Another 4:00 am hackdown. This time I started with a test. Actually I started by installing cygwin. This allowed me to run cvs and download the cwtext project with unix line endings. Cygwin is now installed on a Samba share on my Slackware Linux 9.0 box. While waiting for cygwin to install gcc and make I ssh'ed to slackware and did the building there. How did I get anything done when I only had one computer at home?

Anyhow I got a CVS copy of cwtext and started building the test code. Feed the audio generator a series of frequencies, read back the stored frequency, print both out. Oh yeah, spit out a 1 second sine wave (440 Hz) as a sanity check. I added code to get the stored frequency from the generator and ran the first test.

The output sine wave was ok, but the frequency table showed big problems in my code. I was actually storing the wavelength (1/frequency), as samples per cycle, in an integer. Since my sample rate was 8000 samples per second I started getting big errors at anything over 1000 Hz. Using a higher sample rate would help, but the real fix was to use a float for the wavelength. I also chose to store frequency rather than wavelength. This works with floats, practically imposible with integers. So now I store the frequency in cycles per sample, 1/8 for 1000 Hz and so on.

Build and run, now all my stored frequencies come back as zero. I forgot to cast something to float in the setFrequency. See the changes to pcm.c. With the cast (near line 75) I got the expected numbers, up to 2000Hz. Hmm, 8000/4 = 2000, so my max frequency limit is working too. I could probably get by up to 4000 Hz, some kind of limit at bitrate/4 or bitrate/2.

Anyway, the tests pass (well enough for me), so I check in the code, drop in a notice on the sourceforge news feed, and call it a...morning. Back to bed!

Test First helped me fix a reported issue, and I knew immediately when I broke the code and when I really fixed it. It wasn't too hard to build the first test. I may have to refactor a bit as I expand the tests, but it is good enough for what I need right now.

The other think I think I did right was to modularize the code. I was able to work on the PCM generator without messing up the morse code portions of the project. OK the whole project is a little strange, maybe over-engineered but it does the task with increasingly fewer problems. I'm glad I did it!

Get your CWText Project Updates right here.