Re: Re: 1 last error to fix before the application is done!

On Tue, Oct 14, 2008 at 10:29 AM, Colin Guthrie <gmane@colin.guthr.ie> wrote: > Andrew Ballard wrote: >> >> I've heard that a lot, but I just don't see it. I'm sure some of you >> can come up with better tests than this, but here is what I used: > > ... >> >> Based on these results, I'd hardly use the "language construct versus >> function call" optimization argument to make my decision. I'm not sure >> if this is a testament to improvements in the PHP engine over the last >> couple years, or if equipment has gotten fast enough that the >> differences have become irrelevant. Quite possibly, it's both. > > There are a couple of issues with those benchmarks... > 1. It uses rand()... this is generally a bad thing in benchmarks True. I'm not sure how to test it, but I'm curious if there is a bias toward elements that do or do not exist, so I let it pick values both in and out of range. I suppose I could have preselected values to be consistent among all tests. > 2. It uses internal (wall clock) timing and tests different things in the > same application run. I'm on a Windows development machine. > 3. 50k iterations is easy! Give it something hard :) I've done the same with 100k or more. The average is the same. If I'm ever calling either of these functions 100k times in a production script, I need to rethink the app anyway. > You should really use an external timing function (like "time blah" on *nix) > as it shows you real processor time used, as well as wall-clock time. Again, I'm on Windows and my PHP environment is inside Zend Studio. Is this something I can see with something like the profiler in Xdebug? (I have run similar test scripts on the web servers, but they are also running Windows and I don't have access other than file shares to where we publish apps.) > I've posted a simple benchmark else where on this thread that tests which > prove that for a two index-depth test, !empty is approx 4.5x faster than the > equiv array_key_exists tests. I saw that, and for this case I'd agree with you. To be fair, I think the savings is more due to the difference between calling empty() once to get the desired result versus calling array_key_exists() twice and then testing the actual array value (which I'm sure means scanning the array to get the value and then doing a logical comparison). I will, however, file this information for later use as it looks useful in some situations like this one. > Even for a single depth array test, I still get ~3x speed improvement using > empty() vs. array_key_exists(). Andrew