On 01/16/2012 10:12 AM, Nicholas Clark wrote: > On Fri, Jan 13, 2012 at 12:09:30PM -0700, Karl Williamson wrote: >> I have pushed to blead a series of commits that now use inversion lists >> to store Unicode binary properties, as Jarkko has long advocated. > > Thanks > >> There are a number of changes and a number of implications, all >> hopefully invisible to the end-user except potential performance >> improvements. >> >> Inversion lists allow for a binary search instead of linear. As one of >> the commits details, a program that on my box took 97 seconds now takes >> 1.5. However, this was carefully constructed to be worst case. >> Programs that have significant amounts of ASCII or Latin1 data take much >> less time in the old scheme because those are at the beginning of binary >> searches. > > I presume you meant "at the beginning of linear searches" > > Nicholas Clark > Yes, and in reading this excerpt of mine, I see it could be misconstrued. I meant that under the old scheme ASCII and Latin1 data is fast because it is at the beginning of the linear searches. I did not mean that the old scheme is faster than the new on this data; though there are certainly cases where that might be marginally true.Thread Previous