Refactor logging system and improve device redirection

- Optimized logging with a new TDLogger module
- Simplified logging interface by removing verbosity levels
- Enhanced log message filtering and context tracking
- Improved device redirection and stream/event handling
- Updated README, tests, and demo scripts to reflect changes
- Bumped version to 0.0.2 to reflect significant improvements

Author: unixwzrd

.gitignore

@@ -296,3 +296,5 @@ $RECYCLE.BIN/
 test_projects/cuda-pytorch-template
 .project-planning/
 test_projects/
+backup/
+*.bak*

CHANGELOG.md

@@ -1,5 +1,39 @@
 # CHANGELOG
 
+## 2025-03-08 - 0.0.3
+
+### Logger Improvements
+- Optimized the `_logged_messages` collection by replacing the set with a fixed-size deque
+- This change improves memory management by automatically removing oldest entries when the collection is full
+- Eliminates the need to clear the entire collection, providing more consistent duplicate prevention
+- Consolidated duplicate code by extracting common logic into helper functions:
+  - Added `is_test_environment()` to determine if running in a test environment
+  - Added `is_internal_frame()` to identify frames that should be skipped
+  - Added `should_skip_message()` to centralize message filtering logic
+  - Added `is_setup_or_init()` to identify setup and initialization functions
+- Improved caller identification to ensure log messages show the actual caller
+- Enhanced message filtering to ensure important redirection messages are always logged
+
+### Test Framework Improvements
+- Created a robust test framework with a `PrefixedTestCase` class in `test_utils.py`
+- Fixed test discovery and execution to ensure consistent environment variables
+- Improved logging during tests with better context and error handling
+- Enhanced expected output file management for more reliable test results
+- Fixed issues with program name consistency in log messages during tests
+
+## 2025-03-03 - 0.0.2
+
+### Logging System Improvements
+- Modularized logging: Moved logging functionality into its own module (`TDLogger.py`)
+- Simplified logging interface:
+  - Removed verbosity levels (LOG_VERBOSITY) in favor of a simpler on/off approach
+  - Eliminated different message classes (warning, info, error) for a more streamlined logging experience
+  - Consolidated all logging to use a single `log_message` function as the primary entry point
+  - Removed redundant `log` function since the project hasn't been publicly released yet
+- Improved caller tracking for more accurate log messages
+- Optimized memory usage by limiting the size of the logged messages cache
+- Enhanced test suite with expected output validation
+
 ## 2024-12-30 - 0.0.1
 
 - Added note regarding building NumPy for Apple silicon.

README.md

@@ -141,7 +141,6 @@ device = torch.device('cuda')  # This will be redirected based on available hard
   - If MPS is requested but not available, it will redirect to CUDA if available.
   - If neither is available, it will default to CPU.
 - **Logging**: The module outputs log messages indicating how calls are intercepted and handled. These messages include the caller's filename, function name, and line number.
-- **Verbosity Control**: You can control the verbosity of logging by setting the `TORCHDEVICE_LOG_VERBOSITY` environment variable (0=minimal, 1=verbose, 2=debug).
 - **Log File**: You can direct logs to a file by setting the `TORCHDEVICE_LOG_FILE` environment variable.
 - **Unsupported Functions**: Functions that are not supported on the current hardware are stubbed and will log a warning but allow execution to continue.
 - **Stream and Event Support**: CUDA streams and events are fully supported on MPS devices, allowing for asynchronous operations and event timing.

TODO.md

@@ -0,0 +1,18 @@
+# TorchDevice Project TODO List
+
+## Logger Improvements
+
+- **Simplify the Frame Inspection Logic**
+  - Review the current implementation of `get_caller_info()`
+  - Identify opportunities for simplification
+  - Implement a more direct approach to find the caller frame
+
+- **Implement Caching for Performance**
+  - Identify functions that are called frequently with the same arguments
+  - Add caching for these functions (e.g., using functools.lru_cache)
+  - Measure the performance impact of these changes
+
+- **Use Context Managers for Temporary State**
+  - Identify operations that temporarily modify state
+  - Implement context managers for these operations
+  - Ensure proper cleanup even in case of exceptions