[rvkennedy]

There's no need to trust a comment, it simply tells you what the author was thinking, as best as s/he could express it. It's less imprecise than no comment at all. Suppose you have a very simple function that does something clear - reading the code you can tell what it does and how. And suppose that function is called from several hundred different contexts in the codebase, but one of those contexts causes an exception in your simple function.

You could "fix" it, with a code change. But what if your fix goes against the assumptions of some other context where the same function is called? A comment that describes what the function "should" do allows you to decide whether changing the function, or its caller, will be more in line with the design of the overall codebase. This is particularly true with mature codebases where thousands of tiny fixes and tweaks have accumulated over time. There's a corpus of learning about why the code ended up that way that even in-depth reading of individual functions might miss.

Tests are good, and to be encouraged. But using tests in place of comments is rather like answering a question with a question. Tests provide confidence, they do not provide understanding.

Of course, you could do an in-depth analysis of every function, in every context it's used, and how it interacts with its callers and callees. While you're doing that, the guys who commented their code will be moving onto the next challenge.