XML Charset Encoding detection, how Rss and atOM utilitiEs (ROME) helps getting the right charset encoding

Determining the charset set encoding of an XML document it may prove not as easy as it seems, and when the XML document is served over HTTP things get a little more hairy.

Current Java libraries or utilities don't do this by default, A JAXP SAX parser may detect the charset encoding of a XML document looking at the first bytes of the stream as defined Section 4.3.3 and Appendix F.1 of the in XML 1.0 (Third Edition) specification. But Appendix F.1 is non-normative and not all Java XML parsers do it right now. For example the JAXP SAX parser implementation in J2SE 1.4.2 does not handle Appendix F.1 and Xerces 2.6.2 just added support for it. But still this does not solve the whole problem. JAXP SAX parsers are not aware of the HTTP transport rules for charset encoding resolution as defined by RFC 3023. They are not because they operate on a byte stream or a char stream without any knowledge of the stream transport protocol, HTTP in this case.

Mark Pilgrim did a very good job explaining how the charset encoding should be determined, Determining the character encoding of a feed and XML on the Web Has Failed.

To isolate developers from this issue ROME has a special character stream class, the XmlReader. The XmlReader class is a subclass of the java.io.Reader that detects the charset encoding of XML documents read from files, input streams, URLs and input streams obtained over HTTP. Ideally this should be built into the JAXP SAX classes, most likely in the InputSource class.

Default Lenient Behavior

It is very common for many sites, due to improper configuration or lack of knowledge, to declare an invalid charset encoding. Invalid according to the XML 1.0 specification and the RFC 3023. For example a mismatch between the implicit charset encoding in the HTTP content-type and the explicit charset encoding in the XML prolog.

Because of this, ROME XmlReader by default has a lenient detection. This lenient detection works in the following order:

  • A strict charset encoding detection, as specified by the Algorithms below is attempted
  • If the content type was 'text/html' it replaces it with 'text/xml' and tries strict charset encoding detection again
  • If the XML prolog had a charset encoding that encoding is used
  • If the content type had a charset encoding that encoding is used
  • 'UTF-8' is used

The XmlReader class has 2 constructors that allow a strict (non-lenient) charset encoding detection to be performed. This constructors take a lenient boolean flag, the flag should be set to false for a strict detection.

The Algorithms per XML 1.0 and RFC 3023 specifications

Following it's a detailed explanation on the algorithms the XmlReader uses to determine the charset encoding. These algorithms first appeared in Tucu's Weblog.

Raw XML charset encoding detection

Detection of the charset encoding of a XML document without external information (i.e. reading a XML document from a file). Following Section 4.3.3 and Appendix F.1 of the XML 1.0 specification the charset encoding of an XML document is determined as follows:

BOMEnc     : byte order mark. Possible values: 'UTF-8', 'UTF-16BE', 'UTF-16LE' or NULL
XMLGuessEnc: best guess using the byte representation of the first bytes of XML declaration
             ('<?xml...?>') if present. Possible values: 'UTF-8', 'UTF-16BE', 'UTF-16LE' or NULL
XMLEnc     : encoding in the XML declaration ('<?xml encoding="..."?>'). Possible values: anything or NULL

if BOMEnc is NULL
  if XMLGuessEnc is NULL or XMLEnc is NULL
    encoding is 'UTF-8'                                                                   [1.0]
  else
  if XMLEnc is 'UTF-16' and (XMLGuessEnc is 'UTF-16BE' or XMLGuessEnc is 'UTF-16LE')
    encoding is XMLGuessEnc                                                               [1.1]
  else
    encoding is XMLEnc                                                                    [1.2]
else
if BOMEnc is 'UTF-8'
  if XMLGuessEnc is not NULL and XMLGuessEnc is not 'UTF-8'
    ERROR, encoding mismatch                                                              [1.3]
  if XMLEnc is not NULL and XMLEnc is not 'UTF-8'
    ERROR, encoding mismatch                                                              [1.4]
  encoding is 'UTF-8'
else
if BOMEnc is 'UTF-16BE' or BOMEnc is 'UTF-16LE'
  if XMLGuessEnc is not NULL and XMLGuessEnc is not BOMEnc
    ERROR, encoding mismatch                                                              [1.5]
  if XMLEnc is not NULL and XMLEnc is not 'UTF-16' and XMLEnc is not BOMEnc
    ERROR, encoding mismatch                                                              [1.6]
  encoding is BOMEnc
else
  ERROR, cannot happen given BOMEnc possible values (see above)                           [1.7]

Byte Order Mark encoding and XML guessed encoding detection rules are clearly explained in the XML 1.0 Third Edition Appendix F.1. Note that in this algorithm BOMEnc and XMLGuessEnc are restricted to UTF-8 and UTF-16* encodings.

  • #1.0. There is no BOM, an encoding or encoding family cannot be guessed from the first bytes in the stream, defaulting to UTF-8.
  • #1.1. Strictly following XML 1.0 Third Edition Section 4.3.3 Charset Encoding in Entities (2nd paragraph) no BOM and UTF-16 XML declaration encoding is an error. The BOM is required to identify if the encoding is BE or LE. But if XMLEnc was read it means that the encoding byte order of the stream was guessed from the first bytes in the stream, note that this is possible only if the document starts with a XML declaration. The logic verifies that there is no conflicting encoding information and relaxes the BOM requirement if a XML declaration (that allows guessing the byte order) is present.
  • #1.2. XMLEnc is present, it is not UTF-16 (although it can be UTF-16BE or UTF-16LE), the guessed encoding is used to read the XML declaration encoding. Detecting encoding mismatches here it would require awareness of charset families, instead the algorithm relies on the charset encoding routines that will process the stream to discover and report mismatches. IMPORTANT: To handle other encodings (i.e. USC-4 or EBCDIC encoding families) the algorithm should be extended here.
  • #1.3, #1.4, #1.5, #1.6 There is an explicit encoding mismatch.
  • #1.7. Given the currently assumed BOMEnc values this case cannot happen. IMPORTANT: To handle other encodings (i.e. USC-4 or EBCDIC encoding families) the algorithm should be extended here.

XML over HTTP charset encoding detection

Detection of the charset encoding of a XML document with external information (provided by HTTP). Following Section 4.3.3, Appendix F.1 and Appendix F.2 of the XML 1.0 specification, plus RFC 3023 the charset encoding of an XML document served over HTTP is determined as follows:

ContentType: Content-Type HTTP header
CTMime     : MIME type defined in the ContentType
CTEnc      : charset encoding defined in the ContentType, NULL otherwise
BOMEnc     : byte order mark. Possible values: 'UTF-8', 'UTF-16BE', 'UTF-16LE' or NULL
XMLGuessEnc: best guess using the byte representation of the first bytes of XML declaration
             ('<?xml...?>') if present. Possible values: 'UTF-8', 'UTF-16BE', 'UTF-16LE' or NULL
XMLEnc     : encoding in the XML declaration ('<?xml encoding="..."?>'). Possible values: anything or NULL
APP-XML    : RFC 3023 defined 'application/*xml' types
TEXT-XML   : RFC 3023 defined 'text/*xml' types

if CTMime is APP-XML or CTMime is TEXT-XML
  if CTEnc is NULL
    if CTMime is APP-XML
      encoding is determined using the Raw XML charset encoding detection algorithm      [2.0]
    else
    if CTMime is TEXT-XML
      encoding is 'US-ASCII'                                                             [2.1]
  else
  if (CTEnc is 'UTF-16BE' or CTEnc is 'UTF-16LE') and BOMEnc is not NULL
    ERROR, RFC 3023 explicitly forbids this                                              [2.2]
  else
  if CTEnc is 'UTF-16'
    if BOMEnc is 'UTF-16BE' or BOMEnc is 'UTF-16LE'
      encoding is BOMEnc                                                                 [2.3]
    else
      ERROR, missing BOM or encoding mismatch                                            [2.4]
  else
    encoding is CTEnc                                                                    [2.5]
else
  ERROR, handling for other MIME types is undefined                                      [2.6]

Byte Order Mark encoding and XML guessed encoding detection rules are clearly explained in the XML 1.0 Third Edition Appendix F.1. Note that in this algorithm BOMEnc and XMLGuessEnc are restricted to UTF-8 and UTF-16* encodings.

  • #2.0. HTTP content type declares a MIME of application type and XML sub-type. There is not HTTP Content-type charset encoding information. The charset encoding is determined using the Raw XML charset encoding detection algorithm. Refer to RFC 3023 Section 3.2. Application/xml Registration.
  • #2.1. HTTP content type declares a MIME of text type and XML sub-type. There is not HTTP Content-type charset encoding information. The charset encoding is US-ASCII. Refer to RFC 3023 Section 3.1. Text/xml Registration.
  • #2.2. For text-XML and application-XML MIME types if the HTTP content type declares 'UTF-16BE' or 'UTF-16LE' as the charset encoding, a BOM is prohibited. Refer to RFC 3023 Section 4. The Byte Order Mark (BOM) and Conversions to/from the UTF-16 Charset.
  • #2.3. For text-XML and application-XML MIME types if the HTTP content type declares 'UTF-16' as the charset encoding, a BOM is required and it must be used to determine the byte order. Refer to RFC 3023 Section 4. The Byte Order Mark (BOM) and Conversions to/from the UTF-16 Charset.
  • #2.4. For text-XML and application-XML MIME types if the HTTP content type declares 'UTF-16' as the charset encoding, a BOM must be present and it must be 'UTF-16BE' or 'UTF-16LE'. Refer to RFC 3023 Section 4. The Byte Order Mark (BOM) and Conversions to/from the UTF-16 Charset.
  • #2.5. For text-XML and application-XML MIME types if the HTTP content type declares a charset encoding other than the UTF-16 variants, that charset encoding is used, no other special handling is required.
  • #2.6. There is not defined logic to determine the charset encoding based on the MIME type given by the HTTP content type.