Better tracebacks
This is a more helpful version of Python's built-in exception message: It shows more code context and the current values of nearby variables. That answers many of the questions I'd ask an interactive debugger: Where in the code was the crash, what's in the relevant variables, and why was that function called with those arguments. It either prints to the console or gives you a string for logging.
pip3 install stackprinter
Before
Traceback (most recent call last):
File "demo.py", line 12, in <module>
dangerous_function(somelist + anotherlist)
File "demo.py", line 6, in dangerous_function
return sorted(blub, key=lambda xs: sum(xs))
File "demo.py", line 6, in <lambda>
return sorted(blub, key=lambda xs: sum(xs))
TypeError: unsupported operand type(s) for +: 'int' and 'str'
After
File demo.py, line 12, in <module>
9 somelist = [[1,2], [3,4]]
10 anotherlist = [['5', 6]]
11 spam = numpy.zeros((3,3))
--> 12 dangerous_function(somelist + anotherlist)
13 except:
..................................................
somelist = [[1, 2, ], [3, 4, ], ]
anotherlist = [['5', 6, ], ]
spam = 3x3 array([[0. 0. 0.]
[0. 0. 0.]
[0. 0. 0.]])
..................................................
File demo.py, line 6, in dangerous_function
5 def dangerous_function(blub):
--> 6 return sorted(blub, key=lambda xs: sum(xs))
..................................................
blub = [[1, 2, ], [3, 4, ], ['5', 6, ], ]
..................................................
File demo.py, line 6, in <lambda>
3
4
5 def dangerous_function(blub):
--> 6 return sorted(blub, key=lambda xs: sum(xs))
7
..................................................
xs = ['5', 6, ]
..................................................
TypeError: unsupported operand type(s) for +: 'int' and 'str'
I sometimes use this locally instead of a real debugger, but mostly it helps me sleep when my code runs somewhere where the only debug tool is a log file (though it's not a fully-grown error monitoring system).
By default, it tries to be somewhat polite about screen space (showing only a handful of source lines & the function header, and only the variables in those lines, and only (?) 500 characters per variable). You can configure exactly how verbose things should be.
It outputs plain text normally, which is good for log files. There's also a color mode for some reason
Usage
Exception logging
To replace the default python crash printout, call set_excepthook() somewhere. This will print detailed stacktraces for any uncaught exception except KeyboardInterrupts (to stderr, by default). You could also make this permanent for your python installation.
import stackprinter
stackprinter.set_excepthook(style='darkbg2') # for jupyter notebooks try style='lightbg'
For more control, call show() or format() inside an except block. show() prints to stderr by default, format() returns a string, for custom logging.
try:
something()
except:
# print the current exception to stderr:
stackprinter.show()
# ...or instead, get a string for logging:
logger.error(stackprinter.format())
Or pass specific exceptions explicitly:
try:
something()
except RuntimeError as exc:
tb = stackprinter.format(exc)
logger.error('The front fell off.\n' + tb)
It's also possible to integrate this neatly with standard logging calls through a bit of extra plumbing.
configure_logging() # adds a custom log formatter, see link above
try:
something()
except:
logger.exception('The front fell off.') # Logs a rich traceback along with the given message
For all the config options see the docstring of format().
Printing the current call stack
To see your own thread's current call stack, call show or format anywhere outside of exception handling.
stackprinter.show() # or format()
Printing the stack of another thread
To inspect the call stack of any other running thread:
thread = threading.Thread(target=something)
thread.start()
# (...)
stackprinter.show(thread) # or format(thread)
Making it stick
To permanently replace the crash message for your python installation, you could put a file sitecustomize.py into the site-packages directory under one of the paths revealed by python -c "import site; print(site.PREFIXES)", with contents like this:
# in e.g. some_virtualenv/lib/python3.x/site-packages/sitecustomize.py:
import stackprinter
stackprinter.set_excepthook(style='darkbg2')
That would give you colorful tracebacks automatically every time, even in the REPL.
(You could do a similar thing for IPython, but they have their own method, where the file goes into ~/.ipython/profile_default/startup instead, and also I don't want to talk about what this module does to set an excepthook under IPython.)
Docs
For now, the documentation consists only of some fairly detailed docstrings, e.g. those of format()
Caveats
This displays variable values as they are at the time of formatting. In multi-threaded programs, variables can change while we're busy walking the stack & printing them. So, if nothing seems to make sense, consider that your exception and the traceback messages are from slightly different times. Sadly, there is no responsible way to freeze all other threads as soon as we want to inspect some thread's call stack (...or is there?)
How it works
Basically, this is a frame formatter. For each frame on the call stack, it grabs the source code to find out which source lines reference which variables. Then it displays code and variables in the neighbourhood of the last executed line.
Since this already requires a map of where each variable occurs in the code, it was difficult not to also implement the whole semantic highlighting color thing seen in the screenshots. The colors are ANSI escape codes now, but it should be fairly straightforward™ to render the underlying data without any 1980ies terminal technology. Say, a foldable and clickable HTML page with downloadable pickled variables. For now you'll have to pipe the ANSI strings through ansi2html or something.
The format and everything is inspired by the excellent ultratb in IPython. One day I'd like to contribute the whole "find out which variables in locals and globals are nearby in the source and print only those" machine over there, after trimming its complexity a bit.
Tracing a piece of code
More for curiosity than anything else, you can watch a piece of code execute step-by-step, printing a trace of all calls & returns 'live' as they are happening. Slows everything down though, of course.
with stackprinter.TracePrinter(style='darkbg2'):
dosomething()
or
tp = stackprinter.TracePrinter(style='darkbg2')
tp.enable()
dosomething()
# (...) +1 million lines
tp.disable()
code:
1.5k Jan 6, 2023
795 Dec 18, 2022
13.7k Jan 2, 2023
1.4k Dec 31, 2022
23.7k Jan 1, 2023
41.5k Jan 7, 2023
2.3k Jan 5, 2023
1.4k Jan 4, 2023
778 Dec 26, 2022
496 Dec 30, 2022
2.6k Jan 4, 2023
365 Nov 11, 2022
1.4k Nov 11, 2022
23 Oct 4, 2021
56 Jan 3, 2023
1k Jan 4, 2023
20 Dec 23, 2022
1 Jan 19, 2022