• Poorly conceived versioning. Some things are versioned, some things aren't. No tools (that I've found yet) for analyzing history except on a per-page scope. Also, can only commit changes at the page level, which makes a single "I added alt tags to all my images" commit impossible.
• Browser-only interface. That's right, no scriptable API short of driving a browser through a script. This one is so huge, I'm not really sure what else to say. Surely people occasionaly want to do something that the designers hadn't planned for, like, oh, importing data from a preexisting database.

Anyhow, to overcome the major limitation, I'm driving my browser through a script. Thanks to the magic of WebDriver/Selenium for making this possible.

Some bugs I've run into so far:

• start() doesn't take any args. Simple fix in selenium.py.
• captureNetworkTraffic("xml") returns requestHeaders as responseHeaders. Fixed in source, but you'll need to compile your selenium-server.jar on your own. There's a nice, detailed explaination of building Selenium with Maven here, but for the impatient, a simple mvn install in the svn root directory should do it.
• captureNetworkTraffic("xml") captures invalid XML in URL's (ampersand problem). Another simple patch, but this one's just posted by me, so you'll have to apply it by hand and recompile. You get the feeling that people don't use the captureNetworkTraffic side of Selenium very often ;). Ah well, only 3 open bugs in that area left to bump into ;).

To keep the test suite easy to maintain, I build it on the fly using nose generators, which are awesome :).

More details to come...

Update: Oct. 6. Turns out the code linked from Elsevier is out of date. Rishi over at River-Valley (writers of the elsarticle package) pointed me to some more up to date documentation and source. Version 1.2 fixes most of my problems (with elsarticle-num-names.bst), but I'm still not sure why it doesn't replace elsarticle-num.bst entirely...

$ apt-get install bzr
$ bzr branch http://whatever/remote/repo local_repo
$ cd local_repo
$ be new "The demuxulizer is broken"
Created bug with ID 48f
$ be comment 48f
<Describe bug>
$ bzr send --mail-to "whatever-dev@fancyprojects.com"

mostly just raises eyebrows among non-devs ;). My interface replaces it with

$ cat | mail -s "[be-bug] new" "whatever-dev@fancyprojects.com"
The demuxulizer is broken

<describe bug>
^D

Which will hopefully be more user friendly. Of course, most non-devs I know (and some devs) avoid the command line like the plague, so we will still want a web-frontend. Luckily, web frontends seem popular with other devs, so I don't have to bother with that ;).

Curious parties can branch my public repo:

$ bzr branch http://www.physics.drexel.edu/~wking/code/bzr/be-email be.email

As a nice side effect of this little project, my send_pgp_mime is now much more elegant.

...
make[1]: Warning: File `Makefile' has modification time 1.1e+02 s in the future
make[1]: Nothing to be done for `clean-am'.
make[1]: warning:  Clock skew detected.  Your build may be incomplete.
make[1]: Leaving directory `/thor/wking/src/lab/comedi/comedi-build'
make: warning:  Clock skew detected.  Your build may be incomplete.

“No problem,” thinks I, since I'd just rebooted into the new kernel, but I'd pulled the source in under the old kernel. I'll just touch all the files to make sure their times are in the past.

Nope. Same error. Turns out my touch is from the future:

$ date; ll include/Makefile; touch include/Makefile; ll include/Makefile; date
Fri May  8 10:48:35 EDT 2009
-rw-r--r-- 1 wking wking 15540 2009-05-08 10:48 include/Makefile
-rw-r--r-- 1 wking wking 15540 2009-05-08 10:50 include/Makefile
Fri May  8 10:48:35 EDT 2009

This is because my files are NFS mounted from another computer, and the clocks were out of sync by two minutes. Time to setup NTP for the lab LAN ;).

So I wrote my own. Not very complicated really, just implementing the RFC 3156 specs. There's obviously room for improvement, but it works well enough to make me happy at the momemt. You're welcome to take a look and email me with [encrypted!] feedback.

Besides helping keep track of my bugs, I've enjoyed hacking away at BE. In the process I've learned some cool tricks such as Python decorators (see this tutorial for some really neat examples) and Bash autocompletion.

Meanwhile, the calibration code developed, and I was getting tired of keeping assorted keyword arguments default in sync (how many times do I have to repeat minFreq=500?). The way to avoid repeating yourself in this situation is to use *args or **kwargs (see wxPython's endorsement). I don't trust myself to keep my code in sync enough to use args, so lets focus on *kwargs. Consider the following example:

>>> def A(a=1):
...     return a
>>> def B(b=2):
...     return b
>>> def C(**kwargs):
...     return A(**kwargs)+B(**kwargs)
>>> C(a=3,b=4)
Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
  File "<stdin>", line 2, in C
TypeError: A() got an unexpected keyword argument 'b'

Hmm, how to split the kwargs into those for A and those for B? So I wrote the splittable_kwargs module. See the docstrings for some usage examples and more info. With my module the above example becomes

>>> from splittable_kwargs import splittableKwargsFunction
>>> @splittableKwargsFunction()
... def A(a=1):
...     return a
>>> @splittableKwargsFunction()
... def B(b=2):
...     return b
>>> @splittableKwargsFunction(A,B)
>>> def C(**kwargs):
...     akw,bkw = C._splitargs(C, kwargs)
...     return A(**bkw)+B(**akw)
>>> C(a=3,b=4)
7

In other news, my current side project is a MySQL-based, multi-user, archiving, command-line grade database. Down with web-forms ;). More information to come as the project progresses.

Enter plotpick.py, the raw-data version of clicklock.tk. Hope you like it :).