On Wed, Jan 27, 2016 at 1:17 AM, William A Rowe Jr <wr...@rowe-clan.net> wrote: > > Yann, I didn't pick up our research yet to tweak the SVN implementation, > since we never really tested that. The signed->unsigned transition is a > noop, > so the only question is the fastest way to structure the loop. Hopefully > the > compiler can do a good job for all architectures. In the svn > implementation, > this was through a single character casecmp function/macro that handled > signed chars, and expanded to an int (which "in theory" is an optimal array > index, our mileage may vary). > > Let's compare performance of this implementation and commit the best > enhancement.
I'll do some tests with this new implementation. IIRC, svn_cstring_casecmp() (using svn_ctype_casecmp()) was included in my latest testing. It did not perform as well as the current ap_casecmpstr(), but I need to recheck with these two only now, with latests implementations. Stay tuned, Yann.