When a breakpoint is set anywhere (in which case the global system's
trace function is set), interrupting the debuggee causes Pdb to stop
into its own code. See the following pdb session run with python on
the current head of the default branch from the repo.

=== main.py =================================

import time
def bar():
    i = 1
    while i:
        time.sleep(.100)

bar()

=================================================

$ python -m pdb main.py
> /path_to/main.py(1)<module>()
-> import time
(Pdb) import sys; print(sys.version)
3.3.0a3+ (default:4e9680570be8, May 11 2012, 12:09:15) 
[GCC 4.3.2]
(Pdb) break bar
Breakpoint 1 at /path_to/main.py:2
(Pdb) continue
> /path_to/main.py(3)bar()
-> i = 1
(Pdb) continue
^C
Program interrupted. (Use 'cont' to resume).
--Call--
> /home/xavier/src/cpython/cpython-hg-default/Lib/bdb.py(212)set_trace()
-> def set_trace(self, frame=None):
(Pdb) where
  /home/xavier/src/cpython/cpython-hg-default/Lib/bdb.py(405)run()
-> exec(cmd, globals, locals)
  <string>(1)<module>()
  /path_to/main.py(7)<module>()
-> bar()
  /path_to/main.py(5)bar()
-> time.sleep(.100)
  /home/xavier/src/cpython/cpython-hg-default/Lib/pdb.py(196)sigint_handler()
-> self.set_trace(frame)
> /home/xavier/src/cpython/cpython-hg-default/Lib/bdb.py(212)set_trace()
-> def set_trace(self, frame=None):
(Pdb) quit

=================================================

The attached patch fixes the problem and includes a test case. This
fix is the same as the one proposed at bpo-14743 to fix another
problem, except that the set_step call is moved at the end of the
sigint_handler method for the sake of robustness.

A quick glance tells me the patch is okay, apart from the name "test_issue_XXX", so can we have a formal patch review please. See also bpo-14743.

I think this issue itself might be fixed already / changed since 3.5.

I've came up with something similar in this area though, which is only triggered when using Ctrl-C while pdb is waiting for the next statement: #12880