Re: [ZEND-ENGINE-CVS] cvs: ZendEngine2(PHP_5_3) / zend.c zend_execute_API.c zend_vm_def.h zend_vm_execute.h php-src NEWS

Hi Derick,

Sorry, but I think mainly about PHP performance, and of course I didn't 
know that you trace return_values it in xdebug.

After the patch, in case if return value is not used we set 
EG(return_value_ptr_ptr) to NULL before call to nested execute().
I assume you can just check for (EG(return_value_ptr_ptr)==NULL) before 
write it. Will it work for you?

Thanks. Dmitry.

Derick Rethans wrote:
> On Fri, 11 Apr 2008, Dmitry Stogov wrote:
> 
>> dmitry		Fri Apr 11 09:43:31 2008 UTC
>>
>>   Modified files:              (Branch: PHP_5_3)
>>     /php-src	NEWS 
>>     /ZendEngine2	zend.c zend_execute_API.c zend_vm_def.h 
>>                 	zend_vm_execute.h 
>>   Log:
>>   Optimized ZEND_RETURN opcode to not allocate and copy return value if it is not
>>   used.
> 
> Hmm, I'm not too happy with this, as it breaks some functionality in 
> Xdebug. Xdebug allows you to create a function trace, including all 
> return values. For example the following script:
> 
> <?php
> function foo() { return 42; }
> foo();
> echo foo();
> ?>
> 
> Gave a function trace like
> (with php -d xdebug.auto_trace=1 -dxdebug.collect_return=1 test-ft.php):
> 
> TRACE START [2008-04-11 10:29:49]
>     0.0002     650528   -> {main}() /tmp/test-ft.php:0
>     0.0003     650840     -> foo() /tmp/test-ft.php:8
>                            >=> 42
>     0.0004     650840     -> foo() /tmp/test-ft.php:9
>                            >=> 42
>                          >=> 1
>     0.0013      54536
> TRACE END   [2008-04-11 10:29:49]
> 
> However, with this change, I only get:
> 
> TRACE START [2008-04-11 10:32:11]
>     0.2148     652336   -> {main}() /tmp/test-ft.php:0
>     0.2907     652648     -> foo() /tmp/test-ft.php:8
>     0.3178     652648     -> foo() /tmp/test-ft.php:9
>                            >=> 42
>     0.4802      56344
> TRACE END   [2008-04-11 10:32:12]
> 
> 
> Is it still possible to get the return values of the functions that are 
> *not* used? I currently check if EG(return_value_ptr_ptr) and 
> *EG(return_value_ptr_ptr) are not NULL to see if there is a return 
> value, and now (of course) they are empty (null pointer).
> 
> regards,
> Derick
>