rest/datatypes

Datatypes in HTML

One of the challenges of using HTML as a data transport is that everything, by default, is a string. This page explores ways to use microformats -- specifically, class names -- to encode data type information, e.g., for use with xoxo and rest/ahah, in order to allow lossless import/export from various languages. These could also be used with forms to provide rest/descriptions of the type of data expected.

Contributors

• Dr. Ernie Prabhakar
• Kevin Marks
• Tantek Çelik

Examples

These are the primary datatypes in a range of different languages and formats. Note that we are only concerned with "primitive" datatypes (loosely defined), as structured datatypes (list/array, hash/dictionary) are handled by xoxo.

Datatype comparison table

Language/format	string	float	integer	boolean	data	date/time	null
XML Schema	string	float, double	decimal, integer, etc.	boolean	hexBinary, base64Binary	duration, dateTime, date, time	nil
XML-RPC	string	double	i4, int	boolean	base64	dateTime.iso8601	nil
Mac OS X plists	string	real	integer	true, false	data	date	nil
JSON (JavaScript)	string	number	number	true, false	N/A	Date	nil
YAML tags	str	int	float	bool	null (base 64)	N/A	null
SQL (JDBC)	char,varchar	float, double, real	decimal, numeric	bit	binary	date, time, timestamp	?
C	char[]	float, double	int, long, short	bool, int	char[]	N/A	(void*)0
Java	char, String	float, double	int, long, short, byte	boolean	N/A	util.Date	null
PHP	string	float (double)	integer	boolean	array	N/A	NULL
Perl	array	scalar	scalar	scalar	array	N/A
Python	str	float, complex	int, long	bool	binascii, base64	time,datetime
Ruby + lib	String	Float	Fixnum, Bignum	TrueClass,FalseClass	Hash	Date	NilClass
REBOL	string!	decimal!	integer!	logic!	binary!	date!, time!	none!

Analysis

The most common set of datatypes appears to be those represented by XML-RPC, which (perhaps fortunately) also has historical precedence on the web:

• string
• double
• int [i4] - 4-byte integer (32-bit)
• boolean (0,1)
• base64
  • Lets call this 'binary' as the encoding is in the data: url, and DRY applies
  • RFC 2426 uses "B", which, when lowercased per microformats naming-principles is 'b'. -Tantek
• dateTime[.iso8601]

Whlle not perfect, these certainly cover the 80% case, and are reasonably well-defined. That said, there are a number of open questions about how to use them:

1. should 'string' also be explicitly specified, or can it be assumed?
   • Assumed, and also defined as utf-8. Kevin Marks 16:39, 13 Feb 2006 (PST)
   • Agreed with Kevin. 'string' should be the default if no type is specified. Publishers MAY explicitly specify 'string'.
2. does 'int' always mean 32-bits?
   1. If so, what should be used for 64-bit integers or cryptographic (256-bit+) numbers?
      1. Python's 'long' is simple, but ambiguous.
      2. Ruby's BigNum is clear but much less common.
      3. XML-Schema has so many types it is hard to say.
      4. SQL's "decimal", perhaps?
   2. If not, how should conforming implementations react to longer integers than they can handle?
      • I think integer is fine - we don't have an explict constraint here. Do you want to define +Inf -Inf and NaN behaviour? Certainly when building testcases and examples include these.

1. Is it worth deviating from the standard to allow "dateTime" as an alias? (the one case where XML Schema is actually simpler)
   • See comments below regarding date-time.

Proposal

The proposal is to adopt XML-RPC scalar values as the class names for typed microformats, with the following caveats:

• the alias 'i4' for integer SHOULD not be used
• the name 'long' MAY be used for 64-bit or longer integers
• for 'dateTime'
  • can we make this 'datetime' ? Kevin Marks 16:39, 13 Feb 2006 (PST)
  • microformats don't use camel case. please see naming-principles. alternatives (Tantek):
    • date-time (if you consider it to be two words)
    • datetime (as proposed by Kevin, if you think it is one word)
    • dt (reusing the common prefix shared by existing microformat class names: dtstart, dtend, dtreviewed from hCalendar and hReview. We could also make that a general rule for microformat class names for properties which take ISO8601 datetimes. (Tantek)
  • the trailing '.iso8601' MUST be omitted, as '.' is not (always?) valid in HTML class names
    • '.' is neither a valid HTML class name, nor a valid character (unescaped at least) in a CSS class selector. - Tantek
  • date/time formats SHOULD follow the W3C profile of ISO 8601
  • a more human-readable rendering may be used, with the ISO8601 value in an "abbr"
• binary data SHOULD be encoded in a data: URI, with an explicit ContentType and a human-readable description as the body of the anchor.
• if no datatype is specified, an implementation MAY either attempt to infer a datatype from the syntax of the value, or simply assert that the value is a string. Thus, conforming implementations SHOULD always explicitly label strings.
  • Disagree - either we are labelling datatypes and thus labelling string is redundant, or we are trying to guess from syntax. If the latter this whole spec is unnecessary. Kevin Marks 16:39, 13 Feb 2006 (PST)
  • Agreed with Kevin. Let's keep 'string' as the default, and allow explicit usage of it. -Tantek

To indicate that a particular micforomat uses typed values, precede that microformat with the class name 'typed', as in:

 <ol class="typed xoxo">

Example

 <dl class="typed xoxo">
  <dt>key</dt><dd class="string">value</dd>
  <dt>integer</dt><dd class="int">137</dd>
  <dt>real</dt><dd class="double">3.14159265</dd>
  <dt>date</dt><dd class="dateTime">1994-11-05T13:15:30Z</dd>
  <dt>date(abbr)</dt><dd class="dateTime"><abbr title="1994-11-05">November 5, 1994</abbr></dd>
  <dt>true</dt><dd class="boolean">1</dd>
  <dt>false</dt><dd class="boolean">0</dd>
  <dt>data</dt><dd class="base64"><a href="data:;base64,sdcfo2JTiXE=" type="image/jpg">my image</a></dd>
 </dl>

Example revised with above suggestions:

 <dl class="typed xoxo">
  <dt>key</dt><dd>value</dd>
  <dt>integer</dt><dd class="int">137</dd>
  <dt>real</dt><dd class="double">3.14159265</dd>
  <dt>date</dt><dd class="datetime">1994-11-05T13:15:30Z</dd>
  <dt>date(abbr)</dt><dd class="datetime"><abbr title="1994-11-05">November 5, 1994</abbr></dd>
  <dt>true</dt><dd class="boolean">1</dd>
  <dt>false</dt><dd class="boolean">0</dd>
  <dt>data</dt><dd class="binary"><a href="data:;base64,sdcfo2JTiXE=" type="image/jpg">my image</a></dd>
 </dl>

Note: XOXO always starts with either ol or ul. The dl in XOXO is always used to declare the properties of a specific li. These examples should be updated accordingly. -Tantek

References

• Datatypes in Wikipedia
• Origional datatype discussion
• Original plist datatype mapping proposal
• Revised xoxo datatype proposal

See Also

• xoxo
• naming-principles