php if 的实现

Posted 2020-08-20 taek

 

 简单分析下php中的分支背后的实现

<?php
if($a == 1){
   echo "a";  
}else{
  echo "b";
}

 

1.语法分析

unticked_statement:
        ‘{‘ inner_statement_list ‘}‘
    |    T_IF parenthesis_expr { zend_do_if_cond(&$2, &$1 TSRMLS_CC); } statement { zend_do_if_after_statement(&$1, 1 TSRMLS_CC); }
 elseif_list else_single { zend_do_if_end(TSRMLS_C); }
 
parenthesis_expr:
        ‘(‘ expr ‘)‘        { $$ = $2; }
    |    ‘(‘ yield_expr ‘)‘    { $$ = $2; }
;
elseif_list:
        /* empty */
    |    elseif_list T_ELSEIF parenthesis_expr { zend_do_if_cond(&$3, &$2 TSRMLS_CC); } statement { zend_do_if_after_statement(&$2, 0 TSRMLS_CC); }
;

%token T_ELSEIF    "elseif (T_ELSEIF)"
else_single:
        /* empty */
    |    T_ELSE statement
;
%token T_ELSE      "else (T_ELSE)"

对于上面的php代码来说

if 匹配 T_IF

$a == 1 匹配 parenthesis_expr , 同时语法分析器要执行 zend_do_if_cond

echo "a"  匹配 statement, 同时语法分析器要执行 zend_do_if_after_statement

else 匹配 T_ELSE, 同时语法分析器再执行zend_do_if_end

 

　　由于是先分析 $a==1， 那么它对应的opcode的opline_num 假设为1，发现if ($a == 1) 匹配BNF后，执行下面的函数，得到新的opline_num （这里为2）
,同时得到新的opline, 设置opcode为ZEND_JMPZ, 将op1设置为$a==1对应的opcode, 将op2设置为unused, 为什么呢？后来看代码才知道，
当当前分支不成立时，是要跳转的，在执行下面的函数时，还不知道要跳转的opline_num, 只有当分析完statement后，才知道这个跳转opline_num

void zend_do_if_cond(const znode *cond, znode *closing_bracket_token TSRMLS_DC) /* {{{ */
{
    int if_cond_op_number = get_next_op_number(CG(active_op_array));
    zend_op *opline = get_next_op(CG(active_op_array) TSRMLS_CC);

    opline->opcode = ZEND_JMPZ;
    SET_NODE(opline->op1, cond);
    closing_bracket_token->u.op.opline_num = if_cond_op_number;  
    SET_UNUSED(opline->op2);
    INC_BPC(CG(active_op_array));
}

  　　

　　echo "a"; 这个匹配statement 就不说了，然后执行zend_do_if_after_statement，这里也要获得一个新的opline以及opline_num（这里为3）,

设置opcode为ZEND_JMP, 意思为无条件跳转，同时设置opcodes的数组里第2个元素的属性op2.opline_num设置为5，为什么不是4，是因为本身的无条件跳转ZEND_JMP也算一个opcode

   

void zend_do_if_after_statement(const znode *closing_bracket_token, unsigned char initialize TSRMLS_DC) /* {{{ */
{
    int if_end_op_number = get_next_op_number(CG(active_op_array));
    zend_op *opline = get_next_op(CG(active_op_array) TSRMLS_CC);
    zend_llist *jmp_list_ptr;

    opline->opcode = ZEND_JMP;
    /* save for backpatching */
    if (initialize) {
        zend_llist jmp_list;

        zend_llist_init(&jmp_list, sizeof(int), NULL, 0);
        zend_stack_push(&CG(bp_stack), (void *) &jmp_list, sizeof(zend_llist));
    }
    zend_stack_top(&CG(bp_stack), (void **) &jmp_list_ptr);
    zend_llist_add_element(jmp_list_ptr, &if_end_op_number);

    CG(active_op_array)->opcodes[closing_bracket_token->u.op.opline_num].op2.opline_num = if_end_op_number+1;
    SET_UNUSED(opline->op1);
    SET_UNUSED(opline->op2);
}

   

　　else { echo "b"; }  匹配 T_ELSE statement后，马上执行 zend_do_if_end，得到新的opline_num，注意：现在的opcodes数组里有4个opcode了

第一个是$a==1对应的opcode, opline=1

第二个是ZEND_JMPZ 当  $a!=1时的opcode, opline=5

第三个是 echo "a";对应的opcode, opline =3

第四个是ZEND_JMP对应的opcode, opline=6

第五个是echo "b";对应的opcode  opline=5

 

所以ZEND_JMP 对应的是 else {echo "b";}之后的opcode了

void zend_do_if_end(TSRMLS_D) /* {{{ */ 
{
    int next_op_number = get_next_op_number(CG(active_op_array));
    zend_llist *jmp_list_ptr;
    zend_llist_element *le;

    zend_stack_top(&CG(bp_stack), (void **) &jmp_list_ptr);
    for (le=jmp_list_ptr->head; le; le = le->next) {
        CG(active_op_array)->opcodes[*((int *) le->data)].op1.opline_num = next_op_number;
    }
    zend_llist_destroy(jmp_list_ptr);
    zend_stack_del_top(&CG(bp_stack));
    DEC_BPC(CG(active_op_array));
}

 

 

 

 

pass_two函数，处理op_array中的各个opline， 上面unused掉的op2，在这里又重新赋值，就是当分支不成立时，要跳转的opcode所在的行号

ZEND_API int pass_two(zend_op_array *op_array TSRMLS_DC)
{
    zend_op *opline, *end;

    if (op_array->type!=ZEND_USER_FUNCTION && op_array->type!=ZEND_EVAL_CODE) {
        return 0;
    }
        。。。。

    opline = op_array->opcodes;
    end = opline + op_array->last;
    while (opline < end) {
        if (opline->op1_type == IS_CONST) {
            opline->op1.zv = &op_array->literals[opline->op1.constant].constant;
        }
        if (opline->op2_type == IS_CONST) {
            opline->op2.zv = &op_array->literals[opline->op2.constant].constant;
        }
        switch (opline->opcode) {
            case ZEND_GOTO:
                if (Z_TYPE_P(opline->op2.zv) != IS_LONG) {
                    zend_resolve_goto_label(op_array, opline, 1 TSRMLS_CC);
                }
                /* break omitted intentionally */
            　　　  case ZEND_JMP:
            case ZEND_FAST_CALL:
                opline->op1.jmp_addr = &op_array->opcodes[opline->op1.opline_num];
                break;
            case ZEND_JMPZ:
            case ZEND_JMPNZ:
            case ZEND_JMPZ_EX:
            case ZEND_JMPNZ_EX:
            case ZEND_JMP_SET:
            case ZEND_JMP_SET_VAR:
                opline->op2.jmp_addr = &op_array->opcodes[opline->op2.opline_num];
                break;
            case ZEND_RETURN:
            case ZEND_RETURN_BY_REF:
                if (op_array->fn_flags & ZEND_ACC_GENERATOR) {
                    if (opline->op1_type != IS_CONST || Z_TYPE_P(opline->op1.zv) != IS_NULL) {
                        CG(zend_lineno) = opline->lineno;
                        zend_error(E_COMPILE_ERROR, "Generators cannot return values using \"return\"");
                    }

                    opline->opcode = ZEND_GENERATOR_RETURN;
                }
                break;
        }
        ZEND_VM_SET_OPCODE_HANDLER(opline);
        opline++;
    }

    op_array->fn_flags |= ZEND_ACC_DONE_PASS_TWO;
    return 0;
}

 

 

static int ZEND_FASTCALL  ZEND_JMPZ_SPEC_CONST_HANDLER(ZEND_OPCODE_HANDLER_ARGS)
{
    USE_OPLINE

    zval *val;
    int ret;

    SAVE_OPLINE();
    //opline->op1为 if后面的表达式的值
    val = opline->op1.zv;

    if (IS_CONST == IS_TMP_VAR && EXPECTED(Z_TYPE_P(val) == IS_BOOL)) {
        


ret = Z_LVAL_P(val);
    } else {
        //判断val是否为1 true 0 false,如果为1,执行下一条opcode,如果为0,进行opcode的跳转
        ret = i_zend_is_true(val);

        if (UNEXPECTED(EG(exception) != NULL)) {
            HANDLE_EXCEPTION();
        }
    }
    if (!ret) {
#if DEBUG_ZEND>=2
        printf("Conditional jmp to %d\n", opline->op2.opline_num);
#endif
        ZEND_VM_SET_OPCODE(opline->op2.jmp_addr);
        ZEND_VM_CONTINUE();
    }

    ZEND_VM_NEXT_OPCODE();
}

 

关于i_zend_is_true的实现

 

static zend_always_inline int i_zend_is_true(zval *op)
{
    int result;

    switch (Z_TYPE_P(op)) {
        case IS_NULL:
            result = 0;
            break;
        case IS_LONG:
        case IS_BOOL:
        case IS_RESOURCE:
            result = (Z_LVAL_P(op)?1:0);
            break;
        case IS_DOUBLE:
            result = (Z_DVAL_P(op) ? 1 : 0);
            break;
        case IS_STRING:
            if (Z_STRLEN_P(op) == 0
                || (Z_STRLEN_P(op)==1 && Z_STRVAL_P(op)[0]==‘0‘)) {
                result = 0;
            } else {
                result = 1;
            }
            break;
        case IS_ARRAY:
            result = (zend_hash_num_elements(Z_ARRVAL_P(op))?1:0);
            break;
        case IS_OBJECT:
            if(IS_ZEND_STD_OBJECT(*op)) {
                TSRMLS_FETCH();

                if (Z_OBJ_HT_P(op)->cast_object) {
                    zval tmp;
                    if (Z_OBJ_HT_P(op)->cast_object(op, &tmp, IS_BOOL TSRMLS_CC) == SUCCESS) {
                        result = Z_LVAL(tmp);
                        break;
                    }
                } else if (Z_OBJ_HT_P(op)->get) {
                    zval *tmp = Z_OBJ_HT_P(op)->get(op TSRMLS_CC);
                    if(Z_TYPE_P(tmp) != IS_OBJECT) {
                        /* for safety - avoid loop */
                        convert_to_boolean(tmp);
                        result = Z_LVAL_P(tmp);
                        zval_ptr_dtor(&tmp);
                        break;
                    }
                }
            }
            result = 1;
            break;
        default:
            result = 0;
            break;
    }
    return result;
}