I think this isn't exactly fair, in that it seems to mislocate the problem——Markdown itself is so underspecified (I think Gruber never really expected it to take off the way it did) that the meaning of 'Markdown' can vary wildly from place to place, as a result of which there's no simple, universal test suite, and it's hard to tell on what sort of edge (or not-so-edge) cases your homerolled parser will trip until you actually deploy it (at which point people have come to rely so much on its quirks that it's difficult to fix it).