coredumpy

coredumpy saves your crash site for post-mortem debugging.

Highlights

Easy to use
Native support for unittest, pytest and run-time exceptions
Portable and safe dump
Utilizes pdb interface

Usage

dump

For pytest, you can use coredumpy as a plugin

# Create a dump in "./dumps" when there's a pytest failure/error
pytest --enable-coredumpy --coredumpy-dir ./dumps

For Exception and unittest, you can use coredumpy run command. A dump will be generated when there's an unhandled exception or a test failure

# with no argument coredumpy run will generate the dump in the current dir
coredumpy run my_script.py
coredumpy run my_script.py --directory ./dumps
coredumpy run -m unittest --directory ./dumps

Or you can patch explicitly in your code and execute the script/module as usual

import coredumpy
# Create a dump in "./dumps" when there's an unhandled exception
coredumpy.patch_except(directory='./dumps')
# Create a dump in "./dumps" when there's a unittest failure/error
coredumpy.patch_unittest(directory='./dumps')

Or you can dump the current frame stack manually

import coredumpy

# Without frame argument, top frame will be the caller of coredumpy.dump()
coredumpy.dump()
# Specify a specific frame as the top frame to dump
coredumpy.dump(frame)
# Specify a filename to save the dump, without it a unique name will be generated
coredumpy.dump(path='coredumpy.dump')
# You can use a function for path
coredumpy.dump(path=lambda: f"coredumpy_{time.time()}.dump")
# Specify a directory to keep the dump
coredumpy.dump(directory='./dumps')
# Specify the description of the dump for peek
coredumpy.dump(description="a random dump")

load

Load your dump with

coredumpy load <your_dump_file>

A pdb debugger will be brought up and of course not everything is supported.

peek

If you only need some very basic information of the dump (to figure out which dump you actually need), you can use peek command.

coredumpy peek <your_dump_directory>
coredumpy peek <your_dump_file1> <your_dump_file2>

About the data

Besides a couple of builtin types, coredumpy treats almost every object as an Python object with attributes, and that's what it records in the dump.

It does not use pickle so you don't need to have the same run-time environment when you load the dump. It's also safer to open an arbitrary dump without the unsafe pickling process.

That being said, most of the objects will not be "restored" as they were when being dumped. You are in an observer mode where you can inspect attributes of all objects. None of the methods of the objects would work, nor would any dymanic features.

License

Distributed under the terms of the Apache 2.0 license.

gaogaotiantian/coredumpy