|
|
Posted by Steve on 09/18/07 17:17
"Jerry Stuckle" <jstucklex@attglobal.net> wrote in message
news:EM-dnb1p44ldZ3LbnZ2dnUVZ_rHinZ2d@comcast.com...
> Michael Fesser wrote:
>> .oO(NoDude)
>>
>>> @Michael - I currently use __autoload, which is a neat shortcut,
>>> albeit it has the same speed impact as *_once (in my case, even
>>> greater, because of directory traversing).
>>
>> I also traverse a lot of class directories, but only if the requested
>> class could not be found in the class cache, where the locations of all
>> classes are stored. In such case the cache has to be refreshed.
>>
>>> How I (or Steve for that matter) include our files is not (and never
>>> was) my point however. I was just saying and still am - Using require
>>> over require_once makes you think of what dependencies you'll have in
>>> any given request (every single request is unaware of the dependencies
>>> in the previous request and has its own dependencies).
>>
>> Knowing beforehand which classes will be required to handle a particular
>> request is pretty much impossible in my framework. The request handlers
>> themselves decide which of them will be responsible for answering the
>> request and which other objects might be necessary for doing that. It's
>> even possible that a handler instantiates some objects and then forwards
>> the request to a sub handler, which in turn might need the informations
>> provided by the parent handler.
>>
>
> In a properly designed framework, you can predict not only what classes
> will be required, but what methods in those classes.
not necessarily. what about a c++ framework for creating an STL. the
framework is usually *complete* abstraction where little is known, yes?
Navigation:
[Reply to this message]
|