date-format-xparb spends 12% of TOTAL TIME in localtime_r Holy broken date code, batman! Darin mentioned that he had a rewrite of some of the date code sitting in a patch on his machine at one point in time. Perhaps that also included a fix for this. I'm sure there is a simpler (& faster) way to do getDSTOffset()
Created attachment 17060 [details] here's the old date patch I had lying around -- not sure how good it is
Created attachment 17063 [details] Updated version of darin's date patch (broken, I think) I think the patch somehow broke (hangs) after updating it to TOT. I've not investigated yet.
Created attachment 17064 [details] darin's patch, updated for TOT (0.5% slower on SunSpider)
localtime_r has made it's way all the way to .6% of *total time*. KJS::getDSTOffset(double) is 1.1% of total time on my machine.
Those numbers are for a --shark20 sample across all tests.
Eric, is this bug still valid?
Unless someone has re-written the date code, we are likely still spending too much time here, yes.
The date code has been rewritten; date-format-xparb does not call localtime_r.