%ents; ]>
User-defined Data Transfer This specification proposes a simple mechanism by which applications can transfer data safely, without needing additional protocol design work. It is intended to provide a protocol that is trivial to implement and can be driven with a simple API. &LEGALNOTICE; XXXX ProtoXEP Standards Track Standards XMPP Core udt &dcridland; 0.0.1 2019-12-30 dwd
  • Initial Revision

Applications written on top of XMPP often need to exchange data that has no existing standard. Such applications are often written by developers unfamiliar with best practise in designing new extensions for XMPP, making it hard to achieve this simple design goal without causing longer term problems.

This leads to "solutions" such as stuffing JSON directly in the <body/> element, for example, and recognising this at the receiver either by heuristics or by a special <subject/>. While this works, it's difficult to then migrate to something else, and enforces that custom clients are always used.

Therefore this document proposes a very simple (and simplistic) framework for sending such data which - while very light on features - nevertheless conforms to best practice. Unusually, this specification SHOULD NOT be used as a base upon which to build other standards, and suggests an API for library developers to implement.

Data transferred using this specification is encoded using JSON. The type of the data is given by a URI under the same rules as an XML namespace, and this specification refers to this as the datatype.

Because this document defines mechanisms for sending essentially arbitrary data, no real-world examples are given.

Instead, example namespaces are used within an XML namespace prefixed by &nsx;

Support for this protocol is advertised by the Service Discovery protocol defined in &xep0030; using a feature of &ns;.

Support for a particular datatype is given by concatenating the &ns; feature with a hash character ('#') and the datatype, for example &ns;#&nsx;foo.

This specification provides for two types of user-defined data transfers. Each uses a similar payload syntax, the UDT data payload.

Requests are carried within <iq/> stanzas, either of type "get" or "set", and result in a "result" optionally containing a second UDT data payload.

UDT payloads may also be placed within a <message/> stanza. <message/> stanzas MAY contain multiple UDT payloads, but typical usage is expected to be that there will be only one. The UDT payload may be ancillary data to another message, or a standalone message in its own right.

A UDT payload consists of a single element, <payload/>, qualified by the XML namespace &ns;. It has a single, mandatory attribute of datatype, which MUST contain a string conformant to the requirements for XML namespaces (typically a URI under the control of the application developer).

As with XML namespaces, this URI is never expected to be resolved, and is used solely as an indentifier. Different strings are considered entirely different datatypes, and common prefixes etc MUST be considered irrelevant for the purposes of interpreting the data. There are no common or standard datatypes.

The <payload element contains exactly one mandatory child element, the <json/> element defined in &xep0335;. This in turns contains the JSON data.

{ "annoying-teenager-level": 11 } ]]>

Note that the suggested custom &IQ; query payload of &xep0335; is not used as this would generally require custom handlers within client libraries:

{ "annoying-teenager-percentage": 101 } ]]>

In order to satisfy the goals of this protocol, it is necessary to define an API that can be consistently implemented across APIs. This allows application developers to use the protocol easily, and encourages this over using the ad-hoc techniques described in the introduction.

Therefore, while not imposing hard and fast API definitions, this specification proposes naming conventions for APIs that will hopefully guide application developers toward consistent usage.

While names are specified in "snake_case", API developers are free to use their own naming. This specification also defines APIs as taking session arguments which will often be implied by method calls, and omits types.

Library developers SHOULD make these calls as simple to use as possible. If these are significantly harder to use for inexperienced developers than ad-hoc techniques, then ad-hoc techniques will be used instead.

  • session - The session, connection or stream object or handle for the library. In cases where the library uses a single global connection this is omitted. An OO library is likely to elide this and have the call be a method call on the object instead.
  • message - The message object or handle for the library. An OO library is likely to elide this and have the call be a method call on the object instead.
  • datatype - Typically a string containing an XML-style namespace.
  • jid - A string or JID object. Any form of legal jid might be used here.
  • data - Either a JSON-encoded string, or an object or data structure which can be converted by the library.
  • get_or_set - A flag for indicating whether this is a "get" or "set" request.
  • callback - Some callable object or similar as meets the idiom of the language.

Support is advertised on a session by calling udt_advertise(session, datatype). Calling this MUST explicitly advertise both the &ns; feature and that of the datatype support.

APIs are free to (and encouraged to) implicitly advertise support when other calls are made.

Requests are sent to a particular jid by calling udt_request(session, jid, get_or_set, datatype, data). This might return a UDT payload, or have an additional callback argument to call with the response.

Applications may register to handle such requests by calling udt_request_callback(session, get_or_set, datatype, callback). The callback should be called as callback(session, jid, get_or_set, datatype, data), and returning a payload should send a result containing it. Applications registering this way SHOULD implicitly advertise support for the datatype.

Messages are sent to a particular jid by calling udt_message(session, jid, get_or_set, datatype, data).

Applications may register to handle such requests by calling udt_message_callback(session, get_or_set, datatype, callback). The callback should be called as callback(session, jid, get_or_set, datatype, data). Applications registering this way SHOULD implicitly advertise support for the datatype.

Applications may check to see if any message stanza contains a UDT payload by calling udt_payload(message, datatype), which returns the payload if it exists.

 ]]>

All security implications herein are those of the payload.

This XEP requires no interaction with &IANA;.

None.

The authors wish to share any credit with many members of the community, including Florian Schmaus.