fix(cp): stream stdout tar output instead of accumulating in memory

Use header.Size to write tar header upfront, then stream binary data
directly to tar.Writer. This enables copying large files without
accumulating them entirely in memory.

Author: rgarcia

pkg/cmd/cp.go

@@ -859,7 +859,7 @@ func copyFromInstanceToStdout(ctx context.Context, baseURL, apiKey, instanceID,
 	defer tw.Close()
 
 	var currentHeader *cpFileHeader
-	var fileData []byte
+	var bytesWritten int64
 	var receivedFinal bool
 
 	for {
@@ -881,19 +881,19 @@ func copyFromInstanceToStdout(ctx context.Context, baseURL, apiKey, instanceID,
 
 			switch msgTypeStr {
 			case "header":
-				// Flush previous file if any (but not symlinks - they have no data)
+				// Verify previous file was completely written
 				if currentHeader != nil && !currentHeader.IsDir && !currentHeader.IsSymlink {
-					if err := writeTarEntry(tw, currentHeader, fileData, archive); err != nil {
-						return err
+					if bytesWritten != currentHeader.Size {
+						return fmt.Errorf("file %s: expected %d bytes, got %d", currentHeader.Path, currentHeader.Size, bytesWritten)
 					}
-					fileData = nil
 				}
 
 				var header cpFileHeader
 				if err := json.Unmarshal(message, &header); err != nil {
 					return fmt.Errorf("parse header: %w", err)
 				}
 				currentHeader = &header
+				bytesWritten = 0
 
 				if header.IsDir {
 					// Write directory entry to tar
@@ -928,15 +928,31 @@ func copyFromInstanceToStdout(ctx context.Context, baseURL, apiKey, instanceID,
 					if err := tw.WriteHeader(tarHeader); err != nil {
 						return fmt.Errorf("write tar symlink header: %w", err)
 					}
+				} else {
+					// Write regular file header with known size - enables streaming
+					tarHeader := &tar.Header{
+						Typeflag: tar.TypeReg,
+						Name:     header.Path,
+						Size:     header.Size,
+						Mode:     int64(header.Mode),
+						ModTime:  time.Unix(header.Mtime, 0),
+					}
+					// Only preserve UID/GID in archive mode
+					if archive {
+						tarHeader.Uid = int(header.Uid)
+						tarHeader.Gid = int(header.Gid)
+					}
+					if err := tw.WriteHeader(tarHeader); err != nil {
+						return fmt.Errorf("write tar header: %w", err)
+					}
 				}
 
 			case "end":
-				// Write final file data if any
+				// Verify file was completely written
 				if currentHeader != nil && !currentHeader.IsDir && !currentHeader.IsSymlink {
-					if err := writeTarEntry(tw, currentHeader, fileData, archive); err != nil {
-						return err
+					if bytesWritten != currentHeader.Size {
+						return fmt.Errorf("file %s: expected %d bytes, got %d", currentHeader.Path, currentHeader.Size, bytesWritten)
 					}
-					fileData = nil
 				}
 				currentHeader = nil
 
@@ -953,8 +969,12 @@ func copyFromInstanceToStdout(ctx context.Context, baseURL, apiKey, instanceID,
 				return fmt.Errorf("copy error at %s: %s", cpErr.Path, cpErr.Message)
 			}
 		} else if msgType == websocket.BinaryMessage {
-			// Accumulate file data
-			fileData = append(fileData, message...)
+			// Stream file data directly to tar archive
+			n, err := tw.Write(message)
+			if err != nil {
+				return fmt.Errorf("write tar data: %w", err)
+			}
+			bytesWritten += int64(n)
 		}
 	}
 
@@ -965,26 +985,4 @@ func copyFromInstanceToStdout(ctx context.Context, baseURL, apiKey, instanceID,
 	return nil
 }
 
-// writeTarEntry writes a file entry to the tar archive
-func writeTarEntry(tw *tar.Writer, header *cpFileHeader, data []byte, archive bool) error {
-	tarHeader := &tar.Header{
-		Typeflag: tar.TypeReg,
-		Name:     header.Path,
-		Size:     int64(len(data)),
-		Mode:     int64(header.Mode),
-		ModTime:  time.Unix(header.Mtime, 0),
-	}
-	// Only preserve UID/GID in archive mode
-	if archive {
-		tarHeader.Uid = int(header.Uid)
-		tarHeader.Gid = int(header.Gid)
-	}
-	if err := tw.WriteHeader(tarHeader); err != nil {
-		return fmt.Errorf("write tar header: %w", err)
-	}
-	if _, err := tw.Write(data); err != nil {
-		return fmt.Errorf("write tar data: %w", err)
-	}
-	return nil
-}