[iloveponies]

The idea that code be "self-documenting" is naive at best. Action does not necessarily infer intention, and without explanation can lead to assumption. To reach the code we end up committing a process of thought should have occurred by those who wrote it. For some things such as "boilerplate" code this may not be seen as significant. For others of increased complexity like explaining interactions between two pieces of logic, performing a complex set of logic or the expected usage of an externally provided interface comments are in my opinion essential.

In contrast to what you've said I completely disagree with the notion that the "best code transcends the need for comments". I'd argue the best type of code has the most effective comments in line with code to ensure the next person can gain with the least amount of effort clear understanding of both the rationale and of the functionality that lead to what they see before them.