<p><strong>编辑:</strong>最后添加了Java版本——尽管它本身就很笨拙、不可读、不可维护。在</p>
<hr/>
<h2>“不再有丑陋的图案了!在</h2>
<p>您需要做的第一件<em>事情是<strong>以一种能够经受住任何人类可读性和可维护性的希望的方式编写正则表达式。</strong>
你需要做的第二件事是分析它,看看它到底在做什么。在</p>
<p>这意味着您至少需要在<code>Pattern.COMMENTS</code>模式(或前缀<code>"(?x)"</code>)下编译它,然后添加空格以提供一些可视空间。就我所能理解的,你实际想要匹配的模式是这样的:</p>
<pre><code>^
(?: [^'"\s~:/@\#|^&\[\]()\\{}] # alt 1: unquoted
[^"\s~:/@\#|^&\[\]()\\{}] *
| " (?: [^"]+ | "" )+ " # alt 2: double quoted
| ' (?: [^']+ | '' )+ ' # alt 3: single quoted
)
</code></pre>
<p>如你所见,我在可能的地方引入了垂直和水平空白,以便引导眼睛和大脑作为一种认知分块。我也删除了你所有多余的反斜杠。这些都是要么是彻头彻尾的错误,要么就是让读者迷惑的模糊剂。在</p>
<p>注意,在应用垂直空白时,我将从一行到下一行的相同部分放在同一列中,这样您就可以立即看到哪些部分是相同的,哪些部分是不同的。在</p>
<p>这样做了,我终于可以看到你在这里是一个锚定在开始的比赛,然后是三个备选方案的选择。因此,我将这三种选择标记为描述性注释,这样就不必猜测了。在</p>
<p>我还注意到,您的第一个备选方案有两个微妙的不同(否定)方括号字符类。第二种方法缺少第一种方法中的单引号排除。这是故意的吗?即使是这样,我发现这对于我的口味来说太多的重复;其中的一些或全部应该在一个变量中,这样就不会冒更新一致性问题的风险。在</p>
<hr/>
<h2>分析</h2>
<p>你必须做的两件事中的第二件也是更重要的一件事就是对这件事进行分析。您需要确切地看到该模式被编译到哪个regex程序中,并且您需要跟踪它在数据上运行时的执行情况。在</p>
<p>Java的<code>Pattern</code>类目前还不能做到这一点,尽管我已经与OraSun的当前代码保管人详细地谈过了,他都很想将这个功能添加到Java中,并认为他知道如何做到这一点。他甚至给我寄了一个原型来做第一部分:编译。所以我希望有一天它能上市。在</p>
<p>同时,让我们转而使用一种工具,其中regex是编程语言本身的一个组成部分,而不是一个笨拙的事后诸葛亮。尽管有几种语言符合这一标准,但模式匹配的艺术还没有达到Perl中的复杂程度。在</p>
<p>下面是一个等效程序。在</p>
^{pr2}$
<p>如果我们运行这个程序,就会得到匹配失败的预期答案。问题是为什么和怎么做。在</p>
<p>我们现在要看两件事:我们想看看模式编译成什么样的正则表达式程序,然后我们要跟踪该正则表达式程序的执行情况。在</p>
<p>这两个都是用</p>
<pre><code>use re "debug";
</code></pre>
<p>pragma,也可以通过<code>-Mre=debug</code>在命令行上指定。这就是我们在这里要做的,以避免对源代码的黑客攻击。在</p>
<h2>正则表达式编译</h2>
<p><code>re</code>调试杂注通常会显示模式的编译及其执行。为了分离这些,我们可以使用Perl的“compile only”开关<code>-c</code>,它不尝试执行它编译的程序。这样我们所看到的就是编译后的模式。是否会产生以下36行输出:</p>
<pre><code>$ perl -c -Mre=debug /tmp/bt
Compiling REx "%n ^ (?: [^'%"\s~:/@\#|^&\[\]()\\{}]%n [^%"\s~:/"...
Final program:
1: BOL (2)
2: BRANCH (26)
3: ANYOF[^\x09\x0a\x0c\x0d "#&-)/:@[-\^{-~][^{unicode}+utf8::IsSpacePerl] (14)
14: STAR (79)
15: ANYOF[^\x09\x0a\x0c\x0d "#&()/:@[-\^{-~][^{unicode}+utf8::IsSpacePerl] (0)
26: BRANCH (FAIL)
27: TRIE-EXACT["'] (79)
<"> (29)
29: CURLYX[0] {1,32767} (49)
31: BRANCH (44)
32: PLUS (48)
33: ANYOF[\x00-!#-\xff][{unicode_all}] (0)
44: BRANCH (FAIL)
45: EXACT <""> (48)
47: TAIL (48)
48: WHILEM[1/2] (0)
49: NOTHING (50)
50: EXACT <"> (79)
<'> (55)
55: CURLYX[0] {1,32767} (75)
57: BRANCH (70)
58: PLUS (74)
59: ANYOF[\x00-&(-\xff][{unicode_all}] (0)
70: BRANCH (FAIL)
71: EXACT <''> (74)
73: TAIL (74)
74: WHILEM[2/2] (0)
75: NOTHING (76)
76: EXACT <'> (79)
78: TAIL (79)
79: END (0)
anchored(BOL) minlen 1
/tmp/bt syntax OK
Freeing REx: "%n ^ (?: [^'%"\s~:/@\#|^&\[\]()\\{}]%n [^%"\s~:/"...
</code></pre>
<p>如您所见,已编译的regex程序是用它自己的某种“regex汇编语言”编写的。(它看起来也很像Java原型所有的细节都不是必须的,但是我要指出的是,节点2的指令是一个分支,如果失败,它将继续执行另一个分支分支26。第二个分支是regex程序唯一的另一部分,它由一个TRIE-EXACT节点组成,因为它知道备选方案有不同的起始文本字符串。发生了什么事
关于这两个分支,我们稍后再讨论。在</p>
<h2>正则表达式执行</h2>
<p>现在是时候看看它运行时会发生什么。您使用的文本字符串会导致相当多的回溯,这意味着在最终失败之前,您将有大量的输出要处理。产量是多少?好吧,这么多:</p>
<pre><code>$ perl -Mre=debug /tmp/bt 2>&1 | wc -l
9987
</code></pre>
<p>我假设10000步就是你所说的“灾难性回溯模式”。让我们看看我们不能把它简化成更容易理解的东西。输入字符串的长度为61个字符。为了更好地了解发生了什么,我们可以将其缩减为<code>'pão</code>,它只有4个字符。(好吧,在NFC中,也就是说,NFD中有5个代码点,但在这里没有任何变化)。结果是167行输出:</p>
<pre><code>$ perl -Mre=debug /tmp/bt 2>&1 | wc -l
167
</code></pre>
<p>事实上,以下是当您的字符串有这么多个字符时所得到的regex(compilation plus)执行评测行:</p>
<pre><code>chars lines string
1 63 ‹'›
2 78 ‹'p›
3 109 ‹'pã›
4 167 ‹'pão›
5 290 ‹'pão ›
6 389 ‹'pão d›
7 487 ‹'pão de›
8 546 ‹'pão de ›
9 615 ‹'pão de a›
10 722 ‹'pão de aç›
....
61 9987 ‹'pão de açúcar itaucard mastercard platinum SUSTENTABILIDAD])›
</code></pre>
<p>让我们看看当字符串是四个字符<code>'pão</code>时的调试输出。这次我省略了编译部分,只显示执行部分:</p>
<pre><code>$ perl -Mre=debug /tmp/bt
Matching REx "%n ^ (?: [^'%"\s~:/@\#|^&\[\]()\\{}]%n [^%"\s~:/"... against "'p%x{e3}o"
UTF-8 string...
0 <> <'p%x{e3}o> | 1:BOL(2)
0 <> <'p%x{e3}o> | 2:BRANCH(26)
0 <> <'p%x{e3}o> | 3: ANYOF[^\x09\x0a\x0c\x0d "#&-)/:@[-\^{-~][^{unicode}+utf8::IsSpacePerl](14)
failed...
0 <> <'p%x{e3}o> | 26:BRANCH(78)
0 <> <'p%x{e3}o> | 27: TRIE-EXACT["'](79)
0 <> <'p%x{e3}o> | State: 1 Accepted: N Charid: 2 CP: 27 After State: 3
1 <'> <p%x{e3}o> | State: 3 Accepted: Y Charid: 0 CP: 0 After State: 0
got 1 possible matches
TRIE matched word #2, continuing
only one match left, short-circuiting: #2 <'>
1 <'> <p%x{e3}o> | 55: CURLYX[0] {1,32767}(75)
1 <'> <p%x{e3}o> | 74: WHILEM[2/2](0)
whilem: matched 0 out of 1..32767
1 <'> <p%x{e3}o> | 57: BRANCH(70) 1 <'> <p%x{e3}o> | 58: PLUS(74)
ANYOF[\x00-&(-\xff][{unicode_all}] can match 3 times out of 2147483647...
5 <'p%x{e3}o> <> | 74: WHILEM[2/2](0)
whilem: matched 1 out of 1..32767
5 <'p%x{e3}o> <> | 57: BRANCH(70)
5 <'p%x{e3}o> <> | 58: PLUS(74)
ANYOF[\x00-&(-\xff][{unicode_all}] can match 0 times out of 2147483647...
failed...
5 <'p%x{e3}o> <> | 70: BRANCH(73)
5 <'p%x{e3}o> <> | 71: EXACT <''>(74)
failed...
BRANCH failed...
whilem: failed, trying continuation...
5 <'p%x{e3}o> <> | 75: NOTHING(76)
5 <'p%x{e3}o> <> | 76: EXACT <'>(79)
failed...
failed...
4 <'p%x{e3}> <o> | 74: WHILEM[2/2](0)
whilem: matched 1 out of 1..32767
4 <'p%x{e3}> <o> | 57: BRANCH(70)
4 <'p%x{e3}> <o> | 58: PLUS(74)
ANYOF[\x00-&(-\xff][{unicode_all}] can match 1 times out of 2147483647...
5 <'p%x{e3}o> <> | 74: WHILEM[2/2](0)
whilem: matched 2 out of 1..32767
5 <'p%x{e3}o> <> | 57: BRANCH(70)
5 <'p%x{e3}o> <> | 58: PLUS(74)
ANYOF[\x00-&(-\xff][{unicode_all}] can match 0 times out of 2147483647...
failed...
5 <'p%x{e3}o> <> | 70: BRANCH(73)
5 <'p%x{e3}o> <> | 71: EXACT <''>(74)
failed...
BRANCH failed...
whilem: failed, trying continuation...
5 <'p%x{e3}o> <> | 75: NOTHING(76)
5 <'p%x{e3}o> <> | 76: EXACT <'>(79)
failed...
failed...
failed...
4 <'p%x{e3}> <o> | 70: BRANCH(73)
4 <'p%x{e3}> <o> | 71: EXACT <''>(74)
failed...
BRANCH failed...
whilem: failed, trying continuation...
4 <'p%x{e3}> <o> | 75: NOTHING(76)
4 <'p%x{e3}> <o> | 76: EXACT <'>(79)
failed...
failed...
2 <'p> <%x{e3}o> | 74: WHILEM[2/2](0)
whilem: matched 1 out of 1..32767
2 <'p> <%x{e3}o> | 57: BRANCH(70)
2 <'p> <%x{e3}o> | 58: PLUS(74)
ANYOF[\x00-&(-\xff][{unicode_all}] can match 2 times out of 2147483647...
5 <'p%x{e3}o> <> | 74: WHILEM[2/2](0)
whilem: matched 2 out of 1..32767
5 <'p%x{e3}o> <> | 57: BRANCH(70)
5 <'p%x{e3}o> <> | 58: PLUS(74)
ANYOF[\x00-&(-\xff][{unicode_all}] can match 0 times out of 2147483647...
failed...
5 <'p%x{e3}o> <> | 70: BRANCH(73)
5 <'p%x{e3}o> <> | 71: EXACT <''>(74)
failed...
BRANCH failed...
whilem: failed, trying continuation...
5 <'p%x{e3}o> <> | 75: NOTHING(76)
5 <'p%x{e3}o> <> | 76: EXACT <'>(79)
failed...
failed...
4 <'p%x{e3}> <o> | 74: WHILEM[2/2](0)
whilem: matched 2 out of 1..32767
4 <'p%x{e3}> <o> | 57: BRANCH(70)
4 <'p%x{e3}> <o> | 58: PLUS(74)
ANYOF[\x00-&(-\xff][{unicode_all}] can match 1 times out of 2147483647...
5 <'p%x{e3}o> <> | 74: WHILEM[2/2](0)
whilem: matched 3 out of 1..32767
5 <'p%x{e3}o> <> | 57: BRANCH(70)
5 <'p%x{e3}o> <> | 58: PLUS(74)
ANYOF[\x00-&(-\xff][{unicode_all}] can match 0 times out of 2147483647.
..
failed...
5 <'p%x{e3}o> <> | 70: BRANCH(73)
5 <'p%x{e3}o> <> | 71: EXACT <''>(74)
failed...
BRANCH failed...
whilem: failed, trying continuation...
5 <'p%x{e3}o> <> | 75: NOTHING(76)
5 <'p%x{e3}o> <> | 76: EXACT <'>(79)
failed...
failed...
failed...
4 <'p%x{e3}> <o> | 70: BRANCH(73)
4 <'p%x{e3}> <o> | 71: EXACT <''>(74)
failed...
BRANCH failed...
whilem: failed, trying continuation...
4 <'p%x{e3}> <o> | 75: NOTHING(76)
4 <'p%x{e3}> <o> | 76: EXACT <'>(79)
failed...
failed...
failed...
2 <'p> <%x{e3}o> | 70: BRANCH(73)
2 <'p> <%x{e3}o> | 71: EXACT <''>(74)
failed...
BRANCH failed...
whilem: failed, trying continuation...
2 <'p> <%x{e3}o> | 75: NOTHING(76)
2 <'p> <%x{e3}o> | 76: EXACT <'>(79)
failed...
failed...
failed...
1 <'> <p%x{e3}o> | 70: BRANCH(73)
1 <'> <p%x{e3}o> | 71: EXACT <''>(74)
failed...
BRANCH failed...
failed...
failed...
BRANCH failed...
Match failed
Match failed.
Freeing REx: "%n ^ (?: [^'%"\s~:/@\#|^&\[\]()\\{}]%n [^%"\s~:/"...
</code></pre>
<p>在最后一个引用后,你可以看到最后一个引用的分支。就是这个:</p>
<pre><code> | ' (?: [^']+ | '' )+ ' # alt 3: single quoted
</code></pre>
<p>节点55是trie分支:</p>
<pre><code> <'> (55)
55: CURLYX[0] {1,32767} (75)
57: BRANCH (70)
58: PLUS (74)
59: ANYOF[\x00-&(-\xff][{unicode_all}] (0)
70: BRANCH (FAIL)
71: EXACT <''> (74)
</code></pre>
<p>下面是执行跟踪,显示灾难性退避发生的位置:</p>
<pre><code> 1 <'> <p%x{e3}o> | 74: WHILEM[2/2](0)
whilem: matched 0 out of 1..32767
1 <'> <p%x{e3}o> | 57: BRANCH(70)
1 <'> <p%x{e3}o> | 58: PLUS(74)
ANYOF[\x00-&(-\xff][{unicode_all}] can match 3 times out of 2147483647...
5 <'p%x{e3}o> <> | 74: WHILEM[2/2](0)
whilem: matched 1 out of 1..32767
5 <'p%x{e3}o> <> | 57: BRANCH(70)
5 <'p%x{e3}o> <> | 58: PLUS(74)
ANYOF[\x00-&(-\xff][{unicode_all}] can match 0 times out of 2147483647...
failed...
5 <'p%x{e3}o> <> | 70: BRANCH(73)
5 <'p%x{e3}o> <> | 71: EXACT <''>(74)
failed...
BRANCH failed...
whilem: failed, trying continuation...
5 <'p%x{e3}o> <> | 75: NOTHING(76)
5 <'p%x{e3}o> <> | 76: EXACT <'>(79)
failed...
failed...
4 <'p%x{e3}> <o> | 74: WHILEM[2/2](0)
whilem: matched 1 out of 1..32767
4 <'p%x{e3}> <o> | 57: BRANCH(70)
4 <'p%x{e3}> <o> | 58: PLUS(74)
ANYOF[\x00-&(-\xff][{unicode_all}] can match 1 times out of 2147483647...
5 <'p%x{e3}o> <> | 74: WHILEM[2/2](0)
whilem: matched 2 out of 1..32767
</code></pre>
<p>节点58吞噬了字符串<code>pão</code>中所有剩余的3个字符。这导致单引号的终止精确匹配失败。因此,它尝试您的替代方法,即一对单引号,但也失败了。在</p>
<p>在这一点上,我不得不质疑你的模式。不应该</p>
<pre><code>' (?: [^']+ | '' )+ '
</code></pre>
<p>真的很简单吗?在</p>
<pre><code>' [^']* '
</code></pre>
<p>所以现在的情况是,有很多方法可以让it回溯到逻辑上永远不会发生的事情。你有一个嵌套的量词,这导致了各种无望和无意识的忙碌。在</p>
<p>如果我们将模式简化为:</p>
<pre><code>^ (?: [^'"\s~:/@\#|^&\[\]()\\{}] +
| " [^"]* "
| ' [^']* '
)
</code></pre>
<p>现在,不管输入字符串的大小,它都提供相同数量的跟踪输出行:只有40行,这包括编译。在完整字符串上见证编译和执行:</p>
<pre><code>Compiling REx "%n ^ (?: [^'%"\s~:/@\#|^&\[\]()\\{}]%n [^%"\s~:/"...
Final program:
1: BOL (2)
2: BRANCH (26)
3: ANYOF[^\x09\x0a\x0c\x0d "#&-)/:@[-\^{-~][^{unicode}+utf8::IsSpacePerl] (14)
14: STAR (61)
15: ANYOF[^\x09\x0a\x0c\x0d "#&()/:@[-\^{-~][^{unicode}+utf8::IsSpacePerl] (0)
26: BRANCH (FAIL)
27: TRIE-EXACT["'] (61)
<"> (29)
29: STAR (41)
30: ANYOF[\x00-!#-\xff][{unicode_all}] (0)
41: EXACT <"> (61)
<'> (46)
46: STAR (58)
47: ANYOF[\x00-&(-\xff][{unicode_all}] (0)
58: EXACT <'> (61)
60: TAIL (61)
61: END (0)
anchored(BOL) minlen 1
Matching REx "%n ^ (?: [^'%"\s~:/@\#|^&\[\]()\\{}]%n [^%"\s~:/"... against "'p%x{e3}o de a%x{e7}%x{fa}car itaucard mast
ercard platinum S"...
UTF-8 string...
0 <> <'p%x{e3}o > | 1:BOL(2)
0 <> <'p%x{e3}o > | 2:BRANCH(26)
0 <> <'p%x{e3}o > | 3: ANYOF[^\x09\x0a\x0c\x0d "#&-)/:@[-\^{-~][^{unicode}+utf8::IsSpacePerl](14)
failed...
0 <> <'p%x{e3}o > | 26:BRANCH(60)
0 <> <'p%x{e3}o > | 27: TRIE-EXACT["'](61)
0 <> <'p%x{e3}o > | State: 1 Accepted: N Charid: 2 CP: 27 After State: 3
1 <'> <p%x{e3}o d> | State: 3 Accepted: Y Charid: 0 CP: 0 After State: 0
got 1 possible matches
TRIE matched word #2, continuing
only one match left, short-circuiting: #2 <'>
1 <'> <p%x{e3}o d> | 46: STAR(58)
ANYOF[\x00-&(-\xff][{unicode_all}] can match 60 times out of 2147483647...
failed...
BRANCH failed...
Match failed
Match failed.
Freeing REx: "%n ^ (?: [^'%"\s~:/@\#|^&\[\]()\\{}]%n [^%"\s~:/"...
</code></pre>
<p>我知道你在想占有匹配可能是这里的答案,但我认为真正的问题是原始模式中错误的逻辑。你看现在的情况有多正常?在</p>
<p>如果我们用你的所有格在旧模式上运行它,即使我认为这没有意义,我们仍然可以得到恒定的运行时间,但它需要更多的步骤。用这个图案</p>
<pre><code> ^ (?: [^'"\s~:/@\#|^&\[\]()\\{}] + # alt 1: unquoted
| " (?: [^"]++ | "" )++ " # alt 2: double quoted
| ' (?: [^']++ | '' )++ ' # alt 3: single quoted
)
</code></pre>
<p>编译和执行配置文件如下:</p>
<pre><code>Compiling REx "%n ^ (?: [^'%"\s~:/@\#|^&\[\]()\\{}]%n [^%"\s~:/"...
Final program:
1: BOL (2)
2: BRANCH (26)
3: ANYOF[^\x09\x0a\x0c\x0d "#&-)/:@[-\^{-~][^{unicode}+utf8::IsSpacePerl] (14)
14: STAR (95)
15: ANYOF[^\x09\x0a\x0c\x0d "#&()/:@[-\^{-~][^{unicode}+utf8::IsSpacePerl] (0)
26: BRANCH (FAIL)
27: TRIE-EXACT["'] (95)
<"> (29)
29: SUSPEND (58)
31: CURLYX[0] {1,32767} (55)
33: BRANCH (50)
34: SUSPEND (54)
36: PLUS (48)
37: ANYOF[\x00-!#-\xff][{unicode_all}] (0)
48: SUCCEED (0)
49: TAIL (53)
50: BRANCH (FAIL)
51: EXACT <""> (54)
53: TAIL (54)
54: WHILEM[1/2] (0)
55: NOTHING (56)
56: SUCCEED (0)
57: TAIL (58)
58: EXACT <"> (95)
<'> (63)
63: SUSPEND (92)
65: CURLYX[0] {1,32767} (89)
67: BRANCH (84)
68: SUSPEND (88)
70: PLUS (82)
71: ANYOF[\x00-&(-\xff][{unicode_all}] (0)
82: SUCCEED (0)
83: TAIL (87)
84: BRANCH (FAIL)
85: EXACT <''> (88)
87: TAIL (88)
88: WHILEM[2/2] (0)
89: NOTHING (90)
90: SUCCEED (0)
91: TAIL (92)
92: EXACT <'> (95)
94: TAIL (95)
95: END (0)
anchored(BOL) minlen 1
Matching REx "%n ^ (?: [^'%"\s~:/@\#|^&\[\]()\\{}]%n [^%"\s~:/"... against "'p%x{e3}o de a%x{e7}%x{fa}car itaucard mastercard platinum S"...
UTF-8 string...
0 <> <'p%x{e3}o > | 1:BOL(2)
0 <> <'p%x{e3}o > | 2:BRANCH(26)
0 <> <'p%x{e3}o > | 3: ANYOF[^\x09\x0a\x0c\x0d "#&-)/:@[-\^{-~][^{unicode}+utf8::IsSpacePerl](14)
failed...
0 <> <'p%x{e3}o > | 26:BRANCH(94)
0 <> <'p%x{e3}o > | 27: TRIE-EXACT["'](95)
0 <> <'p%x{e3}o > | State: 1 Accepted: N Charid: 2 CP: 27 After State: 3
1 <'> <p%x{e3}o d>| State: 3 Accepted: Y Charid: 0 CP: 0 After State: 0
got 1 possible matches
TRIE matched word #2, continuing
only one match left, short-circuiting: #2 <'>
1 <'> <p%x{e3}o d>| 63: SUSPEND(92)
1 <'> <p%x{e3}o d>| 65: CURLYX[0] {1,32767}(89)
1 <'> <p%x{e3}o d>| 88: WHILEM[2/2](0)
whilem: matched 0 out of 1..32767
1 <'> <p%x{e3}o d>| 67: BRANCH(84)
1 <'> <p%x{e3}o d>| 68: SUSPEND(88)
1 <'> <p%x{e3}o d>| 70: PLUS(82)
ANYOF[\x00-&(-\xff][{unicode_all}] can match 60 times out of 2147483647...
64 <NTABILIDAD])> <| 82: SUCCEED(0)
subpattern success...
64 <NTABILIDAD])> <| 88: WHILEM[2/2](0)
whilem: matched 1 out of 1..32767
64 <NTABILIDAD])> <| 67: BRANCH(84)
64 <NTABILIDAD])> <| 68: SUSPEND(88)
64 <NTABILIDAD])> <| 70: PLUS(82)
ANYOF[\x00-&(-\xff][{unicode_all}] can match 0 times out of 2147483647...
failed...
failed...
64 <NTABILIDAD])> <| 84: BRANCH(87)
64 <NTABILIDAD])> <| 85: EXACT <''>(88)
failed...
BRANCH failed...
whilem: failed, trying continuation...
64 <NTABILIDAD])> <| 89: NOTHING(90)
64 <NTABILIDAD])> <| 90: SUCCEED(0)
subpattern success...
64 <NTABILIDAD])> <| 92: EXACT <'>(95)
failed...
BRANCH failed...
Match failed
Match failed.
Freeing REx: "%n ^ (?: [^'%"\s~:/@\#|^&\[\]()\\{}]%n [^%"\s~:/"...
</code></pre>
<p>我还是更喜欢我的解决方案。它比较短。在</p>
<hr/>
<h2>编辑</h2>
<p>似乎Java版本比Perl版本的相同模式要多出100倍的步骤,我不知道为什么——除了Perl regex编译器在优化方面比Java regex编译器聪明100倍,Java regex编译器从来没有做过任何事情,而且应该这样做。在</p>
<p>这是等效的Java程序。我已经拆下了主锚,这样我们就可以正常循环了。在</p>
<pre><code>$ cat java.crap
import java.util.regex.*;
public class crap {
public static void
main(String[ ] argv) {
String input = "'pão de açúcar itaucard mastercard platinum SUSTENTABILIDAD])";
String regex = "\n"
+ "(?: [^'\"\\s~:/@\\#|^&\\[\\]()\\\\{}] # alt 1: unquoted \n"
+ " [^\"\\s~:/@\\#|^&\\[\\]()\\\\{}] * \n"
+ " | \" (?: [^\"]++ | \"\" )++ \" # alt 2: double quoted \n"
+ " | ' (?: [^']++ | '' )++ ' # alt 3: single quoted \n"
+ ") \n"
;
System.out.printf("Matching ‹%s› =~ qr{%s}x\n\n", input, regex);
Pattern regcomp = Pattern.compile(regex, Pattern.COMMENTS);
Matcher regexec = regcomp.matcher(input);
int count;
for (count = 0; regexec.find(); count++) {
System.out.printf("Found match: ‹%s›\n", regexec.group());
}
if (count == 0) {
System.out.printf("Match failed.\n");
}
}
}
</code></pre>
<p>运行时,会产生以下结果:</p>
<pre><code>$ javac -encoding UTF-8 crap.java && java -Dfile.encoding=UTF-8 crap
Matching ‹'pão de açúcar itaucard mastercard platinum SUSTENTABILIDAD])› =~ qr{
(?: [^'"\s~:/@\#|^&\[\]()\\{}] # alt 1: unquoted
[^"\s~:/@\#|^&\[\]()\\{}] *
| " (?: [^"]++ | "" )++ " # alt 2: double quoted
| ' (?: [^']++ | '' )++ ' # alt 3: single quoted
)
}x
Found match: ‹pão›
Found match: ‹de›
Found match: ‹açúcar›
Found match: ‹itaucard›
Found match: ‹mastercard›
Found match: ‹platinum›
Found match: ‹SUSTENTABILIDAD›
</code></pre>
<p>如您所见,模式匹配在爪哇有很多话要说,绝对没有一个能通过便溺警察。这简直是皇室之痛</p>