Re: [PHP-DEV] Re: #20993 [Ver]: Element value changeswithoutasking

[Melvyn Sopacua]

At 21:48 15-12-2002, Andi Gutmans wrote:


I don't understand what you're doing here. Are you actually separating on 
every assignment and doing a deep copy?

Well,
like I said: I don't if it's the best solution.

What is nasty, is that:

$a = array(1);
$ref = $a[0];
changeVal($a);
echo $a[0];

function changeVal($arr)
{
$arr[0]=2;
}
?>

prints 2, instead of 1, IE: the array has become global, just by creating a 
reference to one of it's keys.


Andi

At 04:09 PM 12/15/2002 +0900, Moriyoshi Koizumi wrote:


[...]


And the result of a tiny benchmark follows:

[Before patching]
1: 0.263245
2: 0.142505
3: 0.328045
4: 0.137149

[After patching]
1: 0.273811
2: 0.141965
3: 0.699429
4: 0.137010


What's 1,2,3 and 4?


Moriyoshi

> My proposal, was based on 2 things: fix or document. I'm sure 
Zeev/Andi had a
> good reason not to always separate, and that probably is performance.
>
> IF this impacts overall performance very negatively, then maybe the better
> choice is to document it.
>
> I'll try the patch though.
>
>
> With kind regards,
>
> Melvyn Sopacua
> 
>
>
> --
> PHP Development Mailing List 
> To unsubscribe, visit: http://www.php.net/unsub.php
>


--
PHP Development Mailing List 
To unsubscribe, visit: http://www.php.net/unsub.php





With kind regards,

Melvyn Sopacua



--
PHP Development Mailing List 
To unsubscribe, visit: http://www.php.net/unsub.php

Re: [PHP-DEV] Re: #20993 [Ver]: Element value changeswithoutasking

[Andi Gutmans]

I don't understand what you're doing here. Are you actually separating on 
every assignment and doing a deep copy?

Andi

At 04:09 PM 12/15/2002 +0900, Moriyoshi Koizumi wrote:
Oops, the patch was wrong as the runtime occationally segfaults
in a case like:


  $a = 0;
  $a = &$a; /* is_ref=1, refcount=1 */
?>

Attached is the revised patch. Please try it out.

And the result of a tiny benchmark follows:

[Before patching]
1: 0.263245
2: 0.142505
3: 0.328045
4: 0.137149

[After patching]
1: 0.273811
2: 0.141965
3: 0.699429
4: 0.137010

Moriyoshi

> My proposal, was based on 2 things: fix or document. I'm sure Zeev/Andi 
had a
> good reason not to always separate, and that probably is performance.
>
> IF this impacts overall performance very negatively, then maybe the better
> choice is to document it.
>
> I'll try the patch though.
>
>
> With kind regards,
>
> Melvyn Sopacua
> 
>
>
> --
> PHP Development Mailing List 
> To unsubscribe, visit: http://www.php.net/unsub.php
>


--
PHP Development Mailing List 
To unsubscribe, visit: http://www.php.net/unsub.php


--
PHP Development Mailing List 
To unsubscribe, visit: http://www.php.net/unsub.php

Re: [PHP-DEV] Re: #20993 [Ver]: Element value changeswithoutasking

[Moriyoshi Koizumi]

Oops, the patch was wrong as the runtime occationally segfaults
in a case like:



Attached is the revised patch. Please try it out.

And the result of a tiny benchmark follows:

[Before patching]
1: 0.263245
2: 0.142505
3: 0.328045
4: 0.137149

[After patching]
1: 0.273811
2: 0.141965
3: 0.699429
4: 0.137010

Moriyoshi

> My proposal, was based on 2 things: fix or document. I'm sure Zeev/Andi had a
> good reason not to always separate, and that probably is performance.
> 
> IF this impacts overall performance very negatively, then maybe the better
> choice is to document it.
> 
> I'll try the patch though.
> 
> 
> With kind regards,
> 
> Melvyn Sopacua
> 
> 
> 
> -- 
> PHP Development Mailing List 
> To unsubscribe, visit: http://www.php.net/unsub.php
> 

Index: zend_execute.c
===
RCS file: /repository/Zend/zend_execute.c,v
retrieving revision 1.316.2.2
diff -u -r1.316.2.2 zend_execute.c
--- zend_execute.c  17 Nov 2002 22:00:32 -  1.316.2.2
+++ zend_execute.c  15 Dec 2002 06:47:03 -
@@ -65,6 +65,7 @@
 static void zend_extension_statement_handler(zend_extension *extension, zend_op_array 
*op_array TSRMLS_DC);
 static void zend_extension_fcall_begin_handler(zend_extension *extension, 
zend_op_array *op_array TSRMLS_DC);
 static void zend_extension_fcall_end_handler(zend_extension *extension, zend_op_array 
*op_array TSRMLS_DC);
+static int _zval_array_disref(zval **ppz);
 
 #define RETURN_VALUE_USED(opline) (!((opline)->result.u.EA.type & EXT_TYPE_UNUSED))
 
@@ -1816,6 +1817,7 @@
} else if (PZVAL_IS_REF(*param)) {

zend_assign_to_variable_reference(NULL, get_zval_ptr_ptr(&EX(opline)->result, EX(Ts), 
BP_VAR_W), param, NULL TSRMLS_CC);
} else {
+   _zval_array_disref(param);
zend_assign_to_variable(NULL, 
&EX(opline)->result, NULL, *param, IS_VAR, EX(Ts) TSRMLS_CC);
}
}
@@ -2480,4 +2482,59 @@
}
}
zend_error(E_ERROR, "Arrived at end of main loop which shouldn't happen");
+}
+
+static int _zval_ref_check(zval **p, void *flag)
+{
+   if ((*p)->is_ref) {
+   *(int *)flag = 1;
+   }
+   _zval_array_disref(p);
+   return ZEND_HASH_APPLY_KEEP;
+}
+
+static void zval_dis_ref(zval **p)
+{
+   if ((*p)->is_ref) {
+   zval *newzv;
+   ALLOC_ZVAL(newzv);
+   *newzv = **p;
+   zval_copy_ctor(newzv);
+   newzv->refcount = 1;
+   newzv->is_ref = 0;
+
+   if ((*p)->refcount < 1) {
+   zval_dtor((*p));
+   FREE_ZVAL(*p);
+   }
+
+   *p = newzv;
+   } else {
+   ((*p))->refcount++;
+   }
+}
+
+static int _zval_array_disref(zval **ppz)
+{
+   TSRMLS_FETCH();
+   if ((*ppz)->type == IS_ARRAY) {
+   int flag = 0;
+   if ((*ppz)->value.ht == &EG(symbol_table)) {
+   return SUCCESS;
+   }
+   zend_hash_apply_with_argument((*ppz)->value.ht, (apply_func_arg_t) 
+_zval_ref_check, (void *)&flag TSRMLS_CC);
+   if (flag) {
+   zval *newzv, *tmp;
+   ALLOC_ZVAL(newzv);
+   *newzv = **ppz;
+   ALLOC_HASHTABLE(newzv->value.ht);
+   zend_hash_init(newzv->value.ht, 0, NULL, ZVAL_PTR_DTOR, 0);
+   zend_hash_copy(newzv->value.ht, Z_ARRVAL_PP(ppz), 
+(copy_ctor_func_t) zval_dis_ref, (void *) &tmp, sizeof(zval *));
+   zval_ptr_dtor(ppz);
+   *ppz = newzv;
+   newzv->is_ref = 0;
+   newzv->refcount = 1;
+   }
+   } 
+   return SUCCESS;
 }



-- 
PHP Development Mailing List 
To unsubscribe, visit: http://www.php.net/unsub.php