Fix module caching issue causing intermittent empty output (issue #16)

[Copilot]

Fixes #16 - resolves the intermittent issue where atlas-provider-sqlalchemy would sometimes output nothing, particularly after git branch switches.

Problem

The issue was caused by Python's module caching mechanism in the get_metadata() function. When importing SQLAlchemy model files using importlib.util.spec_from_file_location(), the function used non-unique module names (just file_path.stem), which could lead to stale cached modules being returned instead of fresh imports of changed files.

This manifested as:

• Empty output on first run after git branch switches
• Inconsistent behavior where second runs would work correctly
• The workaround PYTHONDONTWRITEBYTECODE=1 fixing the issue (confirming bytecode caching was the culprit)

Solution

The fix implements unique module naming and cache invalidation:

1. Unique module names: Instead of using file_path.stem (e.g., "models"), now uses f"atlas_dynamic_module_{abs(hash(str(file_path.absolute())))}_{file_mtime}" to ensure each import gets a unique module name based on file path and modification time.

2. Cache invalidation: Explicitly clears stale modules from sys.modules and calls importlib.invalidate_caches() to force fresh loading.

3. File change detection: Includes file modification time in the module name to detect when files have changed (e.g., after git branch switches).

Testing

Added comprehensive test suite (tests/test_caching_fix.py) that validates:

• File changes are properly detected and reflected
• Multiple files with same stem names don't interfere
• File modification time tracking works correctly

All existing tests continue to pass, ensuring no regression in functionality.

Example

Before the fix, this scenario could fail:

git checkout feature-branch
atlas-provider-sqlalchemy --path models --dialect postgresql
# Sometimes outputs nothing

atlas-provider-sqlalchemy --path models --dialect postgresql  
# Second run would work

After the fix, both runs consistently produce the correct output.

---

💡 You can make Copilot smarter by setting up custom instructions, customizing its development environment and configuring Model Context Protocol (MCP) servers. Learn more Copilot coding agent tips in the docs.

[QuintenBruynseraede]

This fix seems to work 👍