Everything

php

Internal class, read_property() and type == BP_VAR_W

php.internals

Michael Wallner

20 years ago

Hi,

I'm facing an issue I don't know how to overcome.

Short story:
I've an object which properties are connected to the internal struct
whith get_property() and write_property() object handlers.

Now if BP_VAR_W is used to access a property with read_prop() the
following works fine:

// $this->body is a string
$this->body{0} = 'X';

... but the following does nothing than producing a memleak:

$body = &$this->body;
$body = 'X';

Also, it seems impossible to connect *values* (longs or string lengths for instance) through
that interface, because I cannot tell the zval that the value should point elsewhere...

Any advice very much appreciated :)
Do I have to disallow referenced writing access?

Long story (sorry):

static zval *_http_message_object_read_prop(zval *object, zval *member, int type TSRMLS_DC)
{
getObjectEx(http_message_object, obj, object);
http_message *msg = obj->message;
zval *return_value;
#ifdef WONKY
ulong h = zend_get_hash_value(Z_STRVAL_P(member), Z_STRLEN_P(member)+1);
#else
zend_property_info *pinfo = zend_get_property_info(obj->zo.ce, member, 1 TSRMLS_CC);

if (!pinfo || ACC_PROP_PUBLIC(pinfo->flags)) {
return zend_get_std_object_handlers()->read_property(object, member, type TSRMLS_CC);
}
#endif

if (type == BP_VAR_W) {
return_value = &EG(uninitialized_zval);
return_value->refcount = 1;
return_value->is_ref = 1;
} else {
ALLOC_ZVAL(return_value);
return_value->refcount = 0;
return_value->is_ref = 0;
}

#ifdef WONKY
switch (h)
#else
switch (pinfo->h)
#endif
{
case HTTP_MSG_PROPHASH_TYPE:
case HTTP_MSG_CHILD_PROPHASH_TYPE:
RETVAL_LONG(msg->type);
break;

case HTTP_MSG_PROPHASH_HTTP_VERSION:
case HTTP_MSG_CHILD_PROPHASH_HTTP_VERSION:
RETVAL_DOUBLE(msg->http.version);
break;

case HTTP_MSG_PROPHASH_BODY:
case HTTP_MSG_CHILD_PROPHASH_BODY:
phpstr_fix(PHPSTR(msg));
RETVAL_PHPSTR(PHPSTR(msg), 0, !return_value->is_ref);
break;

case HTTP_MSG_PROPHASH_HEADERS:
case HTTP_MSG_CHILD_PROPHASH_HEADERS:
if (return_value->is_ref) {
Z_TYPE_P(return_value) = IS_ARRAY;
Z_ARRVAL_P(return_value) = &msg->hdrs;
} else {
array_init(return_value);
zend_hash_copy(Z_ARRVAL_P(return_value), &msg->hdrs, (copy_ctor_func_t) zval_add_ref, NULL, sizeof(zval *));
}
break;

case HTTP_MSG_PROPHASH_PARENT_MESSAGE:
case HTTP_MSG_CHILD_PROPHASH_PARENT_MESSAGE:
if (msg->parent) {
RETVAL_OBJVAL(obj->parent);
} else {
RETVAL_NULL();
}
break;

case HTTP_MSG_PROPHASH_REQUEST_METHOD:
case HTTP_MSG_CHILD_PROPHASH_REQUEST_METHOD:
if (HTTP_MSG_TYPE(REQUEST, msg) && msg->http.info.request.method) {
RETVAL_STRING(msg->http.info.request.method, 1);
} else {
RETVAL_NULL();
}
break;

case HTTP_MSG_PROPHASH_REQUEST_URI:
if (HTTP_MSG_TYPE(REQUEST, msg) && msg->http.info.request.URI) {
RETVAL_STRING(msg->http.info.request.URI, !return_value->is_ref);
} else {
RETVAL_NULL();
}
break;

case HTTP_MSG_PROPHASH_RESPONSE_CODE:
case HTTP_MSG_CHILD_PROPHASH_RESPONSE_CODE:
if (HTTP_MSG_TYPE(RESPONSE, msg)) {
RETVAL_LONG(msg->http.info.response.code);
} else {
RETVAL_NULL();
}
break;

case HTTP_MSG_PROPHASH_RESPONSE_STATUS:
case HTTP_MSG_CHILD_PROPHASH_RESPONSE_STATUS:
if (HTTP_MSG_TYPE(RESPONSE, msg) && msg->http.info.response.status) {
RETVAL_STRING(msg->http.info.response.status, !return_value->is_ref);
} else {
RETVAL_NULL();
}
break;

default:
#ifdef WONKY
return zend_get_std_object_handlers()->read_property(object, member, type TSRMLS_CC);
#else
RETVAL_NULL();
#endif
break;
}

return return_value;
}

--
Michael - <mike(@)php.net> http://dev.iworks.at/ext-http/http-functions.html.gz

Dmitry Stogov

20 years ago

Hi Michael,

You should define get_property_ptr_ptr() callback to support such
assignments.

Thanks. Dmitry.

Michael Wallner

20 years ago

Hi Dmitry Stogov, you wrote:

> Hi Michael,
>
> You should define get_property_ptr_ptr() callback to support such
> assignments.

Yeah, I looked at that, but what do I gain? get_property_ptr_ptr() will
be used for both described cases and instead of returning a zval* I'd have
to return a zval**, but I could miss the obvious, and if that's the case,
don't hesitate to drop me a quick pointer in the right direction!

Thanks a lot Dmitry,

--
Michael - <mike(@)php.net> http://dev.iworks.at/ext-http/http-functions.html.gz

Dmitry Stogov

20 years ago

read_property(BP_VAR_W) will never work for all cases.
Probably you can return proxy object with get() and set() callbacks, but I
am not sure if it will work for references too.

So if you can return zval** you should use get_property_ptr_ptr().

Dmitry.

Michael Wallner

20 years ago

Dmitry Stogov wrote:

> read_property(BP_VAR_W) will never work for all cases.
> Probably you can return proxy object with get() and set() callbacks, but I
> am not sure if it will work for references too.
>
> So if you can return zval** you should use get_property_ptr_ptr().

Hm... maybe I made not a good start with my question, let me ty again, please:

typedef struct {
zend_object zo;
int i;
} my_obj;

Would you say it's possible that one can maintain i as class
property with write access through references?

Thank you,

--
Michael - <mike(@)php.net> http://dev.iworks.at/ext-http/http-functions.html.gz

Sara Golemon

20 years ago

> Hm... maybe I made not a good start with my question, let me ty again,

please:

>
> typedef struct {
> zend_object zo;
> int i;
> } my_obj;
>
> Would you say it's possible that one can maintain i as class
> property with write access through references?
>

Nope, because PHP variables are loose typed.

Say you *did* have some way to bind my_obj->i to a userspace variable (e.g.:
$i =& $this->i; ).  The next instruction could say $i = 'foo';, then what is
the engine supposed to do with that?

If you want i to have any visibility in userspace, then it needs to be a
zval, and that means it can be modified, and that means that you'll need to
be prepared to re-cast it back to the right type later.  Kinda sucks, but it
sucks for a good reason...

-Sara

Michael Wallner

20 years ago

Sara Golemon wrote:

>> Would you say it's possible that one can maintain i as class
>> property with write access through references?
>>
> Nope, because PHP variables are loose typed.
>
> Say you *did* have some way to bind my_obj->i to a userspace variable
> (e.g.: $i =& $this->i; ).  The next instruction could say $i =
> 'foo';, then what is the engine supposed to do with that?
>
> If you want i to have any visibility in userspace, then it needs to
> be a zval, and that means it can be modified, and that means that
> you'll need to be prepared to re-cast it back to the right type
> later.  Kinda sucks, but it sucks for a good reason...

Uhm... you're so right, thanks, that cleared my mind and makes me remember my first post to the list ;)

Thank you,

--
Michael - <mike(@)php.net> http://dev.iworks.at/ext-http/http-functions.html.gz

Marcus Börger

20 years ago

Hello Michael,

actually your class could have another object with get/set handlers
implemented, nothing more. This could be mapped through the handlers
to your int i.

marcus

Friday, October 28, 2005, 10:44:42 PM, you wrote:

> Sara Golemon wrote:

>>> Would you say it's possible that one can maintain i as class
>>> property with write access through references?
>>>
>> Nope, because PHP variables are loose typed.
>>
>> Say you *did* have some way to bind my_obj->i to a userspace variable
>> (e.g.: $i =& $this->i; ).  The next instruction could say $i =
>> 'foo';, then what is the engine supposed to do with that?
>>
>> If you want i to have any visibility in userspace, then it needs to
>> be a zval, and that means it can be modified, and that means that
>> you'll need to be prepared to re-cast it back to the right type
>> later.  Kinda sucks, but it sucks for a good reason...

> Uhm... you're so right, thanks, that cleared my mind and makes me
> remember my first post to the list ;)

> Thank you,
> --
> Michael - <mike(@)php.net>
> http://dev.iworks.at/ext-http/http-functions.html.gz

Best regards,
Marcus