fix(index): cap parent index walk at git repository boundary (#56)

Prevents lumen from adopting ancestor indexes outside the current git
repository (e.g. a GOPATH-level index) when EnsureFresh is called from
a subdirectory. Without this cap, the upward walk in findEffectiveRoot
could reach indexes containing the entire workspace, causing excessive
scanning and high GPU/fan usage.

Author: aeneasr

cmd/stdio.go

@@ -30,6 +30,7 @@ import (
 	"github.com/modelcontextprotocol/go-sdk/mcp"
 	"github.com/ory/lumen/internal/config"
 	"github.com/ory/lumen/internal/embedder"
+	"github.com/ory/lumen/internal/git"
 	"github.com/ory/lumen/internal/index"
 	"github.com/ory/lumen/internal/merkle"
 	"github.com/spf13/cobra"
@@ -136,8 +137,21 @@ type indexerCache struct {
 // path passes through a directory in merkle.SkipDirs (e.g. "testdata"). Such
 // a parent index would never contain path's files, so it is not useful.
 func (ic *indexerCache) findEffectiveRoot(path string) string {
+	// Cap the upward walk at the git repository root. This prevents lumen
+	// from adopting a large ancestor index (e.g. a GOPATH index) that
+	// happens to contain path as a subdirectory, which would cause
+	// EnsureFresh to scan the entire ancestor tree.
+	gitRoot, gitErr := git.RepoRoot(path)
+
 	candidate := filepath.Dir(path)
 	for {
+		// Do not walk above the git repo root.
+		if gitErr == nil {
+			if rel, relErr := filepath.Rel(gitRoot, candidate); relErr != nil || strings.HasPrefix(rel, "..") {
+				break
+			}
+		}
+
 		if !pathCrossesSkipDir(candidate, path) {
 			if _, ok := ic.cache[candidate]; ok {
 				return candidate

cmd/stdio_test.go

@@ -17,6 +17,7 @@ package cmd
 import (
 	"context"
 	"os"
+	"os/exec"
 	"path/filepath"
 	"strings"
 	"sync"
@@ -164,6 +165,50 @@ func TestIndexerCache_FindEffectiveRoot(t *testing.T) {
 	})
 }
 
+func TestIndexerCache_FindEffectiveRoot_GitBoundary(t *testing.T) {
+	// Structure: ancestor/ (has an index DB) → repo/ (git root) → subdir/
+	// findEffectiveRoot must not walk above the git repo root to adopt the
+	// ancestor index.
+	tmpDir := t.TempDir()
+	t.Setenv("XDG_DATA_HOME", tmpDir)
+
+	const model = "test-model"
+
+	// Create directory layout.
+	ancestor := filepath.Join(tmpDir, "ancestor")
+	repo := filepath.Join(ancestor, "repo")
+	subdir := filepath.Join(repo, "subdir")
+	for _, d := range []string{ancestor, repo, subdir} {
+		if err := os.MkdirAll(d, 0o755); err != nil {
+			t.Fatal(err)
+		}
+	}
+
+	// Initialise a git repository at repo/.
+	cmd := exec.Command("git", "init", repo)
+	if out, err := cmd.CombinedOutput(); err != nil {
+		t.Fatalf("git init failed: %v\n%s", err, out)
+	}
+
+	// Create a fake index DB for the ancestor directory (above the git root).
+	ancestorDBPath := config.DBPathForProject(ancestor, model)
+	if err := os.MkdirAll(filepath.Dir(ancestorDBPath), 0o755); err != nil {
+		t.Fatal(err)
+	}
+	if err := os.WriteFile(ancestorDBPath, []byte{}, 0o644); err != nil {
+		t.Fatal(err)
+	}
+
+	ic := &indexerCache{
+		cache: make(map[string]cacheEntry),
+		model: model,
+	}
+	root := ic.findEffectiveRoot(subdir)
+	if root != subdir {
+		t.Fatalf("expected findEffectiveRoot to stop at git boundary and return %s, got %s", subdir, root)
+	}
+}
+
 func TestIndexerCache_GetOrCreate_ReusesParentIndex(t *testing.T) {
 	tmpDir := t.TempDir()
 	t.Setenv("XDG_DATA_HOME", tmpDir)

internal/git/worktree.go

@@ -93,6 +93,28 @@ func InternalWorktreePaths(projectPath string) []string {
 	return result
 }
 
+// RepoRoot returns the absolute path of the root directory of the git
+// repository containing projectPath, by running "git rev-parse --show-toplevel".
+// Returns an error if git is not available or projectPath is not inside a git
+// repository.
+func RepoRoot(projectPath string) (string, error) {
+	ctx, cancel := context.WithTimeout(context.Background(), 5*time.Second)
+	defer cancel()
+
+	cmd := exec.CommandContext(ctx, "git", "rev-parse", "--show-toplevel")
+	cmd.Dir = projectPath
+	out, err := cmd.Output()
+	if err != nil {
+		return "", err
+	}
+
+	root := strings.TrimSpace(string(out))
+	if resolved, err := filepath.EvalSymlinks(root); err == nil {
+		root = resolved
+	}
+	return root, nil
+}
+
 // ListWorktrees returns the absolute paths of all worktrees (including the
 // main working tree) for the repository containing projectPath. Returns nil
 // if git is not available or projectPath is not inside a git repository.