##// END OF EJS Templates
protect flush_figures post-exec function from user error...
protect flush_figures post-exec function from user error Invalid matplotlib data may raise inside print_figure. flush_figures() is a post-exec function, so user errors should not raise. Instead, call get_ipython().showtraceback() if called from IPython, raising as before otherwise.

File last commit:

r3917:03c097c7
r5735:c98183ce
Show More
refbug.py
47 lines | 1.5 KiB | text/x-python | PythonLexer
"""Minimal script to reproduce our nasty reference counting bug.
The problem is related to https://github.com/ipython/ipython/issues/141
The original fix for that appeared to work, but John D. Hunter found a
matplotlib example which, when run twice in a row, would break. The problem
were references held by open figures to internals of Tkinter.
This code reproduces the problem that John saw, without matplotlib.
This script is meant to be called by other parts of the test suite that call it
via %run as if it were executed interactively by the user. As of 2011-05-29,
test_run.py calls it.
"""
#-----------------------------------------------------------------------------
# Module imports
#-----------------------------------------------------------------------------
import sys
from IPython.core import ipapi
#-----------------------------------------------------------------------------
# Globals
#-----------------------------------------------------------------------------
# This needs to be here because nose and other test runners will import
# this module. Importing this module has potential side effects that we
# want to prevent.
if __name__ == '__main__':
ip = ipapi.get()
if not '_refbug_cache' in ip.user_ns:
ip.user_ns['_refbug_cache'] = []
aglobal = 'Hello'
def f():
return aglobal
cache = ip.user_ns['_refbug_cache']
cache.append(f)
def call_f():
for func in cache:
print 'lowercased:',func().lower()