On 2/28/14, 1:07 PM, Stefan Fuhrmann wrote: > Things are much worse, in fact. The old implementation > would consider all property lists to be equal (FSFS only) > because the representation_t structs are empty except > for the txn_id. Hence, revision, offset and unifier would > always match between any modified prop lists.
Ouch. > r1573071 documents the problem in the API. Should we > write an API errata? Yes I think so. I don't expect anyone to have expected that old behavior. > This is incorrect. NULL prop list reps are equal to > empty prop lists. Ahh that makes sense. Part of the reason I didn't commit it is what I expected to find in the code so widely divergent from the way the code was that I wasn't sure about what I'd done. > r1573071 fixes the a!=a problem for in-txn items but that's > the best we can do here without a content comparison. Okay thanks!
