Cyrene_v2/internal/services/gateway/subagent_backfill_test.go

package gateway

import (
	"context"
	"errors"
	"io"
	"log/slog"
	"strings"
	"sync"
	"testing"
	"time"

	"git.yeij.top/AskaEth/Cyrene_v2/internal/bus"
	"git.yeij.top/AskaEth/Cyrene_v2/internal/runtime/history"
	"git.yeij.top/AskaEth/Cyrene_v2/internal/runtime/model"
	"git.yeij.top/AskaEth/Cyrene_v2/internal/runtime/orchestration"
	"git.yeij.top/AskaEth/Cyrene_v2/internal/runtime/transcript"
)

// 本文件锁住 M32 的两条面（design/05 §2/§5/§9）与它们**不互相冒充**这条红线：
//
//   - 可观测 / 历史面：父会话多出**恰好一条** `role=system` 的注记行（不是 user、
//     不是 assistant ⇒ 不是「她说的」），并随既有的 `message_persisted` 扇出；
//   - 模型面：同一行经转录播种后在下一回合是一条 **system** 消息，且**绝不**进入
//     记忆抽取的输入面。
//
// 每条断言都对着一个可能被写坏的实现：把角色写成 assistant（人格冒充）、把
// `turn_id` 编一个（冒充某一路回合）、把注记漏掉标记（于是它会被当成对话吞进记忆
// 库）、或者落库失败还照样广播（假的「已回填」）。

// noteStore 是回填落库面的测试替身：记录每次写入并按需注入失败。
type noteStore struct {
	mu      sync.Mutex
	written []history.Message
	err     error
}

func (s *noteStore) AppendFull(_ context.Context, msg history.Message) (history.Message, error) {
	s.mu.Lock()
	defer s.mu.Unlock()
	if s.err != nil {
		return history.Message{}, s.err
	}
	if msg.ID == "" {
		msg.ID = "msg-backfill-1"
	}
	if msg.CreatedAt == 0 {
		msg.CreatedAt = 1
	}
	s.written = append(s.written, msg)
	return msg, nil
}

func (s *noteStore) rows() []history.Message {
	s.mu.Lock()
	defer s.mu.Unlock()
	return append([]history.Message(nil), s.written...)
}

// noteSink 是下行扇出的测试替身。
type noteSink struct {
	mu   sync.Mutex
	sent []history.Message
}

func (s *noteSink) BroadcastPersisted(msg history.Message) {
	s.mu.Lock()
	defer s.mu.Unlock()
	s.sent = append(s.sent, msg)
}

func (s *noteSink) frames() []history.Message {
	s.mu.Lock()
	defer s.mu.Unlock()
	return append([]history.Message(nil), s.sent...)
}

// recordSink 收集结构化日志记录，使「失败必须可见」能被断言而不是靠人看输出。
type recordSink struct {
	mu   sync.Mutex
	recs []slog.Record
}

func (r *recordSink) Enabled(context.Context, slog.Level) bool { return true }

func (r *recordSink) Handle(_ context.Context, rec slog.Record) error {
	r.mu.Lock()
	defer r.mu.Unlock()
	r.recs = append(r.recs, rec.Clone())
	return nil
}

func (r *recordSink) WithAttrs([]slog.Attr) slog.Handler { return r }
func (r *recordSink) WithGroup(string) slog.Handler      { return r }

// has 报告是否有一条指定级别且 msg 含 want 的记录。
func (r *recordSink) has(level slog.Level, want string) bool {
	r.mu.Lock()
	defer r.mu.Unlock()
	for _, rec := range r.recs {
		if rec.Level == level && strings.Contains(rec.Message, want) {
			return true
		}
	}
	return false
}

// backfillFor 构建一个带真 bus（不订阅）的回填订阅者与它的两个替身。
func backfillFor(t *testing.T, log *slog.Logger) (*SubagentBackfill, *noteStore, *noteSink) {
	t.Helper()
	b := bus.New()
	t.Cleanup(func() { _ = b.Close() })
	store, sink := &noteStore{}, &noteSink{}
	if log == nil {
		log = slog.New(slog.NewTextHandler(io.Discard, nil))
	}
	bf, err := NewSubagentBackfill(SubagentBackfillDeps{Bus: b, History: store, Broadcast: sink, Log: log})
	if err != nil {
		t.Fatalf("NewSubagentBackfill: %v", err)
	}
	return bf, store, sink
}

// settledResult 构造一个与 SpawnTool.await 同形同键集的已结算批次载荷。
func settledResult(session string, agents ...orchestration.SpawnAgentResult) *orchestration.SubagentResultEvent {
	return &orchestration.SubagentResultEvent{
		SessionID: session,
		SpawnResult: &orchestration.SpawnResult{
			Status:  "done",
			Summary: "子代理汇总: 共 1 个, 1 成功",
			Agents:  agents,
		},
	}
}

// TestSubagentBackfillWritesSystemNoteOnly 是 M32 的第一条面：结果写成父会话的一条
// **系统注记**，绝不写成她的人格发言，也不冒充任何一路回合。
func TestSubagentBackfillWritesSystemNoteOnly(t *testing.T) {
	t.Parallel()
	bf, store, sink := backfillFor(t, nil)

	payload := settledResult("web", orchestration.SpawnAgentResult{
		ID: "sub-a1", Goal: "调研 X", Role: "cyrene", Mode: "engineer",
		Status: "done", Summary: "结论：X 可行", Text: "结论：X 可行", Steps: 2,
	})
	if err := bf.Handle(context.Background(), bus.Event{Subject: orchestration.SubagentResultSubject, Payload: payload}); err != nil {
		t.Fatalf("Handle: %v", err)
	}

	rows := store.rows()
	if len(rows) != 1 {
		t.Fatalf("persisted %d rows, want exactly one system note (no fabricated speech): %+v", len(rows), rows)
	}
	row := rows[0]
	if row.SessionID != "web" {
		t.Fatalf("row session = %q, want the parent session %q", row.SessionID, "web")
	}
	if row.Role != history.RoleSystem {
		t.Fatalf("row role = %q, want %q (the note is not the owner's words and not the persona's)", row.Role, history.RoleSystem)
	}
	// 显式锁住「不冒充」：既不是 user、也不是 assistant —— 这两条是本条红线的全部内容。
	for _, forbidden := range []string{history.RoleUser, history.RoleAssistant} {
		if row.Role == forbidden {
			t.Fatalf("row role = %q: a sub-agent conclusion must never be written as %s", row.Role, forbidden)
		}
	}
	if !strings.HasPrefix(row.Content, subagentNoteMarker) {
		t.Fatalf("content = %q, want the %q marker (it is what keeps the note out of memory extraction)", row.Content, subagentNoteMarker)
	}
	for _, want := range []string{"sub-a1", "cyrene/engineer", "结论：X 可行"} {
		if !strings.Contains(row.Content, want) {
			t.Fatalf("content = %q, want it to name %q (attribution must be traceable)", row.Content, want)
		}
	}
	if row.TurnID != "" {
		t.Fatalf("turn_id = %q, want empty: a background batch may settle after its turn ended and belongs to no lane", row.TurnID)
	}
	if row.PersonaID != "" {
		t.Fatalf("persona_id = %q, want empty: the note's author is no persona", row.PersonaID)
	}

	frames := sink.frames()
	if len(frames) != 1 || frames[0].ID != row.ID {
		t.Fatalf("broadcast %+v, want exactly the persisted row %q", frames, row.ID)
	}
}

// TestSubagentBackfillNoteIsBoundedAndPerAgent 锁住注记的**形状**：批次逐代理一行、
// 每行点名 id，且整条有界（失控子代理不能淹掉父会话上下文）。
func TestSubagentBackfillNoteIsBoundedAndPerAgent(t *testing.T) {
	t.Parallel()
	long := strings.Repeat("长", maxSubagentNoteBodyRunes*3)
	payload := settledResult("web",
		orchestration.SpawnAgentResult{ID: "sub-a", Status: "done", Summary: "结论A"},
		orchestration.SpawnAgentResult{ID: "sub-b", Status: "failed", Summary: long},
		orchestration.SpawnAgentResult{ID: "sub-c", Status: "aborted"},
	)
	note := subagentNote(payload)
	for _, want := range []string{"[子代理结果] 3 个子代理批次已完成", "- sub-a 已完成：结论A", "- sub-b 失败：", "已中止", subagentNoteEmptyBody} {
		if !strings.Contains(note, want) {
			t.Fatalf("note = %q, want it to contain %q", note, want)
		}
	}
	if got := len([]rune(note)); got > maxSubagentNoteRunes {
		t.Fatalf("note is %d runes, want at most %d", got, maxSubagentNoteRunes)
	}
	// 单代理批次的正文也不能无限长。
	single := subagentNote(settledResult("web", orchestration.SpawnAgentResult{ID: "sub-x", Status: "done", Text: long}))
	if got := len([]rune(single)); got > maxSubagentNoteRunes {
		t.Fatalf("single-agent note is %d runes, want at most %d", got, maxSubagentNoteRunes)
	}
	if !strings.Contains(single, "子代理 sub-x 已完成") {
		t.Fatalf("single-agent note = %q, want the one-line shape", single)
	}
}

// TestSubagentBackfillSkipsUnusablePayloads 锁住两种「不能写」的情形：载荷形状不对，
// 或没有父会话可归属。两者都必须留一条 WARN（绝不静默丢结果）。
func TestSubagentBackfillSkipsUnusablePayloads(t *testing.T) {
	t.Parallel()
	sinkLog := &recordSink{}
	bf, store, sink := backfillFor(t, slog.New(sinkLog))

	cases := []struct {
		name    string
		subject string
		payload any
	}{
		{name: "wrong payload type", subject: orchestration.SubagentResultSubject, payload: map[string]any{"sessionId": "web"}},
		{name: "no spawn result", subject: orchestration.SubagentResultSubject, payload: &orchestration.SubagentResultEvent{SessionID: "web"}},
		{name: "no parent session", subject: orchestration.SubagentResultSubject, payload: settledResult("", orchestration.SpawnAgentResult{ID: "sub-a", Status: "done"})},
		{name: "other subject", subject: "task.done", payload: settledResult("web")},
	}
	for _, tc := range cases {
		if err := bf.Handle(context.Background(), bus.Event{Subject: tc.subject, Payload: tc.payload}); err != nil {
			t.Fatalf("%s: Handle: %v", tc.name, err)
		}
	}
	if rows := store.rows(); len(rows) != 0 {
		t.Fatalf("persisted %+v, want no rows for unusable payloads", rows)
	}
	if frames := sink.frames(); len(frames) != 0 {
		t.Fatalf("broadcast %+v, want no frames for unusable payloads", frames)
	}
	if !sinkLog.has(slog.LevelWarn, "unexpected payload") {
		t.Fatal("a malformed payload must leave a WARN, not vanish silently")
	}
	if !sinkLog.has(slog.LevelWarn, "no parent session") {
		t.Fatal("a result with no parent session must leave a WARN, not vanish silently")
	}
	if !sinkLog.has(slog.LevelWarn, "carries no spawn result") {
		t.Fatal("a payload without a spawn result must leave a WARN (the bus swallows panics)")
	}
}

// TestSubagentBackfillStoreFailureIsNotSilent 锁住「落库失败 ⇒ 既不广播也不静默」：
// 广播一条并不存在的行会让客户端以为回填成立（假成功）。
func TestSubagentBackfillStoreFailureIsNotSilent(t *testing.T) {
	t.Parallel()
	sinkLog := &recordSink{}
	bf, store, sink := backfillFor(t, slog.New(sinkLog))
	store.err = errors.New("db down")

	if err := bf.Handle(context.Background(), bus.Event{
		Subject: orchestration.SubagentResultSubject,
		Payload: settledResult("web", orchestration.SpawnAgentResult{ID: "sub-a", Status: "done", Summary: "结论"}),
	}); err != nil {
		t.Fatalf("Handle: %v", err)
	}
	if frames := sink.frames(); len(frames) != 0 {
		t.Fatalf("broadcast %+v after a failed write, want none (never claim a row that does not exist)", frames)
	}
	if !sinkLog.has(slog.LevelError, "persist result note") {
		t.Fatal("a failed write must leave an ERROR with its source, not vanish silently")
	}
}

// TestSubagentBackfillStartSubscribesBus 锁住「Start 真的挂上总线」：真 bus 上发布
// 一条结果 ⇒ 父会话多出一条注记行。少了这一步，整条回填就只是一个没人调用的方法。
func TestSubagentBackfillStartSubscribesBus(t *testing.T) {
	t.Parallel()
	b := bus.New()
	t.Cleanup(func() { _ = b.Close() })
	store := &noteStore{}
	bf, err := NewSubagentBackfill(SubagentBackfillDeps{Bus: b, History: store, Log: slog.New(slog.NewTextHandler(io.Discard, nil))})
	if err != nil {
		t.Fatalf("NewSubagentBackfill: %v", err)
	}
	if err := bf.Start(context.Background()); err != nil {
		t.Fatalf("Start: %v", err)
	}
	if err := b.Publish(context.Background(), orchestration.SubagentResultSubject,
		settledResult("web", orchestration.SpawnAgentResult{ID: "sub-a", Status: "done", Summary: "结论"})); err != nil {
		t.Fatalf("publish: %v", err)
	}
	deadline := time.Now().Add(2 * time.Second)
	for time.Now().Before(deadline) {
		if len(store.rows()) > 0 {
			break
		}
		time.Sleep(5 * time.Millisecond)
	}
	rows := store.rows()
	if len(rows) != 1 || rows[0].Role != history.RoleSystem {
		t.Fatalf("rows = %+v, want one system note from the bus subscription", rows)
	}
}

// TestSubagentBackfillRequiresStoreAndBus 锁住装配期的诚实失败：没有落库面的订阅者
// 只会把结果丢掉，却让「已接线」看起来成立。
func TestSubagentBackfillRequiresStoreAndBus(t *testing.T) {
	t.Parallel()
	if _, err := NewSubagentBackfill(SubagentBackfillDeps{Bus: bus.New()}); err == nil {
		t.Fatal("NewSubagentBackfill without a history store must fail")
	}
	if _, err := NewSubagentBackfill(SubagentBackfillDeps{History: &noteStore{}}); err == nil {
		t.Fatal("NewSubagentBackfill without a bus must fail")
	}
}

// TestBackfilledNoteNeverReframedAsAssistant 是两条面之间的**接缝**断言：落库行经
// 转录播种（下一回合真正会走的那条路）之后，是一条 `model.RoleSystem` 消息 —— 不是
// assistant，因此模型不会把它读成「她自己说过的话」。
func TestBackfilledNoteNeverReframedAsAssistant(t *testing.T) {
	t.Parallel()
	bf, store, _ := backfillFor(t, nil)
	if err := bf.Handle(context.Background(), bus.Event{
		Subject: orchestration.SubagentResultSubject,
		Payload: settledResult("web", orchestration.SpawnAgentResult{ID: "sub-a", Status: "done", Summary: "结论"}),
	}); err != nil {
		t.Fatalf("Handle: %v", err)
	}
	rows := store.rows()
	if len(rows) != 1 {
		t.Fatalf("rows = %+v, want one", rows)
	}
	seeded := transcript.Message(rows[0])
	if seeded.Role != model.RoleSystem {
		t.Fatalf("seeded role = %q, want system: the next turn must read the note as a system note", seeded.Role)
	}
	if seeded.SenderName != "" {
		t.Fatalf("seeded sender = %q, want empty: no persona may be credited with the sub-agent's words", seeded.SenderName)
	}
}