The question of word boundaries doesn't show up because the example author uses doesn't need it. So all works fine (at least in the article). But in your example that makes a difference.

I know very little about lexers, but I agree that using split causes unexpected behavior (not matching the leftmost rule), but has proven useful in the example of the article, where it creates rules only for what matters (ignoring the = symbol, for example). I don't know how hard/easy would be to do that for leftmost rule matching. split use here is convenient.

And note, I didn't tell that x must be skipped (considered garbage), at least considering the rules I mentioned, but it's matched by WORD, then KEYWORD matches select. HOP::Lexer knows nothing about boundaries, neither give special meaning to \s, you must tell him if you want just match select in " select " or " select, " but not in "selectx".