Joel Roller

Hi,

I've compiled the latest php out of CVS and I'm having the same (wacky)
performance trouble that I've had before. I have cygwin on my machine
that I'm currently using 'time' to test php.exe with my ~100k files of
xes.

The Release_TS_inline numbers appear to be only slightly faster than the
Release_TS numbers. The binary windows php-4.0.0 release is faster
still.

Using the command:
time ./php.exe speed-100k-html.php > dump

For my file with just 100,000 xes, I get this time:
0m7.750s real
0m0.031s user
0m0.031s sys

For my file that <? print ?>s 100,000 chars, I get this time:
0m3.016s real
0m0.046s user
0m0.015s sys

And for php3 I get:
0m0.250s real
0m0.031s user
0m0.031s sys


Perhaps one of you can decipher these numbers... It seems to me that
the user+sys time only adds up to over 0.06 seconds. Which I believe
would be the time that something like this SHOULD actually take.
However, the actual elapsed time is odd because: a) its so slow, b) the
print statements are so much faster than raw characters.

If it would help for me to mail anyone the files I am testing, I will...
I would attatch them here, but they are 200k and it is so easy to
recreate such things in vi ('100ixESCyy999pp').

When I compiled from cvs, the only change I made was adding
ext/standard/scanf.[ch] to the project, because a couple files
referenced php_sscanf_internal.

If you can lend any suggestions for what I might do to get this time
down, please, mail 'em off to me.

-joel