Scope candidate blob readers to validation

internal/validation/tree.go

@@ -18,6 +18,9 @@ type CandidateTree struct {
 	byPath        map[string]int
 	directories   map[string]struct{}
 	blobs         map[string][]byte
+	blobReader    *git.BlobReader
+	closed        bool
+	closeErr      error
 }
 
 // ReadCandidateTree enumerates the captured commit without reading from the
@@ -52,6 +55,9 @@ func (tree *CandidateTree) Entries() []git.TreeEntry {
 // ReadFile returns the committed blob bytes for path. Contents are cached by
 // object ID for later reads from this tree.
 func (tree *CandidateTree) ReadFile(ctx context.Context, path string) ([]byte, error) {
+	if tree.closed {
+		return nil, fmt.Errorf("read committed path %q: candidate tree is closed", path)
+	}
 	index, ok := tree.byPath[path]
 	if !ok {
 		return nil, fmt.Errorf("committed path %q does not exist", path)
@@ -63,14 +69,36 @@ func (tree *CandidateTree) ReadFile(ctx context.Context, path string) ([]byte, e
 	if contents, ok := tree.blobs[entry.ObjectID]; ok {
 		return append([]byte(nil), contents...), nil
 	}
-	contents, err := git.ReadBlob(ctx, tree.repositoryDir, entry.ObjectID)
+	if tree.blobReader == nil {
+		// The reader belongs to the tree and can outlive an individual read
+		// context. Each Read call still applies its own cancellation to the
+		// in-flight exchange.
+		tree.blobReader = git.NewBlobReader(context.Background(), tree.repositoryDir)
+	}
+	contents, err := tree.blobReader.Read(ctx, entry.ObjectID)
 	if err != nil {
 		return nil, fmt.Errorf("read committed path %q: %w", path, err)
 	}
+	if tree.blobs == nil {
+		tree.blobs = make(map[string][]byte)
+	}
 	tree.blobs[entry.ObjectID] = append([]byte(nil), contents...)
 	return contents, nil
 }
 
+// Close closes the scoped local Git blob reader, if content was requested.
+// Metadata-only trees have no child process to close.
+func (tree *CandidateTree) Close() error {
+	if tree.closed {
+		return tree.closeErr
+	}
+	tree.closed = true
+	if tree.blobReader != nil {
+		tree.closeErr = tree.blobReader.Close()
+	}
+	return tree.closeErr
+}
+
 // Entry returns the committed metadata for path when present.
 func (tree *CandidateTree) Entry(path string) (git.TreeEntry, bool) {
 	index, ok := tree.byPath[path]