大约10年前，我从事图像处理工作，通过扫描显微镜视频帧来检测细胞运动。我当时在做一个特别复杂的工作，决定和朋友出去喝一杯。当我回到家时，我工作了一会儿，但没有太多，因为我喝醉了。第二天早上，我发现了一个10行完全混乱的函数，带有以下注释(显然是我自己写的):

/* Ah ah ah! You'll never understand why this one works. */

最奇怪的是，它竟然起作用了。

以下是4个，排名不分先后:

// Father, forgive me, for I am sinning

// heaven help me

// horse string-length into correctitude 
(from a textbook)

// what, me worry?

在给一个朋友看了这篇来自《Coding Horror》的文章几个小时后，我看到了他对代码的评论:

// valdez先生是一个暴力的精神病患者。别惹他生气。

不是注释，而是属性

[ThereBeDragons]

还有一个是我在IHttpHandler的实现中看到的

//What is this?
public bool IsReusable
{
    get{return false;}
}

Re eating one's own dogfood: We have the same term in our workplace (granted, only because I introduced it). My code is peppered with comments that say "TODO" and indicate something that ought to be done eventually, but a comment saying "DOGFOOD" (both keywords are always at the beginning of the comment, in all caps) means something that must be done before this program can be used even internally. It's a handy thing to search for, as the word "dogfood" is never going to appear in a quoted string - if it does, I can always just say "dog-food" or something - so even a case-insensitive search will come up with the right results.

Regarding zeroing the accumulator: I've done exactly the same thing when programming an Intel 80x86 (I started on the 8086 and moved up from there into the modern Pentiums). XORing a register with itself is the quickest and tightest way to clear it. Using "MOV AX,0" requires three bytes (opcode and two bytes of literal 16-bit zero), whereas "XOR AX,AX" is only two; it's even more noticeable with the 386-and-higher extended registers, where "MOV EAX,0" requires five bytes (four bytes of 32-bit zero). My C/C++ compiler always zeroes registers this way, so I'd assume it's still the best way (although I haven't studied opcode timing tables in ages, and probably both XOR reg,reg and MOV reg,imm take one clock).

在LucasArts的电脑游戏《the Eidolon》(游戏邦注:这款游戏本身就很古怪)的源代码中可以看到……

// He's dead, Jim!