Thanks for raising the issue, I think (since issue #2 means tabu is completely broken) I'll push out a 2.10 today without trying to address this now. As you say, it's good to collect the issues here to be handled as anyone finds time.

To correct some bad wording in my post. tabu* does no \scantokens with an active space. It does collect the body with an active space, then executes \scantokens with space again normal. Verbatim will when expanding re-assign the space to active. Each control word (except for the very last one if it ends the body passed over to \scantokens) is now followed by a (normal) space which Verbatim made active. Then each such control word receives an active space as first argument.

It might be that if the "collect body phase" was done with a catcode 12 backslash, then the scantokens will be innocuous and recreate control sequences without an extra space character. If I get time I would have a closer look at tabu to see if it can work to modify the collecting phase this way. (if it works then the tabu patch of fancyvrb for checking end of Verbatim will become unneeded)

My PR #5 implements the "catcode 12 backslash" idea. While keeping close to original code I did need to separate into two distinct branches the body collecting. But thinking about it, I see that simpler code would result from handling tabu* as a verbatim environment i.e. sanitize all catcodes and fetch to the end \end{tabu*} (or \end{longtabu*}). The current documentation I think allows tabu (without a star) to act like tabu* if some boolean is on but says official interfae is tabu*.

But dropping that and requiring either \end{tabu*} or \end{longtabu*}, one thus only needs in this alternative to either fetch all until \end{tabu*} or fetch all until \end{longtabu*}. But it is blocking here that nesting is impossible without extra code, which brings us back into the checking of the balancing of \begin and \end in original. Then we end up being not so different than the current code in PR #5 except that we don't have to worry about \endfoo situation and must count only (verbatimized) \begin{tabu*} occurrences.

@jfbu do I understand correctly that this is an old issue (not a recent one)?

@FrankMittelbach Yes, this is my understanding as well.