Re: [mantisbt-dev] [mantisbt-commits] mantisbt.git branch, master, updated. release-1.2.0rc1-723-g17a2bc4

[Gianluca S. <gi...@gm...>]

On Sun, Aug 29, 2010 at 10:25 AM, David Hicks <hic...@op...> wrote:
> On Sun, 2010-08-29 at 10:06 +0200, Gianluca Sforna wrote:
>> how exactly require_api calls are better than the standard "require_once"?
>
> require_once() uses the library path environment variable to search for
> a file in each of the library paths. This process requires a bunch of
> stat calls to the file system that can be quite expensive.
>
> I gather newer versions of PHP have improved handling of require_* calls
> so the performance differences probably aren't noticeable.
>
> What require_api does do is ensure that there aren't any file naming
> collisions between the different library paths used by MantisBT. We have
> a relocatable 'core' directory, 'plugin' directory, 'lang' directory,
> etc that could be hosted in completely different locations on the disk.
> By calling require_api we can ensure that PHP is only looking in one
> known location rather than attempting to search through multiple paths
> to find the file (that could be the wrong one in case of naming
> collisions).

Thanks David, at least now I know why we have this stuff  :)

>
> Use of require/require_once/etc is far inferior to using __autoload. If
> MantisBT moves towards a more OO code style, __autoload can then be used
> to increase performance by a significant factor.

Uhm. I thought the two mechanisms were not alternative. I see a lot of
"require_once" in the Zend Framerwork class hierarchy, do know why
this apparent contradiction exists?


-- 
Gianluca Sforna

http://morefedora.blogspot.com
http://www.linkedin.com/in/gianlucasforna