2014-09-25 22:02 GMT+02:00 Marc Delisle <marc@infomarc.info>:
Hugues Peccatte a écrit :
> 2014-09-25 20:08 GMT+02:00 Chirayu Chiripal <chirayu.chiripal@gmail.com>:
>
>>
>> On Thu, Sep 25, 2014 at 12:17 AM, Hugues Peccatte <
>> hugues.peccatte@gmail.com> wrote:
>>
>>> 2014-09-24 14:42 GMT+02:00 Chirayu Chiripal <chirayu.chiripal@gmail.com>:
>>>>
>>>> I was just wondering what will be the effect if we make current methods
>>>> as static in those classes?
>>>>
>>>> --
>>>> Regards,
>>>> Chirayu Chiripal
>>>> https://chirayuchiripal.wordpress.com/
>>>>
>>> It's almost the same thing to use a static or a non-static function. This
>>> is 2.5x slower than standard PHP function.
>>>
>>> See:
>>> Time: 1.7506999969482 //standard PHP function
>>> Time: 4.3722500801086 //object
>>> Time: 4.3722498416901 //object with static function
>>> Time: 3.6722099781036 //PMA function, even this one is 2x slower than
>>> standard PHP function…
>>>
>> Are these tests performed on mb_* function or non mb_* one?

I forgot to answer to this question…
So the tests are with non mb_* functions.
 
>>
>> One thing is clear from these results is that even wrapping a built in
>> function in another function (wrapper function) has some significant
>> overhead. And using OOP/Classes is like wrapping that wrapper function
>> further into a another wrapper causing further slowness. So why not just
>> use mb_* function directly everywhere? If we do this, then what to do in
>> case mb_* functions are not available? Maybe we can just define a new mb_*
>> function to use respective non mb_* function for it. For e.g.
>> // Define a function like this
>> function mb_strlen($str) {
>>     return strlen($str);
>> }
>> // or create an alias of strlen() as mb_strlen() (I heard this is possible
>> in PHP 5.6+)
>>
>> By doing this, we can get speed advantage of built in mb_* function
>> whenever it is available by avoiding any wrapper function or class. And in
>> cases where it is not available, a wrapper function with standard non mb
>> function can deliver equivalent performance as non mb function is
>> inherently faster than its mb_* mod.
>>
>>
>>> Hugues.
>>>
> Hi,
>
> This is a very good idea!
>
> Marc, you asked to create PMA_* string functions. Do you agree with the
> Chirayu's proposal? So we won't have PMA_* functions, but only mb_*
> functions.
>
> Hugues.

I have not followed the latest test results in this thread. If it's true
that using wrappers has a big cost, I am in favor but I fear that people
reading the code do not realize that the mb_* function could be a wrapper.

Not sure how to document this.

--
Marc Delisle (phpMyAdmin)

Using wrappers, the time could be doubled.
So with his proposition, the wrappers would be created only when mb_* doesn't exist.
If mb_* exist, we won't redefine it, but the execution would be slower because mb_* are slower.
If mb_* don't exist, the non-mb_* functions would be wrapped, so will be a little bit longer than without wrap.
And we would also be able to force the use of non-mb_* functions if needed.

As you said, the only thing that might be difficult, would be to document the fact that mb_* functions could be pMA functions. We should choose between clarity and "performance".

Hugues.