agents_and_robots/cmd/launcher/registry_test.go

package main

import (
	"context"
	"io"
	"log/slog"
	"os"
	"path/filepath"
	"sync"
	"sync/atomic"
	"testing"
	"time"

	"github.com/enmanuel/agents/devagents"
	"github.com/enmanuel/agents/internal/config"
	"github.com/enmanuel/agents/pkg/command"
	"github.com/enmanuel/agents/pkg/decision"
	"github.com/enmanuel/agents/shell/bus"
)

func TestReadReloadTarget_missing(t *testing.T) {
	got := readReloadTarget(filepath.Join(t.TempDir(), "reload.txt"))
	if got != "" {
		t.Fatalf("expected empty string for missing file, got %q", got)
	}
}

func TestReadReloadTarget_empty(t *testing.T) {
	f := filepath.Join(t.TempDir(), "reload.txt")
	if err := os.WriteFile(f, []byte(""), 0o644); err != nil {
		t.Fatal(err)
	}
	got := readReloadTarget(f)
	if got != "" {
		t.Fatalf("expected empty string for empty file, got %q", got)
	}
}

func TestReadReloadTarget_star(t *testing.T) {
	f := filepath.Join(t.TempDir(), "reload.txt")
	if err := os.WriteFile(f, []byte("*\n"), 0o644); err != nil {
		t.Fatal(err)
	}
	got := readReloadTarget(f)
	if got != "" {
		t.Fatalf("expected empty string for '*', got %q", got)
	}
}

func TestReadReloadTarget_agentID(t *testing.T) {
	f := filepath.Join(t.TempDir(), "reload.txt")
	if err := os.WriteFile(f, []byte("assistant-bot\n"), 0o644); err != nil {
		t.Fatal(err)
	}
	got := readReloadTarget(f)
	if got != "assistant-bot" {
		t.Fatalf("expected 'assistant-bot', got %q", got)
	}
}

func TestReadReloadTarget_whitespace(t *testing.T) {
	f := filepath.Join(t.TempDir(), "reload.txt")
	if err := os.WriteFile(f, []byte("  asistente-2  \n"), 0o644); err != nil {
		t.Fatal(err)
	}
	got := readReloadTarget(f)
	if got != "asistente-2" {
		t.Fatalf("expected 'asistente-2', got %q", got)
	}
}

// ── isSpecialConfig tests ─────────────────────────────────────────────────

func TestIsSpecialConfig_matchesLoadedSpecial(t *testing.T) {
	dir := t.TempDir()
	cfg := filepath.Join(dir, "config.yaml")
	if err := os.WriteFile(cfg, []byte(`
special:
  id: orchestrator
  type: orchestrator
  enabled: true
llm:
  primary:
    provider: openai
`), 0o644); err != nil {
		t.Fatal(err)
	}

	loaded := map[string]bool{"orchestrator": true}
	if !isSpecialConfig(cfg, loaded) {
		t.Fatal("expected isSpecialConfig to return true for loaded orchestrator")
	}
}

func TestIsSpecialConfig_agentConfigNotSpecial(t *testing.T) {
	dir := t.TempDir()
	cfg := filepath.Join(dir, "config.yaml")
	// An AgentConfig doesn't have special.id, so LoadSpecial will fail validation.
	if err := os.WriteFile(cfg, []byte(`
agent:
  id: father-bot
  enabled: true
matrix:
  homeserver: "https://example.com"
  user_id: "@father:example.com"
llm:
  primary:
    provider: claude-code
`), 0o644); err != nil {
		t.Fatal(err)
	}

	loaded := map[string]bool{"orchestrator": true}
	if isSpecialConfig(cfg, loaded) {
		t.Fatal("expected isSpecialConfig to return false for agent config")
	}
}

func TestIsSpecialConfig_emptyLoadedMap(t *testing.T) {
	if isSpecialConfig("any-path", nil) {
		t.Fatal("expected false when no specials loaded")
	}
	if isSpecialConfig("any-path", map[string]bool{}) {
		t.Fatal("expected false when empty specials map")
	}
}

// ── superviseUntilCanceled tests ──────────────────────────────────────────

// fakeRunner implements devagents.Runner for tests. Each Run() call increments
// runs, then closes done after a tiny delay so waitAll observes a "stopped"
// runner. RestartFactory replaces done on each cycle so the supervisor can
// recreate the runner via reload.
type fakeRunner struct {
	mu   sync.Mutex
	done chan struct{}
	runs int32
}

func newFakeRunner() *fakeRunner {
	return &fakeRunner{done: make(chan struct{})}
}

func (f *fakeRunner) Run(_ context.Context) error {
	atomic.AddInt32(&f.runs, 1)
	// Close immediately so the supervisor sees the runner as stopped.
	go func() {
		f.mu.Lock()
		ch := f.done
		f.mu.Unlock()
		// Wait briefly so waitAll has time to subscribe.
		time.Sleep(5 * time.Millisecond)
		close(ch)
	}()
	return nil
}

func (f *fakeRunner) Stop() {}

func (f *fakeRunner) RegisterCommand(_ command.Spec, _ devagents.CommandHandler) {}

func (f *fakeRunner) Done() <-chan struct{} {
	f.mu.Lock()
	defer f.mu.Unlock()
	return f.done
}

func (f *fakeRunner) reset() {
	f.mu.Lock()
	f.done = make(chan struct{})
	f.mu.Unlock()
}

// silentLogger drops everything; keeps tests quiet.
func silentLogger() *slog.Logger {
	return slog.New(slog.NewTextHandler(io.Discard, nil))
}

func TestSuperviseUntilCanceled_ReturnsWhenCtxCanceledFirst(t *testing.T) {
	r := newAgentRegistry(&launchDeps{agentBus: bus.New(silentLogger())})
	ctx, cancel := context.WithCancel(context.Background())
	cancel() // cancel before supervise starts
	done := make(chan struct{})
	go func() {
		r.superviseUntilCanceled(ctx, 50*time.Millisecond, nil, silentLogger())
		close(done)
	}()
	select {
	case <-done:
	case <-time.After(time.Second):
		t.Fatal("superviseUntilCanceled did not return after ctx canceled (empty registry)")
	}
}

func TestSuperviseUntilCanceled_ReturnsAfterCtxCancelDuringBackoff(t *testing.T) {
	// Registry has one fake runner that is already "done". waitAll returns
	// immediately, supervise enters the backoff select, ctx cancel during
	// backoff must unblock it.
	fr := newFakeRunner()
	close(fr.done) // already done

	r := newAgentRegistry(&launchDeps{agentBus: bus.New(silentLogger())})
	r.agents["fake"] = &runningAgent{
		runner:  fr,
		cfg:     &config.AgentConfig{Agent: config.AgentMeta{ID: "fake"}},
		cfgPath: "",
		logger:  silentLogger(),
	}

	ctx, cancel := context.WithCancel(context.Background())
	done := make(chan struct{})
	go func() {
		// Long backoff — only ctx cancel should unblock.
		r.superviseUntilCanceled(ctx, 10*time.Second, nil, silentLogger())
		close(done)
	}()
	// Give supervise a moment to enter the backoff select.
	time.Sleep(20 * time.Millisecond)
	cancel()
	select {
	case <-done:
	case <-time.After(2 * time.Second):
		t.Fatal("superviseUntilCanceled did not return after ctx cancel during backoff")
	}
}

func TestSuperviseUntilCanceled_CallsReloadOnAgentExit(t *testing.T) {
	// Track reloadAll invocations via a custom rulesFor function. reload reads
	// from disk via cfgPath; with cfgPath="" config.Load fails and reload
	// returns early without invoking rulesFor. So instead of inspecting
	// rulesFor calls, we observe that the supervisor loops past the backoff
	// at least once and only stops when ctx is canceled.
	fr := newFakeRunner()
	close(fr.done) // already done — waitAll returns immediately

	r := newAgentRegistry(&launchDeps{agentBus: bus.New(silentLogger())})
	r.agents["fake"] = &runningAgent{
		runner:  fr,
		cfg:     &config.AgentConfig{Agent: config.AgentMeta{ID: "fake"}},
		cfgPath: "/nonexistent/cfg.yaml", // reload will fail loading; supervisor still loops
		logger:  silentLogger(),
	}

	ctx, cancel := context.WithTimeout(context.Background(), 250*time.Millisecond)
	defer cancel()
	rulesForCalls := 0
	rulesFor := func(string, *slog.Logger) []decision.Rule {
		rulesForCalls++
		return nil
	}

	startedAt := time.Now()
	r.superviseUntilCanceled(ctx, 30*time.Millisecond, rulesFor, silentLogger())
	elapsed := time.Since(startedAt)

	// Supervisor must wait until the deadline; never exit early on its own.
	if elapsed < 200*time.Millisecond {
		t.Fatalf("supervisor returned too early: %s", elapsed)
	}
	// rulesFor is not called because reload short-circuits on bad cfgPath, but
	// the loop must have iterated through several backoff cycles (~6 with 30ms
	// backoff over 250ms). The test is best-effort — we simply assert the
	// supervisor stayed alive until ctx deadline.
	_ = rulesForCalls
}