# Fix Remote URI Corruption and Leading Slash Truncation in `storage.join()`

Fix Remote URI Corruption and Leading Slash Truncation in storage.join()

Background & Problem

In src/flyte/storage/_storage.py:

def join(*paths: str) -> str:
    """
    Join multiple paths together. This is a wrapper around os.path.join.
    # TODO replace with proper join with fsspec root etc

    Args:
        paths: Paths to be joined.
    """
    return str(os.path.join(*paths))

1. Windows Path Corruption for Cloud URIs: On Windows, os.path.join("s3://bucket", "prefix", "file.txt") uses \ as separator, outputting "s3://bucket\\prefix\\file.txt". Cloud storage services and object stores reject backslashes in keys.
2. Subpaths with Leading Slash Wipe Out Prefix: In Python, os.path.join treats any argument beginning with / (e.g. storage.join("s3://bucket/dir", "/subfile.txt")) as an absolute root path and discards all preceding components, returning "/subfile.txt".
3. Empty arguments: Calling storage.join() with empty args should return "" cleanly.

User Review Required

[!NOTE] For remote paths (checked via is_remote(paths[0])), components will be joined using forward slashes (/), stripping redundant slashes from child components so prefixes are never dropped. For local paths, standard os.path.join behavior is preserved.

Proposed Changes

Core Library

[MODIFY] src/flyte/storage/_storage.py

• Update join(*paths: str) -> str to check if paths[0] is a remote URI (using is_remote(paths[0])).
• If remote: join using forward slashes /, stripping leading and trailing / from inner segments, preserving the protocol scheme.
• If local: use os.path.join(*paths).

Tests

[MODIFY] tests/internal/storage/test_storage.py

• Add test_storage_join() test covering:
  • S3 / GCS / ABFS / Flyte remote URIs with multiple path components.
  • Remote URIs where subpaths contain leading slashes (/sub/file.txt).
  • Local relative and absolute paths.
  • Empty call storage.join().

Verification Plan

Automated Tests

• Run Python verification tests exercising storage.join across remote URIs, Windows path styles, and leading-slash subpaths.