@debugmcp/mcp-debugger/docs/architecture/task-22-container-test-findings.md

Run-time step-through debugging for LLM agents.

# Task 22: Container Test Failure Investigation

## Summary

Following Task 21's "hands-off" path handling refactor, we have 1 remaining test failure in the container environment.

## Findings

### Test Status
- ✅ Error scenarios test: **FIXED** - Now accepts either 'error' or 'stopped' state with increased timeout
- ❌ Container fibonacci test: **STILL FAILING** - Path truncation issue

### Root Cause Analysis

The container test shows a specific error pattern:
```
FileNotFoundError: [Errno 2] No such file or directory: 'examples/python'
```

When passing `'examples/python/fibonacci.py'`, debugpy is receiving only `'examples/python'` and trying to execute it as a command.

### Investigation Results

1. **Path is correct in our code**: The launch request correctly sends `program: 'examples/python/fibonacci.py'`
2. **Issue is container-specific**: Only happens in the bundled container environment
3. **Truncation pattern**: The path is being truncated at the directory separator
4. **Debugpy interpretation**: It's treating the truncated path as an executable, not a script path

### Hypothesis

The issue appears to be in how the bundled code or the container environment is handling the launch arguments. Possible causes:
1. The bundling process (esbuild) might be mangling the launch arguments
2. The IPC communication between processes might have an issue with path separators
3. The container's working directory setup might be affecting path resolution

### Next Steps

To maintain the hands-off philosophy:
1. The issue appears to be a bug in the container/bundling layer, not in the path handling logic
2. This may require investigation of the bundling process or IPC communication
3. The test failure is legitimate - it's catching a real issue that would affect users

### Recommendation

This appears to be a deeper issue with the container environment that's outside the scope of the path handling refactor. The hands-off approach is correct, but there's a bug in how paths are being communicated to debugpy in the container.