

	stub := newStubConverterClient()
	stub.setError("alice", ErrConverterUnavailable)

	o := NewOwnership(stub, newSilentLogger())

	// 第一次 fetch 失敗
	err := o.EnsureRebuilt(context.Background(), "alice")
	require.Error(t, err)
	assert.True(t, errors.Is(err, ErrConverterUnavailable))
	assert.Equal(t, int32(1), stub.callCount("alice"))

	// 第二次仍會 fetch（不標 rebuilt）
	err = o.EnsureRebuilt(context.Background(), "alice")
	require.Error(t, err)
	assert.Equal(t, int32(2), stub.callCount("alice"),
		"上次失敗後應再次 fetch")

	// 第三次成功 → 後續才會 noop
	stub.setError("alice", nil)
	stub.setJobs("alice", []*ConverterJob{{JobID: "j-1"}})
	require.NoError(t, o.EnsureRebuilt(context.Background(), "alice"))
	assert.Equal(t, int32(3), stub.callCount("alice"))

	require.NoError(t, o.EnsureRebuilt(context.Background(), "alice"))
	assert.Equal(t, int32(3), stub.callCount("alice"), "成功後才標 rebuilt")
}

// TestEnsureRebuilt_ContextCancel：ctx cancel 立即 return（規範必含）。
func TestEnsureRebuilt_ContextCancel(t *testing.T) {
	t.Parallel()

	stub := newStubConverterClient()
	stub.blockUntil = make(chan struct{}) // 永遠不放 → 強迫等 ctx
	stub.setJobs("alice", []*ConverterJob{})

	o := NewOwnership(stub, newSilentLogger())

	ctx, cancel := context.WithCancel(context.Background())

	done := make(chan error, 1)
	go func() {
		done <- o.EnsureRebuilt(ctx, "alice")
	}()

	// 等 50ms 確保 goroutine 已進到 fetch（block 在 blockUntil）
	time.Sleep(50 * time.Millisecond)
	cancel()

	select {
	case err := <-done:
		require.Error(t, err, "ctx cancel 應 return error")
		assert.True(t,
			errors.Is(err, context.Canceled) || errors.Is(err, context.DeadlineExceeded),
			"err 應為 context.Canceled 或 DeadlineExceeded，got: %v", err,
		)
	case <-time.After(2 * time.Second):
		t.Fatal("ctx cancel 後 EnsureRebuilt 沒有及時 return")
	}

	// 不標 rebuilt — 下次重試
	close(stub.blockUntil)              // 解除 block
	stub.blockUntil = nil               // 後續不再 block
	stub.setJobs("alice", []*ConverterJob{{JobID: "j-1"}})
	require.NoError(t, o.EnsureRebuilt(context.Background(), "alice"))
}

// TestEnsureRebuilt_Timeout：rebuild 內部 timeout（converter 慢 > 5s）→ return
// timeout error，不標 rebuilt。
//
// 為避免測試本身跑 5s+，把 fetchDelay 設 100ms 但用 ctx WithTimeout 50ms 模擬同樣語意：
// 驗 ctx cancel path 即可（ownership.go 的 rebuildTimeout 邏輯與此相同）。
func TestEnsureRebuilt_ParentCtxTimeout(t *testing.T) {
	t.Parallel()

	stub := newStubConverterClient()
	stub.fetchDelay = 200 * time.Millisecond
	stub.setJobs("alice", []*ConverterJob{})

	o := NewOwnership(stub, newSilentLogger())

	ctx, cancel := context.WithTimeout(context.Background(), 50*time.Millisecond)
	defer cancel()

	err := o.EnsureRebuilt(ctx, "alice")
	require.Error(t, err)
	assert.True(t, errors.Is(err, context.DeadlineExceeded),
		"parent ctx timeout 應透傳, got: %v", err)
}

// TestEnsureRebuilt_EmptyUserID：空 userID return error。
func TestEnsureRebuilt_EmptyUserID(t *testing.T) {
	t.Parallel()

	stub := newStubConverterClient()
	o := NewOwnership(stub, newSilentLogger())

	err := o.EnsureRebuilt(context.Background(), "")
	require.Error(t, err)
}

// ==========================================================================
// ActiveJobOf
// ==========================================================================

// TestActiveJobOf_AfterRebuild：rebuild 後從 jobToUser 反查到 in_progress 的 job_id（規範必含）。
func TestActiveJobOf_AfterRebuild(t *testing.T) {
	t.Parallel()

	stub := newStubConverterClient()
	stub.setJobs("alice", []*ConverterJob{
		{JobID: "j-active-1", Status: "running"},
	})

	o := NewOwnership(stub, newSilentLogger())

	// rebuild 前 ActiveJobOf 應空（cache 沒資料）
	jobs := o.ActiveJobOf("alice")
	assert.Len(t, jobs, 0)

	require.NoError(t, o.EnsureRebuilt(context.Background(), "alice"))

	// rebuild 後反查
	jobs = o.ActiveJobOf("alice")
	require.Len(t, jobs, 1)
	assert.Equal(t, "j-active-1", jobs[0])
}

// TestActiveJobOf_Empty_NoJobs：user 沒任何 job → 空 slice（規範必含）。
func TestActiveJobOf_Empty_NoJobs(t *testing.T) {
	t.Parallel()

	stub := newStubConverterClient()
	stub.setJobs("alice", []*ConverterJob{}) // 沒 active job

	o := NewOwnership(stub, newSilentLogger())
	require.NoError(t, o.EnsureRebuilt(context.Background(), "alice"))

	jobs := o.ActiveJobOf("alice")
	assert.NotNil(t, jobs, "回非 nil 空 slice 給 caller 安全 range")
	assert.Len(t, jobs, 0)
}

// TestActiveJobOf_OtherUser_NotIncluded：反查只回該 user 的，不會混到別 user。
func TestActiveJobOf_OtherUser_NotIncluded(t *testing.T) {
	t.Parallel()

	stub := newStubConverterClient()
	o := NewOwnership(stub, newSilentLogger())

	o.Set("j-alice", "alice")
	o.Set("j-bob", "bob")
	o.Set("j-alice-2", "alice")

	aliceJobs := o.ActiveJobOf("alice")
	assert.ElementsMatch(t, []string{"j-alice", "j-alice-2"}, aliceJobs)

	bobJobs := o.ActiveJobOf("bob")
	assert.ElementsMatch(t, []string{"j-bob"}, bobJobs)

	// 不存在的 user
	jobs := o.ActiveJobOf("nobody")
	assert.Len(t, jobs, 0)

	// 空 user_id
	jobs = o.ActiveJobOf("")
	assert.Nil(t, jobs)
}

// TestActiveJobOf_AfterDelete：Delete 後反查不回該 job。
func TestActiveJobOf_AfterDelete(t *testing.T) {
	t.Parallel()

	stub := newStubConverterClient()
	o := NewOwnership(stub, newSilentLogger())

	o.Set("j-1", "alice")
	o.Set("j-2", "alice")
	assert.Len(t, o.ActiveJobOf("alice"), 2)

	o.Delete("j-1")
	jobs := o.ActiveJobOf("alice")
	require.Len(t, jobs, 1)
	assert.Equal(t, "j-2", jobs[0])
}

// ==========================================================================
// 壓力測試 — 全 method 併發 race + 不死鎖
// ==========================================================================

// TestStress_AllMethods_Concurrent：所有 method 同時跑，race detector 驗 + 完成不 timeout。
func TestStress_AllMethods_Concurrent(t *testing.T) {
	t.Parallel()

	stub := newStubConverterClient()
	for i := 0; i < 5; i++ {
		uid := "u-" + strconv.Itoa(i)
		stub.setJobs(uid, []*ConverterJob{
			{JobID: fmt.Sprintf("j-%d-a", i)},
		})
	}

	o := NewOwnership(stub, newSilentLogger())

	const ROUNDS = 50
	var wg sync.WaitGroup
	for i := 0; i < ROUNDS; i++ {
		uid := "u-" + strconv.Itoa(i%5)
		jobID := "set-" + strconv.Itoa(i)

		wg.Add(5)
		go func() { defer wg.Done(); o.Set(jobID, uid) }()
		go func() { defer wg.Done(); _, _ = o.Get(jobID) }()
		go func() { defer wg.Done(); _ = o.EnsureRebuilt(context.Background(), uid) }()
		go func() { defer wg.Done(); _ = o.ActiveJobOf(uid) }()
		go func() { defer wg.Done(); o.Delete(jobID) }()
	}

	doneCh := make(chan struct{})
	go func() { wg.Wait(); close(doneCh) }()

	select {
	case <-doneCh:
		// ok
	case <-time.After(5 * time.Second):
		t.Fatal("壓力測試 5s 沒結束 — 疑似 deadlock")
	}
}