这对我来说在Python中已经足够好了:

def fix_filename(name, max_length=255):
    """
    Replace invalid characters on Linux/Windows/MacOS with underscores.
    List from https://stackoverflow.com/a/31976060/819417
    Trailing spaces & periods are ignored on Windows.
    >>> fix_filename("  COM1  ")
    '_ COM1 _'
    >>> fix_filename("COM10")
    'COM10'
    >>> fix_filename("COM1,")
    'COM1,'
    >>> fix_filename("COM1.txt")
    '_.txt'
    >>> all('_' == fix_filename(chr(i)) for i in list(range(32)))
    True
    """
    return re.sub(r'[/\\:|<>"?*\0-\x1f]|^(AUX|COM[1-9]|CON|LPT[1-9]|NUL|PRN)(?![^.])|^\s|[\s.]$', "_", name[:max_length], flags=re.IGNORECASE)

还可以查看这个过时的列表，以获得FAT32中的=等其他遗留内容。

禁止文件名字符的“全面指南”在Windows上不起作用，因为它保留了文件名和字符。是的，像这样的角色 *”?还有一些名字是禁止使用的，但是有无数个名字是只由有效字符组成的，是禁止使用的。例如，空格和点是有效的文件名字符，但仅由这些字符组成的名称是禁止的。

Windows不区分大写字母和小写字母，因此如果已经存在名为a的文件夹，则不能创建名为a的文件夹。更糟糕的是，像PRN和CON这样看似允许的名字，以及许多其他的名字，是被保留和不允许的。Windows也有一些长度限制;在一个文件夹中有效的文件名如果移到另一个文件夹中可能会失效。的规则 命名文件和文件夹 都在微软文档里。

一般来说，不能使用用户生成的文本创建Windows目录名。如果您希望允许用户任意命名，则必须创建安全的名称，如A、AB、A2等，将用户生成的名称及其等效路径存储在应用程序数据文件中，并在应用程序中执行路径映射。

如果您绝对必须允许用户生成文件夹名，那么判断它们是否无效的唯一方法是捕获异常并假定名称无效。即使这样也充满了危险，因为为拒绝访问、脱机驱动器和驱动器空间外抛出的异常与为无效名称抛出的异常重叠。你正在打开一个巨大的伤害罐。

好吧，如果只是为了研究目的，那么你最好的选择是看看维基百科上关于文件名的条目。

如果您想编写一个可移植的函数来验证用户输入并基于此创建文件名，简单的回答是不要。看一看像Perl的File::Spec这样的可移植模块，了解一下完成这样一个“简单”任务所需的所有跳转。

对于任何寻找正则表达式的人:

const BLACKLIST = /[<>:"\/\\|?*]/g;

讨论不同的可能方法

在定义，什么是合法的，什么是不合法的方面的困难已经得到解决，并提出了白名单。但不仅是Windows，很多unix操作系统也支持超过8位的字符，比如Unicode。您还可以在这里讨论编码，例如UTF-8。你可以考虑Jonathan Leffler的评论，他给出了现代Linux的信息，并描述了MacOS的细节。维基百科指出，(例如)

修饰语字母冒号[(见7。有时在Windows文件名中使用，因为它与用于文件名的Segoe UI字体中的冒号相同。[继承的ASCII]冒号本身是不允许的。

因此，我想提出一种更自由的方法，使用Unicode Homoglyph字符替换“非法”字符。我发现在我可比的用例中，结果可读性要强得多，而且它只受所使用字体的限制，它非常广泛，Windows默认为3903个字符。此外，您甚至可以从替换恢复原始内容。

可能的选择和研究笔记

为了保持内容的组织性，我将始终给出字符，它的名称和十六进制数表示。后者不区分大小写，前导零可以自由添加或省略，因此，例如U+002A和U+ 2a是等效的。如果可用，我会尽量指出更多的信息或替代品-请随时向我展示更多或更好的。

Instead of * (U+2A * ASTERISK), you can use one of the many listed, for example U+2217 ∗ (ASTERISK OPERATOR) or the Full Width Asterisk U+FF0A ＊. u+20f0 ⃰ combining asterisk above from combining diacritical marks for symbols might also be a valid choice. You can read 4. for more info about the combining characters. Instead of . (U+2E . full stop), one of these could be a good option, for example ⋅ U+22C5 dot operator. Instead of " (U+22 " quotation mark), you can use “ U+201C english leftdoublequotemark, more alternatives see here. I also included some of the good suggestions of Wally Brockway's answer, in this case u+2036 ‶ reversed double prime and u+2033 ″ double prime - I will from now on denote ideas from that source by ¹³. Instead of / (U+2F / SOLIDUS), you can use ∕ DIVISION SLASH U+2215 (others here), ̸ U+0338 COMBINING LONG SOLIDUS OVERLAY, ̷ COMBINING SHORT SOLIDUS OVERLAY U+0337 or u+2044 ⁄ fraction slash¹³. Be aware about spacing for some characters, including the combining or overlay ones, as they have no width and can produce something like -> ̸th̷is which is ̸th̷is. With added spaces you get -> ̸ th ̷ is, which is ̸ th ̷ is. The second one (COMBINING SHORT SOLIDUS OVERLAY) looks bad in the stackoverflow-font. Instead of \ (U+5C Reverse solidus), you can use ⧵ U+29F5 Reverse solidus operator (more) or u+20E5 ⃥ combining reverse solidus overlay¹³. To replace [ (U+5B [ Left square bracket) and ] (U+005D ] Right square bracket), you can use for example U+FF3B［ FULLWIDTH LEFT SQUARE BRACKET and U+FF3D ］FULLWIDTH RIGHT SQUARE BRACKET (from here, more possibilities here). Instead of : (u+3a : colon), you can use U+2236 ∶ RATIO (for mathematical usage) or U+A789 ꞉ MODIFIER LETTER COLON, (see colon (letter), sometimes used in Windows filenames as it is identical to the colon in the Segoe UI font used for filenames. The colon itself is not permitted ... source and more replacements see here). Another alternative is this one: u+1361 ፡ ethiopic wordspace¹³. Instead of ; (u+3b ; semicolon), you can use U+037E ; GREEK QUESTION MARK (see here). For | (u+7c | vertical line), there are some good substitutes such as: U+2223 ∣ DIVIDES, U+0964 । DEVANAGARI DANDA, U+01C0 ǀ LATIN LETTER DENTAL CLICK (the last ones from Wikipedia) or U+2D4F ⵏ Tifinagh Letter Yan. Also the box drawing characters contain various other options. Instead of , (, U+002C COMMA), you can use for example ‚ U+201A SINGLE LOW-9 QUOTATION MARK (see here). For ? (U+003F ? QUESTION MARK), these are good candidates: U+FF1F ？ FULLWIDTH QUESTION MARK or U+FE56 ﹖ SMALL QUESTION MARK (from here and here). There are also two more from the Dingbats Block (search for "question") and the u+203d ‽ interrobang¹³. While my machine seems to accept it unchanged, I still want to include > (u+3e greater-than sign) and < (u+3c less-than sign) for the sake of completeness. The best replacement here is probably also from the quotation block, such as u+203a › single right-pointing angle quotation mark and u+2039 ‹ single left-pointing angle quotation mark respectively. The tifinagh block only contains ⵦ (u+2D66)¹³ to replace <. The last notion is ⋖ less-than with dot u+22D6 and ⋗ greater-than with dot u+22D7.

对于更多的想法，你也可以在这个块中寻找例子。你还想要更多的想法吗?你可以试着画出你想要的角色，看看这里的建议。

你怎么打这些字符

Say you want to type ⵏ (Tifinagh Letter Yan). To get all of its information, you can always search for this character (ⵏ) on a suited platform such as this Unicode Lookup (please add 0x when you search for hex) or that Unicode Table (that only allows to search for the name, in this case "Tifinagh Letter Yan"). You should obtain its Unicode number U+2D4F and the HTML-code &#11599; (note that 2D4F is hexadecimal for 11599). With this knowledge, you have several options to produce these special characters including the use of

code points to unicode converter or again the Unicode Lookup to reversely convert the numerical representation into the unicode character (remember to set the code point base below to decimal or hexadecimal respectively) a one-liner makro in Autohotkey: :?*:altpipe::{U+2D4F} to type ⵏ instead of the string altpipe - this is the way I input those special characters, my Autohotkey script can be shared if there is common interest Alt Characters or alt-codes by pressing and holding alt, followed by the decimal number for the desired character (more info for example here, look at a table here or there). For the example, that would be Alt+11599. Be aware, that many programs do not fully support this windows feature for all of unicode (as of time writing). Microsoft Office is an exception where it usually works, some other OSes provide similar functionality. Typing these chars with Alt-combinations into MS Word is also the way Wally Brockway suggests in his answer¹³ that was already mentionted - if you don't want to transfer all the hexadecimal values to the decimal asc, you can find some of them there¹³. in MS Office, you can also use ALT + X as described in this MS article to produce the chars if you rarely need it, you can of course still just copy-paste the special character of your choice instead of typing it

虽然唯一非法的Unix字符可能是/和NULL，但应该考虑到命令行解释。

例如，虽然在Unix中将文件命名为1>&2或2>&1是合法的，但在命令行中使用这样的文件名可能会被误解。

类似地，也可以将文件命名为$PATH，但当试图从命令行访问它时，shell将把$PATH转换为其变量值。