Tuesday, September 18, 2012

Tracking down unused variables with Pyflakes and git bisect

I was working on a Python project when I noticed a variable that was defined but never used. Does this indicate that some important code was accidentally deleted? Or was there just a minor oversight during refactoring?

To answer this question, I wanted to see the Git commit which removed the last use of this variable. This sounds like a job for git bisect. And because Pyflakes can detect unused variables, the whole process is completely automatic.

I made a guess that the mystery_variable became unused sometime within the past two weeks. Then I told Git to consider a commit "good" if Pyflakes's list of complaints does not mention mystery_variable.

$ git bisect start master master@{'2 weeks ago'}
Already on 'master'
Bisecting: 150 revisions left to test after this (roughly 7 steps)
[066327622129dbe863f6e2fc4746ff9e869bd049] Synthesize gravity

$ git bisect run bash -c '! pyflakes foo.py | grep mystery_variable'
running bash -c ! pyflakes foo.py | grep mystery_variable
Bisecting: 75 revisions left to test after this (roughly 6 steps)
[d3a5665eff478cccfb86d994a8fc289446325fbf] Model object components
running bash -c ! pyflakes foo.py | grep mystery_variable
Bisecting: 37 revisions left to test after this (roughly 5 steps)
[6ddcfbf27a1a4548acf972a6b817e485743f6bd9] Time-compress simulator clock
...

running bash -c ! pyflakes foo.py | grep mystery_variable
9c2b2f006207ae9f274f9182efeb3e009d18ed04 is the first bad commit
commit 9c2b2f006207ae9f274f9182efeb3e009d18ed04
Author: Ben Bitdiddle <bitdiddle@example.com>
Date:   Fri Sep 14 01:38:31 2012 -0400

    Reticulate splines

Now I can examine this commit and see what happened.

Sunday, September 9, 2012

taralli: Screen edge pointer wrapping for X

The maximum travel distance between points on a desktop is substantially reduced if the pointer is allowed to travel off a screen edge and reappear at the opposite edge. In other words, we change the topology of the desktop to be that of a torus. This is not a new idea: it's implemented in several window managers, programs for Windows and Mac OS X, and is even the subject of a research paper.

I wanted a standalone program to implement mouse pointer wrapping for X Windows on GNU/Linux. Previously I was using Synergy for this task. Synergy is a very cool program, but it's not a good choice for pointer wrapping on a single machine.

  • Correctness: I have several monitors at different resolutions, which means my desktop isn't rectangular. Synergy didn't understand where the edge of each monitor is.

  • Security: I don't need the exposure of 95,000 lines of code for this simple task. Even if there are no bugs in Synergy, I'm one configuration mistake away from running an unencrypted network keylogger.

  • Energy efficiency: Synergy wakes up 40 times per second even when nothing is happening. This can have a significant impact on laptop battery life.

So I wrote taralli as a simple replacement. It doesn't automatically understand non-rectangular desktops, but you can easily customize the pointer position mapping, to implement multi-monitor wrap-around or many other behaviors. (If someone would like to contribute the code to get multi-monitor information from Xinerama or something, I would be happy to add that.)

taralli is under 100 lines of C99. It has been tested on GNU/Linux and is likely portable to other X platforms. You can download it from GitHub, which is also a great place to submit any bug reports or patches.