3.  Relationship between IRIs and URIs

   IRIs are meant to replace URIs in identifying resources for
   protocols, formats, and software components that use a UCS-based
   character repertoire.  These protocols and components may never need
   to use URIs directly, especially when the resource identifier is used
   simply for identification purposes.  However, when the resource
   identifier is used for resource retrieval, it is in many cases
   necessary to determine the associated URI, because currently most
   retrieval mechanisms are only defined for URIs.  In this case, IRIs
   can serve as presentation elements for URI protocol elements.  An
   example would be an address bar in a Web user agent.  (Additional
   rationale is given in section 3.1.)

3.1.  Mapping of IRIs to URIs

   This section defines how to map an IRI to a URI.  Everything in this
   section also applies to IRI references and URI references, as well as
   to components thereof (for example, fragment identifiers).

   This mapping has two purposes:

   Syntaxical. Many URI schemes and components define additional
      syntactical restrictions not captured in section 2.2.
      Scheme-specific restrictions are applied to IRIs by converting
      IRIs to URIs and checking the URIs against the scheme-specific
      restrictions.

   Interpretational. URIs identify resources in various ways.  IRIs also
      identify resources.  When the IRI is used solely for
      identification purposes, it is not necessary to map the IRI to a
      URI (see section 5).  However, when an IRI is used for resource
      retrieval, the resource that the IRI locates is the same as the
      one located by the URI obtained after converting the IRI according
      to the procedure defined here.  This means that there is no need
      to define resolution separately on the IRI level.

   Applications MUST map IRIs to URIs by using the following two steps.

   Step 1.  Generate a UCS character sequence from the original IRI
            format.  This step has the following three variants,
            depending on the form of the input:

            a. If the IRI is written on paper, read aloud, or otherwise
               represented as a sequence of characters independent of
               any character encoding, represent the IRI as a sequence
               of characters from the UCS normalized according to
               Normalization Form C (NFC, [UTR15]).

            b. If the IRI is in some digital representation (e.g., an
               octet stream) in some known non-Unicode character
               encoding, convert the IRI to a sequence of characters
               from the UCS normalized according to NFC.

            c. If the IRI is in a Unicode-based character encoding (for
               example, UTF-8 or UTF-16), do not normalize (see section
               5.3.2.2 for details).  Apply step 2 directly to the
               encoded Unicode character sequence.

   Step 2.  For each character in 'ucschar' or 'iprivate', apply steps
            2.1 through 2.3 below.

       2.1.  Convert the character to a sequence of one or more octets
             using UTF-8 [RFC3629].

       2.2.  Convert each octet to %HH, where HH is the hexadecimal
             notation of the octet value.  Note that this is identical
             to the percent-encoding mechanism in section 2.1 of
             [RFC3986].  To reduce variability, the hexadecimal notation
             SHOULD use uppercase letters.

       2.3.  Replace the original character with the resulting character
             sequence (i.e., a sequence of %HH triplets).

   The above mapping from IRIs to URIs produces URIs fully conforming to
   [RFC3986].  The mapping is also an identity transformation for URIs
   and is idempotent;  applying the mapping a second time will not
   change anything.  Every URI is by definition an IRI.

   Systems accepting IRIs MAY convert the ireg-name component of an IRI
   as follows (before step 2 above) for schemes known to use domain
   names in ireg-name, if the scheme definition does not allow
   percent-encoding for ireg-name:

   Replace the ireg-name part of the IRI by the part converted using the
   ToASCII operation specified in section 4.1 of [RFC3490] on each
   dot-separated label, and by using U+002E (FULL STOP) as a label
   separator, with the flag UseSTD3ASCIIRules set to TRUE, and with the
   flag AllowUnassigned set to FALSE for creating IRIs and set to TRUE
   otherwise.

   The ToASCII operation may fail, but this would mean that the IRI
   cannot be resolved.  This conversion SHOULD be used when the goal is
   to maximize interoperability with legacy URI resolvers.  For example,
   the IRI

   "http://résumé.example.org"

   may be converted to

   "http://xn--rsum-bpad.example.org"

   instead of

   "http://r%C3%A9sum%C3%A9.example.org".

   An IRI with a scheme that is known to use domain names in ireg-name,
   but where the scheme definition does not allow percent-encoding for
   ireg-name, meets scheme-specific restrictions if either the
   straightforward conversion or the conversion using the ToASCII
   operation on ireg-name result in an URI that meets the scheme-
   specific restrictions.

   Such an IRI resolves to the URI obtained after converting the IRI and
   uses the ToASCII operation on ireg-name.  Implementations do not have
   to do this conversion as long as they produce the same result.

   Note: The difference between variants b and c in step 1 (using
      normalization with NFC, versus not using any normalization)
      accounts for the fact that in many non-Unicode character
      encodings, some text cannot be represented directly. For example,
      the word "Vietnam" is natively written "Việt Nam"
      (containing a LATIN SMALL LETTER E WITH CIRCUMFLEX AND DOT BELOW)
      in NFC, but a direct transcoding from the windows-1258 character
      encoding leads to "Việt Nam" (containing a LATIN SMALL
      LETTER E WITH CIRCUMFLEX followed by a COMBINING DOT BELOW).
      Direct transcoding of other 8-bit encodings of Vietnamese may lead
      to other representations.

   Note: The uniform treatment of the whole IRI in step 2 is important
      to make processing independent of URI scheme.  See [Gettys] for an
      in-depth discussion.

   Note: In practice, whether the general mapping (steps 1 and 2) or the
      ToASCII operation of [RFC3490] is used for ireg-name will not be
      noticed if mapping from IRI to URI and resolution is tightly
      integrated (e.g., carried out in the same user agent).  But

      conversion using [RFC3490] may be able to better deal with
      backwards compatibility issues in case mapping and resolution are
      separated, as in the case of using an HTTP proxy.

   Note: Internationalized Domain Names may be contained in parts of an
      IRI other than the ireg-name part.  It is the responsibility of
      scheme-specific implementations (if the Internationalized Domain
      Name is part of the scheme syntax) or of server-side
      implementations (if the Internationalized Domain Name is part of
      'iquery') to apply the necessary conversions at the appropriate
      point.  Example: Trying to validate the Web page at
      http://résumé.example.org would lead to an IRI of
      http://validator.w3.org/check?uri=http%3A%2F%2Frésumé.
      example.org, which would convert to a URI of
      http://validator.w3.org/check?uri=http%3A%2F%2Fr%C3%A9sum%C3%A9.
      example.org.  The server side implementation would be responsible
      for making the necessary conversions to be able to retrieve the
      Web page.

   Systems accepting IRIs MAY also deal with the printable characters in
   US-ASCII that are not allowed in URIs, namely "<", ">", '"', space,
   "{", "}", "|", "\", "^", and "`", in step 2 above.  If these
   characters are found but are not converted, then the conversion
   SHOULD fail.  Please note that the number sign ("#"), the percent
   sign ("%"), and the square bracket characters ("[", "]") are not part
   of the above list and MUST NOT be converted.  Protocols and formats
   that have used earlier definitions of IRIs including these characters
   MAY require percent-encoding of these characters as a preprocessing
   step to extract the actual IRI from a given field.  This
   preprocessing MAY also be used by applications allowing the user to
   enter an IRI.

   Note: In this process (in step 2.3), characters allowed in URI
      references and existing percent-encoded sequences are not encoded
      further.  (This mapping is similar to, but different from, the
      encoding applied when arbitrary content is included in some part
      of a URI.)  For example, an IRI of
      "http://www.example.org/red%09ros&#xE9;#red" (in XML notation) is
      converted to
      "http://www.example.org/red%09ros%C3%A9#red", not to something
      like
      "http%3A%2F%2Fwww.example.org%2Fred%2509ros%C3%A9%23red".

   Note: Some older software transcoding to UTF-8 may produce illegal
      output for some input, in particular for characters outside the
      BMP (Basic Multilingual Plane).  As an example, for the IRI with
      non-BMP characters (in XML Notation):
      "http://example.com/&#x10300;&#x10301;&#x10302";

      which contains the first three letters of the Old Italic alphabet,
      the correct conversion to a URI is
      "http://example.com/%F0%90%8C%80%F0%90%8C%81%F0%90%8C%82"

3.2.  Converting URIs to IRIs

   In some situations, converting a URI into an equivalent IRI may be
   desirable.  This section gives a procedure for this conversion.  The
   conversion described in this section will always result in an IRI
   that maps back to the URI used as an input for the conversion (except
   for potential case differences in percent-encoding and for potential
   percent-encoded unreserved characters).  However, the IRI resulting
   from this conversion may not be exactly the same as the original IRI
   (if there ever was one).

   URI-to-IRI conversion removes percent-encodings, but not all
   percent-encodings can be eliminated.  There are several reasons for
   this:

   1.  Some percent-encodings are necessary to distinguish percent-
       encoded and unencoded uses of reserved characters.

   2.  Some percent-encodings cannot be interpreted as sequences of
       UTF-8 octets.

       (Note: The octet patterns of UTF-8 are highly regular.
       Therefore, there is a very high probability, but no guarantee,
       that percent-encodings that can be interpreted as sequences of
       UTF-8 octets actually originated from UTF-8.  For a detailed
       discussion, see [Duerst97].)

   3.  The conversion may result in a character that is not appropriate
       in an IRI.  See sections 2.2, 4.1, and 6.1 for further details.

   Conversion from a URI to an IRI is done by using the following steps
   (or any other algorithm that produces the same result):

   1.  Represent the URI as a sequence of octets in US-ASCII.

   2.  Convert all percent-encodings ("%" followed by two hexadecimal
       digits) to the corresponding octets, except those corresponding
       to "%", characters in "reserved", and characters in US-ASCII not
       allowed in URIs.

   3.  Re-percent-encode any octet produced in step 2 that is not part
       of a strictly legal UTF-8 octet sequence.

   4. Re-percent-encode all octets produced in step 3 that in UTF-8
      represent characters that are not appropriate according to
      sections 2.2, 4.1, and 6.1.

   5. Interpret the resulting octet sequence as a sequence of characters
      encoded in UTF-8.

   This procedure will convert as many percent-encoded characters as
   possible to characters in an IRI.  Because there are some choices
   when step 4 is applied (see section 6.1), results may vary.

   Conversions from URIs to IRIs MUST NOT use any character encoding
   other than UTF-8 in steps 3 and 4, even if it might be possible to
   guess from the context that another character encoding than UTF-8 was
   used in the URI.  For example, the URI
   "http://www.example.org/r%E9sum%E9.html" might with some guessing be
   interpreted to contain two e-acute characters encoded as iso-8859-1.
   It must not be converted to an IRI containing these e-acute
   characters.  Otherwise, in the future the IRI will be mapped to
   "http://www.example.org/r%C3%A9sum%C3%A9.html", which is a different
   URI from "http://www.example.org/r%E9sum%E9.html".

3.2.1.  Examples

   This section shows various examples of converting URIs to IRIs.  Each
   example shows the result after each of the steps 1 through 5 is
   applied.  XML Notation is used for the final result.  Octets are
   denoted by "<" followed by two hexadecimal digits followed by ">".

   The following example contains the sequence "%C3%BC", which is a
   strictly legal UTF-8 sequence, and which is converted into the actual
   character U+00FC, LATIN SMALL LETTER U WITH DIAERESIS (also known as
   u-umlaut).

   1.  http://www.example.org/D%C3%BCrst

   2.  http://www.example.org/D<c3><bc>rst

   3.  http://www.example.org/D<c3><bc>rst

   4.  http://www.example.org/D<c3><bc>rst

   5.  http://www.example.org/D&#xFC;rst

   The following example contains the sequence "%FC", which might
   represent U+00FC, LATIN SMALL LETTER U WITH DIAERESIS, in the
   iso-8859-1 character encoding.  (It might represent other characters
   in other character encodings.  For example, the octet <fc> in

   iso-8859-5 represents U+045C, CYRILLIC SMALL LETTER KJE.)  Because
   <fc> is not part of a strictly legal UTF-8 sequence, it is
   re-percent-encoded in step 3.

   1.  http://www.example.org/D%FCrst

   2.  http://www.example.org/D<fc>rst

   3.  http://www.example.org/D%FCrst

   4.  http://www.example.org/D%FCrst

   5.  http://www.example.org/D%FCrst

   The following example contains "%e2%80%ae", which is the percent-
   encoded UTF-8 character encoding of U+202E, RIGHT-TO-LEFT OVERRIDE.
   Section 4.1 forbids the direct use of this character in an IRI.
   Therefore, the corresponding octets are re-percent-encoded in step 4.
   This example shows that the case (upper- or lowercase) of letters
   used in percent-encodings may not be preserved.  The example also
   contains a punycode-encoded domain name label (xn--99zt52a), which is
   not converted.

   1.  http://xn--99zt52a.example.org/%e2%80%ae

   2.  http://xn--99zt52a.example.org/<e2><80><ae>

   3.  http://xn--99zt52a.example.org/<e2><80><ae>

   4.  http://xn--99zt52a.example.org/%E2%80%AE

   5.  http://xn--99zt52a.example.org/%E2%80%AE

   Implementations with scheme-specific knowledge MAY convert
   punycode-encoded domain name labels to the corresponding characters
   by using the ToUnicode procedure.  Thus, for the example above, the
   label "xn--99zt52a" may be converted to U+7D0D U+8C46 (Japanese
   Natto), leading to the overall IRI of
   "http://&#x7D0D;&#x8C46;.example.org/%E2%80%AE".

4.  Bidirectional IRIs for Right-to-Left Languages

   Some UCS characters, such as those used in the Arabic and Hebrew
   scripts, have an inherent right-to-left (rtl) writing direction.
   IRIs containing these characters (called bidirectional IRIs or Bidi
   IRIs) require additional attention because of the non-trivial

   relation between logical representation (used for digital
   representation and for reading/spelling) and visual representation
   (used for display/printing).

   Because of the complex interaction between the logical
   representation, the visual representation, and the syntax of a Bidi
   IRI, a balance is needed between various requirements.  The main
   requirements are

   1.  user-predictable conversion between visual and logical
       representation;

   2.  the ability to include a wide range of characters in various
       parts of the IRI; and

   3.  minor or no changes or restrictions for implementations.

4.1.  Logical Storage and Visual Presentation

   When stored or transmitted in digital representation, bidirectional
   IRIs MUST be in full logical order and MUST conform to the IRI syntax
   rules (which includes the rules relevant to their scheme). This
   ensures that bidirectional IRIs can be processed in the same way as
   other IRIs.

   Bidirectional IRIs MUST be rendered by using the Unicode
   Bidirectional Algorithm [UNIV4], [UNI9].  Bidirectional IRIs MUST be
   rendered in the same way as they would be if they were in a
   left-to-right embedding; i.e., as if they were preceded by U+202A,
   LEFT-TO-RIGHT EMBEDDING (LRE), and followed by U+202C, POP
   DIRECTIONAL FORMATTING (PDF).  Setting the embedding direction can
   also be done in a higher-level protocol (e.g., the dir='ltr'
   attribute in HTML).

   There is no requirement to use the above embedding if the display is
   still the same without the embedding.  For example, a bidirectional
   IRI in a text with left-to-right base directionality (such as used
   for English or Cyrillic) that is preceded and followed by whitespace
   and  strong left-to-right characters does not need an embedding.
   Also, a bidirectional relative IRI reference that only contains
   strong right-to-left characters and weak characters and that starts
   and ends with a strong right-to-left character and appears in a text
   with right-to-left base directionality (such as used for Arabic or
   Hebrew) and is preceded and followed by whitespace and strong
   characters does not need an embedding.

   In some other cases, using U+200E, LEFT-TO-RIGHT MARK (LRM), may be
   sufficient to force the correct display behavior.  However, the
   details of the Unicode Bidirectional algorithm are not always easy to
   understand.  Implementers are strongly advised to err on the side of
   caution and to use embedding in all cases where they are not
   completely sure that the display behavior is unaffected without the
   embedding.

   The Unicode Bidirectional Algorithm ([UNI9], section 4.3) permits
   higher-level protocols to influence bidirectional rendering.  Such
   changes by higher-level protocols MUST NOT be used if they change the
   rendering of IRIs.

   The bidirectional formatting characters that may be used before or
   after the IRI to ensure correct display are not themselves part of
   the IRI.  IRIs MUST NOT contain bidirectional formatting characters
   (LRM, RLM, LRE, RLE, LRO, RLO, and PDF).  They affect the visual
   rendering of the IRI but do not appear themselves.  It would
   therefore not be possible to input an IRI with such characters
   correctly.

4.2.  Bidi IRI Structure

   The Unicode Bidirectional Algorithm is designed mainly for running
   text.  To make sure that it does not affect the rendering of
   bidirectional IRIs too much, some restrictions on bidirectional IRIs
   are necessary.  These restrictions are given in terms of delimiters
   (structural characters, mostly punctuation such as "@", ".", ":", and
   "/") and components (usually consisting mostly of letters and
   digits).

   The following syntax rules from section 2.2 correspond to components
   for the purpose of Bidi behavior: iuserinfo, ireg-name, isegment,
   isegment-nz, isegment-nz-nc, ireg-name, iquery, and ifragment.

   Specifications that define the syntax of any of the above components
   MAY divide them further and define smaller parts to be components
   according to this document.  As an example, the restrictions of
   [RFC3490] on bidirectional domain names correspond to treating each
   label of a domain name as a component for schemes with ireg-name as a
   domain name.  Even where the components are not defined formally, it
   may be helpful to think about some syntax in terms of components and
   to apply the relevant restrictions.  For example, for the usual
   name/value syntax in query parts, it is convenient to treat each name
   and each value as a component.  As another example, the extensions in
   a resource name can be treated as separate components.

   For each component, the following restrictions apply:

   1.  A component SHOULD NOT use both right-to-left and left-to-right
       characters.

   2.  A component using right-to-left characters SHOULD start and end
       with right-to-left characters.

   The above restrictions are given as shoulds, rather than as musts.
   For IRIs that are never presented visually, they are not relevant.
   However, for IRIs in general, they are very important to ensure
   consistent conversion between visual presentation and logical
   representation, in both directions.

   Note: In some components, the above restrictions may actually be
      strictly enforced.  For example, [RFC3490] requires that these
      restrictions apply to the labels of a host name for those schemes
      where ireg-name is a host name.  In some other components (for
      example, path components) following these restrictions may not be
      too difficult.  For other components, such as parts of the query
      part, it may be very difficult to enforce the restrictions because
      the values of query parameters may be arbitrary character
      sequences.

   If the above restrictions cannot be satisfied otherwise, the affected
   component can always be mapped to URI notation as described in
   section 3.1.  Please note that the whole component has to be mapped
   (see also Example 9 below).

4.3.  Input of Bidi IRIs

   Bidi input methods MUST generate Bidi IRIs in logical order while
   rendering them according to section 4.1.  During input, rendering
   SHOULD be updated after every new character is input to avoid end-
   user confusion.

4.4.  Examples

   This section gives examples of bidirectional IRIs, in Bidi Notation.
   It shows legal IRIs with the relationship between logical and visual
   representation and explains how certain phenomena in this
   relationship may look strange to somebody not familiar with
   bidirectional behavior, but familiar to users of Arabic and Hebrew.
   It also shows what happens if the restrictions given in section 4.2
   are not followed.  The examples below can be seen at [BidiEx], in
   Arabic, Hebrew, and Bidi Notation variants.

   To read the bidi text in the examples, read the visual representation
   from left to right until you encounter a block of rtl text.  Read the
   rtl block (including slashes and other special characters) from right
   to left, then continue at the next unread ltr character.

   Example 1: A single component with rtl characters is inverted:
   Logical representation: "http://ab.CDEFGH.ij/kl/mn/op.html"
   Visual representation: "http://ab.HGFEDC.ij/kl/mn/op.html"
   Components can be read one by one, and each component can be read in
   its natural direction.

   Example 2: More than one consecutive component with rtl characters is
   inverted as a whole:
   Logical representation: "http://ab.CDE.FGH/ij/kl/mn/op.html"
   Visual representation: "http://ab.HGF.EDC/ij/kl/mn/op.html"
   A sequence of rtl components is read rtl, in the same way as a
   sequence of rtl words is read rtl in a bidi text.

   Example 3: All components of an IRI (except for the scheme) are rtl.
   All rtl components are inverted overall:
   Logical representation: "http://AB.CD.EF/GH/IJ/KL?MN=OP;QR=ST#UV"
   Visual representation: "http://VU#TS=RQ;PO=NM?LK/JI/HG/FE.DC.BA"
   The whole IRI (except the scheme) is read rtl.  Delimiters between
   rtl components stay between the respective components; delimiters
   between ltr and rtl components don't move.

   Example 4: Each of several sequences of rtl components is inverted on
   its own:
   Logical representation: "http://AB.CD.ef/gh/IJ/KL.html"
   Visual representation: "http://DC.BA.ef/gh/LK/JI.html"
   Each sequence of rtl components is read rtl, in the same way as each
   sequence of rtl words in an ltr text is read rtl.

   Example 5: Example 2, applied to components of different kinds:
   Logical representation: "http://ab.cd.EF/GH/ij/kl.html"
   Visual representation: "http://ab.cd.HG/FE/ij/kl.html"
   The inversion of the domain name label and the path component may be
   unexpected, but it is consistent with other bidi behavior.  For
   reassurance that the domain component really is "ab.cd.EF", it may be
   helpful to read aloud the visual representation following the bidi
   algorithm.  After "http://ab.cd." one reads the RTL block
   "E-F-slash-G-H", which corresponds to the logical representation.

   Example 6: Same as Example 5, with more rtl components:
   Logical representation: "http://ab.CD.EF/GH/IJ/kl.html"
   Visual representation: "http://ab.JI/HG/FE.DC/kl.html"
   The inversion of the domain name labels and the path components may
   be easier to identify because the delimiters also move.

   Example 7: A single rtl component includes digits:
   Logical representation: "http://ab.CDE123FGH.ij/kl/mn/op.html"
   Visual representation: "http://ab.HGF123EDC.ij/kl/mn/op.html"
   Numbers are written ltr in all cases but are treated as an additional
   embedding inside a run of rtl characters.  This is completely
   consistent with usual bidirectional text.

   Example 8 (not allowed): Numbers are at the start or end of an rtl
   component:
   Logical representation: "http://ab.cd.ef/GH1/2IJ/KL.html"
   Visual representation: "http://ab.cd.ef/LK/JI1/2HG.html"
   The sequence "1/2" is interpreted by the bidi algorithm as a
   fraction, fragmenting the components and leading to confusion.  There
   are other characters that are interpreted in a special way close to
   numbers; in particular, "+", "-", "#", "$", "%", ",", ".", and ":".

   Example 9 (not allowed): The numbers in the previous example are
   percent-encoded:
   Logical representation: "http://ab.cd.ef/GH%31/%32IJ/KL.html",
   Visual representation (Hebrew): "http://ab.cd.ef/%31HG/LK/JI%32.html"
   Visual representation (Arabic): "http://ab.cd.ef/31%HG/%LK/JI32.html"
   Depending on whether the uppercase letters represent Arabic or
   Hebrew, the visual representation is different.

   Example 10 (allowed but not recommended):
   Logical representation: "http://ab.CDEFGH.123/kl/mn/op.html"
   Visual representation: "http://ab.123.HGFEDC/kl/mn/op.html"
   Components consisting of only numbers are allowed (it would be rather
   difficult to prohibit them), but these may interact with adjacent RTL
   components in ways that are not easy to predict.

5.  Normalization and Comparison

      Note: The structure and much of the material for this section is
      taken from section 6 of [RFC3986]; the differences are due to the
      specifics of IRIs.

   One of the most common operations on IRIs is simple comparison:
   Determining whether two IRIs are equivalent without using the IRIs or
   the mapped URIs to access their respective resource(s).  A comparison
   is performed whenever a response cache is accessed, a browser checks
   its history to color a link, or an XML parser processes tags within a
   namespace.  Extensive normalization prior to comparison of IRIs may
   be used by spiders and indexing engines to prune a search space or
   reduce duplication of request actions and response storage.

   IRI comparison is performed for some particular purpose.  Protocols
   or implementations that compare IRIs for different purposes will
   often be subject to differing design trade-offs in regards to how
   much effort should be spent in reducing aliased identifiers.  This
   section describes various methods that may be used to compare IRIs,
   the trade-offs between them, and the types of applications that might
   use them.

5.1.  Equivalence

   Because IRIs exist to identify resources, presumably they should be
   considered equivalent when they identify the same resource.  However,
   this definition of equivalence is not of much practical use, as there
   is no way for an implementation to compare two resources unless it
   has full knowledge or control of them. For this reason, determination
   of equivalence or difference of IRIs is based on string comparison,
   perhaps augmented by reference to additional rules provided by URI
   scheme definitions.  We use the terms "different" and "equivalent" to
   describe the possible outcomes of such comparisons, but there are
   many application-dependent versions of equivalence.

   Even though it is possible to determine that two IRIs are equivalent,
   IRI comparison is not sufficient to determine whether two IRIs
   identify different resources.  For example, an owner of two different
   domain names could decide to serve the same resource from both,
   resulting in two different IRIs.  Therefore, comparison methods are
   designed to minimize false negatives while strictly avoiding false
   positives.

   In testing for equivalence, applications should not directly compare
   relative references; the references should be converted to their
   respective target IRIs before comparison.  When IRIs are compared to
   select (or avoid) a network action, such as retrieval of a
   representation, fragment components (if any) should be excluded from
   the comparison.

   Applications using IRIs as identity tokens with no relationship to a
   protocol MUST use the Simple String Comparison (see section 5.3.1).
   All other applications MUST select one of the comparison practices
   from the Comparison Ladder (see section 5.3 or, after IRI-to-URI
   conversion, select one of the comparison practices from the URI
   comparison ladder in [RFC3986], section 6.2)

5.2.  Preparation for Comparison

   Any kind of IRI comparison REQUIRES that all escapings or encodings
   in the protocol or format that carries an IRI are resolved.  This is
   usually done when the protocol or format is parsed.  Examples of such

   escapings or encodings are entities and numeric character references
   in [HTML4] and [XML1].  As an example,
   "http://example.org/rosé" (in HTML),
   "http://example.org/ros&#233"; (in HTML or XML), and
   "http://example.org/ros&#xE9"; (in HTML or XML) are all resolved into
   what is denoted in this document (see section 1.4) as
   "http://example.org/ros&#xE9"; (the "é" here standing for the
   actual e-acute character, to compensate for the fact that this
   document cannot contain non-ASCII characters).

   Similar considerations apply to encodings such as Transfer Codings in
   HTTP (see [RFC2616]) and Content Transfer Encodings in MIME
   ([RFC2045]), although in these cases, the encoding is based not on
   characters but on octets, and additional care is required to make
   sure that characters, and not just arbitrary octets, are compared
   (see section 5.3.1).

5.3.  Comparison Ladder

   In practice, a variety of methods are used, to test IRI equivalence.
   These methods fall into a range distinguished by the amount of
   processing required and the degree to which the probability of false
   negatives is reduced.  As noted above, false negatives cannot be
   eliminated.  In practice, their probability can be reduced, but this
   reduction requires more processing and is not cost-effective for all
   applications.

   If this range of comparison practices is considered as a ladder, the
   following discussion will climb the ladder, starting with practices
   that are cheap but have a relatively higher chance of producing false
   negatives, and proceeding to those that have higher computational
   cost and lower risk of false negatives.

5.3.1.  Simple String Comparison

   If two IRIs, when considered as character strings, are identical,
   then it is safe to conclude that they are equivalent.  This type of
   equivalence test has very low computational cost and is in wide use
   in a variety of applications, particularly in the domain of parsing.
   It is also used when a definitive answer to the question of IRI
   equivalence is needed that is independent of the scheme used and that
   can be calculated quickly and without accessing a network.  An
   example of such a case is XML Namespaces ([XMLNamespace]).

   Testing strings for equivalence requires some basic precautions. This
   procedure is often referred to as "bit-for-bit" or "byte-for-byte"
   comparison, which is potentially misleading.  Testing strings for
   equality is normally based on pair comparison of the characters that

   make up the strings, starting from the first and proceeding until
   both strings are exhausted and all characters are found to be equal,
   until a pair of characters compares unequal, or until one of the
   strings is exhausted before the other.

   This character comparison requires that each pair of characters be
   put in comparable encoding form.  For example, should one IRI be
   stored in a byte array in UTF-8 encoding form and the second in a
   UTF-16 encoding form, bit-for-bit comparisons applied naively will
   produce errors.  It is better to speak of equality on a
   character-for-character rather than on a byte-for-byte or bit-for-bit
   basis.  In practical terms, character-by-character comparisons should
   be done codepoint by codepoint after conversion to a common character
   encoding form.  When comparing character by character, the comparison
   function MUST NOT map IRIs to URIs, because such a mapping would
   create additional spurious equivalences.  It follows that an IRI
   SHOULD NOT be modified when being transported if there is any chance
   that this IRI might be used as an identifier.

   False negatives are caused by the production and use of IRI aliases.
   Unnecessary aliases can be reduced, regardless of the comparison
   method, by consistently providing IRI references in an already
   normalized form (i.e., a form identical to what would be produced
   after normalization is applied, as described below). Protocols and
   data formats often limit some IRI comparisons to simple string
   comparison, based on the theory that people and implementations will,
   in their own best interest, be consistent in providing IRI
   references, or at least be consistent enough to negate any efficiency
   that might be obtained from further normalization.

5.3.2.  Syntax-Based Normalization

   Implementations may use logic based on the definitions provided by
   this specification to reduce the probability of false negatives. This
   processing is moderately higher in cost than character-for-character
   string comparison.  For example, an application using this approach
   could reasonably consider the following two IRIs equivalent:

      example://a/b/c/%7Bfoo%7D/rosé
      eXAMPLE://a/./b/../b/%63/%7bfoo%7d/ros%C3%A9

   Web user agents, such as browsers, typically apply this type of IRI
   normalization when determining whether a cached response is
   available.  Syntax-based normalization includes such techniques as
   case normalization, character normalization, percent-encoding
   normalization, and removal of dot-segments.

5.3.2.1.  Case Normalization

   For all IRIs, the hexadecimal digits within a percent-encoding
   triplet (e.g., "%3a" versus "%3A") are case-insensitive and therefore
   should be normalized to use uppercase letters for the digits A - F.

   When an IRI uses components of the generic syntax, the component
   syntax equivalence rules always apply; namely, that the scheme and
   US-ASCII only host are case insensitive and therefore should be
   normalized to lowercase.  For example, the URI
   "HTTP://www.EXAMPLE.com/" is equivalent to "http://www.example.com/".
   Case equivalence for non-ASCII characters in IRI components that are
   IDNs are discussed in section 5.3.3.  The other generic syntax
   components are assumed to be case sensitive unless specifically
   defined otherwise by the scheme.

   Creating schemes that allow case-insensitive syntax components
   containing non-ASCII characters should be avoided. Case normalization
   of non-ASCII characters can be culturally dependent and is always a
   complex operation.  The only exception concerns non-ASCII host names
   for which the character normalization includes a mapping step derived
   from case folding.

5.3.2.2.  Character Normalization

   The Unicode Standard [UNIV4] defines various equivalences between
   sequences of characters for various purposes.  Unicode Standard Annex
   #15 [UTR15] defines various Normalization Forms for these
   equivalences, in particular Normalization Form C (NFC, Canonical
   Decomposition, followed by Canonical Composition) and Normalization
   Form KC (NFKC, Compatibility Decomposition, followed by Canonical
   Composition).

   Equivalence of IRIs MUST rely on the assumption that IRIs are
   appropriately pre-character-normalized rather than apply character
   normalization when comparing two IRIs.  The exceptions are conversion
   from a non-digital form, and conversion from a non-UCS-based
   character encoding to a UCS-based character encoding. In these cases,
   NFC or a normalizing transcoder using NFC MUST be used for
   interoperability.  To avoid false negatives and problems with
   transcoding, IRIs SHOULD be created by using NFC.  Using NFKC may
   avoid even more problems; for example, by choosing half-width Latin
   letters instead of full-width ones, and full-width instead of
   half-width Katakana.

   As an example, "http://www.example.org/résumé.html" (in XML
   Notation) is in NFC.  On the other hand,
   "http://www.example.org/résumé.html" is not in NFC.

   The former uses precombined e-acute characters, and the latter uses
   "e" characters followed by combining acute accents.  Both usages are
   defined as canonically equivalent in [UNIV4].

   Note: Because it is unknown how a particular sequence of characters
      is being treated with respect to character normalization, it would
      be inappropriate to allow third parties to normalize an IRI
      arbitrarily.  This does not contradict the recommendation that
      when a resource is created, its IRI should be as character
      normalized as possible (i.e., NFC or even NFKC).  This is similar
      to the uppercase/lowercase problems.  Some parts of a URI are case
      insensitive (domain name).  For others, it is unclear whether they
      are case sensitive, case insensitive, or something in between
      (e.g., case sensitive, but with a multiple choice selection if the
      wrong case is used, instead of a direct negative result).  The
      best recipe is that the creator use a reasonable capitalization
      and, when transferring the URI, capitalization never be changed.

   Various IRI schemes may allow the usage of Internationalized Domain
   Names (IDN) [RFC3490] either in the ireg-name part or elsewhere.
   Character Normalization also applies to IDNs, as discussed in section
   5.3.3.

5.3.2.3.  Percent-Encoding Normalization

   The percent-encoding mechanism (section 2.1 of [RFC3986]) is a
   frequent source of variance among otherwise identical IRIs.  In
   addition to the case normalization issue noted above, some IRI
   producers percent-encode octets that do not require percent-encoding,
   resulting in IRIs that are equivalent to their non encoded
   counterparts.  These IRIs should be normalized by decoding any
   percent-encoded octet sequence that corresponds to an unreserved
   character, as described in section 2.3 of [RFC3986].

   For actual resolution, differences in percent-encoding (except for
   the percent-encoding of reserved characters) MUST always result in
   the same resource.  For example, "http://example.org/~user",
   "http://example.org/%7euser", and "http://example.org/%7Euser", must
   resolve to the same resource.

   If this kind of equivalence is to be tested, the percent-encoding of
   both IRIs to be compared has to be aligned; for example, by
   converting both IRIs to URIs (see section 3.1), eliminating escape
   differences in the resulting URIs, and making sure that the case of
   the hexadecimal characters in the percent-encoding is always the same
   (preferably uppercase).  If the IRI is to be passed to another

   application or used further in some other way, its original form MUST
   be preserved.  The conversion described here should be performed only
   for local comparison.

5.3.2.4.  Path Segment Normalization

   The complete path segments "." and ".." are intended only for use
   within relative references (section 4.1 of [RFC3986]) and are removed
   as part of the reference resolution process (section 5.2 of
   [RFC3986]).  However, some implementations may incorrectly assume
   that reference resolution is not necessary when the reference is
   already an IRI, and thus fail to remove dot-segments when they occur
   in non-relative paths.  IRI normalizers should remove dot-segments by
   applying the remove_dot_segments algorithm to the path, as described
   in section 5.2.4 of [RFC3986].

5.3.3.  Scheme-Based Normalization

   The syntax and semantics of IRIs vary from scheme to scheme, as
   described by the defining specification for each scheme.
   Implementations may use scheme-specific rules, at further processing
   cost, to reduce the probability of false negatives.  For example,
   because the "http" scheme makes use of an authority component, has a
   default port of "80", and defines an empty path to be equivalent to
   "/", the following four IRIs are equivalent:

      http://example.com
      http://example.com/
      http://example.com:/
      http://example.com:80/

   In general, an IRI that uses the generic syntax for authority with an
   empty path should be normalized to a path of "/".  Likewise, an
   explicit ":port", for which the port is empty or the default for the
   scheme, is equivalent to one where the port and its ":" delimiter are
   elided and thus should be removed by scheme-based normalization.  For
   example, the second IRI above is the normal form for the "http"
   scheme.

   Another case where normalization varies by scheme is in the handling
   of an empty authority component or empty host subcomponent.  For many
   scheme specifications, an empty authority or host is considered an
   error; for others, it is considered equivalent to "localhost" or the
   end-user's host.  When a scheme defines a default for authority and
   an IRI reference to that default is desired, the reference should be
   normalized to an empty authority for the sake of uniformity, brevity,

   and internationalization.  If, however, either the userinfo or port
   subcomponents are non-empty, then the host should be given explicitly
   even if it matches the default.

   Normalization should not remove delimiters when their associated
   component is empty unless it is licensed to do so by the scheme
   specification.  For example, the IRI "http://example.com/?" cannot be
   assumed to be equivalent to any of the examples above.  Likewise, the
   presence or absence of delimiters within a userinfo subcomponent is
   usually significant to its interpretation.  The fragment component is
   not subject to any scheme-based normalization; thus, two IRIs that
   differ only by the suffix "#" are considered different regardless of
   the scheme.

   Some IRI schemes may allow the usage of Internationalized Domain
   Names (IDN) [RFC3490] either in their ireg-name part or elsewhere.
   When in use in IRIs, those names SHOULD be validated by using the
   ToASCII operation defined in [RFC3490], with the flags
   "UseSTD3ASCIIRules" and "AllowUnassigned".  An IRI containing an
   invalid IDN cannot successfully be resolved.  Validated IDN
   components of IRIs SHOULD be character normalized by using the
   Nameprep process [RFC3491]; however, for legibility purposes, they
   SHOULD NOT be converted into ASCII Compatible Encoding (ACE).

   Scheme-based normalization may also consider IDN components and their
   conversions to punycode as equivalent.  As an example,
   "http://résumé.example.org" may be considered equivalent to
   "http://xn--rsum-bpad.example.org".

   Other scheme-specific normalizations are possible.

5.3.4.  Protocol-Based Normalization

   Substantial effort to reduce the incidence of false negatives is
   often cost-effective for web spiders. Consequently, they implement
   even more aggressive techniques in IRI comparison.  For example, if
   they observe that an IRI such as

      http://example.com/data

   redirects to an IRI differing only in the trailing slash

      http://example.com/data/

   they will likely regard the two as equivalent in the future.  This
   kind of technique is only appropriate when equivalence is clearly
   indicated by both the result of accessing the resources and the

   common conventions of their scheme's dereference algorithm (in this
   case, use of redirection by HTTP origin servers to avoid problems
   with relative references).