[validation] Which characters make a URL invalid?

Which characters make a URL invalid?

Are these valid URLs?

• example.com/file[/].html
• http://example.com/file[/].html

In general URIs as defined by RFC 3986 (see Section 2: Characters) may contain any of the following 84 characters:

ABCDEFGHIJKLMNOPQRSTUVWXYZabcdefghijklmnopqrstuvwxyz0123456789-._~:/?#[]@!$&'()*+,;=

Note that this list doesn't state where in the URI these characters may occur.

Any other character needs to be encoded with the percent-encoding (%hh). Each part of the URI has further restrictions about what characters need to be represented by an percent-encoded word.

All valid characters that can be used in a URI (a URL is a type of URI) are defined in RFC 3986.

All other characters can be used in a URL provided that they are "URL Encoded" first. This involves changing the invalid character for specific "codes" (usually in the form of the percent symbol (%) followed by a hexadecimal number).

This link, HTML URL Encoding Reference, contains a list of the encodings for invalid characters.

Most of the existing answers here are impractical because they totally ignore the real-world usage of addresses like:

• https://en.wikipedia.org/wiki/Möbius_strip or
• https://zh.wikipedia.org/wiki/Wikipedia:????????/en.

First, a digression into terminology. What are these addresses? Are they valid URLs?

Historically, the answer was "no". According to RFC 3986, from 2005, such addresses are not URIs (and therefore not URLs, since URLs are a type of URIs). Per the terminology of 2005 IETF standards, we should properly call them IRIs (Internationalized Resource Identifiers), as defined in RFC 3987, which are technically not URIs but can be converted to URIs simply by percent-encoding all non-ASCII characters in the IRI.

Per modern spec, the answer is "yes". The WHATWG Living Standard simply classifies everything that would previously be called "URIs" or "IRIs" as "URLs". This aligns the specced terminology with how normal people who haven't read the spec use the word "URL", which was one of the spec's goals.

What characters are allowed under the WHATWG Living Standard?

Per this newer meaning of "URL", what characters are allowed? In many parts of the URL, such as the query string and path, we're allowed to use arbitrary "URL units", which are

URL code points and percent-encoded bytes.

What are "URL code points"?

The URL code points are ASCII alphanumeric, U+0021 (!), U+0024 ($), U+0026 (&), U+0027 ('), U+0028 LEFT PARENTHESIS, U+0029 RIGHT PARENTHESIS, U+002A (*), U+002B (+), U+002C (,), U+002D (-), U+002E (.), U+002F (/), U+003A (:), U+003B (;), U+003D (=), U+003F (?), U+0040 (@), U+005F (_), U+007E (~), and code points in the range U+00A0 to U+10FFFD, inclusive, excluding surrogates and noncharacters.

(Note that the list of "URL code points" doesn't include %, but that %s are allowed in "URL code units" if they're part of a percent-encoding sequence.)

The only place I can spot where the spec permits the use of any character that's not in this set is in the host, where IPv6 addresses are enclosed in [ and ] characters. Everywhere else in the URL, either URL units are allowed or some even more restrictive set of characters.

What characters were allowed under the old RFCs?

For the sake of history, and since it's not explored fully elsewhere in the answers here, let's examine was allowed under the older pair of specs.

First of all, we have two types of RFC 3986 reserved characters:

• :/?#[]@, which are part of the generic syntax for a URI defined in RFC 3986
• !$&'()*+,;=, which aren't part of the RFC's generic syntax, but are reserved for use as syntactic components of particular URI schemes. For instance, semicolons and commas are used as part of the syntax of data URIs, and & and = are used as part of the ubiquitous ?foo=bar&qux=baz format in query strings (which isn't specified by RFC 3986).

Any of the reserved characters above can be legally used in a URI without encoding, either to serve their syntactic purpose or just as literal characters in data in some places where such use could not be misinterpreted as the character serving its syntactic purpose. (For example, although / has syntactic meaning in a URL, you can use it unencoded in a query string, because it doesn't have meaning in a query string.)

RFC 3986 also specifies some unreserved characters, which can always be used simply to represent data without any encoding:

• abcdefghijklmnopqrstuvwxyzABCDEFGHIJKLMNOPQRSTUVWXYZ0123456789-._~

Finally, the % character itself is allowed for percent-encodings.

That leaves only the following ASCII characters that are forbidden from appearing in a URL:

• The control characters (chars 0-1F and 7F), including new line, tab, and carriage return.
• "<>\^`{|}

Every other character from ASCII can legally feature in a URL.

Then RFC 3987 extends that set of unreserved characters with the following unicode character ranges:

  %xA0-D7FF / %xF900-FDCF / %xFDF0-FFEF
/ %x10000-1FFFD / %x20000-2FFFD / %x30000-3FFFD
/ %x40000-4FFFD / %x50000-5FFFD / %x60000-6FFFD
/ %x70000-7FFFD / %x80000-8FFFD / %x90000-9FFFD
/ %xA0000-AFFFD / %xB0000-BFFFD / %xC0000-CFFFD
/ %xD0000-DFFFD / %xE1000-EFFFD

These block choices from the old spec seem bizarre and arbitrary given the latest Unicode block definitions; this is probably because the blocks have been added to in the decade since RFC 3987 was written.

Finally, it's perhaps worth noting that simply knowing which characters can legally appear in a URL isn't sufficient to recognise whether some given string is a legal URL or not, since some characters are only legal in particular parts of the URL. For example, the reserved characters [ and ] are legal as part of an IPv6 literal host in a URL like http://[1080::8:800:200C:417A]/foo but aren't legal in any other context, so the OP's example of http://example.com/file[/].html is illegal.

To add some clarification and directly address the question above, there are several classes of characters that cause problems for URLs and URIs.

There are some characters that are disallowed and should never appear in a URL/URI, reserved characters (described below), and other characters that may cause problems in some cases, but are marked as "unwise" or "unsafe". Explanations for why the characters are restricted are clearly spelled out in RFC-1738 (URLs) and RFC-2396 (URIs). Note the newer RFC-3986 (update to RFC-1738) defines the construction of what characters are allowed in a given context but the older spec offers a simpler and more general description of which characters are not allowed with the following rules.

Excluded US-ASCII Characters disallowed within the URI syntax:

   control     = <US-ASCII coded characters 00-1F and 7F hexadecimal>
   space       = <US-ASCII coded character 20 hexadecimal>
   delims      = "<" | ">" | "#" | "%" | <">

The character "#" is excluded because it is used to delimit a URI from a fragment identifier. The percent character "%" is excluded because it is used for the encoding of escaped characters. In other words, the "#" and "%" are reserved characters that must be used in a specific context.

List of unwise characters are allowed but may cause problems:

   unwise      = "{" | "}" | "|" | "\" | "^" | "[" | "]" | "`"

Characters that are reserved within a query component and/or have special meaning within a URI/URL:

  reserved    = ";" | "/" | "?" | ":" | "@" | "&" | "=" | "+" | "$" | ","

The "reserved" syntax class above refers to those characters that are allowed within a URI, but which may not be allowed within a particular component of the generic URI syntax. Characters in the "reserved" set are not reserved in all contexts. The hostname, for example, can contain an optional username so it could be something like ftp://user@hostname/ where the '@' character has special meaning.

Here is an example of a URL that has invalid and unwise characters (e.g. '$', '[', ']') and should be properly encoded:

http://mw1.google.com/mw-earth-vectordb/kml-samples/gp/seattle/gigapxl/$[level]/r$[y]_c$[x].jpg

Some of the character restrictions for URIs and URLs are programming language-dependent. For example, the '|' (0x7C) character although only marked as "unwise" in the URI spec will throw a URISyntaxException in the Java java.net.URI constructor so a URL like http://api.google.com/q?exp=a|b is not allowed and must be encoded instead as http://api.google.com/q?exp=a%7Cb if using Java with a URI object instance.

Several of Unicode character ranges are valid HTML5, although it might still not be a good idea to use them.

E.g., href docs say http://www.w3.org/TR/html5/links.html#attr-hyperlink-href:

The href attribute on a and area elements must have a value that is a valid URL potentially surrounded by spaces.

Then the definition of "valid URL" points to http://url.spec.whatwg.org/, which says it aims to:

Align RFC 3986 and RFC 3987 with contemporary implementations and obsolete them in the process.

That document defines URL code points as:

ASCII alphanumeric, "!", "$", "&", "'", "(", ")", "*", "+", ",", "-", ".", "/", ":", ";", "=", "?", "@", "_", "~", and code points in the ranges U+00A0 to U+D7FF, U+E000 to U+FDCF, U+FDF0 to U+FFFD, U+10000 to U+1FFFD, U+20000 to U+2FFFD, U+30000 to U+3FFFD, U+40000 to U+4FFFD, U+50000 to U+5FFFD, U+60000 to U+6FFFD, U+70000 to U+7FFFD, U+80000 to U+8FFFD, U+90000 to U+9FFFD, U+A0000 to U+AFFFD, U+B0000 to U+BFFFD, U+C0000 to U+CFFFD, U+D0000 to U+DFFFD, U+E1000 to U+EFFFD, U+F0000 to U+FFFFD, U+100000 to U+10FFFD.

The term "URL code points" is then used in the statement:

If c is not a URL code point and not "%", parse error.

in a several parts of the parsing algorithm, including the schema, authority, relative path, query and fragment states: so basically the entire URL.

Also, the validator http://validator.w3.org/ passes for URLs like "??", and does not pass for URLs with characters like spaces "a b"

Of course, as mentioned by Stephen C, it is not just about characters but also about context: you have to understand the entire algorithm. But since class "URL code points" is used on key points of the algorithm, it that gives a good idea of what you can use or not.

See also: Unicode characters in URLs

I need to select character to split urls in string, so I decided to create list of characters which could not be found in URL by myself:

>>> allowed = "-_.~!*'();:@&=+$,/?%#[]?@ABCDEFGHIJKLMNOPQRSTUVWXYZabcdefghijklmnopqrstuvwxyz0123456789"
>>> from string import printable
>>> ''.join(set(printable).difference(set(allowed)))
'`" <\x0b\n\r\x0c\\\t{^}|>'

So, the possible choices are the newline, tab, space, backslash and "<>{}^|. I guess I'll go with the space or newline. :)

In your supplementary question you asked if www.example.com/file[/].html is a valid URL.

That URL isn't valid because a URL is a type of URI and a valid URI must have a scheme like http: (see RFC 3986).

If you meant to ask if http://www.example.com/file[/].html is a valid URL then the answer is still no because the square bracket characters aren't valid there.

The square bracket characters are reserved for URLs in this format: http://[2001:db8:85a3::8a2e:370:7334]/foo/bar (i.e. an IPv6 literal instead of a host name)

It's worth reading RFC 3986 carefully if you want to understand the issue fully.

Not really an answer to your question but validating url's is really a serious p.i.t.a You're probably just better off validating the domainname and leave query part of the url be. That is my experience. You could also resort to pinging the url and seeing if it results in a valid response but that might be too much for such a simple task.

Regular expressions to detect url's are abundant, google it :)

I am implementing old http (0.9, 1.0, 1.1) request and response reader/writer. Request URI is the most problematic place.

You can't just use RFC 1738, 2396 or 3986 as it is. There are many old HTTP clients and servers that allows more characters. So I've made research based on accidentally published webserver access logs: "GET URI HTTP/1.0" 200.

I've found that the following non-standard characters are often used in URI:

\ { } < > | ` ^ "

These characters were described in RFC 1738 as unsafe.

If you want to be compatible with all old HTTP clients and servers - you have to allow these characters in request URI.

Please read more information about this research in oghttp-request-collector.

I came up with a couple regular expressions for PHP that will convert urls in text to anchor tags. (First it converts all www. urls to http:// then converts all urls with https?:// to a href=... html links

$string = preg_replace('/(https?:\/\/)([!#$&-;=?\-\[\]_a-z~%]+)/sim', '<a href="$1$2">$2</a>', preg_replace('/(\s)((www\.)([!#$&-;=?\-\[\]_a-z~%]+))/sim', '$1http://$2', $string) );