A quick story about the hardest bug I ever debugged. My first job in high school was working at a Houston-based ISP called NeoSoft. I was writing a multi-platform web server in Tcl/Tk (w/ OTcl) called NeoWebScript 1/
I noticed that the server was crashing, and the stack trace was really unhelpful (an exception but no actual error or error message). After debugging for several hours, I showed it to my coworkers and it magically started working! Hurray! But wait... 2/
After a few productive hours of working on the server, I went home. The next morning, the same thing happened all over again--a hard crash, a vague exception, and no stack trace. After debugging for a couple hours, it just started working again. ARGH. 3/
My coworkers were perplexed. It worked fine for them, but not for me. But it did work for me when I showed them the code. They had me add a bunch of print statements and assertions. Everything looked good, until the next morning, when it blew up again 4/
On day three I found the problem. See, in Tcl/Tk, everything is a "string" and types are inferred at runtime by looking at the contents of the string. I had a variable that held the time, I think to implement the If-modified-since header. 5/
See, I got to work at 8am. My coworkers came into work at noon. 6/
In Tcl/Tk, a string that starts with "08" or "09" (the hours I was working alone before they came in) were incorrectly interpreted as octal numbers. And 08 and 09 are not valid octal numbers. 7/
Of course, when they came in, the variable held "12" or "01" or "02", and so everything worked fine. Yes, this program worked 22 hours a day, and crashed between 8:00am and 9:59am. *head banging on desk* 8/
I'm a huge fan of static types now! Also, the reason this error didn't surface to me was that someone had wrapped the entire web server in a giant "try... except (Exception e) { }" block that caught every exception and just quietly threw it away. Yeah please don't do that. 9/9
