Re: include vs require

On Mon, 24 Feb 2003, Pierre-Alain Joye wrote: > On Sun, 23 Feb 2003 15:59:02 -0800 (Pacific Standard Time) > Rasmus Lerdorf <rasmus@lerdorf.com> wrote: > > > I haven't analyzed it in detail, but switching it in a Yahoo! app gave > > me another couple of requests per second. I'll add it to my strace > > stuff and have a closer look sometime in the next couple of days. > > Perhaps someone did sneak something nasty in there. Our syscall bloat > > has been really bad lately. > > If possible, keep us informed, it may be very interesting to get real > exemples of hi-load site using the different include methods. (to close > the debat for one day or two ;-) ). I checked the code, and to verify, did the strace test as well. No syscall differences between the two. But, it also wasn't a much of a difference I saw. You gain way more by eliminating an include by combining two files than by going from include_once to include. Especially in the amazingly broken 4.3 code right now. The double realpath() calls and the triple fstat() calls are deadly. Well, we only get a double realpath() on the primary script, but each included file gets a single realpath() which turns into a stat() for each directory level. The really performance-paranoid should put all their includes in the / directory (half-joking ;), but the triple fstat() calls happen on every included file so including files in 4.3 is much more expensive than it needs to be right now. -Rasmus