Dreaming in Code : Two Dozen Programmers, Three Years, 4,732 Bugs, and One Quest for Transcendent Software

CHAPTER 1

DOOMED

[JULY 2003]

Michael Toy places his palms on his cheeks, digs his chin into his wrists, squints into his PowerBook, and begins the litany.

"John is doomed. He has five hundred hours of work scheduled between now and the next release. . . . Katie's doomed. She has way more hours than there are in the universe. Brian is majorly doomed. Plus he's only half time. Andy--Andy is the only one who doesn't look doomed. There are no hundreds on his list."

They don't look doomed, these programmers sitting around a nondescript conference room table in Belmont, California, on a summer day. They listen quietly to their manager. Toy is a tall man with an impressive gut and a ponytail, but he seems to shrink into a space of dejection as he details how far behind schedule the programmers have fallen. It's July 17, 2003, and he's beginning to feel doomed himself about getting everything done in the less than two months before they are supposed to finish another working version of their project.

"Everybody who has a list with more time than there is in the universe needs to sit down with me and go over it."

These lists are the bug lists--rosters of unsolved or "open" problems or flaws. Together they provide a full accounting of everything these software developers know must be fixed in their product. The bug lists live inside a program called Bugzilla. Toy's programmers are also using Bugzilla to track all the programming tasks that must be finished in order to complete a release of the project; each one is responsible for entering his or her list into Bugzilla along with an estimate of how long each task will take to complete.

"Now let's talk about why we're behind. Does anyone have a story to tell?"

There's silence for a minute. John Anderson, a lanky programming veteran whose title is systems architect and who is, in a de facto sort of way, the project's lead coder, finally speaks up, in a soft voice. "There's a bunch of reasons. In order to build something, you have to have a blueprint. And we don't always have one. Then you hit unexpected problems. It's hard to know how long something's going to take until you know for sure you can build it."

"But you can't just throw up your hands and say, I quit." Toy usually prefers to check things off his agenda fast, running his developers' meetings with a brisk attitude of "let's get out of here as fast as we can" that's popular among programmers. But today he's persistent. He won't let the scheduling problems drop. "We need to make guesses and then figure out what went wrong with our guesses."

Jed Burgess, one of the project's younger programmers, speaks up. "There's a compounding of uncertainty: Your estimates are based on someone else's estimates."

Toy begins reviewing Anderson's bugs. "The famous flicker-free window resizing problem. What's up with that?"

Officially, this was bug number 44 in Bugzilla, originally entered on January 19, 2003, and labeled "Flicker Free window display when resizing windows." I had first heard of the flicker-free window resizing problem at a meeting in February 2003 when the Open Source Applications Foundation (OSAF), whose programmers Toy was managing, had completed the very earliest version of its project, Chandler--an internal release not for public unveiling that came even before the 0.1 edition. Ultimately, Chandler was supposed to grow up into a powerful "personal information manager" (PIM) for organizing and sharing calendars, email, to-do lists, and all the other stray information in our lives. Right now, the program remained barely embryonic.

At that February meeting, Anderson had briefly mentioned the flicker bug--when you changed the size of a window on the Chandler screen, everything flashed for a second--as a minor matter, something he wanted to investigate and resolve because, though it did not stop the program from working, it offended him aesthetically. Now, nearly six months later, he still hasn't fixed it.

Today Anderson explains that the problem is thornier than he had realized. It isn't simply a matter of fixing code that he or his colleagues have written; its roots lie in a body of software called wxWidgets that the Chandler team has adopted as one of the building blocks of their project. Anderson must either wait for the programmers who run wxWidgets to fix their own code or find a way to work around their flaw.

"So you originally estimated that this would take four hours of work," Toy says. "That seems to have been off by an order of magnitude."

"It's like a treasure hunt," Anderson, unflappable, responds. "You have to find the first thing. You have to get the first clue before you're on your way, and you don't know how long it will take."

"So you originally estimated four hours on this bug. You now have eight hours."

"Sometimes," Anderson offers philosophically, "you just wake up in the morning, an idea pops into your head, and it's done--like that."

Mitchell Kapor has been sitting quietly during the exchange. Kapor is the founder and funder of the Open Source Applications Foundation, and Chandler is his baby. Now he looks up from his black Thinkpad. "Would it be useful to identify issues that have this treasure-hunt aspect? Is there a certain class of task that has this uncertainty?"

"Within the first hour of working on the bug," Burgess volunteers, "you know which it's going to be."

So it is agreed: Bugs that have a black hole-like quality--bugs that you couldn't even begin to say for sure how long they would take to fix--would be tagged in Bugzilla with a special warning label.

Shortly after the meeting, Toy sits down at his desk, calls up the Bugzilla screen, and enters a new keyword for bug number 44, "Flicker Free window display when resizing windows": scary.



Toy's fatalistic language wasn't just a quirk of personality: Gallows humor is a part of programming culture, and he picked up his particular vocabulary during his time at Netscape. Though today Netscape is remembered as the Web browser company whose software and stock touched off the Internet boom, its developers had always viewed themselves as a legion of the doomed, cursed with impossible deadlines and destined to fail.

There was, in truth, nothing especially doomed about OSAF's programmers: Several of them had just returned from a conference where they presented their work to an enthusiastic crowd of their peers--who told them that their vision could be "crisper" but who mostly looked at the blueprint for Chandler and said, "I want it now!" Though the software industry had been slumping for three straight years, they were working for a nonprofit organization funded by $5 million from Kapor. Their project was ambitious, but their ranks included veteran programmers with estimable achievements under their belts. Andy Hertzfeld had written central chunks of the original Macintosh operating system. John Anderson had written one of the first word processors for the Macintosh and later managed the software team at Steve Jobs's Next. Lou Montulli, another Chandler programmer who was not at the meeting, had written key parts of the Netscape browser. They'd all looked doom in the eye before.

Similarly, there was nothing especially scary about bug number 44. It was a routine sort of problem that programmers had accepted responsibility for ever since computer software had migrated from a text-only, one-line-at-a-time universe to today's graphic windows-and-mouse landscape. What scared Toy was not so much the nature of Bug 44 but the impossibility of knowing how...