4  Uniform Resource Identifiers

4 Uniform Resource Identifiers

At the time of writing this document, in October 2020, there are two major standards concerning Universal Resource Identifiers and Universal Resource Locators:

The former is a classical standard with a proper formal syntax, using the so called Augmented Backus-Naur Form (ABNF) for describing the grammar, while the latter is a living document describing the current pratice, that is, how a majority of Web browsers work with URIs. WHAT WG URL is Web focused and it has no formal grammar but a plain english description of the algorithms that should be followed.

What is the difference between them, if any? They provide an overlapping definition for resource identifiers and they are not compatible. The uri_string module implements RFC 3986 and the term URI will be used throughout this document. A URI is an identifier, a string of characters that identifies a particular resource.

For a more complete problem statement regarding the URIs check the URL Problem Statement and Directions.

Let's start with what it is not. It is not the text that you type in the address bar in your Web browser. Web browsers do all possible heuristics to convert the input into a valid URI that could be sent over the network.

A URI is an identifier consisting of a sequence of characters matching the syntax rule named URI in RFC 3986.

It is crucial to clarify that a character is a symbol that is displayed on a terminal or written to paper and should not be confused with its internal representation.

A URI more specifically, is a sequence of characters from a subset of the US ASCII character set. The generic URI syntax consists of a hierarchical sequence of components referred to as the scheme, authority, path, query, and fragment. There is a formal description for each of these components in ABNF notation in RFC 3986:

    URI         = scheme ":" hier-part [ "?" query ] [ "#" fragment ]
    hier-part   = "//" authority path-abempty
                   / path-absolute
                   / path-rootless
                   / path-empty
    scheme      = ALPHA *( ALPHA / DIGIT / "+" / "-" / "." )
    authority   = [ userinfo "@" ] host [ ":" port ]
    userinfo    = *( unreserved / pct-encoded / sub-delims / ":" )

    reserved    = gen-delims / sub-delims
    gen-delims  = ":" / "/" / "?" / "#" / "[" / "]" / "@"
    sub-delims  = "!" / "$" / "&" / "'" / "(" / ")"
                / "*" / "+" / "," / ";" / "="

    unreserved  = ALPHA / DIGIT / "-" / "." / "_" / "~"

As producing and consuming standard URIs can get quite complex, Erlang/OTP provides a module, uri_string, to handle all the most difficult operations such as parsing, recomposing, normalizing and resolving URIs against a base URI.

The API functions in uri_string work on two basic data types uri_string() and uri_map(). uri_string() represents a standard URI, while uri_map() is a wider datatype, that can represent URI components using Unicode characters. uri_map() is a convenient choice for enabling operations such as producing standard compliant URIs out of components that have special or Unicode characters. It is easier to explain this by an example.

Let's say that we would like to create the following URI and send it over the network: http://cities/örebro?foo bar. This is not a valid URI as it contains characters that are not allowed in a URI such as "ö" and the space. We can verify this by parsing the URI:

  1> uri_string:parse("http://cities/örebro?foo bar").

The URI parser tries all possible combinations to interpret the input and fails at the last attempt when it encounters the colon character ":". Note, that the inital fault occurs when the parser attempts to interpret the character "ö" and after a failure back-tracks to the point where it has another possible parsing alternative.

The proper way to solve this problem is to use uri_string:recompose/1 with a uri_map() as input:

  2> uri_string:recompose(#{scheme => "http", host => "cities", path => "/örebro",
  query => "foo bar"}).

The result is a valid URI where all the special characters are encoded as defined by the standard. Applying uri_string:parse/1 and uri_string:percent_decode/1 on the URI returns the original input:

  3> uri_string:percent_decode(uri_string:parse("http://cities/%C3%B6rebro?foo%20bar")).
  #{host => "cities",path => "/örebro",query => "foo bar",
  scheme => "http"}

This symmetric property is heavily used in our property test suite.

As you have seen in the previous chapter, a standard URI can only contain a strict subset of the US ASCII character set, moreover the allowed set of characters is not the same in the different URI components. Percent-encoding is a mechanism to represent a data octet in a component when that octet's corresponding character is outside of the allowed set or is being used as a delimiter. This is what you see when "ö" is encoded as %C3%B6 and space as %20. Most of the API functions are expecting UTF-8 encoding when handling percent-encoded triplets. The UTF-8 encoding of the Unicode character "ö" is two octets: OxC3 0xB6. The character space is in the first 128 characters of Unicode and it is encoded using a single octet 0x20.


Unicode is backward compatible with ASCII, the encoding of the first 128 characters is the same binary value as in ASCII.

It is a major source of confusion exactly which characters will be percent-encoded. In order to make it easier to answer this question the library provides a utility function, uri_string:allowed_characters/0 , that lists the allowed set of characters in each major URI component, and also in the most important standard character sets.

    1> uri_string:allowed_characters().

If a URI component has a character that is not allowed, it will be percent-encoded when the URI is produced:

    2> uri_string:recompose(#{scheme => "https", host => "local#host", path => ""}).

Consuming a URI containing percent-encoded triplets can take many steps. The following example shows how to handle an input URI that is not normalized and contains multiple percent-encoded triplets. First, the input uri_string() is to be parsed into a uri_map(). The parsing only splits the URI into its components without doing any decoding:

    3> uri_string:parse("http://%6C%6Fcal%23host/%F6re%26bro%20").
    #{host => "%6C%6Fcal%23host",path => "/%F6re%26bro%20",
      scheme => "http"}}

The input is a valid URI but how can you decode those percent-encoded octets? You can try to normalize the input with uri_string:normalize/1. The normalize operation decodes those percent-encoded triplets that correspond to a character in the unreserved set. Normalization is a safe, idempotent operation that converts a URI into its canonical form:

    4> uri_string:normalize("http://%6C%6Fcal%23host/%F6re%26bro%20").
    5> uri_string:normalize("http://%6C%6Fcal%23host/%F6re%26bro%20", [return_map]).
    #{host => "local%23host",path => "/%F6re%26bro%20",
      scheme => "http"}

There are still a few percent-encoded triplets left in the output. At this point, when the URI is already parsed, it is safe to apply application specific decoding on the remaining character triplets. Erlang/OTP provides a function, uri_string:percent_decode/1 for raw percent decoding that you can use on the host and path components, or on the whole map:

    6> uri_string:percent_decode("local%23host").
    7> uri_string:percent_decode("/%F6re%26bro%20").
    {error,invalid_utf8,<<"/öre&bro ">>}
    8> uri_string:percent_decode(#{host => "local%23host",path => "/%F6re%26bro%20",
    scheme => "http"}).
    {error,{invalid,{path,{invalid_utf8,<<"/öre&bro ">>}}}}

The host was successfully decoded but the path contains at least one character with non-UTF-8 encoding. In order to be able to decode this, you have to make assumptions about the encoding used in these triplets. The most obvious choice is latin-1, so you can try uri_string:transcode/2, to transcode the path to UTF-8 and run the percent-decode operation on the transcoded string:

    9> uri_string:transcode("/%F6re%26bro%20", [{in_encoding, latin1}]).
    10> uri_string:percent_decode("/%C3%B6re%26bro%20").
    "/öre&bro "

It is important to emphasize that it is not safe to apply uri_string:percent_decode/1 directly on an input URI:

    11> uri_string:percent_decode("http://%6C%6Fcal%23host/%C3%B6re%26bro%20").
    "http://local#host/öre&bro "
    12> uri_string:parse("http://local#host/öre&bro ").

Percent-encoding is implemented in uri_string:recompose/1 and it happens when converting a uri_map() into a uri_string(). There is no equivalent to a raw percent-encoding function as percent-encoding shall be applied on the component level using different sets of allowed characters. Applying percent-encoding directly on an input URI would not be safe just as in the case of uri_string:percent_decode/1, the output could be an invalid URI.

Normalization is the operation of converting the input URI into a canonical form and keeping the reference to the same underlying resource. The most common application of normalization is determining whether two URIs are equivalent without accessing their referenced resources.

Normalization has 6 distinct steps. First the input URI is parsed into an intermediate form that can handle Unicode characters. This datatype is the uri_map(), that can hold the components of the URI in map elements of type unicode:chardata(). After having the intermediate form, a sequence of normalization algorithms are applied to the individual URI components:

Case normalization

Converts the scheme and host components to lower case as they are not case sensitive.

Percent-encoding normalization

Decodes percent-encoded triplets that correspond to characters in the unreserved set.

Scheme-based normalization

Applying rules for the schemes http, https, ftp, ssh, sftp and tftp.

Path segment normalization

Converts the path into a canonical form.

After these steps, the intermediate data structure, an uri_map(), is fully normalized. The last step is applying uri_string:recompose/1 that converts the intermediate structure into a valid canonical URI string.

Notice the order, the uri_string:normalize(URIMap, [return_map]) that we used many times in this user guide is a shortcut in the normalization process returning the intermediate datastructure, and allowing us to inspect and apply further decoding on the remaining percent-encoded triplets.

    13> uri_string:normalize("hTTp://LocalHost:80/%c3%B6rebro/a/../b").
    14> uri_string:normalize("hTTp://LocalHost:80/%c3%B6rebro/a/../b", [return_map]).
    #{host => "localhost",path => "/%C3%B6rebro/b",
      scheme => "http"}

The current URI implementation provides support for producing and consuming standard URIs. The API is not meant to be directly exposed in a Web browser's address bar where users can basically enter free text. Application designers shall implement proper heuristics to map the input into a parsable URI.