tiny Go library to normalize URLs
Clone or download
Fetching latest commit…
Cannot retrieve the latest commit at this time.



Purell is a tiny Go library to normalize URLs. It returns a pure URL. Pure-ell. Sanitizer and all. Yeah, I know...

Based on the wikipedia paper and the RFC 3986 document.

build status


go get github.com/PuerkitoBio/purell


  • 2016-11-14 (v1.1.0) : IDN: Conform to RFC 5895: Fold character width (thanks to @beeker1121).
  • 2016-07-27 (v1.0.0) : Normalize IDN to ASCII (thanks to @zenovich).
  • 2015-02-08 : Add fix for relative paths issue (PR #5) and add fix for unnecessary encoding of reserved characters (see issue #7).
  • v0.2.0 : Add benchmarks, Attempt IDN support.
  • v0.1.0 : Initial release.


From example_test.go (note that in your code, you would import "github.com/PuerkitoBio/purell", and would prefix references to its methods and constants with "purell."):

package purell

import (

func ExampleNormalizeURLString() {
  if normalized, err := NormalizeURLString("hTTp://someWEBsite.com:80/Amazing%3f/url/",
    FlagLowercaseScheme|FlagLowercaseHost|FlagUppercaseEscapes); err != nil {
  } else {
  // Output: http://somewebsite.com:80/Amazing%3F/url/

func ExampleMustNormalizeURLString() {
  normalized := MustNormalizeURLString("hTTpS://someWEBsite.com:443/Amazing%fa/url/",

  // Output: http://somewebsite.com/Amazing%FA/url

func ExampleNormalizeURL() {
  if u, err := url.Parse("Http://SomeUrl.com:8080/a/b/.././c///g?c=3&a=1&b=9&c=0#target"); err != nil {
  } else {
    normalized := NormalizeURL(u, FlagsUsuallySafeGreedy|FlagRemoveDuplicateSlashes|FlagRemoveFragment)

  // Output: http://someurl.com:8080/a/c/g?c=3&a=1&b=9&c=0


As seen in the examples above, purell offers three methods, NormalizeURLString(string, NormalizationFlags) (string, error), MustNormalizeURLString(string, NormalizationFlags) (string) and NormalizeURL(*url.URL, NormalizationFlags) (string). They all normalize the provided URL based on the specified flags. Here are the available flags:

const (
	// Safe normalizations
	FlagLowercaseScheme           NormalizationFlags = 1 << iota // HTTP://host -> http://host, applied by default in Go1.1
	FlagLowercaseHost                                            // http://HOST -> http://host
	FlagUppercaseEscapes                                         // http://host/t%ef -> http://host/t%EF
	FlagDecodeUnnecessaryEscapes                                 // http://host/t%41 -> http://host/tA
	FlagEncodeNecessaryEscapes                                   // http://host/!"#$ -> http://host/%21%22#$
	FlagRemoveDefaultPort                                        // http://host:80 -> http://host
	FlagRemoveEmptyQuerySeparator                                // http://host/path? -> http://host/path

	// Usually safe normalizations
	FlagRemoveTrailingSlash // http://host/path/ -> http://host/path
	FlagAddTrailingSlash    // http://host/path -> http://host/path/ (should choose only one of these add/remove trailing slash flags)
	FlagRemoveDotSegments   // http://host/path/./a/b/../c -> http://host/path/a/c

	// Unsafe normalizations
	FlagRemoveDirectoryIndex   // http://host/path/index.html -> http://host/path/
	FlagRemoveFragment         // http://host/path#fragment -> http://host/path
	FlagForceHTTP              // https://host -> http://host
	FlagRemoveDuplicateSlashes // http://host/path//a///b -> http://host/path/a/b
	FlagRemoveWWW              // http://www.host/ -> http://host/
	FlagAddWWW                 // http://host/ -> http://www.host/ (should choose only one of these add/remove WWW flags)
	FlagSortQuery              // http://host/path?c=3&b=2&a=1&b=1 -> http://host/path?a=1&b=1&b=2&c=3

	// Normalizations not in the wikipedia article, required to cover tests cases
	// submitted by jehiah
	FlagDecodeDWORDHost           // http://1113982867 ->
	FlagDecodeOctalHost           // http://0102.0146.07.0223 ->
	FlagDecodeHexHost             // http://0x42660793 ->
	FlagRemoveUnnecessaryHostDots // http://.host../path -> http://host/path
	FlagRemoveEmptyPortSeparator  // http://host:/path -> http://host/path

	// Convenience set of safe normalizations
	FlagsSafe NormalizationFlags = FlagLowercaseHost | FlagLowercaseScheme | FlagUppercaseEscapes | FlagDecodeUnnecessaryEscapes | FlagEncodeNecessaryEscapes | FlagRemoveDefaultPort | FlagRemoveEmptyQuerySeparator

	// For convenience sets, "greedy" uses the "remove trailing slash" and "remove www. prefix" flags,
	// while "non-greedy" uses the "add (or keep) the trailing slash" and "add www. prefix".

	// Convenience set of usually safe normalizations (includes FlagsSafe)
	FlagsUsuallySafeGreedy    NormalizationFlags = FlagsSafe | FlagRemoveTrailingSlash | FlagRemoveDotSegments
	FlagsUsuallySafeNonGreedy NormalizationFlags = FlagsSafe | FlagAddTrailingSlash | FlagRemoveDotSegments

	// Convenience set of unsafe normalizations (includes FlagsUsuallySafe)
	FlagsUnsafeGreedy    NormalizationFlags = FlagsUsuallySafeGreedy | FlagRemoveDirectoryIndex | FlagRemoveFragment | FlagForceHTTP | FlagRemoveDuplicateSlashes | FlagRemoveWWW | FlagSortQuery
	FlagsUnsafeNonGreedy NormalizationFlags = FlagsUsuallySafeNonGreedy | FlagRemoveDirectoryIndex | FlagRemoveFragment | FlagForceHTTP | FlagRemoveDuplicateSlashes | FlagAddWWW | FlagSortQuery

	// Convenience set of all available flags
	FlagsAllGreedy    = FlagsUnsafeGreedy | FlagDecodeDWORDHost | FlagDecodeOctalHost | FlagDecodeHexHost | FlagRemoveUnnecessaryHostDots | FlagRemoveEmptyPortSeparator
	FlagsAllNonGreedy = FlagsUnsafeNonGreedy | FlagDecodeDWORDHost | FlagDecodeOctalHost | FlagDecodeHexHost | FlagRemoveUnnecessaryHostDots | FlagRemoveEmptyPortSeparator

For convenience, the set of flags FlagsSafe, FlagsUsuallySafe[Greedy|NonGreedy], FlagsUnsafe[Greedy|NonGreedy] and FlagsAll[Greedy|NonGreedy] are provided for the similarly grouped normalizations on wikipedia's URL normalization page. You can add (using the bitwise OR | operator) or remove (using the bitwise AND NOT &^ operator) individual flags from the sets if required, to build your own custom set.

The full godoc reference is available on gopkgdoc.

Some things to note:

  • FlagDecodeUnnecessaryEscapes, FlagEncodeNecessaryEscapes, FlagUppercaseEscapes and FlagRemoveEmptyQuerySeparator are always implicitly set, because internally, the URL string is parsed as an URL object, which automatically decodes unnecessary escapes, uppercases and encodes necessary ones, and removes empty query separators (an unnecessary ? at the end of the url). So this operation cannot not be done. For this reason, FlagRemoveEmptyQuerySeparator (as well as the other three) has been included in the FlagsSafe convenience set, instead of FlagsUnsafe, where Wikipedia puts it.

  • The FlagDecodeUnnecessaryEscapes decodes the following escapes (from -> to): - %24 -> $ - %26 -> & - %2B-%3B -> +,-./0123456789:; - %3D -> = - %40-%5A -> @ABCDEFGHIJKLMNOPQRSTUVWXYZ - %5F -> _ - %61-%7A -> abcdefghijklmnopqrstuvwxyz - %7E -> ~

  • When the NormalizeURL function is used (passing an URL object), this source URL object is modified (that is, after the call, the URL object will be modified to reflect the normalization).

  • The replace IP with domain name normalization ( → http://www.example.com/) is obviously not possible for a library without making some network requests. This is not implemented in purell.

  • The remove unused query string parameters and remove default query parameters are also not implemented, since this is a very case-specific normalization, and it is quite trivial to do with an URL object.

Safe vs Usually Safe vs Unsafe

Purell allows you to control the level of risk you take while normalizing an URL. You can aggressively normalize, play it totally safe, or anything in between.

Consider the following URL:


Normalizing with the FlagsSafe gives:


With the FlagsUsuallySafeGreedy:


And with FlagsUnsafeGreedy:



  • Add a class/default instance to allow specifying custom directory index names? At the moment, removing directory index removes (^|/)((?:default|index)\.\w{1,4})$.

Thanks / Contributions

@rogpeppe @jehiah @opennota @pchristopher1275 @zenovich @beeker1121


The BSD 3-Clause license.