I refactored the blocklocation code significantly a few commits ago. I see this code adds most of the change back in.

Is there a bug this is intended to fix? Can you rebase and try the fix on the refactored version?

In a stripe-based volume, a block is located on just a particular stripe. The single pass regex in the #122 cannot parse the pathinfo correctly as it always returns all matching <POSIX..> stripes for any block which is not true. I have tried the "2-step rake" committed previously but it fails in many test cases. Also, the code was difficult to understand (at least for me). So, I decide not to fix the "2-step rake" but extend from the refactored (single pass) code. My code is just to replace the simple regex matching by a small BNF parser of the pathinfo grammar.