Agent-Native/runtime/loop_event_fidelity_test.go

package runtime

import (
	"context"
	"encoding/json"
	"strings"
	"sync"
	"testing"
	"time"

	"agent-native/runtime/hooks"
	"agent-native/runtime/llm"
	"agent-native/runtime/tools"
)

type failTool struct{ reason string }

func (f failTool) Name() string                { return "echo" }
func (f failTool) Description() string         { return "echo that fails" }
func (f failTool) Parameters() json.RawMessage { return json.RawMessage(`{"type":"object"}`) }
func (f failTool) Execute(context.Context, json.RawMessage) (tools.Result, error) {
	return tools.Result{Content: f.reason, IsError: true}, nil
}

// TestLoopSameNameToolsCarryDistinctCallIDs 同名工具连续/并行调用时，
// start/end 的 CallID 必须互异且可按 ID 配对（不得按 Tool 名合并）。
func TestLoopSameNameToolsCarryDistinctCallIDs(t *testing.T) {
	for _, mode := range []string{ExecSequential, ExecParallel} {
		t.Run(mode, func(t *testing.T) {
			hits := 0
			reg := tools.NewRegistry(t.TempDir())
			reg.Register(hitTool{hits: &hits})
			var starts, ends []Event
			var mu sync.Mutex
			_, err := Run(context.Background(), Config{
				Model:         "fake",
				Tools:         reg,
				ToolExecution: mode,
				Stream: script(
					llm.Message{Role: llm.RoleAssistant, StopReason: llm.StopToolCalls, ToolCalls: []llm.ToolCall{
						{ID: "call-a", Name: "echo", Arguments: `{}`},
						{ID: "call-b", Name: "echo", Arguments: `{}`},
					}},
					llm.Message{Role: llm.RoleAssistant, Content: "done", StopReason: llm.StopEnd},
				),
				OnEvent: func(ev Event) {
					mu.Lock()
					defer mu.Unlock()
					switch ev.Type {
					case "tool_execution_start":
						starts = append(starts, ev)
					case "tool_execution_end":
						ends = append(ends, ev)
					}
				},
			}, llm.Message{Role: llm.RoleUser, Content: "go"})
			if err != nil {
				t.Fatal(err)
			}
			if hits != 2 || len(starts) != 2 || len(ends) != 2 {
				t.Fatalf("hits=%d starts=%d ends=%d", hits, len(starts), len(ends))
			}
			byStart := map[string]Event{}
			for _, ev := range starts {
				if ev.CallID == "" || ev.Tool != "echo" {
					t.Fatalf("start missing identity: %+v", ev)
				}
				if _, dup := byStart[ev.CallID]; dup {
					t.Fatalf("duplicate start CallID %q", ev.CallID)
				}
				byStart[ev.CallID] = ev
			}
			for _, ev := range ends {
				if ev.Status != "ok" {
					t.Fatalf("end status = %q, want ok: %+v", ev.Status, ev)
				}
				if _, ok := byStart[ev.CallID]; !ok {
					t.Fatalf("end CallID %q has no matching start", ev.CallID)
				}
				delete(byStart, ev.CallID)
			}
			if len(byStart) != 0 {
				t.Fatalf("unpaired starts: %+v", byStart)
			}
			if _, ok := map[string]Event{ends[0].CallID: {}, ends[1].CallID: {}}["call-a"]; !ok {
				t.Fatalf("missing call-a in ends %+v", ends)
			}
			if ends[0].CallID == ends[1].CallID {
				t.Fatalf("CallIDs must differ, both %q", ends[0].CallID)
			}
		})
	}
}

func TestLoopFailingToolEndStatusError(t *testing.T) {
	reg := tools.NewRegistry(t.TempDir())
	reg.Register(failTool{reason: "boom-detail"})
	var end Event
	_, err := Run(context.Background(), Config{
		Model: "fake",
		Tools: reg,
		Stream: script(
			llm.Message{Role: llm.RoleAssistant, StopReason: llm.StopToolCalls, ToolCalls: []llm.ToolCall{{
				ID: "c-fail", Name: "echo", Arguments: `{}`,
			}}},
			llm.Message{Role: llm.RoleAssistant, Content: "done", StopReason: llm.StopEnd},
		),
		OnEvent: func(ev Event) {
			if ev.Type == "tool_execution_end" {
				end = ev
			}
		},
	}, llm.Message{Role: llm.RoleUser, Content: "go"})
	if err != nil {
		t.Fatal(err)
	}
	if end.CallID != "c-fail" || end.Status != "error" || end.ErrorReason != "boom-detail" {
		t.Fatalf("end = %+v", end)
	}
}

func TestLoopDeniedToolEndStatusDenied(t *testing.T) {
	hits := 0
	reg := tools.NewRegistry(t.TempDir())
	reg.Register(hitTool{hits: &hits})
	bus := hooks.New()
	bus.On(func(c hooks.Call) hooks.Outcome {
		if c.Event == hooks.PreToolUse {
			return hooks.Outcome{Decision: hooks.Deny, Reason: "策略拒绝"}
		}
		return hooks.Outcome{}
	})
	var end Event
	_, err := Run(context.Background(), Config{
		Model: "fake",
		Tools: reg,
		Hooks: bus,
		Stream: script(
			llm.Message{Role: llm.RoleAssistant, StopReason: llm.StopToolCalls, ToolCalls: []llm.ToolCall{{
				ID: "c-deny", Name: "echo", Arguments: `{}`,
			}}},
			llm.Message{Role: llm.RoleAssistant, Content: "done", StopReason: llm.StopEnd},
		),
		OnEvent: func(ev Event) {
			if ev.Type == "tool_execution_end" {
				end = ev
			}
		},
	}, llm.Message{Role: llm.RoleUser, Content: "go"})
	if err != nil {
		t.Fatal(err)
	}
	if hits != 0 {
		t.Fatalf("tool ran %d times", hits)
	}
	if end.CallID != "c-deny" || end.Status != "denied" || end.ErrorReason != "策略拒绝" {
		t.Fatalf("end = %+v", end)
	}
	if end.Status == "ok" {
		t.Fatal("denied must not be reported ok")
	}
}

func TestLoopCanceledToolEndStatusCanceled(t *testing.T) {
	bt := &blockingTool{canceled: make(chan struct{})}
	reg := tools.NewRegistry(t.TempDir())
	reg.Register(bt)
	ctx, cancel := context.WithCancel(context.Background())
	firstCall := true
	stream := func(ctx context.Context, model string, messages []llm.Message, decls []llm.ToolDecl, opts llm.StreamOptions) <-chan llm.Event {
		ch := make(chan llm.Event, 4)
		go func() {
			defer close(ch)
			if firstCall {
				firstCall = false
				ch <- llm.Event{Kind: llm.KindMessageEnd, Message: llm.Message{
					Role: llm.RoleAssistant, StopReason: llm.StopToolCalls,
					ToolCalls: []llm.ToolCall{{ID: "c-cancel", Name: "block_forever", Arguments: `{}`}},
				}}
				return
			}
			select {
			case <-ctx.Done():
				ch <- llm.Event{Kind: llm.KindError, Err: ctx.Err(), Message: llm.Message{StopReason: llm.StopAborted}}
			case <-time.After(3 * time.Second):
				ch <- llm.Event{Kind: llm.KindMessageEnd, Message: llm.Message{Role: llm.RoleAssistant, Content: "done", StopReason: llm.StopEnd}}
			}
		}()
		return ch
	}
	var ends []Event
	var mu sync.Mutex
	go func() {
		time.Sleep(100 * time.Millisecond)
		cancel()
	}()
	_, err := Run(ctx, Config{
		Model:  "fake",
		Tools:  reg,
		Stream: stream,
		OnEvent: func(ev Event) {
			if ev.Type == "tool_execution_end" {
				mu.Lock()
				ends = append(ends, ev)
				mu.Unlock()
			}
		},
	}, llm.Message{Role: llm.RoleUser, Content: "block"})
	if err != nil {
		t.Fatal(err)
	}
	mu.Lock()
	defer mu.Unlock()
	if len(ends) == 0 {
		t.Fatal("expected tool_execution_end")
	}
	end := ends[0]
	if end.CallID != "c-cancel" {
		t.Fatalf("CallID = %q", end.CallID)
	}
	if end.Status != "canceled" {
		t.Fatalf("Status = %q, want canceled (not ok); end=%+v", end.Status, end)
	}
	if end.ErrorReason == "" {
		t.Fatalf("ErrorReason empty on canceled: %+v", end)
	}
}

type cancelBatchTool struct {
	cancel context.CancelFunc
}

func (c cancelBatchTool) Name() string                { return "cancel_batch" }
func (c cancelBatchTool) Description() string         { return "cancels ctx then returns" }
func (c cancelBatchTool) Parameters() json.RawMessage { return json.RawMessage(`{"type":"object"}`) }
func (c cancelBatchTool) Execute(ctx context.Context, args json.RawMessage) (tools.Result, error) {
	c.cancel()
	return tools.Result{Content: "cancelled-siblings"}, nil
}

// TestLoopSkippedCancelEmitsCanceledNotOK：串行批内前一个调用取消 ctx 后，
// 后续未执行的调用必须发 Status=canceled，不得报 ok。
func TestLoopSkippedCancelEmitsCanceledNotOK(t *testing.T) {
	ctx, cancel := context.WithCancel(context.Background())
	reg := tools.NewRegistry(t.TempDir())
	reg.Register(cancelBatchTool{cancel: cancel})
	hits := 0
	reg.Register(hitTool{hits: &hits})
	var ends []Event
	_, err := Run(ctx, Config{
		Model:         "fake",
		Tools:         reg,
		ToolExecution: ExecSequential,
		Stream: script(
			llm.Message{Role: llm.RoleAssistant, StopReason: llm.StopToolCalls, ToolCalls: []llm.ToolCall{
				{ID: "c-first", Name: "cancel_batch", Arguments: `{}`},
				{ID: "c-skip", Name: "echo", Arguments: `{}`},
			}},
			llm.Message{Role: llm.RoleAssistant, Content: "done", StopReason: llm.StopEnd},
		),
		OnEvent: func(ev Event) {
			if ev.Type == "tool_execution_end" {
				ends = append(ends, ev)
			}
		},
	}, llm.Message{Role: llm.RoleUser, Content: "go"})
	_ = err
	if hits != 0 {
		t.Fatalf("echo should not run after cancel, hits=%d", hits)
	}
	var skip *Event
	for i := range ends {
		if ends[i].CallID == "c-skip" {
			skip = &ends[i]
		}
	}
	if skip == nil {
		t.Fatalf("missing end for skipped call: %+v", ends)
	}
	if skip.Status != "canceled" {
		t.Fatalf("skipped end Status=%q want canceled: %+v", skip.Status, skip)
	}
	if skip.ErrorReason == "" || strings.Contains(skip.Status, "ok") {
		t.Fatalf("skipped end malformed: %+v", skip)
	}
}