"Lines" is short for "lines of code" I would say so comments and whitespace would not count. But holistically speaking you are right of course.
If we're trying to measure code complexity I think we have to ask "complexity to whom". If it's to humans, we should measure the code as it appears to humans. If it's to machines, then "line" is meaningless and we should look at tokens, the way PG suggested years ago.
Doesn't work for every language, though =P
A tight definition of SLOC isn't something we have. We do have sloccount, or we can install it, different tool from `wc -l` (we can sed the blank lines out first).
Really we should be calculating cyclomatic complexity like adults. I'm told this isn't straightforward! But what is the use of heaven if a man's reach cannot exceed his grasp.