Debugging with Jupyter

This had been bugging me for a long while. Jupyter (or IPython) has decent support for debugging via the magic command %debug.

However, it does not quite cut it for me. The problems with it are well known:

  • Cannot run code elsewhere without quitting pdb first.
  • Refreshing the notebook would leave you with a non-responsive kernel.
  • gdb like syntax, which can be conflict with variable names and introduces surprises.
  • Tab completion doesn’t work.
  • Command recall (using arrow keys) doesn’t work reliably, etc.

Bring out the local variables

One of the most useful features of %debug is that it allows me to inspect the current values of the local variables in each frame and I believe that information itself would be sufficient for me to debug my problems.

Force out the local variables

Hence, I wrote a quick Python function which will allow me to explode out of my functions and will spill all the local variables in my global namespace:

import inspect as I

class Exploded(BaseException): pass

def explode():
  for k, v in I.currentframe.f_back.f_locals.items():
    globals()[k] = v

  raise Exploded()

To use it, consider the following:

In [42]: def foo():
    ...:     a, b = 101, 42
    ...:     explode()

In [43]: a, b = 1, 1  # Setting the global values here, not necessary

In [44]: foo()
Exploded                                  Traceback (most recent call last)
<ipython-input-44-624891b0d01a> in <module>()
----> 1 foo()

<ipython-input-34-d9d09911e62e> in foo()
      1 def foo():
      2     a, b = 101, 42
----> 3     explode()

<ipython-input-41-eae3be1cadaa> in explode()
      3         globals()[k] = v
----> 5     raise Exploded()


In [45]: a, b   # These are the local variables from `foo`
Out[45]: 101, 42

If this helps me out with my debugging as much as I think it will, I will release it as a simple Python package with some more bells and whistles.