I went through the first chapter of the famous proc-macro workshop, and also checked it (canonical) solution. (Btw, the workshop is updated, solution is not.)
Since I chose to do it in a slightly different way I would be glad if you could point out what I got wrong and where should improve things.
The main point of this solution is that I liked it more using single move along the token stream, since there's no complex logic/dependencies which would require to analyze input code more than one time.