Re: com php-src: Fix class constant fetching: Zend/tests/class_constant_to_reference_cached.phpt Zend/zend_execute.c Zend/zend_vm_def.h
Zend/zend_vm_execute.h
| From: | Dmitry Stogov | Date: | Fri, 30 May 2014 05:02:47 +0000 |
| Subject: | Re: com php-src: Fix class constant fetching: Zend/tests/class_constant_to_reference_cached.phpt Zend/zend_execute.c Zend/zend_vm_def.h Zend/zend_vm_execute.h |
||
| References: | 1 2 3 4 5 | Groups: | php.cvs |
| Request: | Send a blank email to php-cvs+get-78645@lists.php.net to get a copy of this message | ||
Yes. Lua and most compiled languages use this approach.
I agree, func_get_arg() and backtrace won't work out of the box, but may be
we can collect additional arguments into an array in the similar way as you
did it for foo(...$a).
Thanks. Dmitry.
On Fri, May 30, 2014 at 2:27 AM, Nikita Popov <nikita.ppv@gmail.com> wrote:
> On Thu, May 29, 2014 at 11:34 AM, Dmitry Stogov <dmitry@zend.com> wrote:
>
>> Thanks. I saw the patch, but didn't check it yet.
>>
>> BTW: I have one big improvement idea, that might correlate with the work
>> you made.
>>
>> Currently, when we pass an argument to user php function, it's always
>> copied twice.
>> Once by SEND instruction, and then by RECV.
>> It means that we not only do the double work, but also keep two copies of
>> the same zval on VM stack.
>> I started thinking (in background yet) how it might be improved.
>> It's not going to be easy and may require significant redesign, but it
>> must make significant gain.
>> what do you think?
>>
>
> I think this is a good idea in principle (and iirc this is what LuaJIT
> does), but I'm not sure whether this is really feasible for PHP. If we
> share the storage for args pushed on the stack and the CVs used to refer to
> them in functions, this means that any change to the CV will also change
> the passed argument. That usually doesn't matter, but for example
> func_get_arg(s) should always return the original argument(s), which
> wouldn't work anymore with such an approach. Same applies to backtraces for
> exceptions etc.
>
> Nikita
>