Re: Comments about PHPTAL package proposal?

Kornel Lesiński wrote: > I've got one more related issue: Is it OK to include files within > methods (conditionally)? yes, many packages do this to include drivers > > 99% of the time PHPTAL runs compiled, cached copies of templates, and > only requires small fraction of classes to run. > By including only necessary minimum, I get 3-fold performance increase > without opcode cache, and 2-fold with opcode cache (on simple > benchmark that runs PHPTAL with empty template). > > Currently I have: > > class PHPTAL { > … > function parse() > { > require_once 'PHPTAL/Dom/DocumentBuilder.php'; > … > } > > This isn't pretty and probably APC-unfriendly as well. Is there a > better solution? The best solution allowed by PEAR CS is to determine which files are always needed, and to always require_once them. Obviously, since you observe a real-world performance improvement, this means you've found files that are not needed, and it is very smart not to load them. Because the require_once is in a method, it is already conditional (processed at run-time), and so you might find even more performance enhancement if you were to use: if (!class_exists('PHPTAL_Dom_DocumentBuilder')) { include 'PHPTAL/Dom/DocumentBuilder.php'; } I am wary of this kind of construct, as it is hugely slow in the top-level, but I have not benchmarked it inside a method. However, a better system might be to put all of the lazy loading into a separate method function loadClasses() { static $loaded = 0; if ($loaded) { return; } include 'PHPTAL/Random/Class1.php'; include 'PHPTAL/Random/Class2.php'; //etc. } and call loadClasses() from parse(). Even better yet, split the parsing into a completely separate class, so that the parser is loaded only if a template needs recompiling. function parse() { static $loaded = null; if (!$loaded) { // this may be slightly faster if several templates need parsing, but who knows include 'PHPTAL/Parser.php'; $parser = new PHPTAL_Parser(); } return $parser->parse($this); } Again, I haven't benchmarked any of these, so would be interested to hear if any solution results in an improvement for you. Thanks, Greg