Fix every failed clone being reported as a timeout

The context was cancelled before its error was checked, so any failure
looked like a timeout and private repositories never offered sign-in.
Clone handling moves into cloneRepo, which checks for DeadlineExceeded
specifically, with tests for success, failure with token masking, and
timeout.

github.go

@@ -195,6 +195,21 @@ func gitClone(ctx context.Context, repoURL, dest, token string) *exec.Cmd {
 	return cmd
 }
 
+// cloneRepo clones with a timeout and returns git's output with the token masked.
+func cloneRepo(repoURL, dest, token string) (string, error) {
+	ctx, cancel := context.WithTimeout(context.Background(), cloneTimeout)
+	defer cancel()
+	output, err := gitClone(ctx, repoURL, dest, token).CombinedOutput()
+	text := string(output)
+	if errors.Is(ctx.Err(), context.DeadlineExceeded) {
+		text = "timed out after " + cloneTimeout.String()
+	}
+	if token != "" {
+		text = strings.ReplaceAll(text, token, "***")
+	}
+	return text, err
+}
+
 // cloneFailure turns git output into a user-facing message and whether signing in (or
 // granting the app access) could fix it.
 func (g githubApp) cloneFailure(output string, signedIn bool) (message string, authRequired string) {

github_test.go

@@ -6,6 +6,7 @@ import (
 	"net/http/httptest"
 	"net/url"
 	"os/exec"
+	"path/filepath"
 	"strings"
 	"testing"
 	"time"
@@ -173,3 +174,29 @@ func TestCloneFailureMessages(t *testing.T) {
 		}
 	}
 }
+
+func TestCloneRepo(t *testing.T) {
+	if _, err := exec.LookPath("git"); err != nil {
+		t.Skip("git not installed")
+	}
+	source := t.TempDir()
+	for _, args := range [][]string{{"init", "-q"}, {"-c", "user.name=t", "-c", "user.email=t@t", "commit", "-q", "--allow-empty", "-m", "x"}} {
+		if out, err := exec.Command("git", append([]string{"-C", source}, args...)...).CombinedOutput(); err != nil {
+			t.Fatalf("git %v: %v %s", args, err, out)
+		}
+	}
+	if out, err := cloneRepo(source, filepath.Join(t.TempDir(), "ok"), ""); err != nil {
+		t.Fatalf("local clone failed: %v %s", err, out)
+	}
+
+	out, err := cloneRepo(filepath.Join(source, "missing-ghu_secret"), filepath.Join(t.TempDir(), "x"), "ghu_secret")
+	if err == nil || strings.Contains(out, "timed out") || strings.Contains(out, "ghu_secret") || !strings.Contains(out, "***") {
+		t.Fatalf("failed clone = %v %q; want a masked, non-timeout error", err, out)
+	}
+
+	defer func(old time.Duration) { cloneTimeout = old }(cloneTimeout)
+	cloneTimeout = time.Nanosecond
+	if out, err := cloneRepo(source, filepath.Join(t.TempDir(), "slow"), ""); err == nil || !strings.HasPrefix(out, "timed out") {
+		t.Fatalf("expired clone = %v %q; want timeout", err, out)
+	}
+}

main.go

@@ -3,7 +3,6 @@ package main
 import (
 	"bufio"
 	"bytes"
-	"context"
 	"crypto/rand"
 	"embed"
 	"encoding/json"
@@ -50,10 +49,13 @@ const (
 	idleExpiry   = time.Hour // snapshots unused and jobs untouched for this long are dropped
 	maxSnapshots = 8         // ponytail: fixed caps sized for the 2GB container; make them configurable if memory changes
 	maxIndexing  = 2         // concurrent index runs; each already uses every CPU
-	cloneTimeout = 10 * time.Minute
 )
 
-var sseKeepAlive = 15 * time.Second // variable so tests can shorten it
+// Variables so tests can shorten them.
+var (
+	sseKeepAlive = 15 * time.Second
+	cloneTimeout = 10 * time.Minute
+)
 
 type Server struct {
 	jobsMu      sync.Mutex
@@ -188,17 +190,7 @@ func (s *Server) createJob(w http.ResponseWriter, r *http.Request) {
 		token := s.githubAuthFor(owner).token
 		go func() {
 			checkout := filepath.Join(root, "repo")
-			ctx, cancel := context.WithTimeout(context.Background(), cloneTimeout)
-			output, err := gitClone(ctx, url, checkout, token).CombinedOutput()
-			cancel()
-			if err != nil {
-				text := string(output)
-				if ctx.Err() != nil {
-					text = "timed out after " + cloneTimeout.String()
-				}
-				if token != "" {
-					text = strings.ReplaceAll(text, token, "***")
-				}
+			if text, err := cloneRepo(url, checkout, token); err != nil {
 				message, authRequired := s.github.cloneFailure(text, token != "")
 				s.failJob(id, message, authRequired)
 				s.removeWorkspace(root)