git workspace逻辑

internal/infra/git/errors.go

@@ -23,7 +23,13 @@ func (e *GitError) Error() string {
 	for i, a := range e.Args {
 		redacted[i] = redactURLUserinfo(a)
 	}
-	return fmt.Sprintf("git %v: exit %d: %s", redacted, e.ExitCode, e.Stderr)
+	// stderr 为空说明进程多半没跑起来（chdir 失败、二进制缺失、超时被 kill），
+	// 此时 Cause 才是唯一有效信息，不能丢。
+	detail := e.Stderr
+	if detail == "" && e.Cause != nil {
+		detail = e.Cause.Error()
+	}
+	return fmt.Sprintf("git %v: exit %d: %s", redacted, e.ExitCode, detail)
 }
 
 // redactURLUserinfo 剥离形如 scheme://user[:pass]@host 中的 userinfo，

internal/infra/git/runner_test.go

@@ -54,6 +54,20 @@ func TestDefaultRunner_exec_Timeout(t *testing.T) {
 	require.Error(t, err)
 }
 
+func TestGitError_Error_EmptyStderrFallsBackToCause(t *testing.T) {
+	// 进程没跑起来时（chdir 失败 / 二进制缺失）stderr 为空，Error() 必须带上
+	// Cause，否则落库的 last_sync_error 只剩 "exit -1: " 无法定位。
+	t.Parallel()
+	ge := &GitError{Args: []string{"fetch"}, ExitCode: -1,
+		Cause: errors.New("chdir /data/x: no such file or directory")}
+	require.Contains(t, ge.Error(), "no such file or directory")
+	// stderr 非空时仍以 stderr 为准，不重复拼 Cause。
+	ge2 := &GitError{Args: []string{"fetch"}, ExitCode: 128,
+		Stderr: "fatal: boom", Cause: errors.New("exit status 128")}
+	require.Contains(t, ge2.Error(), "fatal: boom")
+	require.NotContains(t, ge2.Error(), "exit status 128")
+}
+
 func TestIsAuthError(t *testing.T) {
 	t.Parallel()
 	cases := []struct {

internal/workspace/workspace_service.go

@@ -228,6 +228,21 @@ func (s *workspaceService) Sync(ctx context.Context, c Caller, wsID uuid.UUID) (
 	}
 	// 状态回写与请求生命周期解耦：客户端断开/请求超时也要能写回状态，避免永久卡 syncing。
 	writeCtx := context.WithoutCancel(ctx)
+	// main 仓库目录缺失（历史 clone 从未成功 / 数据卷丢失）时 fetch 注定失败，
+	// 转为重新 clone 自愈：状态切回 cloning 并调度 CloneRunner，由其落 idle/error。
+	if _, statErr := os.Stat(ws.MainPath); errors.Is(statErr, os.ErrNotExist) {
+		if s.clones == nil {
+			_, _ = s.repo.UpdateWorkspaceSyncStatus(writeCtx, wsID, SyncStatusError, ws.LastSyncedAt, "main repo missing on disk; clone scheduler unavailable")
+			return nil, errs.New(errs.CodeGitCmdFailed, "main repo missing on disk")
+		}
+		updated, err := s.repo.UpdateWorkspaceSyncStatus(writeCtx, wsID, SyncStatusCloning, ws.LastSyncedAt, "")
+		if err != nil {
+			return nil, err
+		}
+		s.audit(writeCtx, &c.UserID, "workspace.reclone", "workspace", wsID.String(), map[string]any{"reason": "main path missing"})
+		s.clones.Schedule(wsID)
+		return updated, nil
+	}
 	fetchCtx, cancel := context.WithTimeout(ctx, s.fetchTimout)
 	defer cancel()
 	if fetchErr := s.gitr.Fetch(fetchCtx, ws.MainPath, cred); fetchErr != nil {

internal/workspace/workspace_service_test.go

@@ -4,6 +4,7 @@ import (
 	"context"
 	"errors"
 	"log/slog"
+	"path/filepath"
 	"sync"
 	"testing"
 	"time"
@@ -79,7 +80,21 @@ func (f *fakeRepo) ResetStuckSyncStatuses(_ context.Context) ([]uuid.UUID, error
 	return nil, nil
 }
 func (f *fakeRepo) ListFetchTargets(_ context.Context) ([]FetchTarget, error) { return nil, nil }
-func (f *fakeRepo) MarkSyncing(_ context.Context, _ uuid.UUID) (bool, error)  { return false, nil }
+
+// MarkSyncing 模拟真实实现的 CAS 语义：仅 idle/error 可抢到锁并置为 syncing。
+func (f *fakeRepo) MarkSyncing(_ context.Context, id uuid.UUID) (bool, error) {
+	f.mu.Lock()
+	defer f.mu.Unlock()
+	w, ok := f.wss[id]
+	if !ok {
+		return false, errors.New("not found")
+	}
+	if w.SyncStatus != SyncStatusIdle && w.SyncStatus != SyncStatusError {
+		return false, nil
+	}
+	w.SyncStatus = SyncStatusSyncing
+	return true, nil
+}
 func (f *fakeRepo) MarkFetchResult(_ context.Context, _ uuid.UUID, _ bool, _ string) error {
 	return nil
 }
@@ -312,6 +327,24 @@ func TestWorkspaceService_Sync_BlockedWhenCloning(t *testing.T) {
 	require.Error(t, err)
 }
 
+func TestWorkspaceService_Sync_MissingMainPath_SchedulesReclone(t *testing.T) {
+	// main 仓库目录不存在（历史 clone 失败 / 数据卷丢失）时，Sync 不做 fetch，
+	// 而是状态切回 cloning 并调度 CloneRunner 重新 clone 自愈。
+	t.Parallel()
+	pa := &fakePA{pid: uuid.New(), canWrite: true, canRead: true}
+	sched := &fakeSched{}
+	svc, repo := newSvc(t, pa, sched, &fakeGit{})
+	id := uuid.New()
+	repo.wss[id] = &Workspace{
+		ID: id, ProjectID: pa.pid, SyncStatus: SyncStatusError,
+		MainPath: filepath.Join(t.TempDir(), "gone", "main"),
+	}
+	ws, err := svc.Sync(context.Background(), Caller{UserID: uuid.New()}, id)
+	require.NoError(t, err)
+	require.Equal(t, SyncStatusCloning, ws.SyncStatus)
+	require.Equal(t, []uuid.UUID{id}, sched.called)
+}
+
 func TestCloneRunner_HappyPath(t *testing.T) {
 	t.Parallel()
 	pa := &fakePA{pid: uuid.New(), canWrite: true, canRead: true}