Heim  >  Artikel  >  Backend-Entwicklung  >  深入剖析php执行原理(4):函数的调用,深入剖析php_PHP教程

深入剖析php执行原理(4):函数的调用,深入剖析php_PHP教程

WBOY
WBOYOriginal
2016-07-12 08:50:28948Durchsuche

深入剖析php执行原理(4):函数的调用,深入剖析php

本章开始研究php中函数的调用和执行,先来看函数调用语句是如何被编译的。

我们前面的章节弄明白了函数体会被编译生成哪些zend_op指令,本章会研究函数调用语句会生成哪些zend_op指,等后面的章节再根据这些op指令,来剖析php运行时的细节。

源码依然取自php5.3.29。

函数调用

回顾之前用的php代码示例:

<?<span>php
</span><span>function</span> foo(<span>$arg1</span><span>)
{
    </span><span>print</span>(<span>$arg1</span><span>);
}

</span><span>$bar</span> = 'hello php'<span>;
foo(</span><span>$bar</span>);

在函数编译一章里已经分析过,函数foo最终会编译生成对应的zend_function,存放于函数表(CG(function_table))中。

现在开始看 foo($bar); 一句,这应该是最简单的函数调用语句了。其他还有一些形式更为复杂的函数调用,例如以可变变量作为函数名,例如导入的函数以别名进行调用(涉及到命名空间),再例如以引用作为参数,以表达式作为参数,以函数调用本身作为参数等等。

我们从简单的来入手,弄清楚调用语句的编译过程及产出,对于复杂的一些调用,下文也争取都能谈到一些。

1、语法推导

就 foo($bar); 而言,其主要部分语法树为:

绿色的节点表示最后对应到php代码中的字面。红色的部分是语法推导过程中最重要的几步,特别是function_call。

我们从语法分析文件zend_language_parser.y中挑出相关的:

function_call:
		namespace_name '(' { $2.u.opline_num = zend_do_begin_function_call(&$1, 1 TSRMLS_CC); }
				function_call_parameter_list
				')' { zend_do_end_function_call(&$1, &$$, &$4, 0, $2.u.opline_num TSRMLS_CC); zend_do_extended_fcall_end(TSRMLS_C); }
	|	T_NAMESPACE T_NS_SEPARATOR namespace_name '(' { $1.op_type = IS_CONST; ZVAL_EMPTY_STRING(&$1.u.constant);  zend_do_build_namespace_name(&$1, &$1, &$3 TSRMLS_CC); $4.u.opline_num = zend_do_begin_function_call(&$1, 0 TSRMLS_CC); }
				function_call_parameter_list
				')' { zend_do_end_function_call(&$1, &$$, &$6, 0, $4.u.opline_num TSRMLS_CC); zend_do_extended_fcall_end(TSRMLS_C); }
	|	T_NS_SEPARATOR namespace_name '(' { $3.u.opline_num = zend_do_begin_function_call(&$2, 0 TSRMLS_CC); }
				function_call_parameter_list
				')' { zend_do_end_function_call(&$2, &$$, &$5, 0, $3.u.opline_num TSRMLS_CC); zend_do_extended_fcall_end(TSRMLS_C); }
	|	class_name T_PAAMAYIM_NEKUDOTAYIM T_STRING '(' { $4.u.opline_num = zend_do_begin_class_member_function_call(&$1, &$3 TSRMLS_CC); }
			function_call_parameter_list
			')' { zend_do_end_function_call($4.u.opline_num?NULL:&$3, &$$, &$6, $4.u.opline_num, $4.u.opline_num TSRMLS_CC); zend_do_extended_fcall_end(TSRMLS_C);}
	|	class_name T_PAAMAYIM_NEKUDOTAYIM variable_without_objects '(' { zend_do_end_variable_parse(&$3, BP_VAR_R, 0 TSRMLS_CC); zend_do_begin_class_member_function_call(&$1, &$3 TSRMLS_CC); }
			function_call_parameter_list
			')' { zend_do_end_function_call(NULL, &$$, &$6, 1, 1 TSRMLS_CC); zend_do_extended_fcall_end(TSRMLS_C);}
	|	variable_class_name T_PAAMAYIM_NEKUDOTAYIM T_STRING '(' { zend_do_begin_class_member_function_call(&$1, &$3 TSRMLS_CC); }
			function_call_parameter_list
			')' { zend_do_end_function_call(NULL, &$$, &$6, 1, 1 TSRMLS_CC); zend_do_extended_fcall_end(TSRMLS_C);}
	|	variable_class_name T_PAAMAYIM_NEKUDOTAYIM variable_without_objects '(' { zend_do_end_variable_parse(&$3, BP_VAR_R, 0 TSRMLS_CC); zend_do_begin_class_member_function_call(&$1, &$3 TSRMLS_CC); }
			function_call_parameter_list
			')' { zend_do_end_function_call(NULL, &$$, &$6, 1, 1 TSRMLS_CC); zend_do_extended_fcall_end(TSRMLS_C);}
	|	variable_without_objects  '(' { zend_do_end_variable_parse(&$1, BP_VAR_R, 0 TSRMLS_CC); zend_do_begin_dynamic_function_call(&$1, 0 TSRMLS_CC); }
			function_call_parameter_list ')'
			{ zend_do_end_function_call(&$1, &$$, &$4, 0, 1 TSRMLS_CC); zend_do_extended_fcall_end(TSRMLS_C);}
;<br /><br />function_call_parameter_list:<br />        non_empty_function_call_parameter_list    { $$ = $1; }<br />    |    /* empty */                        { Z_LVAL($$.u.constant) = 0; }<br />;<br /><br /><br />non_empty_function_call_parameter_list:<br />        expr_without_variable    { Z_LVAL($$.u.constant) = 1;  zend_do_pass_param(&$1, ZEND_SEND_VAL, Z_LVAL($$.u.constant) TSRMLS_CC); }<br />    |    variable                { Z_LVAL($$.u.constant) = 1;  zend_do_pass_param(&$1, ZEND_SEND_VAR, Z_LVAL($$.u.constant) TSRMLS_CC); }<br />    |    '&' w_variable          { Z_LVAL($$.u.constant) = 1;  zend_do_pass_param(&$2, ZEND_SEND_REF, Z_LVAL($$.u.constant) TSRMLS_CC); }<br />    |    non_empty_function_call_parameter_list ',' expr_without_variable    { Z_LVAL($$.u.constant)=Z_LVAL($1.u.constant)+1;  zend_do_pass_param(&$3, ZEND_SEND_VAL, Z_LVAL($$.u.constant) TSRMLS_CC); }<br />    |    non_empty_function_call_parameter_list ',' variable                 { Z_LVAL($$.u.constant)=Z_LVAL($1.u.constant)+1;  zend_do_pass_param(&$3, ZEND_SEND_VAR, Z_LVAL($$.u.constant) TSRMLS_CC); }<br />    |    non_empty_function_call_parameter_list ',' '&' w_variable           { Z_LVAL($$.u.constant)=Z_LVAL($1.u.constant)+1;  zend_do_pass_param(&$4, ZEND_SEND_REF, Z_LVAL($$.u.constant) TSRMLS_CC); }<br />;

其结构并不复杂:

1)function_call这条推导,代表了一个完整的函数调用。

2)namespace_name是指经过命名空间修饰过之后的函数名,由于我们的例子中,函数foo并没有处于任何一个命名空间里,所以namespace_name其实就是foo。如果我们的函数定义在命名空间中,则namespace_name是一个类似“全路径”的fullname。

<span>namespace MyProject
{
</span><span>function</span> foo(<span>$arg1</span><span>)
{
    </span><span>print</span>(<span>$arg1</span><span>);
}
}

namespace
{
</span><span>$bar</span> = 'hello php'<span>;</span>
MyProject\foo(<span>$bar</span><span>);<span>//<span> 以类似&ldquo;全路径&rdquo;的fullname来调用函数,则namespace_name为MyProject\foo</span></span> 
}</span>

3)function_call_parameter_list是函数的参数列表,而non_empty_function_call_parameter_list则代表了非空参数列表。

4)从这些推导产生式里,我们还能看出编译时的所运用的一些关键处理:

zend_do_begin_function_call-->zend_do_pass_param-->zend_do_end_function_call<br /><br />        开始                        解析参数                     结束

和编译function语句块时的几步(zend_do_begin_function_declaration->zend_do_receive_arg->zend_do_end_function_declaration等)顺序上比较类似。

上面提到语法树我们仅仅画了一部分,准确讲,没有将namespace以及function_call_parameter_list以下的推导过程进一步画出来。原因一是namespace的推导比较简单。第二,由于function_call_parameter_list-->variable这步会回到variable上,而variable经过若干步一直到产生变量$bar的推导比较复杂,也不是本文的重点,所以这里就不一进步探究了。

2、开始编译

看下function_call的推导式,一开始,zend vm会执行zend_do_begin_function_call做一些函数调用的准备。

2.1、 zend_do_begin_function_call

代码注解如下:

zend_function *<span>function;
</span><span>char</span> *<span>lcname;
</span><span>char</span> *is_compound = memchr(Z_STRVAL(function_name->u.constant), <span>'</span><span>\\</span><span>'</span>, Z_STRLEN(function_name-><span>u.constant));

</span><span>//</span><span> 将函数名进行修正,例如带上命名空间作为前缀等</span>
<span>zend_resolve_non_class_name(function_name, check_namespace TSRMLS_CC);

</span><span>//</span><span> 能进入该分支,说明在一个命名空间下以shortname调用函数,会生成一条DO_FCALL_BY_NAME指令</span>
<span>if</span> (check_namespace && CG(current_namespace) && !<span>is_compound) {
        </span><span>/*</span><span> We assume we call function from the current namespace
        if it is not prefixed. </span><span>*/</span>

        <span>/*</span><span> In run-time PHP will check for function with full name and
        internal function with short name </span><span>*/</span><span>
        zend_do_begin_dynamic_function_call(function_name, </span><span>1</span><span> TSRMLS_CC);
        </span><span>return</span> <span>1</span><span>;
} 

</span><span>//</span><span> 转成小写,因为CG(function_table)中的函数名都是小写</span>
lcname = zend_str_tolower_dup(function_name->u.constant.value.str.val, function_name-><span>u.constant.value.str.len);

</span><span>//</span><span> 如果function_table中找不到该函数,则也尝试生成DO_FCALL_BY_NAME指令</span>
<span>if</span> ((zend_hash_find(CG(function_table), lcname, function_name->u.constant.value.str.len+<span>1</span>, (<span>void</span> **) &function) == FAILURE) ||<span>
    ((CG(compiler_options) </span>& ZEND_COMPILE_IGNORE_INTERNAL_FUNCTIONS) && (function->type ==<span> ZEND_INTERNAL_FUNCTION))) {
        zend_do_begin_dynamic_function_call(function_name, </span><span>0</span><span> TSRMLS_CC);
        efree(lcname);
        </span><span>return</span> <span>1</span>; <span>/*</span><span> Dynamic </span><span>*/</span><span>
} 
efree(function_name</span>-><span>u.constant.value.str.val);
function_name</span>->u.constant.value.str.val =<span> lcname;

</span><span>//</span><span> 压入CG(function_call_stack)</span>
zend_stack_push(&CG(function_call_stack), (<span>void</span> *) &function, <span>sizeof</span>(zend_function *<span>));
zend_do_extended_fcall_begin(TSRMLS_C);
</span><span>return</span> <span>0</span>;

有几点需要理解的:

1,zend_resolve_non_class_name。由于php支持命名空间、也支持别名/导入等特性,因此首先要做的是将函数名称进行修正,否则在CG(function_table)中找不到。例如,函数处于一个命名空间中,则可能需要将函数名添加上命名空间作为前缀,最终形成完整的函数名,也就是我们前文提到的以一种类似“全路径”的fullname作为函数名。再例如,函数名只是一个设置的别名,它实际指向了另一个命名空间中的某个函数,则需要将其改写成真正被调用函数的名称。这些工作,均由zend_resolve_non_class_name完成。命名空间添加了不少复杂度,下面是一些简单的例子:

<?<span>php
namespace MyProject;

</span><span>function</span> foo(<span>$arg1</span><span>)
{
    </span><span>print</span>(<span>$arg1</span><span>);
}

</span><span>$bar</span> = 'hello php'<span>;<br />
foo(</span><span>$bar</span>);                 <span>//</span><span> zend_resolve_non_class_name会将foo处理成MyProject\foo</span><span>
namespace\foo(</span><span>$bar</span>);       <span>//</span><span> 在进入zend_do_begin_function_call之前,<span>函数名已经</span>被扩展成<span><span>\<span>MyProject\foo</span></span></span>,再经过zend_resolve_non_class_name,<span>将\<span>MyProject\foo</span></span>处理成MyProject\foo</span><span>
\MyProject\foo(</span><span>$bar</span>);      <span>//</span><span> zend_resolve_non_class_name会将<span><span>\<span>MyProject\foo</span></span></span>处理成MyProject\foo</span>

总之,zend_resolve_non_class_name是力图生成一个最精确、最完整的函数名。

2,CG(current_namespace)存储了当前的命名空间。check_namespace和!is_compound一起说明被调用函数在当前命名空间下的,并且以shortname名称被调用。所谓shortname,是和上述的fullname相对,shorname的函数名,不存在"\"。

就像上面的例子中,我们在MyProject命名空间下,以foo为函数名来调用。这种情况下,check_namespace=1,is_compound = NULL,CG(current_namespace) = MyProject。因此,会走到zend_do_begin_dynamic_function_call里进一步处理。zend_do_begin_dynamic_function_call我们下面再具体描述。

<?<span>php
namespace MyProject\sub;

</span><span>function</span> foo(<span>$arg1</span><span>)
{
    </span><span>print</span>(<span>$arg1</span><span>);
}

namespace MyProject;
</span><span>$bar</span> = 'hello php'<span>;
sub\foo(</span><span>$bar</span>);      <span>//</span><span> 以sub\foo调用函数,并不算shortname,因为存在\</span>

注意上述例子,我们以sub\foo来调用函数。zend_resolve_non_class_name会将函数名处理成MyProject\sub\foo。不过is_compound是在zend_resolve_non_class_name之前算的,由于sub\foo存在"\",所以is_compound为"\foo",!is_compound是false,因而不能进入zend_do_begin_dynamic_function_call。

3,同样,如果CG(function_table)中找不到函数,也会进入zend_do_begin_dynamic_function_call进一步处理。为什么在函数表中找不到函数,因为php允许我们先调用,再去定义函数。例如:

<?<span>php
</span><span>$bar</span> = 'hello php'<span>;

</span><span>//</span><span> 先调用</span>
foo(<span>$bar</span><span>);

</span><span>//</span><span> 后定义</span>
<span>function</span> foo(<span>$arg1</span><span>)
{
    </span><span>print</span>(<span>$arg1</span><span>);
}</span>

4,在zend_do_begin_function_call的最后,我们将函数压入CG(function_call_stack)。这是一个栈,因为在后续对传参的编译,我们仍然需要用到函数,所以这里将其压亚入栈中,方便后面获取使用。之所以用栈,是因为调用函数传递的参数,可能是另一次函数调用。为了确保参数总是能找到对应的函数,所以用栈。

<?<span>php
</span><span>function</span> foo(<span>$arg1</span><span>)
{
    </span><span>print</span>(<span>$arg1</span><span>);
}

</span><span>$bar</span> = 'hello php'<span>;
foo(</span><span>strlen</span>(<span>$bar</span>));   <span>//</span><span> 首先foo入栈,然后分析参数strlen($bar),发现依然是个函数,于是strlen入栈,再分析参数$bar,此时弹出对应的函数正好为strlen。</span>

2.2、 zend_do_begin_dynamic_function_call

前面提到,正常的调用,会先执行zend_do_begin_function_call,在zend_do_begin_function_call中有两种情况会进一步调用zend_do_begin_dynamic_function_call来处理。

一是,在命名空间中,以shortname调用函数;

二是,在调用函数时,尚未定义函数。

其实还有第三种情况会走到zend_do_begin_dynamic_function_call,就是当我们调用函数的时候,函数名并非直接写成字面,而是通过变量等形式来间接确定。这种情况下,zend vm会直接执行zend_do_begin_dynamic_function_call。

举例1:

<?<span>php
</span><span>function</span> foo(<span>$arg1</span><span>)
{
    </span><span>print</span>(<span>$arg1</span><span>);
}

</span><span>$bar</span> = 'hello php'<span>;
</span><span>$func</span> = 'foo'<span>;
</span><span>$func</span>(<span>$bar</span>);          <span>//</span><span> 我们以变量$func作为函数名,试图调用函数foo,$func类型是IS_CV</span>

此时, $func($bar) 对应function_call语法推导式的最后一条:

function_call:<br />        ...
	|	variable_without_objects  '(' { zend_do_end_variable_parse(&$1, BP_VAR_R, 0 TSRMLS_CC); zend_do_begin_dynamic_function_call(&$1, 0 TSRMLS_CC); }
			function_call_parameter_list ')'
			{ zend_do_end_function_call(&$1, &$$, &$4, 0, 1 TSRMLS_CC); zend_do_extended_fcall_end(TSRMLS_C);}

推导式中的variable_without_objects对应的就是变量 $func 。$func其实是一个compiled_variable,并且在op_array->vars数组中索引为1,索引为0的是在它之前定义的变量 $bar

举例2:

<span>function</span> foo(<span>$arg1</span><span>)
{
    </span><span>print</span>(<span>$arg1</span><span>);
}

</span><span>$bar</span> = 'hello php'<span>;
</span><span>$func</span> = 'foo'<span>;
</span><span>$ref_func</span> = 'func'<span>;
$</span><span>$ref_func</span>(<span>$bar</span>); <span>//</span><span> 以可变变量的形式来调用函数,$$ref_func类型是IS_VAR</span>

该例是以可变变量来调用函数,和例1一样, $$ref_func($bar)也是对应function_call语法推导式的最后一条,所以不会走进zend_do_begin_function_call,而是直接进入zend_do_begin_dynamic_function_call。不同的点在于 $$ref_func 节点类型不再是compiled_variable,而是普通的variable,标识为IS_VAR。

下面的图画出了5种case,第1种不经过zend_do_begin_dynamic_function_call,而后4种会调用zend_do_begin_dynamic_function_call处理,注意最后2种不经过zend_do_begin_function_call:

具体看下zend_do_begin_dynamic_function_call的代码:

<span>void</span> zend_do_begin_dynamic_function_call(znode *function_name, <span>int</span> ns_call TSRMLS_DC) <span>/*</span><span> {{{ </span><span>*/</span><span>
{
    unsigned </span><span>char</span> *ptr =<span> NULL;
    zend_op </span>*opline, *<span>opline2;

    </span><span>//</span><span> 拿一条zend_op</span>
    opline =<span> get_next_op(CG(active_op_array) TSRMLS_CC);
    
    </span><span>//</span><span> 参数ns_call表名是否以shortname在命名空间中调用函数</span>
    <span>if</span><span> (ns_call) {
        </span><span>char</span> *<span>slash;
        </span><span>int</span><span> prefix_len, name_len;
        </span><span>/*</span><span> In run-time PHP will check for function with full name and
           internal function with short name </span><span>*/</span>
        
        <span>//</span><span> 第一条指令是ZEND_INIT_NS_FCALL_BY_NAME</span>
        opline->opcode =<span> ZEND_INIT_NS_FCALL_BY_NAME;
        opline</span>->op2 = *<span>function_name;
        opline</span>->extended_value = <span>0</span><span>;
        opline</span>->op1.op_type =<span> IS_CONST;
        Z_TYPE(opline</span>->op1.u.constant) =<span> IS_STRING;
        Z_STRVAL(opline</span>->op1.u.constant) = zend_str_tolower_dup(Z_STRVAL(opline->op2.u.constant), Z_STRLEN(opline-><span>op2.u.constant));
        Z_STRLEN(opline</span>->op1.u.constant) = Z_STRLEN(opline-><span>op2.u.constant);
        opline</span>->extended_value = zend_hash_func(Z_STRVAL(opline->op1.u.constant), Z_STRLEN(opline->op1.u.constant) + <span>1</span><span>);
        
        </span><span>//</span><span> 再拿一条zend_op,指令为ZEND_OP_DATA</span>
        slash = zend_memrchr(Z_STRVAL(opline->op1.u.constant), <span>'</span><span>\\</span><span>'</span>, Z_STRLEN(opline-><span>op1.u.constant));
        prefix_len </span>= slash-Z_STRVAL(opline->op1.u.constant)+<span>1</span><span>;
        name_len </span>= Z_STRLEN(opline->op1.u.constant)-<span>prefix_len;
        opline2 </span>=<span> get_next_op(CG(active_op_array) TSRMLS_CC);
        opline2</span>->opcode =<span> ZEND_OP_DATA;
        opline2</span>->op1.op_type =<span> IS_CONST;
        Z_TYPE(opline2</span>->op1.u.constant) =<span> IS_LONG;
        </span><span>if</span>(!<span>slash) {
            zend_error(E_CORE_ERROR, </span><span>"</span><span>Namespaced name %s should contain slash</span><span>"</span>, Z_STRVAL(opline-><span>op1.u.constant));
        }
        </span><span>/*</span><span> this is the length of namespace prefix </span><span>*/</span><span>
        Z_LVAL(opline2</span>->op1.u.constant) =<span> prefix_len;
        </span><span>/*</span><span> this is the hash of the non-prefixed part, lowercased </span><span>*/</span><span>
        opline2</span>->extended_value = zend_hash_func(slash+<span>1</span>, name_len+<span>1</span><span>);
        SET_UNUSED(opline2</span>-><span>op2);
    } </span><span>else</span><span> {
        </span><span>//</span><span> 第一条指令是ZEND_INIT_FCALL_BY_NAME</span>
        opline->opcode =<span> ZEND_INIT_FCALL_BY_NAME;
        opline</span>->op2 = *<span>function_name;
        
        </span><span>//</span><span> 先调用,再定义</span>
        <span>if</span> (opline->op2.op_type ==<span> IS_CONST) {
            opline</span>->op1.op_type =<span> IS_CONST;
            Z_TYPE(opline</span>->op1.u.constant) =<span> IS_STRING;
            Z_STRVAL(opline</span>->op1.u.constant) = zend_str_tolower_dup(Z_STRVAL(opline->op2.u.constant), Z_STRLEN(opline-><span>op2.u.constant));
            Z_STRLEN(opline</span>->op1.u.constant) = Z_STRLEN(opline-><span>op2.u.constant);
            opline</span>->extended_value = zend_hash_func(Z_STRVAL(opline->op1.u.constant), Z_STRLEN(opline->op1.u.constant) + <span>1</span><span>);
        }
        </span><span>//</span><span> 以变量当函数名来调用</span>
        <span>else</span><span> {
            opline</span>->extended_value = <span>0</span><span>;
            SET_UNUSED(opline</span>-><span>op1);
        }
    }

    </span><span>//</span><span> 将NULL压入CG(function_call_stack)</span>
    zend_stack_push(&CG(function_call_stack), (<span>void</span> *) &ptr, <span>sizeof</span>(zend_function *<span>));
    zend_do_extended_fcall_begin(TSRMLS_C);
}</span>

ns_call参数取值为0或者1。如果在命名空间中,以shortname调用函数,则ns_call = 1,并且会生成2条指令。如果是先调用再定义,或者以变量作函数名,则ns_call = 0,并且只会生成1条指令。

以ns_call = 1为例:

<?<span>php
namespace MyProject;

</span><span>function</span> foo(<span>$arg1</span><span>)
{
    </span><span>print</span>(<span>$arg1</span><span>);
}

</span><span>$bar</span> = 'hello php'<span>;
foo(</span><span>$bar</span>);

 生成的op指令如下所示:

深入剖析php执行原理(4):函数的调用,深入剖析php_PHP教程php $bar = 'hello php'; foo($bar); function foo($arg1) { print($arg1); }

生成的op指令如下所示:

深入剖析php执行原理(4):函数的调用,深入剖析php_PHP教程php function foo($arg1) { print($arg1); } $bar = 'hello php'; $func = 'foo'; $func($bar);

生成的op指令如下所示:

深入剖析php执行原理(4):函数的调用,深入剖析php_PHP教程void zend_do_pass_param(znode *param, zend_uchar op, int offset TSRMLS_DC) /* {{{ */ { zend_op *opline; int original_op = op; zend_function **function_ptr_ptr, *function_ptr; int send_by_reference; int send_function = 0; // 从CG(function_call_stack)获取当前函数,注意可能拿出的是NULL zend_stack_top(&CG(function_call_stack), (void **) &function_ptr_ptr); function_ptr = *function_ptr_ptr; // 调用的地方以引用传参,但是php.ini中配置不允许这样,则抛错 if (original_op == ZEND_SEND_REF && !CG(allow_call_time_pass_reference)) { if (function_ptr && function_ptr->common.function_name && function_ptr->common.type == ZEND_USER_FUNCTION && !ARG_SHOULD_BE_SENT_BY_REF(function_ptr, (zend_uint) offset)) { zend_error(E_DEPRECATED, "Call-time pass-by-reference has been deprecated; " "If you would like to pass it by reference, modify the declaration of %s(). " "If you would like to enable call-time pass-by-reference, you can set " "allow_call_time_pass_reference to true in your INI file", function_ptr->common.function_name); } else { zend_error(E_DEPRECATED, "Call-time pass-by-reference has been deprecated"); } }

1,首先是从CG(function_call_stack)中获取当前参数对应的函数。注意,可能拿到的只是一个NULL。因为php的语法允许我们先函数调用,再接着对函数进行定义。如前文所述,这种情况下zend_do_begin_function_call中会向CG(function_call_stack)中压入NULL,同时会产生DO_FCALL_BY_NAME指令。

2,在传参的语法推导式中,op可能会有3种,分别是ZEND_SEND_VAL、ZEND_SEND_VAR、ZEND_SEND_REF。

expr_without_variable    { Z_LVAL($$.u.constant) = <span>1</span>;  zend_do_pass_param(&$<span>1</span><span>, ZEND_SEND_VAL, Z_LVAL($$.u.constant) TSRMLS_CC); }
variable                 { Z_LVAL($$.u.constant) </span>= <span>1</span>;  zend_do_pass_param(&$<span>1</span><span>, ZEND_SEND_VAR, Z_LVAL($$.u.constant) TSRMLS_CC); }
</span><span>'</span><span>&</span><span>'</span> w_variable           { Z_LVAL($$.u.constant) = <span>1</span>;  zend_do_pass_param(&$<span>2</span>, ZEND_SEND_REF, Z_LVAL($$.u.constant) TSRMLS_CC); }

这三种op分别对应的语法是expr_without_variable、variable、'&'w_variable,简单来说就是“不含变量的表达式”、“变量”、“引用”。

zend_do_pass_param会判断,如果用户传递的是引用,但同时在php.INI中配置了形如 allow_call_time_pass_reference = Off ,则需要产生一条E_DEPRECATED错误信息,告知用户传递的时候不建议强制写成引用。

其实,还有第4种传参的opcode,即ZEND_SEND_VAR_NO_REF。我们接下来会提到。

<span>//</span><span> 函数已定义,则根据函数的定义,来决定send_by_reference是否传引用</span>
<span>if</span><span> (function_ptr) {
    </span><span>if</span><span> (ARG_MAY_BE_SENT_BY_REF(function_ptr, (zend_uint) offset)) {
        ...
    } </span><span>else</span><span> {
        </span><span>//</span><span> 要么为0,要么为ZEND_ARG_SEND_BY_REF</span>
        send_by_reference = ARG_SHOULD_BE_SENT_BY_REF(function_ptr, (zend_uint) offset) ? ZEND_ARG_SEND_BY_REF : <span>0</span><span>;
    }
}
</span><span>//</span><span> 函数为定义,先统一将send_by_reference置为0</span>
<span>else</span><span> {
    send_by_reference </span>= <span>0</span><span>;
}

</span><span>//</span><span> 如果用户传递的参数,本身就是一次函数调用,则将op改成ZEND_SEND_VAR_NO_REF</span>
<span>if</span> (op == ZEND_SEND_VAR &&<span> zend_is_function_or_method_call(param)) {
    </span><span>/*</span><span> Method call </span><span>*/</span><span>
    op </span>=<span> ZEND_SEND_VAR_NO_REF;
    send_function </span>=<span> ZEND_ARG_SEND_FUNCTION;
}
</span><span>//</span><span> 如果用户传递的参数,是一个表达式,并且结果会产生中间变量,则也将op改成ZEND_SEND_VAR_NO_REF</span>
<span>else</span> <span>if</span> (op == ZEND_SEND_VAL && (param->op_type & (IS_VAR|<span>IS_CV))) {
    op </span>=<span> ZEND_SEND_VAR_NO_REF;
}</span>

1,send_by_reference表示根据函数的定义,参数是不是引用。ARG_MAY_BE_SENT_BY_REF和ARG_SHOULD_BE_SENT_BY_REF两个宏这里就不具体叙述了,感兴趣的朋友可以自己阅读代码。

2,op == ZEND_SEND_VAR对应的是variable,假如参数是一个函数调用,也可能会被编译成variable,但是函数调用并不存在显式定义的变量,所以不能直接编译成SEND_VAR指令,因此这里就涉及到了上文提到的第4种opcode,即ZEND_SEND_VAR_NO_REF。例如:

3,op == ZEND_SEND_VAL对应的是一个表达式,如果该表达式产生了一个变量作为结果,则也需要将op改成ZEND_SEND_VAR_NO_REF。例如:

继续来看zend_do_pass_param:

<span>//</span><span> 如果根据函数定义需要传递引用,且实际传递的参数是变量,则将op改成ZEND_SEND_REF</span>
<span>if</span> (op!=ZEND_SEND_VAR_NO_REF && send_by_reference==<span>ZEND_ARG_SEND_BY_REF) {
    </span><span>/*</span><span> change to passing by reference </span><span>*/</span>
    <span>switch</span> (param-><span>op_type) {
        </span><span>case</span><span> IS_VAR:
        </span><span>case</span><span> IS_CV:
            op </span>=<span> ZEND_SEND_REF;
            </span><span>break</span><span>;
        </span><span>default</span><span>:
            zend_error(E_COMPILE_ERROR, </span><span>"</span><span>Only variables can be passed by reference</span><span>"</span><span>);
            </span><span>break</span><span>;
    }
}

</span><span>//</span><span> 如果实际传递的参数是变量,调用zend_do_end_variable_parse处理链式调用</span>
<span>if</span> (original_op ==<span> ZEND_SEND_VAR) {
    </span><span>switch</span><span> (op) {
        </span><span>case</span><span> ZEND_SEND_VAR_NO_REF:
            zend_do_end_variable_parse(param, BP_VAR_R, </span><span>0</span><span> TSRMLS_CC);
            </span><span>break</span><span>;
        </span><span>case</span><span> ZEND_SEND_VAR:
            </span><span>if</span><span> (function_ptr) {
                zend_do_end_variable_parse(param, BP_VAR_R, </span><span>0</span><span> TSRMLS_CC);
            } </span><span>else</span><span> {
                zend_do_end_variable_parse(param, BP_VAR_FUNC_ARG, offset TSRMLS_CC);
            }
            </span><span>break</span><span>;
        </span><span>case</span><span> ZEND_SEND_REF:
            zend_do_end_variable_parse(param, BP_VAR_W, </span><span>0</span><span> TSRMLS_CC);
            </span><span>break</span><span>;
    }
}</span>

这里注意param->op_type是传递的参数经过编译得到znode的op_type,如果不属于变量(IS_VARIS_CV),就直接报错了。举例来说:

<span>function</span> foo(&<span>$a</span><span>)
{
    </span><span>print</span>(<span>$a</span><span>);
}

foo(</span><span>$bar</span> == 1);  // 抛错 <span>"<span>Only variables can be passed by reference<span>"</span></span></span>

上面 $bar == 1 表达式的编译结果,op_type为IS_TMP_VAR,可以看做一种临时的中间结果,并非IS_VAR,IS_CV,因此无法编译成功。看着逻辑有点绕,其实很好理解。因为我们传递引用,实际目的是希望能够在函数中,对这个参数的值进行修改,需要参数是可写的。然而 $bar == 1 产生的中间结果,我们无法做出修改,是只读的。

来看zend_do_pass_param的最后一段:

<span>//</span><span> 获取下一条zend op指令</span>
opline =<span> get_next_op(CG(active_op_array) TSRMLS_CC);

</span><span>//</span><span> extended_value加上不同的附加信息</span>
<span>if</span> (op ==<span> ZEND_SEND_VAR_NO_REF) {
    </span><span>if</span><span> (function_ptr) {
        opline</span>->extended_value = ZEND_ARG_COMPILE_TIME_BOUND | send_by_reference |<span> send_function;
    } </span><span>else</span><span> {
        opline</span>->extended_value =<span> send_function;
    }
} </span><span>else</span><span> {
    </span><span>if</span><span> (function_ptr) {
        opline</span>->extended_value =<span> ZEND_DO_FCALL;
    } </span><span>else</span><span> {
        opline</span>->extended_value =<span> ZEND_DO_FCALL_BY_NAME;
    }
}

</span><span>//</span><span> 设置opcode、op1、op2等</span>
opline->opcode =<span> op;
opline</span>->op1 = *<span>param;
opline</span>->op2.u.opline_num =<span> offset;
SET_UNUSED(opline</span>->op2);

上面这段代码生成了一条SEND指令。如果我们调用函数时候传递了多个参数,则会调用多次zend_do_pass_param,最终会生成多条SEND指令。

至于指令具体是SEND_VAR,SEND_VAL,还是SEND_RE,亦或是ZEND_SEND_VAR_NO_REF,则依靠zend_do_pass_param中的判断。zend_do_pass_param中的逻辑分支比较多,一下子不能弄明白所有分支也没关系,最重要的是知道它会根据函数的定义以及实际传递的参数,产生最合适的SEND指令。

还是回到我们开始的例子,对于 foo($bar) ,则经过zend_do_pass_param之后,产生的SEND指令细节如下:

 

4、结束编译

结束函数调用是通过zend_do_end_function_call来完成的。根据前文所述,zend_do_begin_function_call并不产生一条实际的调用指令,但它确定了最终函数调用走的是DO_FCALL还是DO_FCALL_BY_NAME,并且据此来生成ZEND_INIT_NS_FCALL_BY_NAME或ZEND_INIT_FCALL_BY_NAME指令。

实际的调用指令是放在zend_do_end_function_call中来生成的。

具体分析下zend_do_end_function_call

zend_op *<span>opline;

</span><span>//</span><span> 这段逻辑分支现在已经走不到了</span>
<span>if</span> (is_method && function_name && function_name->op_type ==<span> IS_UNUSED) {
    </span><span>/*</span><span> clone </span><span>*/</span>
    <span>if</span> (Z_LVAL(argument_list->u.constant) != <span>0</span><span>) {
        zend_error(E_WARNING, </span><span>"</span><span>Clone method does not require arguments</span><span>"</span><span>);
    }
    opline </span>= &CG(active_op_array)->opcodes[Z_LVAL(function_name-><span>u.constant)];
} </span><span>else</span><span> {
    opline </span>=<span> get_next_op(CG(active_op_array) TSRMLS_CC);
    
    </span><span>//</span><span> 函数,名称确定,非dynamic_fcall,函数则生成ZEND_DO_FCALL指令</span>
    <span>if</span> (!is_method && !is_dynamic_fcall && function_name->op_type==<span>IS_CONST) {
        opline</span>->opcode =<span> ZEND_DO_FCALL;
        opline</span>->op1 = *<span>function_name;
        ZVAL_LONG(</span>&opline->op2.u.constant, zend_hash_func(Z_STRVAL(function_name->u.constant), Z_STRLEN(function_name->u.constant) + <span>1</span><span>));
    }
    </span><span>//</span><span> 否则生成ZEND_DO_FCALL_BY_NAME指令</span>
    <span>else</span><span> {
        opline</span>->opcode =<span> ZEND_DO_FCALL_BY_NAME;
        SET_UNUSED(opline</span>-><span>op1);
    }
}

</span><span>//</span><span> 生成临时变量索引,函数的调用,返回的znode必然是IS_VAR</span>
opline->result.u.<span>var</span> =<span> get_temporary_variable(CG(active_op_array));
opline</span>->result.op_type =<span> IS_VAR;
</span>*result = opline-><span>result;
SET_UNUSED(opline</span>-><span>op2);

</span><span>//</span><span> 从CG(function_call_stack)弹出当前被调用的函数</span>
zend_stack_del_top(&<span>CG(function_call_stack));

</span><span>//</span><span> 传参个数</span>
opline->extended_value = Z_LVAL(argument_list->u.constant);

其中有一段if逻辑分支已经走不到了,可以忽略。

具体考据:这段逻辑在462eff3中被添加,主要用于当调用__clone魔术方法时传参进行抛错,但在8e30d96中,已经不允许直接调用__clone方法了,在进入zend_do_end_function_call之前便会终止编译,所以实际上已经再也走不到该分支了。

直接看else部分,else生成了一条zend op指令。如果函数名确定,函数已被定义,并且不属于动态调用等,则生成的op指令为ZEND_DO_FCALL,否则生成ZEND_DO_FCALL_BY_NAME。对于ZEND_DO_FCALL指令,其操作数比较明确,为函数名,但是对于ZEND_DO_FCALL_BY_NAME来说,由于被调的函数尚未明确,所以将操作数置为UNUSED。

5、总结

用一张图总结一下函数调用大致的编译流程:

红色的方框为生成的op指令。特别是编译传参的地方,情况比较多,可能会产出4种SEND指令。

 

www.bkjia.comtruehttp://www.bkjia.com/PHPjc/1133568.htmlTechArticle深入剖析php执行原理(4):函数的调用,深入剖析php 本章开始研究php中函数的调用和执行,先来看函数调用语句是如何被编译的。 我们前...
Stellungnahme:
Der Inhalt dieses Artikels wird freiwillig von Internetnutzern beigesteuert und das Urheberrecht liegt beim ursprünglichen Autor. Diese Website übernimmt keine entsprechende rechtliche Verantwortung. Wenn Sie Inhalte finden, bei denen der Verdacht eines Plagiats oder einer Rechtsverletzung besteht, wenden Sie sich bitte an admin@php.cn