【性能为王】从PHP源码剖析array_keys和array_unique，arraykeys

【性能为王】从PHP源码剖析array_keys和array_unique，arraykeys

之前在[译]更快的方式实现PHP数组去重这篇文章里讨论了使用array_flip后再调用array_keys函数替换直接调用array_unique函数实现数组去重性能较好。由于原文没有给出源码分析和测试的结果，导致给读者造成迷惑，在此说声抱歉。为了解开读者的疑惑，笔者承诺了会补上源码的分析，于是花了一些时间去研究PHP的源码，现在此补上详细的说明。

 

性能分析 

从运行性能上分析，看看下面的测试代码：

<span>$test</span>=<span>array</span><span>();
</span><span>for</span>(<span>$run</span>=0; <span>$run</span><10000; <span>$run</span>++<span>)
</span><span>$test</span>[]=<span>rand</span>(0,100<span>);

</span><span>$time</span>=<span>microtime</span>(<span>true</span><span>);

</span><span>$out</span> = <span>array_unique</span>(<span>$test</span><span>);

</span><span>$time</span>=<span>microtime</span>(<span>true</span>)-<span>$time</span><span>;
</span><span>echo</span> 'Array Unique: '.<span>$time</span>."\n"<span>;

</span><span>$time</span>=<span>microtime</span>(<span>true</span><span>);

</span><span>$out</span>=<span>array_keys</span>(<span>array_flip</span>(<span>$test</span><span>));

</span><span>$time</span>=<span>microtime</span>(<span>true</span>)-<span>$time</span><span>;
</span><span>echo</span> 'Keys Flip: '.<span>$time</span>."\n"<span>;

</span><span>$time</span>=<span>microtime</span>(<span>true</span><span>);

</span><span>$out</span>=<span>array_flip</span>(<span>array_flip</span>(<span>$test</span><span>));

</span><span>$time</span>=<span>microtime</span>(<span>true</span>)-<span>$time</span><span>;
</span><span>echo</span> 'Flip Flip: '.<span>$time</span>."\n";

 

运行结果如下：

从上图可以看到，使用array_unique函数需要0.069s;使用array_flip后再使用array_keys函数需要0.00152s;使用两次array_flip函数需要0.00146s。

测试结果表明，使用array_flip后再调用array_keys函数比array_unique函数快。那么，具体原因是什么呢？让我们看看在PHP底层，这两个函数是怎么实现的。

 

源码分析

<span> 1</span> <span>/*</span><span> {{{ proto array array_keys(array input [, mixed search_value[, bool strict]])
</span><span> 2</span> <span>   Return just the keys from the input array, optionally only for the specified search_value </span><span>*/</span>
<span> 3</span> <span>PHP_FUNCTION(array_keys)
</span><span> 4</span> <span>{
</span><span> 5</span>     <span>//</span><span>变量定义</span>
<span> 6</span>     zval *input,                <span>/*</span><span> Input array </span><span>*/</span>
<span> 7</span>          *search_value = NULL,    <span>/*</span><span> Value to search for </span><span>*/</span>
<span> 8</span>          **entry,                <span>/*</span><span> An entry in the input array </span><span>*/</span>
<span> 9</span>            res,                    <span>/*</span><span> Result of comparison </span><span>*/</span>
<span>10</span>           *new_val;                <span>/*</span><span> New value </span><span>*/</span>
<span>11</span>     <span>int</span>    add_key;                <span>/*</span><span> Flag to indicate whether a key should be added </span><span>*/</span>
<span>12</span>     <span>char</span>  *string_key;            <span>/*</span><span> String key </span><span>*/</span>
<span>13</span>     <span>uint</span><span>   string_key_len;
</span><span>14</span>     <span>ulong</span>  num_key;                <span>/*</span><span> Numeric key </span><span>*/</span>
<span>15</span>     zend_bool strict = <span>0</span>;        <span>/*</span><span> do strict comparison </span><span>*/</span>
<span>16</span> <span>    HashPosition pos;
</span><span>17</span>     <span>int</span> (*is_equal_func)(zval *, zval *, zval * TSRMLS_DC) =<span> is_equal_function;
</span><span>18</span> 
<span>19</span>     <span>//</span><span>程序解析参数</span>
<span>20</span>     <span>if</span> (zend_parse_parameters(ZEND_NUM_ARGS() TSRMLS_CC, <span>"</span><span>a|zb</span><span>"</span>, &input, &search_value, &strict) ==<span> FAILURE) {
</span><span>21</span>         <span>return</span><span>;
</span><span>22</span> <span>    }
</span><span>23</span> 
<span>24</span>     <span>//</span><span> 如果strict是true，则设置is_equal_func为is_identical_function，即全等比较</span>
<span>25</span>     <span>if</span><span> (strict) {
</span><span>26</span>         is_equal_func =<span> is_identical_function;
</span><span>27</span> <span>    }
</span><span>28</span> 
<span>29</span>     <span>/*</span><span> 根据search_vale初始化返回的数组大小 </span><span>*/</span>
<span>30</span>     <span>if</span> (search_value !=<span> NULL) {
</span><span>31</span> <span>        array_init(return_value);
</span><span>32</span>     } <span>else</span><span> {
</span><span>33</span> <span>        array_init_size(return_value, zend_hash_num_elements(Z_ARRVAL_P(input)));
</span><span>34</span> <span>    }
</span><span>35</span>     add_key = <span>1</span><span>;
</span><span>36</span> 
<span>37</span>     <span>/*</span><span> 遍历输入的数组参数，然后添加键值到返回的数组 </span><span>*/</span>
<span>38</span>     zend_hash_internal_pointer_reset_ex(Z_ARRVAL_P(input), &pos);<span>//</span><span>重置指针
</span><span>39</span>     <span>//</span><span>循环遍历数组</span>
<span>40</span>     <span>while</span> (zend_hash_get_current_data_ex(Z_ARRVAL_P(input), (<span>void</span> **)&entry, &pos) ==<span> SUCCESS) {
</span><span>41</span>         <span>//</span><span> 如果search_value不为空</span>
<span>42</span>         <span>if</span> (search_value !=<span> NULL) {
</span><span>43</span>             <span>//</span><span> 判断search_value与当前的值是否相同，并将比较结果保存到add_key变量</span>
<span>44</span>             is_equal_func(&res, search_value, *<span>entry TSRMLS_CC);
</span><span>45</span>             add_key = zval_is_true(&<span>res);
</span><span>46</span> <span>        }
</span><span>47</span> 
<span>48</span>         <span>if</span><span> (add_key) {
</span><span>49</span>             <span>//</span><span> 创建一个zval结构体</span>
<span>50</span> <span>            MAKE_STD_ZVAL(new_val);
</span><span>51</span> 
<span>52</span>             <span>//</span><span> 根据键值是字符串还是整型数字将值插入到return_value中</span>
<span>53</span>             <span>switch</span> (zend_hash_get_current_key_ex(Z_ARRVAL_P(input), &string_key, &string_key_len, &num_key, <span>1</span>, &<span>pos)) {
</span><span>54</span>                 <span>case</span><span> HASH_KEY_IS_STRING:
</span><span>55</span>                     ZVAL_STRINGL(new_val, string_key, string_key_len - <span>1</span>, <span>0</span><span>);
</span><span>56</span>                     <span>//</span><span> 此函数负责将值插入到return_value中，如果键值已存在，则使用新值更新对应的值，否则直接插入</span>
<span>57</span>                     zend_hash_next_index_insert(Z_ARRVAL_P(return_value), &new_val, <span>sizeof</span>(zval *<span>), NULL);
</span><span>58</span>                     <span>break</span><span>;
</span><span>59</span> 
<span>60</span>                 <span>case</span><span> HASH_KEY_IS_LONG:
</span><span>61</span>                     Z_TYPE_P(new_val) =<span> IS_LONG;
</span><span>62</span>                     Z_LVAL_P(new_val) =<span> num_key;
</span><span>63</span>                     zend_hash_next_index_insert(Z_ARRVAL_P(return_value), &new_val, <span>sizeof</span>(zval *<span>), NULL);
</span><span>64</span>                     <span>break</span><span>;
</span><span>65</span> <span>            }
</span><span>66</span> <span>        }
</span><span>67</span> 
<span>68</span>         <span>//</span><span> 移动到下一个</span>
<span>69</span>         zend_hash_move_forward_ex(Z_ARRVAL_P(input), &<span>pos);
</span><span>70</span> <span>    }
</span><span>71</span> <span>}
</span><span>72</span> <span>/*</span><span> }}} </span><span>*/</span>

以上是array_keys函数底层的源码。为方便理解，笔者添加了一些中文注释。如果需要查看原始代码，可以点击查看。这个函数的功能就是新建一个临时数组，然后将键值对重新复制到新的数组，如果复制过程中有重复的键值出现，那么就用新的值替换。这个函数的主要步骤是地57和63行调用的zend_hash_next_index_insert函数。该函数将元素插入到数组中，如果出现重复的值，则使用新的值更新原键值指向的值，否则直接插入，时间复杂度是O(n)。

 

<span> 1</span> <span>/*</span><span> {{{ proto array array_flip(array input)
</span><span> 2</span> <span>   Return array with key <-> value flipped </span><span>*/</span>
<span> 3</span> <span>PHP_FUNCTION(array_flip)
</span><span> 4</span> <span>{
</span><span> 5</span>     <span>//</span><span> 定义变量</span>
<span> 6</span>     zval *array, **entry, *<span>data;
</span><span> 7</span>     <span>char</span> *<span>string_key;
</span><span> 8</span>     <span>uint</span><span> str_key_len;
</span><span> 9</span>     <span>ulong</span><span> num_key;
</span><span>10</span> <span>    HashPosition pos;
</span><span>11</span> 
<span>12</span>     <span>//</span><span> 解析数组参数</span>
<span>13</span>     <span>if</span> (zend_parse_parameters(ZEND_NUM_ARGS() TSRMLS_CC, <span>"</span><span>a</span><span>"</span>, &array) ==<span> FAILURE) {
</span><span>14</span>         <span>return</span><span>;
</span><span>15</span> <span>    }
</span><span>16</span> 
<span>17</span>     <span>//</span><span> 初始化返回数组</span>
<span>18</span> <span>    array_init_size(return_value, zend_hash_num_elements(Z_ARRVAL_P(array)));
</span><span>19</span> 
<span>20</span>     <span>//</span><span> 重置指针</span>
<span>21</span>     zend_hash_internal_pointer_reset_ex(Z_ARRVAL_P(array), &<span>pos);
</span><span>22</span>     <span>//</span><span> 遍历每个元素，并执行键<->值交换操作</span>
<span>23</span>     <span>while</span> (zend_hash_get_current_data_ex(Z_ARRVAL_P(array), (<span>void</span> **)&entry, &pos) ==<span> SUCCESS) {
</span><span>24</span>         <span>//</span><span> 初始化一个结构体</span>
<span>25</span> <span>        MAKE_STD_ZVAL(data);
</span><span>26</span>         <span>//</span><span> 将原数组的值赋值为新数组的键</span>
<span>27</span>         <span>switch</span> (zend_hash_get_current_key_ex(Z_ARRVAL_P(array), &string_key, &str_key_len, &num_key, <span>1</span>, &<span>pos)) {
</span><span>28</span>             <span>case</span><span> HASH_KEY_IS_STRING:
</span><span>29</span>                 ZVAL_STRINGL(data, string_key, str_key_len - <span>1</span>, <span>0</span><span>);
</span><span>30</span>                 <span>break</span><span>;
</span><span>31</span>             <span>case</span><span> HASH_KEY_IS_LONG:
</span><span>32</span>                 Z_TYPE_P(data) =<span> IS_LONG;
</span><span>33</span>                 Z_LVAL_P(data) =<span> num_key;
</span><span>34</span>                 <span>break</span><span>;
</span><span>35</span> <span>        }
</span><span>36</span> 
<span>37</span>         <span>//</span><span> 将原数组的键赋值为新数组的值，如果有重复的，则使用新值覆盖旧值</span>
<span>38</span>         <span>if</span> (Z_TYPE_PP(entry) ==<span> IS_LONG) {
</span><span>39</span>             zend_hash_index_update(Z_ARRVAL_P(return_value), Z_LVAL_PP(entry), &data, <span>sizeof</span><span>(data), NULL);
</span><span>40</span>         } <span>else</span> <span>if</span> (Z_TYPE_PP(entry) ==<span> IS_STRING) {
</span><span>41</span>             zend_symtable_update(Z_ARRVAL_P(return_value), Z_STRVAL_PP(entry), Z_STRLEN_PP(entry) + <span>1</span>, &data, <span>sizeof</span><span>(data), NULL);
</span><span>42</span>         } <span>else</span><span> {
</span><span>43</span>             zval_ptr_dtor(&data); <span>/*</span><span> will free also zval structure </span><span>*/</span>
<span>44</span>             php_error_docref(NULL TSRMLS_CC, E_WARNING, <span>"</span><span>Can only flip STRING and INTEGER values!</span><span>"</span><span>);
</span><span>45</span> <span>        }
</span><span>46</span> 
<span>47</span>         <span>//</span><span> 下一个</span>
<span>48</span>         zend_hash_move_forward_ex(Z_ARRVAL_P(array), &<span>pos);
</span><span>49</span> <span>    }
</span><span>50</span> <span>}
</span><span>51</span> <span>/*</span><span> }}} </span><span>*/</span>

 

上面就是是array_flip函数的源码。点击链接查看原始代码。这个函数主要的做的事情就是创建一个新的数组，遍历原数组。在26行开始将原数组的值赋值为新数组的键，然后在37行开始将原数组的键赋值为新数组的值，如果有重复的，则使用新值覆盖旧值。整个函数的时间复杂度也是O(n)。因此，使用了array_flip之后再使用array_keys的时间复杂度是O(n)。

 

接下来，我们看看array_unique函数的源码。点击链接查看原始代码。

<span> 1</span> <span>/*</span><span> {{{ proto array array_unique(array input [, int sort_flags])
</span><span> 2</span> <span>   Removes duplicate values from array </span><span>*/</span>
<span> 3</span> <span>PHP_FUNCTION(array_unique)
</span><span> 4</span> <span>{
</span><span> 5</span>     <span>//</span><span> 定义变量</span>
<span> 6</span>     zval *array, *<span>tmp;
</span><span> 7</span>     Bucket *<span>p;
</span><span> 8</span>     <span>struct</span><span> bucketindex {
</span><span> 9</span>         Bucket *<span>b;
</span><span>10</span>         unsigned <span>int</span><span> i;
</span><span>11</span> <span>    };
</span><span>12</span>     <span>struct</span> bucketindex *arTmp, *cmpdata, *<span>lastkept;
</span><span>13</span>     unsigned <span>int</span><span> i;
</span><span>14</span>     <span>long</span> sort_type =<span> PHP_SORT_STRING;
</span><span>15</span> 
<span>16</span>     <span>//</span><span> 解析参数</span>
<span>17</span>     <span>if</span> (zend_parse_parameters(ZEND_NUM_ARGS() TSRMLS_CC, <span>"</span><span>a|l</span><span>"</span>, &array, &sort_type) ==<span> FAILURE) {
</span><span>18</span>         <span>return</span><span>;
</span><span>19</span> <span>    }
</span><span>20</span> 
<span>21</span>     <span>//</span><span> 设置比较函数</span>
<span>22</span> <span>    php_set_compare_func(sort_type TSRMLS_CC);
</span><span>23</span> 
<span>24</span>     <span>//</span><span> 初始化返回数组</span>
<span>25</span> <span>    array_init_size(return_value, zend_hash_num_elements(Z_ARRVAL_P(array)));
</span><span>26</span>     <span>//</span><span> 将值拷贝到新数组</span>
<span>27</span>     zend_hash_copy(Z_ARRVAL_P(return_value), Z_ARRVAL_P(array), (copy_ctor_func_t) zval_add_ref, (<span>void</span> *)&tmp, <span>sizeof</span>(zval*<span>));
</span><span>28</span> 
<span>29</span>     <span>if</span> (Z_ARRVAL_P(array)->nNumOfElements <= <span>1</span>) {    <span>/*</span><span> 什么都不做 </span><span>*/</span>
<span>30</span>         <span>return</span><span>;
</span><span>31</span> <span>    }
</span><span>32</span> 
<span>33</span>     <span>/*</span><span> 根据target_hash buckets的指针创建数组并排序 </span><span>*/</span>
<span>34</span>     arTmp = (<span>struct</span> bucketindex *) pemalloc((Z_ARRVAL_P(array)->nNumOfElements + <span>1</span>) * <span>sizeof</span>(<span>struct</span> bucketindex), Z_ARRVAL_P(array)-><span>persistent);
</span><span>35</span>     <span>if</span> (!<span>arTmp) {
</span><span>36</span> <span>        zval_dtor(return_value);
</span><span>37</span> <span>        RETURN_FALSE;
</span><span>38</span> <span>    }
</span><span>39</span>     <span>for</span> (i = <span>0</span>, p = Z_ARRVAL_P(array)->pListHead; p; i++, p = p-><span>pListNext) {
</span><span>40</span>         arTmp[i].b =<span> p;
</span><span>41</span>         arTmp[i].i =<span> i;
</span><span>42</span> <span>    }
</span><span>43</span>     arTmp[i].b =<span> NULL;
</span><span>44</span>     <span>//</span><span> 排序</span>
<span>45</span>     zend_qsort((<span>void</span> *) arTmp, i, <span>sizeof</span>(<span>struct</span><span> bucketindex), php_array_data_compare TSRMLS_CC);
</span><span>46</span> 
<span>47</span>     <span>/*</span><span> 遍历排序好的数组，然后删除重复的元素 </span><span>*/</span>
<span>48</span>     lastkept =<span> arTmp;
</span><span>49</span>     <span>for</span> (cmpdata = arTmp + <span>1</span>; cmpdata->b; cmpdata++<span>) {
</span><span>50</span>         <span>if</span><span> (php_array_data_compare(lastkept, cmpdata TSRMLS_CC)) {
</span><span>51</span>             lastkept =<span> cmpdata;
</span><span>52</span>         } <span>else</span><span> {
</span><span>53</span>             <span>if</span> (lastkept->i > cmpdata-><span>i) {
</span><span>54</span>                 p = lastkept-><span>b;
</span><span>55</span>                 lastkept =<span> cmpdata;
</span><span>56</span>             } <span>else</span><span> {
</span><span>57</span>                 p = cmpdata-><span>b;
</span><span>58</span> <span>            }
</span><span>59</span>             <span>if</span> (p->nKeyLength == <span>0</span><span>) {
</span><span>60</span>                 zend_hash_index_del(Z_ARRVAL_P(return_value), p-><span>h);
</span><span>61</span>             } <span>else</span><span> {
</span><span>62</span>                 <span>if</span> (Z_ARRVAL_P(return_value) == &<span>EG(symbol_table)) {
</span><span>63</span>                     zend_delete_global_variable(p->arKey, p->nKeyLength - <span>1</span><span> TSRMLS_CC);
</span><span>64</span>                 } <span>else</span><span> {
</span><span>65</span>                     zend_hash_quick_del(Z_ARRVAL_P(return_value), p->arKey, p->nKeyLength, p-><span>h);
</span><span>66</span> <span>                }
</span><span>67</span> <span>            }
</span><span>68</span> <span>        }
</span><span>69</span> <span>    }
</span><span>70</span>     pefree(arTmp, Z_ARRVAL_P(array)-><span>persistent);
</span><span>71</span> <span>}
</span><span>72</span> <span>/*</span><span> }}} </span><span>*/</span>

可以看到，这个函数初始化一个新的数组，然后将值拷贝到新数组，然后在45行调用排序函数对数组进行排序，排序的算法是zend引擎的块树排序算法。接着遍历排序好的数组，删除重复的元素。整个函数开销最大的地方就在调用排序函数上，而快排的时间复杂度是O(n*logn)，因此，该函数的时间复杂度是O(n*logn)。

结论

因为array_unique底层调用了快排算法，加大了函数运行的时间开销，导致整个函数的运行较慢。这就是为什么array_keys比array_unique函数更快的原因。

 

原创文章，文笔有限，才疏学浅，文中若有不正之处，万望告知。

如果本文对你有帮助，请点下推荐，写文章不容易。