URI Normalization Inconsistency Causes Workspace Cache Misses on Windows

Summary

pylsp-rope refactoring operations fail on Windows when the LSP client sends URIs with uppercase drive letters (RFC 8089 standard file:///C:/). The root cause is inconsistent URI normalization: uris.from_fs_path() normalizes drive letters to lowercase, but workspace.put_document() stores URIs without normalization, causing dictionary key mismatches.

Environment

• OS: Windows 10/11
• pylsp version: 1.14.0
• pylsp-rope version: 0.1.17
• LSP Client: Any client following RFC 8089 standard (uppercase drive letters)

Steps to Reproduce

1. Setup: Use an LSP client that sends RFC 8089 compliant URIs with uppercase drive letters (file:///C:/)
2. Open: A Python file on Windows
3. First refactoring: Perform any refactoring operation (e.g., "Extract Variable") - works fine
4. DO NOT SAVE the file - leave it with unsaved changes from the refactoring
5. Second refactoring: Attempt another refactoring operation on the same file
6. Observe: Second refactoring fails because it operates on the old saved content from disk instead of the current unsaved content in workspace

Root Cause Analysis

The Problem

File: pylsp/uris.py (lines 135-137)

# Normalize drive paths to lower case
if RE_DRIVE_LETTER_PATH.match(path):
    path = path[0] + path[1].lower() + path[2:]

This function normalizes Windows drive letters to lowercase (C: → c:).

File: pylsp/workspace.py

• put_document() stores URIs exactly as received from client (NO normalization)
• get_maybe_document() retrieves URIs exactly as provided (NO normalization)

The Flow That Breaks

1. Client sends didOpen: file:///C:/test.py (uppercase C, RFC 8089 standard)
2. pylsp stores: workspace._docs["file:///C:/test.py"] (uppercase, as received)
3. pylsp-rope needs file content:
   • Calls uris.from_fs_path("C:\\test.py")
   • Returns file:///c:/test.py (lowercase c, normalized)
   • Calls workspace.get_maybe_document("file:///c:/test.py")
   • Returns None (key mismatch: "C:" vs "c:")
4. Fallback: Reads from filesystem (stale content)
5. Result: Refactoring operates on outdated code

Evidence from Logs

Server log shows the problem:

pylsp_rope.project - reading from filesystem: "C:\Users\...\file.py"

Instead of the expected:

pylsp_rope.project - reading from workspace: "C:\Users\...\file.py"

Client-server communication:

Client → Server (didOpen): file:///C:/Users/test.py
Server stores in _docs:    file:///C:/Users/test.py  (uppercase)
Rope generates URI:        file:///c:/Users/test.py  (lowercase)
Workspace lookup:          _docs.get("file:///c:/...") → None

Impact

• Subsequent refactoring operations fail when file has unsaved changes from previous refactoring
• First refactoring works, but second/third/etc. fail because they operate on stale disk content
• Affects all Windows users using LSP clients that follow RFC 8089 (uppercase drive letters)
• Works by accident if client sends lowercase drive letters (non-standard)
• Silent failure: No error message, refactoring just produces incorrect results based on old content

Affected LSP Clients

• Any LSP client following RFC 8089 standard (uppercase drive letters)
• Potentially: Sublime Text, Vim/Neovim, Emacs, custom editors
• VSCode likely unaffected (may send lowercase URIs)

Cross-Platform Considerations

⚠️ CRITICAL: Any fix must preserve case-sensitivity on Linux/macOS where filesystems are case-sensitive. The normalization should ONLY affect Windows drive letters (C: vs c:), not entire paths.

Safe on Windows: C:\Users\File.py and c:\users\file.py refer to the same file (case-insensitive filesystem)

NOT safe on Linux: /home/User/File.py and /home/user/file.py are DIFFERENT files (case-sensitive filesystem)

Workaround for Users

LSP client developers can work around this by normalizing drive letters to lowercase before sending URIs to pylsp:

# Client-side workaround (Python example)
if sys.platform == 'win32' and uri[8:9].isupper():
    uri = f"{uri[:8]}{uri[8].lower()}{uri[9:]}"

However, this violates RFC 8089 standard and should not be necessary.

Related Issues

This may also affect other pylsp functionality that relies on workspace document cache, not just pylsp-rope.

References

• RFC 8089 (File URI Scheme): https://tools.ietf.org/html/rfc8089
• LSP Specification: https://microsoft.github.io/language-server-protocol/
• Python pathlib.Path.as_uri() generates uppercase drive letters on Windows (standard behavior)

Thank you for maintaining pylsp! This is a subtle but impactful bug that affects Windows users following LSP standards. Happy to provide additional information or testing if needed.