Re: [fw-general] ZF and Autoloading

[Bill Karwin]

Lee Saferite wrote:
Anyway, could someone smarter than me give an intelligent reason we do 
not use this function?

Tony Hoare is a very smart man, and he said:  "Premature optimization is 
the root of all evil."
This is known as Hoare's Dictum.  See 
http://en.wikiquote.org/wiki/Tony_Hoare

Donald Knuth is also very smart, and he has quoted the above principle 
frequently.

If a given optimization only improves performance in a fraction of 
cases, and it introduces potential complexity or breaks in certain 
environments, is it worth pursuing the optimization?  It may become 
apparent that a different optimization is much simpler, gives more 
benefit, and works in more environments.  But we won't know which are 
the best improvements until the code is more functionally complete.

The Zend Framework, like any component library, should be 
environment-neutral wherever possible.  We are sure that require_once() 
works in all environments.

You mentioned that autoload may not be compatible with environments that 
do bytecode caching.  I don't know the answer to that, but certainly it 
deserves some testing before we adopt autoloading in the Zend 
Framework.  Feel free to use spl_register_autoload() in your application 
that employs the Zend Framework if it works for you, but I would not be 
in favor of using autoload within ZF until we're sure it won't break in 
some people's environment.

We must also be sure that it will give performance benefit that 
justifies this risk.  Educated conjecture is not a good basis for making 
this design decision.  We must run tests that show quantitatively where 
the performance bottleneck in Zend Framework lies, using a typical 
application as the test case.  Then try the test again with the code 
modified to use autoload.  Quantify how much improvement you get by 
using autoload.  Then compare this solution to the improvement given by 
other potential solutions, for instance bytecode caching.

If the bottleneck in Zend Framework is revealed through testing to be 
unrelated to loading classes, then autoload won't offer much benefit.  
If that is the case, I would favor using the most clear coding practice, 
which is require_once().

Regards,
Bill Karwin