On Mon, Nov 03, 2003 at 10:16:06AM +0000, Alan Burlison wrote: > What is your intention for how 5.8.n where 'n' is >= 3 will handle the hash > key binary compatibility issue? At what point will we break backwards > binary compatibility and just have *all* hashing use the seed with no 'Oh > dear I'm under attack, switch behaviour' logic? Obviously there is a > performance hit for this 5.8.0/5.8.1 binary compatibility stuff, and I did > get the impression that at some point we would switch over entirely to the > new mechanism. > > I think it needs to be clearly documented in 5.8.2 somewhere that we will > at some point break binary compatibility, so people can prepare for it. 5.10.0 :-) [I think that answers the 5.8.n question - 5.8.n for all n > 1] > Also, I don't actually care about 5.8.0/5.8.1 binary compatibility, as we > didn't ship either version. I'd therefore like to permenantly enable the > new behaviour so that I don't have to take a potential prrformance hit, and > so I don't have to break binary compatibility in the future. Is there > going to be a Configure flag to allow me to do this? There isn't one currently, and I can't see one going in before release. Given that the cut in threshold is a linked list of 14 elements, it shouldn't happen with regular data, and even with the rehashing may well be faster than 5.8.1 I'd suggest benchmarking. Nicholas ClarkThread Previous | Thread Next