thinktoomuch.net

Traditional blogging wisdom would say I should have more than one blog, to keep the subject matter more “focused”. Well, oops. What can I say. I’ll sort that out when my thesis is done. (Like the rest of my life. ) In the mean time, I’ll always feel some urge to not post more technical stuff, like this. (Also, I suppose it doesn’t make sense to have this kind of post imported into Facebook - if you’re reading this in Facebook, would you prefer not to see “notes” like these?)

I found a cute call graph generator for python, pycallgraph. Despite still needing some attention (which it doesn’t seem to be getting?), it is usable, once you fix a major defect by adding a single line:

Index: pycallgraph.py
===================================================================
--- pycallgraph.py      (revision 23)
+++ pycallgraph.py      (working copy)
@@ -182,6 +182,7 @@
     if event == 'return':
         if call_stack:
             call_stack.pop(-1)
+    return tracer

 def get_dot(stop=True):
     """Returns a string containing a DOT file. Setting stop to True will cause

I’m not sure why this problem exists… this bug (and fix) has been known for quite some time? The actual graph is drawn with graphviz. There are some examples on the website, but of course, it’s most fun to run it on your own code that you know well. Usage (once fixed and installed) is as simple as adding four lines of code:

import pycallgraph
pycallgraph.start_trace()
...your code here...
pycallgraph.stop_trace()
pycallgraph.make_dot_graph('test.png')

This will use “dot”. You can use other graphviz binaries and adjust the output format:

pycallgraph.make_dot_graph('test.jpg', format='jpg', tool='neato')

I’m not sure what the best way would be to collect a trace over all your unit tests. I placed start_trace(reset=False) and stop_trace() in each test (placing them in setUp() and tearDown() would require filtering out the testing framework if I want a clean graph, and I’m a little lazy to use pycallgraph’s exclusion mechanism right now). Then I subclassed unittest.TextTestRunner to include make_dot_graph(...) as follows:

class CallGraphTextTestRunner(TextTestRunner):
    def __init__(self, callgraph_filename, *args, **kwargs):
        TextTestRunner.__init__(self, *args, **kwargs)
        self._callgraph_filename = callgraph_filename

    def run(self, test):
        result = TextTestRunner.run(self, test)
        pycallgraph.make_dot_graph(self._callgraph_filename)
        return result

main(testRunner=CallGraphTextTestRunner('test_callgraph.png'))

What an unimaginative class name. Any better suggestions?

(If you are new to unit testing in Python, you might need to look at more than just a basic example of unit testing in Python.)