coredump-debugger

Dpdb - Python Core Dump System

A comprehensive Python "core dump" system that captures execution state (variables, frames, stack traces) for post-mortem debugging, similar to C++ core dumps.

Features

• Automatic Core Dumps: Automatically create core dumps when unhandled exceptions occur
• Manual Core Dumps: Create core dumps at any point in your code
• Post-Mortem Debugger: Interactive debugger to analyze core dumps with rich terminal UI
• Command-Line Interface: Run programs with automatic crash handling or debug existing core dumps
• Adaptive Theme Detection: Automatically detects light/dark terminal themes for optimal color schemes
• Complete State Capture: Captures local variables, global variables, stack frames, and code context
• Persistent Storage: Core dumps are saved as pickle files for later analysis
• pdb-like Interface: Familiar debugging commands for easy use

Quick Start

0. Installation

Requirements:

• Python 3.8+
• Rich - For terminal UI and formatting
• term_background (optional) - For improved terminal theme detection

# Basic installation
pip install coredump-debugger

# With optional dependencies for better theme detection
pip install coredump-debugger[dev]

The debugger will work without term_background but will have more accurate theme detection when it's available.

After installation, the dpdb command will be available in your system PATH. The package automatically creates a console script entry point that installs the dpdb executable.

1. Command-Line Usage

Dpdb provides a command-line interface similar to pdb for running programs with automatic crash handling:

# Run a Python program with automatic core dump generation
dpdb my_program.py arg1 arg2

# Debug an existing core dump file
dpdb crash_dump.pkl

# Get help
dpdb --help

Mode 1: Program Execution with Crash Handling

dpdb my_program.py arg1 arg2

This automatically installs the global core dump handler and runs your program. If the program crashes, a core dump file will be created automatically.

Mode 2: Core Dump Debugging

dpdb crash_dump.pkl

This loads an existing core dump file and starts the interactive debugger.

2. Programmatic API

Global Exception Handler

Just add dpdb.install_exception_handler("path/to/core_dump.pkl") to your code, and a core dump will be created and saved automatically when an exception occurs.

import dpdb

# Install automatic core dump generation
dpdb.install_exception_handler("my_app_crash.pkl")

# Your application code here
def my_function():
    x = 42
    y = "hello"
    # This will create a core dump if an exception occurs
    result = 1 / 0  # ZeroDivisionError

my_function()

Manual Core Dump

import dpdb

def debug_point():
    local_var = "debug info"
    data = [1, 2, 3, 4, 5]
    
    # Create a manual core dump
    dump = dpdb.CoreDumpGenerator.create_from_current_stack()
    dpdb.save_core_dump(dump, "debug_dump.pkl")

debug_point()

3. Post-Mortem Debugging

Attach Dpdb to a core dump file:

# Debug a core dump file
dpdb my_app_crash.pkl

# Or using the module
python -m dpdb my_app_crash.pkl

Command-Line Interface

The dpdb command provides a powerful interface for both running programs with crash protection and debugging existing core dumps.

Usage Patterns

# Run a program with automatic crash handling
dpdb my_script.py arg1 arg2

# Debug an existing core dump
dpdb crash_dump.pkl

# Get help
dpdb --help
dpdb -h

Examples

Running a Program with Crash Protection:

# Run a script that might crash
dpdb my_data_processor.py input.csv output.json

# Run with multiple arguments
dpdb my_web_server.py --port 8080 --debug

# The program runs normally, but if it crashes, a core dump is created

Debugging a Core Dump:

# After a crash, debug the generated core dump
dpdb main_process_12345_crash.pkl

# This starts the interactive debugger with full source code access

How It Works

• Program Execution Mode: When you run dpdb my_program.py, the system:

  • Installs the global core dump handler
  • Loads and executes your program in the same process
  • Preserves all source code information for debugging
  • Creates a core dump file if the program crashes

• Core Dump Debugging Mode: When you run dpdb crash_dump.pkl, the system:

  • Loads the core dump file
  • Starts the interactive debugger
  • Provides full access to the crash state and source code

Advantages Over Subprocess Approaches

• Source Code Preservation: Full source code context is available in the debugger
• Same Process Execution: The global handler works correctly because everything runs in the same process
• Clean Implementation: No temporary files or complex subprocess management
• Argument Handling: Programs receive the correct sys.argv arguments
• Elegant Design: Direct module loading and execution

Post-Mortem Debugger Commands

The interactive debugger provides the following commands with rich terminal formatting:

Navigation Commands

• up [count] or u [count] - Move up the stack (towards caller)
• down [count] or d [count] - Move down the stack (towards callee)
• where or w or bt - Show stack trace with current frame indicator
• frame [num] or f [num] - Select frame by number
• frames - Show detailed frame information

Information Commands

• list or l - Show source code context for current frame
• longlist or ll - Show extended source code for current frame
• locals - Show local variables in current frame
• globals - Show global variables in current frame
• args or a - Show function arguments
• info - Show core dump information (timestamp, Python version, etc.)
• source [object] - Show source code for object

Evaluation Commands

• p <expression> - Print expression value
• pp <expression> - Pretty-print expression value
• whatis <expression> - Show expression type
• <statement> - Execute Python code directly

Display Commands

• display <expression> - Add expression to auto-display list
• undisplay [expression] - Remove from auto-display list

Advanced Commands

• interact - Start interactive Python interpreter

Control Commands

• help or h - Show help information
• quit or q - Exit the debugger

Theme Detection and Customization

The debugger automatically detects your terminal's theme (light or dark) and adjusts colors accordingly for optimal readability. The detection uses multiple methods:

Automatic Detection Methods

• term_background library - Most reliable method when available
• Rich's color system detection - Leverages Rich's built-in terminal capabilities
• COLORFGBG environment variable - Standard terminal background indicator
• Terminal program identification - Detects known terminals with light defaults
• Various theme environment variables - Checks for theme-related settings

Manual Theme Override

You can override the automatic detection by setting an environment variable:

# Force light theme
export DPDB_THEME=light

# Force dark theme  
export DPDB_THEME=dark

# Let the system auto-detect (default)
unset DPDB_THEME

Color Schemes

• Dark theme: Uses bright colors (cyan, yellow, green, magenta) on dark backgrounds
• Light theme: Uses darker colors (blue, dark_orange, dark_green, purple) on light backgrounds

API Reference

Command-Line Interface

• dpdb <program.py> [args...] - Run program with automatic crash handling
• dpdb <dump_file> - Debug an existing core dump
• dpdb --help - Show usage information

CoreDumpGenerator

• create_from_exception(exc_type, exc_value, exc_traceback) - Create core dump from exception
• create_from_current_stack() - Create core dump from current call stack

Functions

• save_core_dump(core_dump, filename) - Save core dump to file
• load_core_dump(filename) - Load core dump from file
• install_exception_handler(dump_filename) - Install automatic exception handler
• install_global_handler() - Install global handler for all processes
• uninstall_global_handler() - Uninstall global handler and restore normal behavior

PostMortemDebugger

• __init__(core_dump) - Initialize with a core dump
• run() - Start interactive debugging session

Multiprocessing Support

The core dump system provides excellent support for multiprocessing applications through a global handler that automatically applies core dump handling to ALL processes, including those created by frameworks you can't modify:

import dpdb
import multiprocessing as mp

# Install global handler once at the start of your program
dpdb.install_global_handler()

# Now ALL processes will automatically generate core dumps on crashes
# No code modification needed!

def worker_function(data):
    # Your worker code here
    result = process_data(data)
    return result

# Standard multiprocessing - no changes needed
process = mp.Process(target=worker_function, args=(data,))
process.start()
process.join()

# Works with frameworks like accelerate, torch.distributed, etc.
# that control process creation

Framework Compatibility

The global handler works seamlessly with existing frameworks:

import dpdb

# Install once at the start of your program
dpdb.install_global_handler()

# Now accelerate, torch.distributed, etc. will automatically
# generate core dumps for crashed processes
from accelerate import Accelerator
accelerator = Accelerator()
# All processes launched by accelerate will have core dump support

Limitations

• Some objects cannot be serialized (e.g., file handles, network connections)
• Very large objects may consume significant disk space
• Core dumps capture state at the time of creation, not live debugging
• Thread-local storage may not be fully captured in multi-threaded applications

Security Considerations

Core dumps may contain sensitive information (passwords, tokens, etc.), so

• Store core dump files securely
• Be cautious when sharing core dump files

License

This code is provided under GPLv3.