RDF Patch

This page describes RDF Patch. An RDF Patch is a set of changes to an RDF dataset. The change are for triples, quads and prefixes.

Changes to triples involving blank nodes are handled by using their system identifier which uniquely identifies a blank node. Unlike RDF syntaxes, blank nodes are not generated afresh each time the document is parsed.


This example ensures certain prefixes are in the dataset and adds some basic triples for a new subclass of <http://example/SUPER_CLASS>.

TX .
PA "rdf" "http://www.w3.org/1999/02/22-rdf-syntax-ns#" .
PA "owl" "http://www.w3.org/2002/07/owl#" .
PA "rdfs" "http://www.w3.org/2000/01/rdf-schema#" .
A <http://example/SubClass> <http://www.w3.org/1999/02/22-rdf-syntax-ns#type> <http://www.w3.org/2002/07/owl#Class> .
A <http://example/SubClass> <http://www.w3.org/2000/01/rdf-schema#subClassOf> <http://example/SUPER_CLASS> .
A <http://example/SubClass> <http://www.w3.org/2000/01/rdf-schema#label> "SubClass" .
TC .


The text format for an RDF Patch is N-Triples-like: it is a series of rows, each row ends with a . (DOT). The tokens on a row are keywords, URIs, blank nodes, writen with their label (see below) or RDF Literals, in N-triples syntax. A keyword follows the same rules as Turtle prefix declarations without a trailing :.

A line has an operation code, then some number of items depending on the operation.

H Header
Change block: transactions
Change: Prefix add and delete
Change: Add and delete triples and quads

The general structure of an RDF patch is a header (possible empty), then a number of change blocks.

Each change block is a transaction. Transactions can be explicit recorded (‘TX’ start, TC commit) to include multiple transaction in one patch. They are not required. If not present, the patch should be applied atomically to the data.

Quad, triple or prefix changes
TC or TA

Multiple transaction blocks are allowed for multiple sets of changes in one patch.

A binary version based on RDF Thrift is provided. Parsing binary compared to text for N-triples achieves a x3-x4 increase in throughput.

The header provides for basic information about patch. It is a series of (key, value) pairs.

It is better to put complex metadata in a separate file and link to it from the header, but certain information is best kept with the patch. If patches are given an identifier, and als refer to the exp[ected previous patch, it create a log and patches can be applied in the right order.

A header section can be used to provide additional information. In this example a patch has an identifier and refers to a previous patch. This might be used to create a log of patches, a log being a sequnce of chnages to apply in-order.

H id <uuid:0686c69d-8f89-4496-acb5-744f0157a8db> .
H prev <uuid:3ee0eca0-6d5f-4b4d-85db-f69ab1167eb1> .
TX .
PA "rdf" "http://www.w3.org/1999/02/22-rdf-syntax-ns#" .
PA "owl" "http://www.w3.org/2002/07/owl#" .
PA "rdfs" "http://www.w3.org/2000/01/rdf-schema#" .
A <http://example/SubClass> <http://www.w3.org/1999/02/22-rdf-syntax-ns#type> <http://www.w3.org/2002/07/owl#Class> .
A <http://example/SubClass> <http://www.w3.org/2000/01/rdf-schema#subClassOf> <http://example/SUPER_CLASS> .
A <http://example/SubClass> <http://www.w3.org/2000/01/rdf-schema#label> "SubClass" .
TC .

Header format:

H word RDFTerm .

where word is a string in quotes, or an unquoted string (no spaces, starts with a letter, same as a prefix without the colon).

The header is ended by the first non H line or the end of the patch.


TX .
TC .

These delimit a block of quad, triple and prefix changes.

Abort, TA is provided so that changes can be streamed, not obliging the application to buffer change and wait to confirm the action is committed.

Transactions should be applied atomically when a patch is applied.


A change is an add or delete of a quad or a prefix.


Prefixes do not apply to the data of the patch. They are changes to the data the patch is applied to.

The prefix name is without the trailing colon. It can be given as a quoted string or unquoted string (keyword) with the same limitations as Turtle on the prefix name.

PA rdf <http://www.w3.org/1999/02/22-rdf-syntax-ns#> .

PA is adding a prefix, PD is deleting a prefix.

Quads and Triples

Triples and quads are written like N-Quads, 3 or 4 RDF terms, with the addition of an initial A or D for “add” or “delete”. Triples are in the order S-P-O, quads are S-P-O-G.

Add a triple:

A  <http://example/SubClass> <http://www.w3.org/1999/02/22-rdf-syntax-ns#type> <http://www.w3.org/2002/07/owl#Class> .

Blank nodes

In order to synchronize datasets, changes involving blank nodes may need to refer to a blank node already in the data. RDF Patch deals with this by making blank node labels refer to the “system identifier” for the blank node.

In this way, RDF Patch is not an “RDF Format”. In all syntaxes for RDF (Turtle, TriG, RDF/XML etc), blank nodes are “document scoped” meaning that the blank node is unique to that one time reading of the document. A new blank node is generated every time the file is read into a graph or dataset, and that blank node does not appear in the existing data.

In practice, most RDF triplestores have some kind of internal identifier that identifies the blank node. RDF Patch requires a “system identifier” for blank nodes so that change can refer to an existing blank node in the data.

These can be written as _:label or <_:label> (the latter provides a wider set of permissible characters in the label). Note that _ is illegal as an IRI scheme to highlight the fact this is not, strictly, an IRI.

RDF 1.1 describes skolemization where blank nodes are replaced by a URI. A system could use those for RDF Patch if it also meets the additional requirements to be able to receive and reverse the mapping back to the internal blank node object and also that all system generating patches can safely generate new, fresh skolem IRIs that will become new blank nodes in the RDF dataset then a patch is applied to it.

Preferred Style

The preferred style is to write patch rows on a single line, single space between tokens on a row and a single space before the terminal .. No comments should be included (comments start # and run to end of line).

Headers should be placed before the item they refer to; for information used by an RDF Patch Log, the metadata is about the whole patch and should be at the start of the file, before any TX.