Hi all, 2 announcements in fact : 1) the ReverseDiffSource package has moved to the JuliaDiff org, 2) the new 0.1.0 version brings a refreshed API and new abilities : automated code simplifications, for loops, and derivation for orders 2 and above.
There are still many bugs and annoyances that need to be corrected but you may find it usable for your needs (I do use it on a regular basis for gradient based optimizations). Simply put, the package provides a function `rdiff` that takes an expression and returns another expression that computes all derivatives up to the order requested. That expression can be edited if needed or used to make a function (using eval). More details at http://reversediffsourcejl.readthedocs.org/en/master/index.html
