通用解决方案

Mathias Bynens建议遵循 UTS18 建议，因此应遵循Unicode意识如下所示:

[^\p{Alphabetic}\p{Mark}\p{Decimal_Number}\p{Connector_Punctuation}\p{Join_Control}]

请注意有关建议的Unicode属性类组合的注释:

这仅是单词边界的近似值(请参见下面的 b ) .这 添加了连接器标点符号以用于编程语言 标识符，因此将"_"添加到和类似的字符.

更多注意事项

\w构造(以及与之对应的\W)在支持Unicode的上下文中进行匹配时，会在正则表达式引擎中匹配相似但略有不同的字符集.

例如，这里是文档)， [^\p{Alpha}\p{gc=Mn}\p{gc=Me}\p{gc=Mc}\p{Digit}\p{gc=Pc}\p{IsJoin_Control}]，其中\p{gc=Mn}\p{gc=Me}\p{gc=Mc}可以写为\p{M}.

在PHP PCRE中，\W与[^\p{L}\p{N}_]匹配.

Rexegg速查表将Python 3 \w定义为" Unicode字母，表意文字，数字或下划线"，即[\p{L}\p{Mn}\p{Nd}_].

您可以将\W大致分解为[^\p{L}\p{N}\p{M}\p{Pc}]:

/[^\p{L}\p{N}\p{M}\p{Pc}]/gu

其中

• [^-是否定字符类的开头，该字符类与除以下以外的单个字符相匹配:
  • \p{L}-任何Unicode字母
  • \p{N}-任意Unicode数字
  • \p{M}-变音符号
  • \p{Pc}-连接器标点符号
• ]-字符类的结尾.

请注意，这是与下划线匹配的\p{Pc}类.

注意，\p{Alphabetic}(\p{Alpha})包括所有与\p{L}匹配的字母，以及由>匹配的字母数字(例如，Ⅻ –罗马数字12的字符，以及与\p{Other_Alphabetic}(\p{OAlpha})匹配的其他一些符号.

其他版本:

• /[^\p{L}0-9_]/gu-仅使用仅识别Unicode字母的\W
• /[^\p{L}\p{N}_]/gu-(PCRE \W样式)仅使用仅识别Unicode字母和数字的\W.

请注意，Java的(?U)\W将匹配PCRE，Python和.NET中的\W匹配项.

In python or PHP a simple regex such as /\W/gu matches any non-word character in any script, in javascript however it matches [^A-Za-z0-9_], what are the correct ranges to match the same characters as python and PHP?

https://regex101.com/r/yhNF8U/1/

Generic solution

Mathias Bynens suggests to follow the UTS18 recommendation and thus a Unicode-aware \W will look like:

[^\p{Alphabetic}\p{Mark}\p{Decimal_Number}\p{Connector_Punctuation}\p{Join_Control}]

Please note the comment for the suggested Unicode property class combination:

This is only an approximation to Word Boundaries (see b below). The Connector Punctuation is added in for programming language identifiers, thus adding "_" and similar characters.

More considerations

The \w construct (and thus its \W counterpart), when matching in a Unicode-aware context, matches similar, but somewhat different set of characters across regex engines.

For example, here is Non-word character: \W .NET definition: [^\p{Ll}\p{Lu}\p{Lt}\p{Lo}\p{Nd}\p{Mn}\p{Pc}\p{Lm}], where \p{Ll}\p{Lu}\p{Lt}\p{Lo} can be contracted to a sheer \p{L} and the pattern is thus equal to [^\p{L}\p{Nd}\p{Mn}\p{Pc}].

In Android (see documentation), [^\p{Alpha}\p{gc=Mn}\p{gc=Me}\p{gc=Mc}\p{Digit}\p{gc=Pc}\p{IsJoin_Control}], where \p{gc=Mn}\p{gc=Me}\p{gc=Mc} can be just written as \p{M}.

In PHP PCRE, \W matches [^\p{L}\p{N}_].

Rexegg cheat sheet defines Python 3 \w as "Unicode letter, ideogram, digit, or underscore", i.e. [\p{L}\p{Mn}\p{Nd}_].

You may roughly decompose \W as [^\p{L}\p{N}\p{M}\p{Pc}]:

/[^\p{L}\p{N}\p{M}\p{Pc}]/gu

where

• [^ - is the start of the negated character class that matches a single char other than:
  • \p{L} - any Unicode letter
  • \p{N} - any Unicode digit
  • \p{M} - a diacritic mark
  • \p{Pc} - a connector punctuation symbol
• ] - end of the character class.

Note it is \p{Pc} class that matches an underscore.

NOTE that \p{Alphabetic} (\p{Alpha}) includes all letters matched by \p{L}, plus letter numbers matched by \p{Nl} (e.g. Ⅻ – a character for the roman number 12), plus some other symbols matched with \p{Other_Alphabetic} (\p{OAlpha}).

Other variations:

• /[^\p{L}0-9_]/gu - to just use \W that is aware of Unicode letters only
• /[^\p{L}\p{N}_]/gu - (PCRE \W style) to just use \W that is aware of Unicode letters and digits only.

Note that Java's (?U)\W will match a mix of what \W matches in PCRE, Python and .NET.