Table of Contents

Conceptual Triple-Store Structure

Basic Triple-Store Operations

Managing a triple-store

Adding Triples

Triple Manipulation

Querying Triples

Modifying triples

Deleting triples

Serializing Triples

Indexing Triples

Advanced Triple-Store Operations

The !-reader macro and future-parts

Freetext Indexing

Data-type and Predicate Mapping

Federation

Simple Reification

SPARQL Query Language

Prolog Select

Select Query Planner

RDFS++ Reasoning

Specialized Datatypes

Social Network Analysis

Geospatial Primitives

Temporal Primitives

REPL Interface

Miscellaneous Utilities

Indices

Function and Macro index

Functor index

Variable index

This is the reference guide for AllegroGraph 4.7. See the server installation and Lisp Quick Start for an overview. An introduction to AllegroGraph covering all of its many features at a high-level is in the AllegroGraph Introduction.

Conceptual Triple-Store Structure

AllegroGraph is a graph database that makes an excellent RDF triple-store. This is a conceptual diagram of the data AllegroGraph manages:

AllegroGraph's logical structure

Triple-data generally comes into AllegroGraph as strings either from pure RDF/XML or as the more verbose but simpler N-Triple format. Once the triples are in AllegroGraph, it is possible to manipulate them, perform logical inference, and execute extremely fast queries.

The string dictionary manages efficient storage and fast lookup of the strings in the triple-store. We call the process of hashing a string into its UPI and storing it in the string dictionary interning.

Basic Triple-Store Operations

Managing a triple-store

These functions are used to create, delete, and examine triple-stores. For convenience, many operations act by default on the current triple-store which is kept in a special variable named *db*. The with-triple-store macro makes it easy to call other code with a particular store marked current. 1

*db*
variable

The default triple-store instance. API functions that take a db keyword argument will use this value by default.

Note that the *db* value should not be shared among multiple processes because add, rollback and commit events from independent processes will be executed on the same connection with unexpected results. Since *db* is the default value for many functions that have a :db keyword, this problem can easily arise in a multi-process application. To avoid it, each process should make its own connection to the database by using open-triple-store.

*default-ag-http-port*
variable

The default port to use when attaching to an AllegroGraph server.

This is the port where the AllegroGraph server listens for connections from clients. It is used to start a session by triple-store functions like create-triple-store, open-triple-store, and so on. The server's port can be set explicitly in the configuration file or found by looking at the file the server creates when it starts. The location of this file is specified using the PortFile directive. For more details see the server configuration guide.

catalogs &optional  port
function

List the catalogs on the local AllegroGraph server running on port.

This function is deprecated, please refer to catalog-names instead.

close-triple-store &key  db  if-closed  commit  ensure-not-lingering
function

Close the triple-store db

  • :db - defaults to the value of *db* and specifies which triple-store to close. If the db being closed is eq to *db*, then *db* is set to nil.

  • :if-closed - controls the behavior when the db is either not open or nil. If if-closed is :error then an error will be signaled in this situation. If it is :ignore, then close-triple-store will just return without signaling an error. The argument defaults to :ignore.

  • :commit is true, then commit-triple-store will be called before the store is closed. This defaults to nil

  • :ensure-not-lingering is true, then call ensure-not-lingering after the store is closed.
ensure-not-lingering name  &rest  args  &key  triple-store-class
function

To conserve resources, makes the database instance and its child processes exit if the instance is unused (nobody has the corresponding triple store open). This function returns nothing.

Normally unused database instances linger for InstanceTimeout seconds to speed up subsequent open operations.

You can use the triple-store-class argument to specify the kind of store you are trying to check. Like open-triple-store and create-triple-store, ensure-not-lingering takes additional keyword arguments that vary depending on this class. For example, checking a remote-triple-store may require specifying a username and password. See open-triple-store or create-triple-store for additional details. Also, see the ensure-not-lingering keyword argument to close-triple-store.

commit-triple-store &key  db
function
Ensure that all persistent data needed by the triple store is written to disk. db defaults to *db*.
copy-triples target-db  source  &key  source-db
function

Copy triples from source to target-db.

source can be a triple-store, a cursor, a list or an array. If the source-db can be determined from source (e.g., if source is a triple-store or cursor), then source-db is ignored. If source is a list or array then the triples are assumed to exist in source-db, which defaults to *db*.

target-db should be an open and writable triple-store.

create-triple-store name  &rest  args  &key  if-exists  catalog  port  user  password  exclude-standard-ontology-uris  triple-store-class  with-indices
function

Create a new triple store with the given name.

create-triple-store takes numerous keyword arguments (not all of which appear in the argument list, which does have an &allow-other-keys). The allowed arguments vary depending on the class of triple-store being created. For example, a remote-triple-store usually requires a :user a :password, a :port, and a :server (ecsept for anonymous access if it is allowed) whereas an instance of triple-db does not. :user, :password, and :server can be specified with the individual arguments or combined in a :url argument. (See Lisp Quick Start for an example of creating a remote triple store and other examples creating and opening triple stores.)

The possible arguments include (but are not limited to):

  • :if-exists - controls what to do if the triple-store already exists. The default value, :supersede, will cause create-triple-store to attempt to delete the old triple store and create a new one; :error causes create-triple-store to signal an error. If the triple-store is opened by another connection, then :supersede will fail because the store cannot be deleted.

  • :catalog - the catalog to use in the server's configuration file. If left unspecified, the RootCatalog will be used.

  • :port - the port to use to find the AllegroGraph server. The default is 10035.

  • :user - the username to use when accessing the server (anonymous access, if supported, does not require a user value)

  • :password - the password to use when accessing the server (anonymous access, if supported, does not require a password)

  • :server - the server for a remote-triple-store. The default is "127.0.0.1" (which is the local machine).

  • :url - this argument can combine username, password, server, and port, for example http:newuser:xyzzy@//mymachine.franz.com:10422/ (user newuser, password xyzzy, server mymachine.franz.com, port 10422).

  • :exclude-standard-ontology-uris - if true, then AllegroGraph will not add the set of usual strings to the triple-store at creation time. Examples of these strings are: rdf:type, owl:sameAs, owl:inverseOf, rdfs:subPropertyOf, rdfs:subClassOf, rdfs:range, rdfs:domain, owl:transitiveProperty. If not specified, then this will default to nil.

  • :triple-store-class - the kind of triple-store to create. If left unspecified, then triple-db will be used.

  • :with-indices - a list of index flavors that this store will use. If left unspecified, the store will use the standard indices of :gospi, :gposi, :gspoi, :ospgi, :posgi, :spogi, and :i.

  • :https-verification - the value should be a list of keywords and value pairs, as follows: :method allows control over the SSL protocol handshake process; can be: :sslv23 (the default), :sslv2, :sslv3, :sslv3+, :tlsv1; :certificate names a file which contains one or more PEM-encoded certificates. :key should be a string or pathname naming a file containing the private RSA key corresponding the the public key in the certificate. The file must be in PEM format. :certificate-password, if specified, should be a string. :verify: nil means that no automatic verification will occur, :optional means that the server's certificate (if supplied) will be automatically verified during SSL handshake, :required means that the server's certificate will be automatically verified during SSL handshake. :maxdepth must be an integer (which defaults to 10) which indicates the maximum allowable depth of the certificate verification chain; :ca-file specifies the name of a file containing a series of trusted PEM-encoded Intermediate CA or Root CA certificates that will be used during peer certificate verification; :ca-directory specifies the name of a directory containing a series of trusted Intermediate CA or Root CA certificate files that will be used during peer certificate verification; :ciphers should be a string which specifies an OpenSSL cipher list. these values are passed as arguments to the function socket:make-ssl-client-stream. Here is an example:

    (create-triple-store "v4.4-clienttest"  
    	 :triple-store-class 'remote-triple-store  
    	 :port 10398 :server "localhost"  
    	 :scheme :https  
    	 ;; new options:  
    	 :https-verification  
    	 '(;; Server must identify to the client and must  
    	   ;; pass certificate verification. Other sensible  
    	   ;; option is :optional; :nil suppresses  
    	   ;; verification altogether and should never be used.  
    	   :verify :required  
     
    	   ;; The CA certificate used to sign the server's  
    	   ;; certificate.  
    	   :ca-file "/path/to/ca.cert"  
     
    	   ;; The certificate and key file to authenticate  
    	   ;; this client with.  
    	   :certificate "/path/to/test.cert")) 

Create-triple-store returns a triple-store object and sets the value of the variable *db* to that object.

db-name db
function

The name of the triple-store db (read-only).

This is set when the triple-store is created or opened.

default-graph-upi db
function
Returns the UPI of the default-graph of db. This is the graph that will be assigned to any triples added to db unless a different graph is specified.
delete-triple-store db-or-name  &rest  args  &key  if-does-not-exist  triple-store-class
function

Delete an existing triple store. Returns t if the deletion was successful and nil if it was not.

The db-or-name argument can be either a triple store instance or the name of a triple store. If it is an instance, then the triple store associated with the instance will be closed if it is open and then it will be deleted (assuming that it is not open by any other processes). If db-or-name is eq to *db*, then *db* will be set to nil once the triple-store is deleted.

The :if-does-not-exist keyword argument specifies what to do if the triple-store does not exist. The default value, :error, causes delete-triple-store to signal an error. The value :ignore will cause delete-triple-store to do nothing and return nil.

You can use the triple-store-class argument to specify the kind of store you are trying to delete. Like open-triple-store and create-triple-store, delete-triple-store takes additional keyword arguments that vary depending on this class. For example, deleting a remote-triple-store may require specifying a username and password. See open-triple-store or create-triple-store for additional details.

make-tutorial-store &key  apply-reasoner-p
function

Close any current triple-store and create a new empty one.

  • :apply-reasoner-p - If true (the default), then the new triple-store will use RDFS++ reasoning. If nil, then the triple-store will have no reasoning enabled.

The new triple-store will be bound to *db* and is also returned by make-tutorial-store.

open-triple-store name  &rest  args  &key  read-only  catalog  port  user  password  triple-store-class
function

Open an existing triple store (previously created using create-triple-store) with the given name.

open-triple-store takes numerous keyword arguments. The allowed arguments vary depending on the class of triple-store being opened. For example, a remote-triple-store usually requires a :user and :password (though not for anonymous access if it is allowed) and perhaps a port and server (though those arguments do have defaults which may suffice) whereas a local triple-db does not. :user, :password, and :server can be specified with the individual arguments or combined in a :url argument.

The possible arguments include (but are not limited to):

  • :catalog - the catalog to use in the server's configuration file. If left unspecified, the RootCatalog will be used.

  • :port - the port to use to find the AllegroGraph server.

  • :user - the username to use when accessing the server (anonymous access, if supported, does not require a user value)

  • :password - the password to use when accessing the server (anonymous access, if supported, does not require a password)

  • :server - the server for a remote-triple-store. The default is "127.0.0.1" (which is the local machine).

  • :url - this argument can combine username, password, server, and port, for example http:newuser:xyzzy@//mymachine.franz.com:10422/ (user newuser, password xyzzy, server mymachine.franz.com, port 10422).

  • :triple-store-class - the kind of triple-store to open. If left unspecified, then triple-db will be used. Depending on the class of the triple-store opened, other parameters may be possible or required.

  • :read-only - if true, then the triple-store will be opened read-only.

  • :https-verification - See the description of this argument in the documentation for create-triple-store.

Returns a triple store object and sets the value of the variable *db* to that object. If the named triple store is already open, open-triple-store returns a new object connected to the same store.

See Lisp Quick Start for an example of creating and opening triple stores.

rollback-triple-store &key  db
function

Undo any local changes to db and synchronize with the most up-to-date server version.

  • :db - the triple-store to rollback; defaults to the current value of *db*.
triple-count &key  db
function

Returns the number of triples in a triple store. The :db keyword argument specifies the triple store to use, either by name or a triple store object. It defaults to the value of *db*.

Note that reasoning triple-stores do not report accurate triple-counts (doing so might take an inordinate amount of time!). The count returned will be a count of the triples in the ground store.

triple-store-exists-p name  &rest  args  &key  triple-store-class
function

Returns true if the triple-store with the given name exists.

You can use the triple-store-class argument to specify the kind of store you are trying to check. Like open-triple-store and create-triple-store, triple-store-exists-p takes additional keyword arguments that vary depending on this class. For example, checking a remote-triple-store may require specifying a username and password. See open-triple-store or create-triple-store for additional details.

triple-store-id db
function
This returns the ID of the triple-store. The ID is generated randomly when the triple-store is created. It is an array of four octets.
with-triple-store (var  store  &key  state  read-only-p  errorp  verifyp)  &body  body
macro

Binds both var and *db* to the triple-store designated by store. The following keyword arguments can also be used:

  • errorp - controls whether or not with-triple-store signals an error if the specified store cannot be found.

  • read-only-p - if specified then with-triple-store will signal an error if the specified triple-store is writable and read-only-p is nil or if the store is read-only and read-only-p is t.

  • state - can be :open, :closed or nil. If :open or :closed, an error will be signaled unless the triple-store is in the same state.

Adding Triples

You can import data into your triple-store using the client API (such as the Lisp functions below) or the command-line program agload.

Triples stored in files using the N-Triples 2 , RDF/XML 3 , and other formats can be loaded into the triple-store with the following functions.

Bulk loading and commit-triple-store

All of AllegroGraph bulk load functions take a :commit parameter that controls when the data being loaded is actually committed. This parameter can be:

Note also that if you are loading many small files, you will find that it is more efficient to use the method specialized on lists and not iterate through the the list of files yourself:

;; pathnames is a list of files  
;; let load-ntriples do the work of iterating over them  
(load-ntriples pathnames)  
 
;; Much slower!!  
;; probably _NOT_ what you want to do  
(dolist (pathname pathnames)  
  (load-ntriples pathname)) 

The Bulk Loading Functions

load-nquads source  &rest  args  &key  db  graph  verbose  preserve-strings  external-format  relax-syntax  commit  continue-on-errorp  &allow-other-keys
function

Add triples from source (in N-Quads format) to the triple store.

Returns (as multiple values) the count of triples loaded and the UPI of the graph into which the triples were loaded.

  • source - can be a stream, a pathname to an N-Triples file, a file URI, an HTTP URI or a string that can be coerced into a pathname. Source can also be a list of any of these things. In this case, each item in the list will be imported in turn.

In the case where a list of sources is being loaded and the graph argument is :source, each item in the list will be associated with the URI specifying it. In this case, the second return value will be nil.

The following keyword parameters can be used to control the loading process:

  • :db - specifies the triple-store into which triples will be loaded. This defaults to the value of *db*.

  • :graph - the graph to which the triples from source will be placed. It defaults to nil which is interpreted as db's default graph. If supplied, it can be:

    • a string representing a URIref, or a UPI or future-part encoding a URIref, which adds the triples in source to a graph named by that URI

    • the keyword :source (in which case the source argument will be interned as a URI and the loaded triples added to a graph named by that URI). This has the effect of associating the file or URL of source with the new triples.

  • :verbose - specifies whether or not progress information is printed as triples are loaded. It defaults to nil.

  • :preserve-strings - determine whether or not to save the strings of a triple's object field when the object can be encoded directly into the triple. If true (the default) then the strings will be saved. If false, then only the encoded values will be preserved (this may prevent exact round-trips if data is coerced during the encoding process).

  • :continue-on-error-p - determine how the parser should behave if it encounters an error parsing or adding a triple. It can be one of:

    • nil - meaning to raise the error normally,

    • t - meaning to ignore the error and continue parsing with the next line in the source,

    • a function of four arguments: a parsing helper data structure, the line in the source where the problem was encountered, the condition that caused the error and the arguments (if any) to that condition.

  • You can use continue-on-error-p to print all of the problem lines in a file using something like

    (load-ntriples  
     "source"  
     :continue-on-errorp  
     (lambda (helper source-line condition &rest args)  
       (declare (ignore helper args))  
       (format t "~&Error at line ~:6d - ~a" source-line condition))) 
  • :commit - controls whether and how often commit-triple-store is called while loading. This can be nil for no commit, t for commit at the end of the load or a whole number to commit whenever more than this number of triples have been added.
load-nquads-from-string string  &rest  args  &key  db  graph  verbose  preserve-strings  commit  continue-on-errorp
function

Add the triples in string (in N-Quads format) to the triple store.

See load-nquads for details on the parameters.

load-ntriples source  &rest  args  &key  db  graph  verbose  preserve-strings  external-format  relax-syntax  continue-on-errorp  commit  &allow-other-keys
function

Add triples from source (in N-Triples format) to the triple store.

Returns (as multiple values) the count of triples loaded and the UPI of the graph into which the triples were loaded.

  • source - can be a stream, a pathname to an N-Triples file, a file URI, an HTTP URI or a string that can be coerced into a pathname. Source can also be a list of any of these things. In this case, each item in the list will be imported in turn.

In the case where a list of sources is being loaded and the graph argument is :source, each item in the list will be associated with the URI specifying it. In this case, the second return value will be nil.

The following keyword parameters can be used to control the loading process:

  • :db - specifies the triple-store into which triples will be loaded. This defaults to the value of *db*.

  • :graph - the graph to which the triples from source will be placed. It defaults to nil which is interpreted as db's default graph. If supplied, it can be:

    • a string representing a URIref, or a UPI or future-part encoding a URIref, which adds the triples in source to a graph named by that URI

    • the keyword :source (in which case the source argument will be interned as a URI and the loaded triples added to a graph named by that URI). This has the effect of associating the file or URL of source with the new triples.

  • :verbose - specifies whether or not progress information is printed as triples are loaded. It defaults to nil.

  • :preserve-strings - determine whether or not to save the strings of a triple's object field when the object can be encoded directly into the triple. If true (the default) then the strings will be saved. If false, then only the encoded values will be preserved (this may prevent exact round-trips if data is coerced during the encoding process).

  • :continue-on-error-p - determine how the parser should behave if it encounters an error parsing or adding a triple. It can be one of:

    • nil - meaning to raise the error normally,

    • t - meaning to ignore the error and continue parsing with the next line in the source,

    • a function of four arguments: a parsing helper data structure, the line in the source where the problem was encountered, the condition that caused the error and the arguments (if any) to that condition.

  • You can use continue-on-error-p to print all of the problem lines in a file using something like

    (load-ntriples  
     "source"  
     :continue-on-errorp  
     (lambda (helper source-line condition &rest args)  
       (declare (ignore helper args))  
       (format t "~&Error at line ~:6d - ~a" source-line condition))) 
  • :commit - controls whether and how often commit-triple-store is called while loading. This can be nil for no commit, t for commit at the end of the load or a whole number to commit whenever more than this number of triples have been added.
load-ntriples-from-string string  &rest  args  &key  db  graph  verbose  preserve-strings  continue-on-errorp  commit
function

Add the triples in string (in N-Triples format) to the triple store.

See load-ntriples for details on the parameters.

load-rdf-manifest manifest  destination-db  &key  verbosep
function
Load the file named by manifest, loading the triples from the graphs it references.
load-rdf/xml source  &rest  args  &key  db  base-uri  graph  verbose  commit  use-rapper-p  &allow-other-keys
function

Add triples from the named RDF/XML file to the triple-store. The additional arguments are:

  • :db - specifies the triple-store into which triples will be loaded; defaults to the value of *db*.

  • :base-uri - this defaults to the name of the file from which the triples are loaded. It is used to resolve relative URI references during parsing. To use no base-uri, use the empty string "".

  • :graph - the graph to which the triples from source will be placed. It defaults to nil which is interpreted as db's default graph. If supplied, it can be:

    • a string representing a URIref, or a UPI or future-part encoding a URIref, which adds the triples in source to a graph named by that URI

    • the keyword :source (in which case the source argument will be interned as a URI and the loaded triples added to a graph named by that URI). This has the effect of associating the file or URL of source with the new triples.

  • use-rapper-p - If use-rapper-p is true, then the RDF/XML file will be piped through the open source tool rapper using run-shell-command. Obviously, rapper must be both installed and in your path for this to work. If rapper is not in your path, you can supply it explicitly as the value of use-rapper-p.
load-rdf/xml-from-string string  &rest  args  &key  db  base-uri  graph  commit
function

Treat string as an RDF/XML data source and add it to the triple-store. For example:

   (load-rdf/xml-from-string  
    "<?xml version=\"1.0\"?>  
<rdf:RDF xmlns:rdf=\"http://www.w3.org/1999/02/22-rdf-syntax-ns#\"  
         xmlns:ex=\"http://example.org/stuff/1.0/\">  
  <rdf:Description rdf:about=\"http://example.org/item01\">  
    <ex:prop rdf:parseType=\"Literal\"  
             xmlns:a=\"http://example.org/a#\"><a:Box required=\"true\">  
         <a:widget size=\"10\" />  
         <a:grommit id=\"23\" /></a:Box>  
    </ex:prop>  
  </rdf:Description>  
</rdf:RDF>  
") 

See load-rdf/xml for details on the parser and the other arguments to this function.

RDF in the form of named graphs can also be loaded using load-trix, which understands the TriX format.

load-trix source  &key  db  default-graph  verbose  commit
function

Load a TriX document into the triple store named by db.

  • source - a string or pathname identifying a file, or a stream.

  • :db - specifies the triple-store into which triples will be loaded; defaults to the value of *db*.

  • :verbose - if true, information about the load progress will be printed to *standard-output*.

  • :default-graph - a future-part or UPI that identifies a graph, or nil. If it is non-null, any graphs in the TriX document that are equal to the given URI will be treated as the default graph in db.

  • default-graph can also be the symbol :source, in which case namestring is called on the source and treated as a URI.

  • :commit - controls whether and how often commit-triple-store is called while loading. This can be nil for no commit, t for commit at the end of the load or a whole number to commit whenever more than this number of triples have been added.

We have implemented a few extensions to TriX to allow it to represent richer data:

  • Graphs can be named by <id>, not just <uri>. (A future revision might also permit literals.)

  • The default graph can be denoted in two ways: by providing <default/> as the name for the graph; or by providing a graph URI as an argument to load-trix.

  • In the interest of generality, the predicate position of a triple is not privileged: it can be a literal or blank node (just like the subject and object), not just a URI.
load-trix-from-string string  &key  db  default-graph  verbose  commit
function

Load a string containing data in TriX format into db

  • string - a string containing TriX data

  • :db - specifies the triple-store into which triples will be loaded; defaults to the value of *db*.

  • :verbose - if true, information about the load progress will be printed to *standard-output*.

  • :default-graph - a future-part or UPI that identifies a graph, or nil. If it is non-null, any graphs in the TriX document that are equal to the given URI will be treated as the default graph in db.

  • :commit - controls whether and how often commit-triple-store is called while loading. This can be nil for no commit, t for commit at the end of the load or a whole number to commit whenever more than this number of triples have been added.

See load-trix for more details.

load-turtle source  &key  db  graph  base-uri  use-rapper-p  commit
function

Add triples from the named Turtle file to the triple-store. The additional arguments are:

  • :db - specifies the triple-store into which triples will be loaded; defaults to the value of db.

  • :base-uri - this defaults to the name of the file from which the triples are loaded. It is used to resolve relative URI references during parsing. To use no base-uri, use the empty string "".

  • :graph - the graph to which the triples from source will be placed. It defaults to nil which is interpreted as db's default graph. If supplied, it can be:

    • a string representing a URIref, or a UPI or future-part encoding a URIref, which adds the triples in source to a graph named by that URI

    • the keyword :source (in which case the source argument will be interned as a URI and the loaded triples added to a graph named by that URI). This has the effect of associating the file or URL of source with the new triples.
load-turtle-from-string string  &key  db  graph  base-uri  commit
function

Treat string as a Turtle data source and add it to the triple-store.

See load-turtle for details.

Loading other data formats

If you need to load a format that AllegroGraph does not yet support, we suggest you use the excellent free tool rapper from http://librdf.org/raptor/rapper.html to create either an N-Triples or RDF/XML file that AllegroGraph can load.

Adding Programmatically

You can also add triples to a triple-store programatically with the function add-triple. The three required arguments, representing the subject, predicate, and object of the triple to be added can be expressed either as:

add-triple s  p  o  &key  db  g  preserve-strings
function

Add a triple to the db

The added triple will have the given subject, predicate and object.

  • :db - the triple-store into which to add the triple; this defaults to *db*.

  • :g - the graph for the new triple. If not specified, then the triple will be added to the default-graph (see default-graph-upi).

  • :preserve-strings - controls whether typed-literals will have only their encoded value added to the store or whether an additional hashed version of the triple will also be added. The defaults to nil. For example, if I evaluate:

    (add-triple !ex:gary !ex:age !"46"^^xsd:byte) 
  • then only a single triple will be added to the store. If I instead

    (add-triple !ex:gary !ex:age !"46"^^xsd:byte :preserve-strings t) 

    then there will be two triples added: one whose object is the encoded value 46 and the other whose object is a hashed UPI which points to the string "46"^^<http://www.w3.org/2001/XMLSchema#byte>.

add-triple returns the triple-id of the new triple.

Triple Manipulation

Triple parts: Resources, Literals, UPIs and more

Each triple has five parts (!), a subject, a predicate, an object, a graph and a (unique, AllegroGraph assigned) ID. In RDF, the subject must be a "resource", i.e., a URI or a blank node. The predicate must be a URI. The object may be a URI, a blank node or a "literal". Literals are represented as strings with an optional type indicated by a URI or with a (human) language tag such as en or jp. 4 Blank nodes are anonymous parts whose identity is only meaningful within a given triple-store.

Resources, literals and blank nodes are represented as strings in RDF/XML or N-Triple syntax. AllegroGraph stores these strings in a string dictionary and hashes them to compute a Unique Part Identifier (UPI) for each string. A UPI is a length 12 octet array. One byte of the array is used to identify its type (e.g., is it a resource, a literal, or a blank node). The other 11-bytes are used to either store a hash of the string or to store an encoding of the UPIs contents (see type mapping below for more information about encoded UPIs).

From Strings to Parts

Resources and literals can be denoted with plain Lisp strings in the syntax used in N-Triples files. However this isn't entirely convenient since the N-Triples syntax for literals requires quotation marks which then need to be escaped when writing a Lisp string. For instance the literal whose value is "foo" must be written in N-Triples syntax as "\"foo\"". Similarly -- though not quite as irksome -- URIs must be written enclosed in angle brackets. The string "http://www.franz.com/simple#lastName", passed as an argument to add-triple will be interpreted as a literal, not as the resource indicated by the URI. To refer to the resource in N-Triples syntax you must write "<http://www.franz.com/simple#lastName>". Literals with datatypes or language codes are even more cumbersome to write as strings, requiring both escaped quotation marks and other syntax.

To make it easier to produce correctly formatted N-Triple strings we provide two functions resource and literal. (The ! reader macro, discussed below, can also be used to produce future-parts and UPIs suitable to use as arguments for most of AllegroGraph's API.):

literal string  &key  language  datatype
function

Create a new future-part with the provided values.

  • string - the string out of which to create the part.

  • :language - If provided, language should be a valid RDF language tag.

  • :datatype - If provided, the datatype must be a resource. I.e., it can be a string representation of a URI (e.g., "http://foo.com/") or a future-part specifying a resource (e.g., !<http://foo.com>). If it does not specify a resource, a condition of type invalid-datatype-for-literal-error is signaled. An overview of RDF datatypes can be found in the W3C's RDF concepts guide.

Only one of datatype and language can be used at any one time. If both are supplied, a condition of type datatype-and-language-specified-error will be signaled.

resource string  &optional  namespace
function

Return the provided string as a future-part naming a resource.

If namespace is provided, then string will be treated as a fragment and the future-part returned will be the URIref whose prefix is the string to which namespace maps and whose fragment is string. I.e., if the namespace prefix rdf maps to <http://www.w3.org/1999/02/22-rdf-syntax-ns#>, then the parts created by

(resource "Car" "rdf") 

and

(resource "http://www.w3.org/1999/02/22-rdf-syntax-ns#Car") 

will be the same.

Some examples (we will describe and explain the ! notation below):

> (resource "http://www.franz.com/")  
!<http://www.franz.com/>  
 
> (literal "Peter")  
!"Peter"  
 
> (literal "10" :datatype  
  "http://www.example.com/datatypes#Integer")  
!"10"^^<http://www.example.com/datatypes#Integer>  
 
> (literal "Lisp" :language "EN")  
!"Lisp"@en 

Another issue with using Lisp strings to denote literals and resources is that the strings must, at some point, be translated to the UPIs used internally by the triple-store. This means that if you are going to add a large number of triples containing the same resource or literal and you pass the resource or literal value as a string, add-triple will have to repeatedly convert the string into its UPI.

To prevent this repeated computation, you can use functions like intern-resource or intern-literal to compute the UPI of a string outside of the add-triple loop. The function new-blank-node (or the macro with-blank-nodes) can be used to produce the UPI of a new anonymous node for use as the subject or object of a triple. You can also use encoded ids in place of blank nodes. See Encoded ids for more information

blank-node-p upi
function

Returns true if upi is a blank node and nil otherwise. For example:

> (blank-node-p (new-blank-node))  
t  
> (blank-node-p (literal "hello"))  
nil 
intern-resource uri  &key  namespace  db  upi
function

Compute the UPI of uri, make sure that it is stored in the string dictionary, and return the UPI.

  • :db - specifies the triple-store into which uri will be interned. This defaults to the value of *db*.

  • :upi - if supplied, then this upi will be used to store the uri's UPI; otherwise, a new UPI will be created using make-upi.

  • :namespace - If namespace is provided, then uri will be treated as a fragment and the UPI returned will encode the URIref whose prefix is the string to which namespace maps and whose fragment is uri.

See also resource.

intern-literal value  &key  language  datatype  db  upi
function

Compute the UPI of value treating it as an untyped literal, possibly with a language tag. Ensure that the literal is in the store's string dictionary and return the UPI.

  • :db - specifies the triple-store into which the uri will be interned. This defaults to the value of *db*.

  • :upi - if supplied, then this upi will be used to store the uri's UPI; otherwise, a new UPI will be created using make-upi.

  • :language - if supplied, then this language will be associated with the literal value. See rfc-3066 for details on language tags.

  • :datatype - If supplied, the datatype must be a resource. I.e., it can be a string representation of a URI (e.g., "http://foo.com/") or a future-part specifying a resource (e.g., !<http://foo.com>). If it does not specify a resource, a condition of type invalid-datatype-for-literal-error is signaled. An overview of RDF datatypes can be found in the W3C's RDF concepts guide.

Only one of datatype and language can be used at any one time. If both are supplied, a condition of type datatype-and-language-specified-error will be signaled.

new-blank-node &key  db  upi
function
Create a new blank-node in the triple-store db and return the UPI. If a upi is not passed in with the :upi parameter, then a new UPI structure will be created. db defaults to *db*.
with-blank-nodes blank-nodes  &body  body
macro

This convenience macro binds one or more variables to new blank nodes within the body of the form. For example:

(with-blank-nodes (b1 b2)  
  (add-triple b1 !rdf:type !ex:Person)  
  (add-triple b1 !ex:firstName "Gary")  
  (add-triple b2 !rdf:type !ex:Dog)  
  (add-triple b2 !ex:firstName "Abbey")  
  (add-triple b2 !ex:petOf b1)) 

The following example demonstrates the use of these functions. We use intern-resource to avoid repeatedly translating the URIs used as predicates into UPIs and then use new-blank-node to create a blank node representing each employee and intern-literal to translate the strings in the list employee-data into UPIs. (We could also use literal to convert the strings but using intern-literal is more efficient.)

(defun add-employees (company employee-data)  
  (let ((first-name (intern-resource "http://www.franz.com/simple#firstName"))  
    (last-name (intern-resource "http://www.franz.com/simple#lastName"))  
    (salary (intern-resource "http://www.franz.com/simple#salary"))  
    (employs (intern-resource "http://www.franz.com/simple#employs"))  
    (employed-by (intern-resource "http://www.franz.com/simple#employed-by")))  
    (loop for (first last sal) in employee-data do  
     (let ((employee (new-blank-node)))  
       (add-triple company employs employee)  
       (add-triple employee employed-by company)  
       (add-triple employee first-name (intern-literal first))  
       (add-triple employee last-name (intern-literal last))  
       (add-triple  
        employee salary  
        (intern-literal sal :datatype "http://www.franz.com/types#dollars")))))) 

Note that the only difference between resource and intern-resource is that the former only computes the UPI of a string whereas the latter both does this computation and ensures that the string (and its UPI) are present in the triple-store's string dictionary.

Encoded ids

Encoded ids, described in detail in Encoded ids, are analogous to blank nodes but have certain advantages, such as they can be located using their URI rather than through another node that connects to them (you can only get a handle on a blank node by finding another node which points to it).

register-encoded-id-prefix prefix  format  &key  db
function

Sets up an encoding for prefix, which must be a string, using format, which should be a template (also a string) indicating with brackets, braces, etc. the allowable suffixes. Returns two values: the index of this encoding (a positive integer) and t or nil. t is returned if the encoded-id definition is registered or if the template is modified. nil is returned if the prefix is already registered and the template is unchanged. (This allows the same registration form to be evaluated more than once.) You can redefine the template by calling register-encoded-id-prefix with the same prefix and the new template. But an error will be signaled if you do that after triples have been created using an encoded-id from that prefix.

See Encoded IDs for further information.

Examples:

(register-encoded-id-prefix   
    "http://www.franz.com/managers"     
    "[0-9]{3}-[a-z]{3}-[0-9]{2}")  
RETURNS  
3 ;; you will likely see a different value  
t  
 
If you run the same form again:  
 
(register-encoded-id-prefix   
    "http://www.franz.com/managers"     
    "[0-9]{3}-[a-z]{3}-[0-9]{2}")  
RETURNS  
3 ;; same value as before  
nil ;; meaning no action was taken  
 
If you modify the template (you can only do this when no stored  
triple uses en encoded-id from the prefix):  
 
(register-encoded-id-prefix   
    "http://www.franz.com/managers"     
    "[0-9]{3}-[a-z]{3}-[0-9]{4}") ;; last value is 4 rather than 2  
RETURNS  
3 ;; same value as before  
t ;; meaning made the change 
unregister-encoded-id-prefix prefix  &key  db
function
Unregisters the previously registered encoded id prefix.
encoded-id-prefix-registered-p prefix  &key  db
function
Returns nil if the prefix is not registered and its ID format if it is.
next-encoded-upi-for-prefix prefix  upi  &key  db
function

Returns upi after modifying it to be the next id for the specified encoded-id prefix. The system maintains an internal order of encoded-ids for a prefix. (This order may not be the obvious one, but it will run over all possible ids for a prefix.) upi should be a upi and will be modified. If you do not want to modify an existing upi, create a new one with (make-upi). Do not use this function if you have created an id for prefix directly using the @# encoding as doing so may create duplicate ids (which you may think are distinct).

db defaults to *db*.

Example:

triple-store-user(59): (register-encoded-id-prefix   
    "http://www.franz.com/department"     
    "[0-9]{4}-[a-z]{3}-[0-9]{2}")  
5  
t  
triple-store-user(60): (setq d1  
                            (next-encoded-upi-for-prefix  
                               "http://www.franz.com/department"  
                               (make-upi)))  
{http://www.franz.com/department@#0000-aaa-00}  
triple-store-user(61): (setq bu (make-upi))  
#(49 2 238 3 16 0 0 0 161 210 ...)  
triple-store-user(62): (setq d2  
                         (next-encoded-upi-for-prefix  
                            "http://www.franz.com/department" bu))  
{http://www.franz.com/department@#0000-aaa-01}  
triple-store-user(63): bu        ;; bu is modified  
{http://www.franz.com/department@#0000-aaa-01}  
triple-store-user(64): d2  
{http://www.franz.com/department@#0000-aaa-01}  
triple-store-user(65): (setq d3  
                          (next-encoded-upi-for-prefix  
                             "http://www.franz.com/department" bu))  
{http://www.franz.com/department@#0000-aaa-02}  
triple-store-user(66): bu  ;; bu is modified  
{http://www.franz.com/department@#0000-aaa-02}  
triple-store-user(67): d2  ;; so is d2.  
{http://www.franz.com/department@#0000-aaa-02} 
encoded-id-prefix-in-use-p prefix  &key  db
function
Returns true if there are any triples that use the prefix.
collect-encoded-id-prefixes &key  db
function
Returns a list of 3-element lists of encoded-id specifications. The list elements are the prefix and the template (both strings) and the internal index of the encoded id.
map-encoded-id-prefixes fn  &key  db
function

Gets the list of encoded ids (as returned by collect-encoded-id-prefixes) and applies fn to each element of that list. Since each list element is itself a list of three elements, fn must be or name a function that accepts three arguments.

Example:

(defun foo (&rest args)  
  (dolist (i args)  
  (print i)))  
 
(map-encoded-id-prefixes 'foo)  
 
"http://www.franz.com/employees"  
"[0-9]{3}-[a-z]{3}-[0-9]{2}"  
2  
"http://www.franz.com/managers"  
"[0-9]{4}-[a-z]{3}-[0-9]{2}"  
3  
nil 
Working with Triples

The functions subject, predicate, object, graph, and triple-id provide access to the part UPIs of triples returned by the cursor functions cursor-row and cursor-next-row or collected by get-triples-list.

graph triple  &optional  upi
function

Get the graph UPI of a triple.

Use (setf graph) to change the UPI. If the upi argument is supplied, then the triple's graph will be copied into it. If it is not supplied then a new UPI will be created.

object triple  &optional  upi
function

Get the object UPI of a triple.

Use (setf object) to change the UPI. If the upi argument is supplied, then the triple's object will be copied into it. If it is not supplied then a new UPI will be created.

predicate triple  &optional  upi
function

Get the predicate UPI of a triple.

Use (setf predicate) to change the UPI. If the upi argument is supplied, then the triple's predicate will be copied into it. If it is not supplied then a new UPI will be created.

subject triple  &optional  upi
function

Get the subject UPI of a triple.

Use (setf subject) to change the UPI. If the upi argument is supplied, then the triple's subject will be copied into it. If it is not supplied then a new UPI will be created.

triple-id triple
function
Returns the triple-id of triple
Working with UPIs

You can also create UPIs and triples programmatically; determine the type-code of a UPI, and compare them. You may want to make your own copy of a triple or UPI when using a cursor since AllegroGraph does not save triples automatically (see, e.g., iterate-cursor for more details).

If using encoded-triples or UPIs, then the functions upi->value and value->upi will be essential. You can also check the type of a UPI with either upi-type-code or upi-typep (see type-code->type-name and its inverse type-name->type-code for additional information.)

copy-triple triple  &optional  new
function

Copy a triple.

  • triple - the triple to copy

  • new - (optional) If supplied, then this must be a triple and the copy of triple will be placed into it. Otherwise, a new triple will be created.

This function is useful if you want to keep a reference to a triple obtained from a cursor returned by query functions such as get-triples since the cursor reuses the triple data structure for efficiency reasons.

copy-upi upi
function

Create a copy of a UPI.

  • upi - the UPI to copy.
make-triple
function

Creates and returns a new triple data structure.

The subject, predicate, object and graph slots will have UPIs whose upi-type-code is +unencoded-upi+ and the triple-id will be zero.

make-upi
function
Creates and returns a new UPI data structure. The initial contents of the UPI returned are unspecified.
triple-spog= triple1  triple2
function
Compares triple1 and triple2 and returns true if they have the same contents and false otherwise. Only the subject, predicate, object and graph slots of each triple will be compared. See triple= if you need to compare triples including their IDs.
triple= triple1  triple2
function
Compares triple1 and triple2 and returns true if they have the same contents and triple-id and false otherwise. See triple-spog= if you need to compare triples regardless of their IDs.
upi->value upi  &key  db  errorp
function

Decodes UPI and returns the value, the type-code and any extra information as multiple values.

  • upi - the UPI to decode

  • :db - specifies the triple-store from which the UPI originates.

The value, type-code, and extra information are interpreted as follows:

  • value - a string, integer, or float representing the contents of the UPI
  • type-code - an integer corresponding to one of the defined UPI types (see supported-types for more information). You can use type-code->type-name to see the English name of the type.
  • extra - There are three disjoint sets of literals in RDF:
    1. "simple literals" -- a string.
    2. literals with datatypes
    3. literals with language annotations.
  • A literal cannot have both a language and a datatype; if a literal has type-name :literal-typed (i.e., its type-code is 2), then the extra value is a datatype URI; if its type-code is 3, then extra will be the language code.

Examples:

> (upi->value (value->upi 22 :byte))  
22  
18  
nil  
 
> (upi->value (upi (literal "hello" :language "en")))  
"hello"  
3  
"en" 

See value->upi for additional information.

upi->values upi  &key  db
function
Decodes UPI and returns the multiple values encoded by the upi. This is intended for UPIs that intrinsically carry more than a single value, such as longitude and latitude for a geospatial UPI. Operates the same as upi->value for hashed UPIs.
upi-type-code upi
function
Returns the type-code of a UPI. This is a one-byte tag that describes how the rest of the bytes in the UPI should be interpreted. Some UPIs are hashed and their representation is stored in a string-table. Other UPIs encode their representation directly (see upi->value and value->upi for additional details).
upi-typep upi  type-code
function

Returns true if upi has type-code type-code.

  • upi - a UPI

  • type-code - either a numeric UPI type-code or a keyword representing a type-code. See type-code->type-name and type-name->type-code for more information on type-codes and their translations.
upi< upi-1  upi-2
function
Compare upi-1 and upi-2 bytewise and return true if upi-1 is less than upi-2.
upi<= u1  u2
function
Compare upi-1 and upi-2 bytewise and return true if upi-1 is less or equal than upi-2.
upi= upi-1  upi-2
function
Test if two UPIs are equal. UPIs are equal if they are eq (i.e., the same object) or if they are equal as octet arrays.
upip thing
function
Returns true if thing appears to be a UPI. Recall that every UPI is a octet array 12 bytes in length but not every length 12 octet array is a UPI. It is possible, therefore, that upip will return true even if thing is not a UPI.
value->upi value  encode-as  &optional  upi
function

Returns a UPI that encodes value using type encode-as.

encode-as can be a type-code or a type-name (see supported-types and type-name->type-code for details. If a upi keyword argument is not supplied, then a new UPI will be created. See upi->value for information on retrieving the original value back from an encoded UPI.

Note that value->upi can not create hashed UPIs (i.e., UPIs whose value must be stored in the string dictionary. To create these, use intern-resource and intern-literal.

Comparing parts of triples
graph-part= triple  part
function
Return true if and only if the graph field of triple is the same as part using part= for comparison. The part may be a UPI, a string (in N-Triples syntax) that can be converted to a UPI, or a future-part. See graph-upi= if you know that you will be comparing UPIs.
object-part= triple  part
function
Return true if and only if the object field of triple is the same as part using part= for comparison. The part may be a UPI, a string (in N-Triples syntax) that can be converted to a UPI, or a future-part. See object-upi= if you know that you will be comparing UPIs.
predicate-part= triple  part
function
Return true if and only if the predicate field of triple is the same as part using part= for comparison. The part may be a UPI, a string (in N-Triples syntax) that can be converted to a UPI, or a future-part. See predicate-upi= if you know that you will be comparing UPIs.
subject-part= triple  part
function
Return true if and only if the subject field of triple is the same as part using part= for comparison. The part may be a UPI, a string (in N-Triples syntax) that can be converted to a UPI, or a future-part. See subject-upi= if you know that you will be comparing UPIs.
graph-upi= triple  upi
function
Return true if and only if the graph field of triple is the same as upi using upi= for comparison. The upi must be a UPI. See graph-part= if you need to compare future-parts or convert strings into UPIs.
object-upi= triple  upi
function
Return true if and only if the object field of triple is the same as upi using upi= for comparison. The upi must be a UPI. See object-part= if you need to compare future-parts or convert strings into UPIs.
predicate-upi= triple  upi
function
Return true if and only if the predicate field of triple is the same as upi using upi= for comparison. The upi must be a UPI. See predicate-part= if you need to compare future-parts or convert strings into UPIs.
subject-upi= triple  upi
function
Return true if and only if the subject field of triple is the same as upi using upi= for comparison. The upi must be a UPI. See subject-part= if you need to compare future-parts or convert strings into UPIs.
Future-Parts and UPIs

Future-parts (which are discussed in detail in their own section) can take the place of UPIs in many of the functions above. For efficiencies sake, functions like upi= assume that they are called with actual UPIs. AllegroGraph provides more general functions for the UPI only variants when it makes sense to do so. For example, future-part= works only with future-parts whereas part= works equally well with any combination of UPIs, future-parts or even strings.

future-part
structure
A future-part defers namespace resolution and string interning until it is required. Along with the !-reader, they make writing and using UPIs significantly more readable and easier.
future-part-extra future-part
function

Return whatever is extra in the string of the future-part. The meaning of extra depends on the type of the part:

  • resource - the prefix (if applicable)
  • literal-typed - the datatype
  • literal-language - the language
  • other - the value will always be nil

See future-part-type and future-part-value if you need to access the part's other parts.

future-part-type future-part
function
Returns the name (a keyword) of the future-part. See type-name->type-code for a list of names.
future-part-value future-part
function

Return the value of the future-part. The meaning of value depends on the type of the part:

  • resource - the namespace (if any)
  • literal-typed - the literal without the datatype
  • literal-language - the literal without the language
  • other - the value will always be nil

See future-part-type and future-part-extra if you need to access the others parts (no pun intended) of the part.

future-part= part-1  part-2
function
Test if two future-parts are equal. Future-parts are equal if they both resolve to the same UPI or, if they cannot yet be resolved, if their strings are string=.
make-future-part &key  value  type  extra
function

Create a 'future' part that will intern itself in new triple stores as necessary.

  • :type - one of :node, :literal, :literal/lang or :literal-typed.

  • :value and :extra - the contents of the part. Their meaning changes depending on type. If type is:

    • :node - value is the string representation of the URIref and extra must be nil.

    • :literal - value is the string contents of the literal and extra must be nil.

    • :literal/lang - value is the string contents of the literal and extra specifies the language.

    • :literal-typed - value is the string contents of the literal and extra specifies the datatype.
part= part-1  part-2
function
Compare any two things that can be coerced into UPIs as UPIs.
upi future-part  &key  errorp
function

Returns the UPI associated with the future-part future-part.

If the future-part uses namespaces, then calling upi will resolve the namespace mapping. An error will be signaled if upi is called and there is no namespace mapping defined. You can use the errorp keyword parameter to disable the error and return nil instead.

Querying Triples

You can get triples out of a triple-store as a list or a cursor. The list structure is convenient but unwieldy if your query returns millions of triples (since every triple must be returned before you will see any of them). A cursor is like a database cursor from the RDBMS-world. It lets you traverse through the results of your query one step at a time.

Cursors

Cursors supply the functions cursor-next and cursor-row for basic forward iteration. For convenience we include cursor-next-row which advances the cursor and returns the next row immediately. Cursors reuse the triple data-structure as they move through the result set so if you want to accumulate triples, make sure to use the copy-triple function.

cursor-next cursor
function

Moves cursor forward to the next record in the collection.

Returns t if there was another row on which to move and nil if the cursor is exhausted. I.e., if cursor-next returns t, then cursor-row will return a record.

Note that cursor-row and cursor-next are lower-level cursor manipulation routines. You may be better served by using collect-cursor, count-cursor, map-cursor and iterate-cursor.

cursor-next-row cursor
function

Returns the next record from the cursor. Data structures holding the record might be reused, so if you want to hold onto, for example, a triple for use after you advance the cursor, use the function copy-triple to make a copy of the value returned by cursor-next-row

Note that cursor-row, cursor-next and cursor-next-row are lower-level cursor manipulation routines. You may be better served by using collect-cursor, count-cursor, iterate-cursor and map-cursor.

cursor-row cursor
function

Returns the record that cursor is currently pointing at.

If the cursor is exhausted, then cursor-row returns nil.

Note that cursor-row and cursor-next are lower-level cursor manipulation routines. You may be better served by using collect-cursor, count-cursor, map-cursor and iterate-cursor.

There are several natural cursor idioms, the following functions handle many of them. We suggest building your own functions using these as building blocks since there may be internal optimizations made possible only through these. 5

collect-cursor cursor  &key  transform
function

Iterate over cursor and collect a list of its rows.

The :transform keyword can be used to modify the rows as they are collected. Rows may be re-used so the caller must ensure that copies are made if necessary. As an example, consider the difference between

(collect-cursor (get-triples)) 

and

(collect-cursor (get-triples) :transform #'copy-triple) 

Because the rows are re-used, the former will return a list of the same triple repeated as many times as there are triples in the store. The later will return a list of all of the triples in the store (which is probably what you intend!).

Note that the idiom above is so common that collect-triples is already defined.

count-cursor cursor
function
Returns the number of triples remaining in a cursor and also exhausts cursor in the process.
iterate-cursor (var  cursor  &key  count)  &body  body
macro

Iterate over the triples in cursor binding var to each triple. Use the :count keyword with a fixnum value to limit the maximum number of triples that iterate-cursor visits. The binding to var is the same eq triple in each iteration. Make sure to use copy-triple if you are retaining any of the triples that you visit.

The macro creates a block nil around its expansion. If iteration finishes normally by satisfying the count or by exhausting the cursor, iterate-cursor returns the number of triples visited.

map-cursor count  fn  cursor  &rest  args
function

Iterate over the triples in cursor while applying the function fn to each one. Use the count keyword to limit the maximum number of triples that map-cursor visits. In each iteration fn is applied to the current triple and to the list of arguments in args.

map-cursor reuses a single triple as it iterates so make sure to use copy-triple if you want to retain any of the triples you visit.

The function returns the number of triples visited.

collect-triples cursor
function

Collect the triples from cursor into a list.

Uses collect-cursor with a transform of copy-triple to return a fresh list of the triples in cursor. collect-triples assumes that cursor iterates over triples. Returns the list.

There are many other functions that either query a triple-store directly or return information about a triple-store. For example, count-query determines the number of matches to a query very quickly (because it doesn't bother to iterate over them) and pprint-subject lets you easily explore more about particular resources.

count-query &key  s  p  o  g  s-end  p-end  o-end  g-end  db
function

Return the number of triples in db matching the query. db defaults to *db*.

The keyword arguments s, p, o, g, s-end, p-end, o-end and g-end are used as in get-triples.

Note that the count includes deleted triples, encoded triples and unencoded triples. Finally, no filter can be used with a count query.

estimated-count-query &key  s  p  o  g  s-end  p-end  o-end  g-end  db
function

Return an estimated count of the number of triples in db matching the query. db defaults to *db*.

The keyword arguments s, p, o, g, s-end, p-end, o-end and g-end are used as in get-triples.

Note that the estimated count may include deleted triples.

get-triple &key  s  p  o  g  db  return-encoded-triples  return-non-encoded-triples  filter  triple
function

Searches in db for a triple matching the search pattern specified by s, p, o and g (perhaps restricting the search to encoded triples only if return-non-encoded-triples is nil, or to non-encoded triples only if return-encoded-triples is nil). filter, if supplied, should be a predicate of one parameter, a triple. If the filter function returns nil on a possible match, that match is skipped. Searching is stopped when a match (which passes the filter, if supplied) is found.

If a match is found and triple is supplied (it must be of type triple), the match is copied to triple and triple is returned. If triple is not supplied, a new triple is created (as if with make-triple), the match is copied onto the new triple and the new triple is returned.

If no match is found, nil is returned.

db defaults to *db*.

get-triple-by-id id  &key  db  triple
function

Locate the triple whose triple-id is id and return it. Return nil if there is no triple associated with id. The keyword argument db can be used to specify the triple-store in which to search. It defaults to the current triple-store, *db*. get-triple-by-id allocates a new triple (using make-triple). You can prevent this by passing in your own triple using the keyword argument :triple. The data in the triple you pass in will by overwritten.

Note that this function can return a triple and then later (after you've done a commit or rollback) return nil because another user removed it. These removals are transactional, which means the behavior can change (from returning a triple to returning nil) only after a commit or a rollback.

get-triples &key  s  p  o  g  s-end  p-end  o-end  g-end  db  return-encoded-triples  return-non-encoded-triples  filter
function

Query a triple store for triples matching the given subject, predicate, object, and graph. These can be specified either as UPIs, future-parts, strings in N-Triple format, or the wildcard nil. Returns a cursor object that can be used with cursor-row, cursor-next-row and other cursor functions.

The following example finds every triple that starts with !ub:Kevin.

> (add-triple !ub:Kevin !ub:isa !"programmer")  
8523645  
> !ub:Kevin  
!<http://www.w3.org/1999/02/22-rdf-syntax-ns#Kevin>  
> (get-triples :s !ub:Kevin)  
#<row-cursor #<triple-record-file @ #x13c1a87a> 2019 [1 - 2018] @  
  #x14942bba>  
> (print-triples *)  
<http://www.w3.org/1999/02/22-rdf-syntax-ns#Kevin>  
  <http://www.w3.org/1999/02/22-rdf-syntax-ns#isa>  
  "programmer" . 

The function get-triples takes the following arguments:

  • s, p, o, g - specify the query pattern. Use nil as a wildcard. Each of these can be a UPI, a future-part, or a string that can be converted into a part. These can also take on the value :minimum if a corresponding ?-end parameter is specified. If :minimum is used, then the range query will run from the smallest value up to the ending value.

  • s-end, p-end, o-end, g-end - Allows for range queries over encoded triples (triples whose parts are encoded UPIs). Each ?-end parameter may only be used in conjunction with its corresponding starting value parameter. Each of these can be a UPI, a future-part, or a string that can be converted into a part. These can also take on the value :maximum if a corresponding starting parameter is specified. If :maximum is used, then the range query will run from the starting value up to the maximum value in the triple-store.

  • The :db keyword argument specifies the triple store to query and defaults to the value of *db*.

  • filter - if supplied, the filter should be a predicate of one parameter, a triple. If the filter function returns nil, then the triple will not be included in the result set.

  • return-encoded-triples - If true, then get-triples returns triples with encoded parts; i.e., triples that use directly encoded UPIs rather than strings stored in the dictionary. The default value is true.

  • return-non-encoded-triples - if true, then get-triples will return triples all of whose UPIs are stored as strings. This is set to true unless overridden.

The return value is a cursor object. The functions collect-cursor, count-cursor and cursor-next-row can be used to step through the cursor.

If it can be determined that one of the search parameters is not interned in the triple-store, then get-triples will return a null-cursor (i.e., a cursor that has no rows) and a second value of :part-not-found.

Note on tracing: get-triples can be inlined by the compiler, so if you want to trace get-triples calls, you should declare the function notinline. Alternatively, you can trace db.agraph::get-triples-positional, the non-exported internal function, which the compiler macro uses.

get-triples-list &rest  args  &key  s  p  o  g  s-end  p-end  o-end  g-end  limit  db  return-encoded-triples  return-non-encoded-triples  cursor  filter
function

Query a triple store for triples matching the given subject, predicate, object, and graph, specified either as part IDs (UPIs), future-parts, strings in N-Triples format, or the wildcard nil. Returns a list of matching triples. The get-triples-list function supports a multitude of options:

  • db - This keyword argument specifies the triple store to query and defaults to the value of *db*.

  • s, p, o, g - controls the actual query pattern. Use nil as a wildcard.

  • s-end, p-end, o-end, g-end - Allows for range queries over encoded triples (triples whose parts are encoded UPIs). Each ?-end parameter may only be used in conjunction with its corresponding starting value parameter and only one range may be specified per query.

  • cursor - if cursor is supplied then AllegroGraph will use it to return more triples rather than building a new cursor (i.e. the other query parameters will be ignored).

  • limit - This keyword argument can be used to place a cap on the maximum number of triples returned. It defaults to the value of the special variable *get-triples-list-limit*. If supplied, then
    get-triples-list will return no more than :limit triples.
    If it is nil, then get-triples-list will return all of the triples found by the query. Warning: setting limit to nil can cause get-triples-list to return every triple in the triple-store;
    this will take a very long time if the triple-store is large.

  • return-encoded-triples - If true, then get-triples-list returns triples with encoded parts; i.e., triples that use directly encoded UPIs rather than strings stored in the dictionary. The is set to true unless overridden.

  • return-non-encoded-triples - if true, then get-triples-list will return triples all of whose UPIs are stored as strings. This is set to true unless overridden.

  • Note that most of the arguments to get-triples-list do not make sense when reasoning is turned on. AllegroGraph will signal an error if you try to combine reasoning with other parameters that it cannot use.

If it can be determined that one of the search parameters is not interned in the triple-store, then get-triples-list will return immediately return nil and :part-not-found as multiple values.

get-triples-list will return

  • a list and a cursor if triples are found.

  • nil and t as multiple values if all search parameters are interned and no triples are found.

  • nil and :part-not-found as multiple values if one of the search parameters is not interned in the triple-store.

If nil is returned, the second return value allows you to determine why.

pprint-object part  &key  maximum-depth  format  db
function

Print information about part down to a maximum depth of maximum-depth using the format format. Triples for which part is an object and their children will be printed. See part->string for information about part printing. See pprint-subject to display information based on objects rather than subjects.

Note that pprint-object is designed to show the actual triples in a store. It uses ground-triple-store to remove one layer of reasoning from db. db defaults to *db*.

pprint-subject part  &key  maximum-depth  format  db
function

Print information about part down to a maximum depth of maximum-depth using the format format. Triples for which part is a subject and their children will be printed. See part->string for information about part printing. See pprint-object to display information based on subjects rather than objects.

Note that pprint-subject is designed to show the actual triples in a store. It uses ground-triple-store to remove one layer of reasoning from db. db defaults to *db*.

triple-exists-p s  p  o  &key  g  db  filter
function
Returns true if a triple matching s, p and o (and optionally g) can be found in the designated triple-store. If left unspecified, the triple-store designated by *db* will be searched. This is useful when you care only about the presence of a triple and not its contents. If you want to use the triple, then use get-triple instead.
*get-triples-list-limit*
variable

The default number of triples to return from get-triples-list.

If nil, then all triples will be returned. This defaults to nil.

Range Queries

Both get-triples and get-triples-list support fast inclusive range queries using AllegroGraph's encoded data types. Though RDF requires that subjects be resources or blank nodes and predicates by resources, AllegroGraph allows you to store and query for encoded data types in any of a triple's fields. 6 In each case, to execute a range query, you must supply a starting and ending encoded UPI. You can only make a range query on a single field in each query but you can include other fields which act as additional filters. For example:

> (get-triples :o (value->upi "12:01" :time)  
               :o-end (value->upi "12:59" :time)  
               :p !ex:startTime) 

will return a cursor that iterates only all of the triples whose predicate is !ex:startTime and whose object falls on or between 12:01 and 12:59. 7

SPARQL Overview

AllegroGraph includes twinql, an implementation of the powerful SPARQL query language. SPARQL is covered both in this reference guide and in the SPARQL reference and tutorial.

Prolog Select Overview

With pure Lisp as the retrieval language, you use a combination of functional and procedural approaches to query the database. With Prolog, you can specify queries in a much more declarative manner. Allegro CL Prolog and AllegroGraph work very well together. The reference guide includes more details and the tutorial provides an introduction to using Prolog and AllegroGraph with many complete examples.

Query analysis

There are tools for seeing what indexes would be best to use when answering a query and what indexes are actually used (these may not be the same because the optimal indexes may not be available). See Query Analysis. The following function does such analysis:

analyze-query-index-usage query  &key  db  language  technique  duration
function

Analyze the index usage for query using technique.

Determine which indices in query language language will access using technique technique.

Currently supports only SPARQL with one of the following techniques:

  • static - analyze the query algebracially; fast but can be wrong.
  • executed - execute the query and monitor actual index usage; correct but can be slow.

The results are returned as lists of sub-lists where each sub-list looks like:

(:desired (flavor mode) :actual (flavor mode)) 

where

  • flavor - an index flavor
  • mode - optimal, suboptimal or full

The first part of the list (:desired) is what AllegroGraph would most like assuming that every index flavor was available. The second part (:actual) is what AllegroGraph actually got given the current triple-store.

If supplied, duration specifies how long to run the analysis. The standard behavior is to run for as long as necessary.

RDFS++ Reasoning Overview

See the Reasoner tutorial for more details on using AllegroGraph's RDFS++ reasoner. It works with get-triples, get-triples-list, SPARQL and the Prolog q functor.

SPOGI Caching

If you are running queries that gather a lot of triples about each particular subject, then AllegroGraph's SPOGI cache can help you to achieve significant performance improvements. Once enabled, the cache operates in the background and saves the triples associated with every subject / predicate pair that it encounters. This means that when you first execute a query like:

> (get-triples-list :s !ex:gary :p !ex:telephoneNumbers) 

AllegroGraph will scan all of the triples associated with the subject !ex:gary and cache the list of triples for each predicate. If you then execute another query involving !ex:gary, it will return the results almost instantly. The cache tracks changes to your triple-stores and will recompute its results whenever it changes.

spogi-cache-size db
function

Gets the size of the spogi cache. .

The cache size determines the maximum number of subjects that can be help in the cache at once. The value of *default-spogi-cache-size* is used for newly created triple-stores.

You can use (setf spogi-cache-size) to change the size of the cache. For example:

(setf (spogi-cache-size *db*) 10000) 
spogi-cache-enabled-p db
function

Returns true if the spogi cache is enabled for db.

You can use (setf spogi-cache-enabled-p) to turn caching on and off. When using the setf method, you can use a boolean or a whole number. If you use a whole number, it has the same effect as turning caching on and setting the spogi-cache-size to that value. For example:

(setf (spogi-cache-enable *db*) t)	;turn the cache on  
(setf (spogi-cache-enable *db*) 5000)	;turn the cache on and set it to 5000 
*default-spogi-cache-enabled*
variable

Determines whether the SPOGI-cache is enabled for newly created triple-stores.

By default, this variable will be set to nil so that SPOGI-caching will not be in effect for newly created triple-stores. You will need to experiment with your data and queries to determine whether or not the extra RAM used by the cache is worth the performance improvements.

*default-spogi-cache-size*
variable

Sets the default size of the SPOGI-cache for newly created triple-stores.

By default, the SPOGI-cache-size is 100000. Note that caching will not be enabled for new triple-stores unless *default-spogi-cache-enabled* is set to true.

Modifying triples

AllegroGraph stores triples in five parts: subject, predicate, object, graph and ID. (Triple refers to the subject, predicate, and object.) RDF operations ignore the ID but it is still there under the hood. Once a triple has been added it cannot be changed (there is no in-place modification of triples). You make modifications by deleting (see delete-triple) and re-adding (see add-triple) a triple but the new triple will have a different ID from the old one. This is true regardless of the client used to effect the changes.

Note also that if original triple is committed and you evaluate:

(delete-triple id)      ;; deleting the original triple  
(add-triple s p o :g g) ;; adding the modified triple  
(commit-triple-store) 

or

(add-triple s p o :g g) ;; adding the modified triple  
(delete-triple id)      ;; deleting the original triple  
(commit-triple-store) 

then every other user will see the original triple until the commit is processed and they rollback and then they will see the new triple. Other users will never see neither triple nor both triples. So from the point of view of other users, the triple will simply be modified.

Note that SPARQL update has no notion of IDs (because it is an RDF language). Even though a SPARQL update may not be explicit about implementing triple modification by deleting and adding, that is what is in fact happening so there is no efficiency gain from using update though the code might be easier to understand.

Again, the new triple will always have a different id regardless of how the "modification" is actually done.

Deleting triples

AllegroGraph lets us delete triples. The function delete-triples deletes triples from the triple store, using the same query syntax as get-triples. Once a transaction has committed (see commit-triple-store), there is no way to restore a deleted triple. If you have not committed, then you can rollback (see rollback-triple-store) the transaction to restore any deleted triples and also undo any other changes made since the last commit (including additions).

delete-triple id  &key  db
function

Deletes the triple whose id is id.

The :db keyword argument specifies the triple store in which the triple should be deleted. It defaults to *db*.

Note delete-triple requires the i index in order to run efficiently. See add-index for details on how to manage a triple-store's indices.

delete-triples &key  s  p  o  g  s-end  p-end  o-end  g-end  filter  db
function

Delete triples matching the given triple-pattern.

The triple-pattern is specified as in get-triples where each of s, p, o, g, s-end, p-end, o-end and g-end may be specified as part IDs, strings in N-Triples format, or the wildcard nil. Returns the number of triples deleted.

The :db keyword argument specifies the triple store to query and defaults to the value of *db*.

The :filter keyword can be used to limit the triples deleted. If specified it should be a function predicate of one argument, a triple, which returns true if and only if the triple should be deleted.

Deleting Duplicate Triples

Duplicate triples take up disk space and cause queries to take longer. To remove duplicates from a store, run delete-duplicate-triples and commit. You can also arrange to have duplicates removed at commit time, as specified by duplicate-suppression-strategy.

If your store uses any triple-id based reification, note that only the triple with the lowest triple-id will be left alive after running this operation. In releases prior to 4.4, duplicate deletion was done in the background. Now it is only done with the explicit call to delete-duplicate-triples.

delete-duplicate-triples &key  db  mode
function
Deletes duplicated triples from a store inside the current transaction. The mode keyword argument specifies the definition of a triple: It can be :spog or :spo.
duplicate-suppression-strategy &key  db
function

Returns a keyword (or nil) specifying whether the store suppresses duplicate triples at commit time.

This setting specifies the way AllegroGraph manages duplicates at commit time:

  • nil - duplicate triples are not suppressed, they are committed as usual.

  • :spo - do not commit any triples whose subject, predicate and object parts are duplicates of other triples.

  • :spog - do not commit any triples whose subject, predicate, object and graph parts are duplicates of other triples.

With a non-nil setting, calling commit-triple-store will prevent any additional duplicate triples from being added to the store at commit time.

Changing this setting does not affect any potential duplicates that may already be present in the store. See delete-duplicate-triples.

Serializing Triples

There are several methods by which you can create a textual representation of your triple-store:

The print-triples function provides a simple mechanism to output triples to *standard-output* or a stream. It's easy to build an export function on top of it:

(defun export-triples (triples file)  
  (with-open-file (output file  
                          :direction :output  
                          :if-does-not-exist :create  
                          :if-exists :error)  
    (print-triples triples  
                   :limit nil :stream output :format :ntriple))) 

The other techniques provide more control over the output format.

print-triple triple  &key  format  stream  db
function

Print a triple returned by cursor-next-row or get-triples-list.

Returns the triple printed.

The keyword argument :format, which defaults to :ntriples, specifies how the triple should be printed. The value :ntriples specifies that it should be printed in N-Triples syntax. The value :long indicates that the string value of the part should be used. And the value :concise causes it to use a more concise, but possibly ambiguous, human-readable format.

print-triples triple-container  &key  limit  format  stream  db
function
Display the triples in triple-container which can be either a triple store object, a list of triples such as is returned by get-triples-list, or a cursor such as is returned by get-triples. If the keyword argument :limit is supplied, then at most that many triples will be displayed. The :format keyword argument controls how the triples will be displayed, in either :ntriples, :long, or :concise format. It defaults to :ntriples. The stream` argument can be used to send output to the stream of your choice. If left unspecified, then output will go to standard-output.
serialize-ntriples from  to  &key  nquads  base-uri  if-exists  if-does-not-exist
function

Write an N-Triples representation of from to to. from should be a triple store, a list of triples, or a cursor. to should be a stream, a file path to open for writing, t (meaning *standard-output*), or nil (return a string).

if-exists and if-does-not-exist are arguments to the Common Lisp open function which apply when to is a filename.

If base-uri is provided, it is used to generate a “baseURI:” comment at the start of the serialization. This comment is expected by TopBraid Composer.

base-uri can be t, in which case a generated UUID URN is used, or a string, UPI, or future-part naming a URI. It can also be a triple-store instance; in this case, a UUID URN associated with the store is used. For persistent stores this value should remain constant.

The nquads argument causes this function to emit N-Quads instead of N-Triples. It is used by the function serialize-nquads.

In other respects this function behaves like print-triples.

serialize-nquads from  to  &key  base-uri  if-exists  if-does-not-exist
function

Just like serialize-ntriples but emits quads in N-Quads format.

See serialize-ntriples for more information on the arguments to this function.

serialize-rdf/xml from  to  &key  memoize-abbrev-lookups-p  prepare-namespaces-p  output-types-p  nestp  error-on-invalid-p  indent  xml-base  if-exists  if-does-not-exist
function

Write from, which should be a triple store, a list, or a cursor, to to, which should be a stream, file path, t (print to *standard-output*), or nil (return a string).

If from is a triple-store or a list of triples, and prepare-namespaces-p is t, it is first traversed to build a hash of namespaces to prefixes for all properties in the store. The value of db.agraph::*namespaces* is used as a seed.

If you can ensure that every property used in the triple-store has a defined prefix, you can pass nil for prepare-namespaces-p to gain a speed improvement from omitting this phase.

If error-on-invalid-p is t, the serializer will throw an error if it encounters a type or predicate that it cannot abbreviate for RDF/XML. Ordinarily a local namespace declaration is included in such a property element instead.

If a namespace prefix table is built, it will be returned as the second value.

If from is a cursor, it cannot be traversed multiple times, so prepare-namespaces-p is ignored. If a property is encountered that cannot be abbreviated with the current available prefixes, an error will be signaled or a local namespace declaration generated instead.

Such local namespace declarations are verbose, and so providing a thorough namespace hash is preferred.

If a predicate cannot be abbreviated as an XML QName, your graph cannot be serialized in RDF/XML. An error will be raised in this case.

if-exists and if-does-not-exist are arguments to the Common Lisp open function which apply when to is a filename.

xml-base can be t, in which case a generated UUID xml:base attribute is written into the serialized RDF/XML, or a string, UPI, or future-part naming a URI. It can also be a triple-store instance; in this case, a UUID URN associated with the store is used.

For persistent stores this value should remain constant.

If xml-base is nil, no xml:base attribute will appear in the output.

Note that xml-base values will be escaped, but they will not be checked for validity (i.e., you can successfully provide an invalid URI to this function).

If memoize-abbrev-lookups-p is t, a upi-hash-table is built to store the mappings between resources in the store and string abbreviations. This hash-table will contain as many entries as there are types and properties in the data to be serialized. For some datasets disabling this caching will yield a significant reduction in space usage in exchange for a possible loss of speed.

If indent is non-nil, then it specifies the initial indentation of elements.

If nestp is t, then (subject to the order of triples in the triple store) some nesting of RDF/XML elements will be applied. nestp of nil will cause a flat tree to be produced, where each resource is its own top-level element. You should not rely on this alone for the purposes of processing RDF data with XML tools.

If output-types-p is t, then additional queries will be performed for each resource to decorate the RDF/XML with types. If nil, then rdf:type elements alone will be used.

Because of rounding errors and other conversion issues, the RDF/XML serializer is not guaranteed to round-trip correctly with value-encoded literals (i.e. literals stored as encoded UPIs. See the reference guide section on type mapping for more information on encoded UPIs).

serialize-rdf-manifest source  directory  &key  verbosep  single-stream-p
function

Serialize source according to exchange-of-named-rdf-graphs. Serialization will probably open at least as many files as there are graphs in the source.

Returns the manifest path and the number of graphs saved.

If single-stream-p only a single file is open at any one time. This is slower, but guaranteed not to fail with large numbers of graphs.

serialize-rdf-n3 from  to  &key  indent  if-exists  if-does-not-exist  error-on-invalid-p
function

Write from, which should be a triple-store, a list, or a cursor, to to, which should be a stream, file path, t (print to *standard-output*), or nil (return a string).

  • :if-exists and :if-does-not-exist - apply when from is a filename. They are passed along to the open function are used there.

  • :indent - if non-nil, then this specifies the initial indentation of elements.
serialize-trix source  output  &key  constraints  default-graph  if-exists  if-does-not-exist
function

Write source, which should be a triple store, cursor, or list of triples, to output. output must be t (equivalent to *standard-output*), nil (which returns the serialization as a string), a string or pathname, which will be opened for output respecting the if-exists and if-does-not-exist arguments, or a stream.

The resulting output is produced according to the value of constraints. This must be one of the following symbols:

  • :canonical: each graph appears once in the output. All graphs and triples are sorted lexicographically. All graphs are named, even when a name must be generated.

  • :collated: each graph appears once in the output.

  • :lenient (or nil, the default): graphs can appear multiple times in the output. No sorting occurs. load-trix can consume this, but not all other TriX parsers necessarily can.

The value of the default-graph argument applies when the default graph has to be serialized. If it is nil, the default, the AllegroGraph-specific <default/> element is used. If it is non-null, it should be a URI string or a valid part that will be substituted into the output.

Indexing Triples

AllegroGraph automatically manages a set of indices that it uses to speed queries. See Triples Indices.

The standard indices that AllegroGraph uses are spogi, posgi, ospgi, gspoi, gposi, and gospi. Here are the sort of queries where these different indices are applicable:

spogi    get-triples _s_, ---, ---, ---  
         get-triples _s_, _p_, ---, ---  
         get-triples _s_, _p_, _o_, ---  
 
posgi    get-triples ---, _p_, ---, ---  
         get-triples ---, _p_, _o_, ---  
 
ospgi    get-triples ---, ---, _o_, ---  
         get-triples _s_, ---, _o_, ---  
 
gspoi    get-triples _s_, ---, ---, _g_  
         get-triples _s_, _p_, ---, _g_  
         get-triples _s_, _p_, _o_, _g_  
 
gposi    get-triples ---, _p_, ---, _g_  
         get-triples ---, _p_, _o_, _g_  
 
gospi    get-triples ---, ---, _o_, _g_  
         get-triples _s_, ---, _o_, _g_ 

Having the standard set available means that any query can be answered quickly. There are cases, however, when some of these become unnecesary. For example, if you are not using the graph for any of your triples, then none of the g indices will ever by used in a query and maintaining them will use unnecessary CPU cycles and waste disk space.

You can customize the indices used in a triple-store when it is created (with the with-indices option of create-triple-store). You can also use the following commands to view and modify the triple-store's indices:

add-index flavor  &key  db
function

Ensure that db indexes triples on flavor. db defaults to *db*.

Note that the new index will not be available until after commit (see commit-triple-store).

drop-index flavor  &key  db
function

Ensure that db does not index triples on flavor. db defaults to *db*.

Note that the index will not be dropped until after commit.

triple-store-indices &key  db
function

Return a list of the index flavors currently active for db. db defaults to *db*.

Use add-index and drop-index to manipuate a store's active indices.

As more triples are added, the index can become less efficient. The system performs index optimization in the background. The optimize-indices function can force this optimization to occur immediately.

optimize-indices &key  level  wait-p  db
function

Optimize the indices in the db at aggressiveness level. level must be a non-negative integer or nil. nil means the same as not specifying a value: use the system default. A larger integer means be more aggressive, which can take more time. Specific integers translate to specific action, and currently only 0 and 1 are meaningful, with higher values being equivalent to 1.

If wait-p is true, don't return until the operation is finished.

db defaults to *db*.

Note that the function delete-duplicates-during-merge-p is no longer supported (starting in release 4.2.1). The function duplicate-deletion-strategy is no longer supported starting in release 4.4.

Advanced Triple-Store Operations

The !-reader macro and future-parts

When working with the triple-store at the REPL (the lisp listener) it's nice to have a more concise way to refer to resources and literals than with calls to resource, literal or the part interning functions. It's also handy to be able to abbreviate the many long URIs with a common prefix such as http://www.w3.org/2000/01/rdf-schema#. Namespaces and the !-reader macro provide a concise syntax for both resources and literals.

The first thing the !-reader macro allows you to do is write N-Triples strings without quotation marks (except for those required by the N-Triples syntax itself!). Thus instead of writing:

"<http://www.franz.com/>"  
"\"foo\""  
"\"foo\"^^<http://www.w3.org/2000/01/rdf-schema#integer>"  
"\"foo\"@en" 

you can simply write:

!<http://www.franz.com/>  
!"foo"  
!"foo"^^<http://www.w3.org/2000/01/rdf-schema#integer>  
!"foo"@en 

In addition, the !-reader macro uses namespaces to abbreviate long URIs. Use the register-namespace function to assign an abbreviation to any prefix used in URIs. For instance you can register s as an abbreviation for the URI prefix http://www.franz.com/simple# like this:

(register-namespace "s" "http://www.franz.com/simple#") 

Then you can use that prefix with the !-reader macro to write URIs starting with that prefix:

!s:jans => !<http://www.franz.com/simple#jans> 

You have probably noticed that the !-reader macro does not seem to be doing anything:

!"hello" => !"hello" 

This is because ! is converting the string "hello" into what AllegroGraph calls a future-part and the future-part prints itself using the !-notation. If we describe the future-part, then we will see all of the additional structure:

> (describe !"hello")  
!"hello" is a structure of type future-part.  It has these slots:  
 type               :literal  
 value-prefix       nil  
 value-fragment     "hello"  
 extra-prefix       nil  
 extra-fragment     nil  
 value              "hello"  
 extra              nil  
 extra-done-p       t  
 upi                #(5 0 0 0 0 0 111 108 108 101 104 7)  
 triple-db          nil 

Now it's clear that AllegroGraph has parsed the string and performed the computations to determine the part's UPI.

future-parts are called future-parts because they cache some of the information (e.g., the namespace prefix) and wait to resolve until the namespace mappings are available which may be in the future. Before we finish describing resolution, however, here are some examples. First, literals:

The story for resources is very similar:

Future parts are resolved when it is necessary to determine their UPI. If the part uses a namespace (e.g., is something like !a:b), then the namespace will be resolved first. It is an error to try and determine a part's UPI if the necessary namespace mapping has not been registered. Once the namespace of a part is resolved, then it will not be resolved again (during the current Lisp session). After namespace resolution, a future-part is equivalent to a particular string which can be interned into a triple-store (i.e., stored in the store's string dictionary).

Future-parts make working with AllegroGraph much simpler but they do contain some machinery and can be confusing. Remember that you can always tell what is happening by using the Lisp describe function.

The following are the functions used for enabling the ! reader macro and for managing namespaces.

clear-namespaces &key  keep-standard-namespaces
function
Delete all existing namespace mappings. If :keep-standard-namespaces is true (the default) then the namespace mappings in standard-namespaces will not be removed.
collect-namespaces &optional  filter
function

Returns a list of namespace mappings.

filter can be nil or a string. If it is a string, then only namespaces that contain that string will be returned.

display-namespaces &optional  filter
function

Print out all namespace mappings in a human-readable format.

filter can be nil or a string. If it is a string, then only namespaces that contain that string will be returned.

dump-namespaces &optional  file
function
Dump all registered namespaces as calls to register-namespace that can be used to recreate the original namespaces if needed later. If the optional file argument is provided, output is written to that file. Otherwise the output is sent to *standard-output*.
namespace-redefinition-error
class
The error is signaled when an attempt is made to alter the mapping between a namespace prefix and its associated URI reference.
lookup-namespace namespace-prefix
function
Returns the uriref associated with namespace-prefix or nil if there is no association.
map-namespaces fn  &optional  filter
function

Applies fn to each namespace mapping. Fn must be a function of two arguments: the prefix and the uri.

filter can be nil to map all namespaces; a string to map only namespaces that include that string or a function to map only namespaces for which the function returns true.

register-namespace namespace-prefix  uri-reference  &key  errorp
function
Create a mapping between namespace-prefix and uri-reference, both of which should be strings. If the errorp keyword argument, which defaults to t, is true, then defining a mapping for an existing prefix will signal a continuable namespace-redefinition-error condition.
register-standard-namespaces
function
Add standard namespaces such as rdf, rdfs and owl. See the variable standard-namespaces for more details.
remove-namespace prefix
function
*standard-namespaces*
variable

The standard-namespaces is a list of (name prefix) pairs representing namespace mappings. For example:

'(("rdf" "http://www.w3.org/1999/02/22-rdf-syntax-ns#")  
  ("rdfs" "http://www.w3.org/2000/01/rdf-schema#")  
  ("owl" "http://www.w3.org/2002/07/owl#")) 

It is used by the function register-standard-namespaces to create standard mappings.

enable-!-reader &optional  readtable
function

Enables the use of the !-reader macro.

The !-reader makes it easier to type URIs and literals at the Read-Eval-Print-Loop.

enable-!-reader returns the previous definition of the #! in the current readtable.

Note: enable-!-reader modifies the readtable specified. If left unspecified, then enable-!-reader modifies the current *readtable*. There is no disable-!-reader command because Common Lisp does not provide a portable mechanism to cleanly compose and operate with multiple readtables. For information on working with readtables, please consult that Allegro Common Lisp reference.

Note: Because of its global effect, if enable-!-reader is called on the default readtable (*readtable*), then use of the exclamation mark character in other Lisp packages and systems may break. Generally speaking, it is possible to work around these issues. Please contact support for additional help.

enable-print-decoded boolean  &key  pprint-dispatch
function

Triples and parts are represented by (simple-array (unsigned-byte 8) 56) and (simple-array (unsigned-byte 8) 12) respectively. By default the Lisp printer prints these as vectors of bytes which is not very informative when using AllegroGraph interactively. This function modifies a pprint-dispatch table to print triples and parts interpretively if this can be done in the current environment. Specifically, the value of *db* must be an open triple store, and the vector being printed must print without error in the manner of print-triple or upi->value. If any of these conditions don't hold the vector is printed normally.

If the boolean argument is true, the informative printing is enabled, otherwise disabled. For example:

> (enable-print-decoded t)  
t  
> (upi !"Dürst"^^xsd:string)  
{Dürst}  
> (enable-print-decoded nil)  
nil  
> (upi !"Dürst"^^xsd:string)  
#(48 61 125 164 117 231 137 163 64 100 17 2) 

The :pprint-dispatch argument specifies the dispatch table to modify, by default the value of print-pprint-dispatch. Once enable-print-decoded has been turned on, you can also use the special-variable *print-decoded* to enable fine-grained control of triple and UPI printing. See the AllegroGraph tutorial for more information on the various notations AllegoGraph uses.

Freetext Indexing

Note that freetext indexing is orthogonal to triple indices like spogi. Freetext indexes let you quickly map from words and phrases to triples; triple-indexes let you quickly find triples that match particular patterns.

AllegroGraph supports freetext indexing the content of triples. See Freetext Indices for a general discussion.

Freetext indexing is done by creating one or more indexes, each of which can be configured to determine which triples, and which fields of those triples, it indexes. Indexed triples can be queried using a simple but robust query language.

Indices are managed using the following functions:

create-freetext-index name  &key  db  predicates  index-fields  index-resources  index-literals  minimum-word-size  stop-words  border-chars  inner-chars  tokenizer  word-filters
function

This creates a new index with the given name, which should be a string. The arguments have the following meaning:

predicates
A list of future-parts or UPIs that determine the predicates this index indexes. Default is (), which means index all predicates.
index-fields
A list containing any subset of (:subject :predicate :object :graph), which determines which fields of the triples are indexed. Default is (:object).
index-literals
Determines whether literals are indexed. Can be t (yes), nil (no), or a list of strings, UPIs or future-parts, each of which specifies either a type for typed-literals or a language (only indexes those literals whose type or language matches any of them). Default is t.
index-resources
Whether and how resources are indexed. The default is nil (don't index them). t means index resources fully, and :short means index only the part after the last # or /.
minimum-word-size
The minimum size a word must have to be indexed. Default is 3, which is suitable for European languages using the default tokenizer. If you use :simple-cjk or :japanese tokenizer, set this to 2 or 1.
stop-words
A list of strings, which indicates the set of words that should not be indexed. Defaults to a short list of English stop-words.
border-chars
If given, a freetext-charset object specifying characters that can appear at word boundaries. How this is used depends on the tokenizer; see tokenizer below.
inner-chars
If given, a freetext-charset object specifying characters that can appear within a word (but may not be at its boundaries). How this is used depends on the tokenizer; see tokenizer below.
word-filters
A list of string, naming filters that should be applied to words before they are indexed and before they are searched for, allowing terms to be normalized. Currently "stem.english", "drop-accents" and "soundex" are supported.
tokenizer
A keyword to name the tokenization scheme. Currently, :default, :japanese and :simple-cjk are supported. If unspecified, the the default tokenizer is used, which is sufficient for European languages. With the default tokenizer, each word begins with one of border-chars, followed by zero or more inner-chars, and ends with one of border-chars. The :simple-cjk tokenizer can be used to tokenize Chinese, Japanese or Korean text. This tokenizer first separates the text into sequences of CJK characters and sequences of other characters. For CJK parts, each bigram of characters is taken as a word; that is, if you have "XYZ" where X, Y and Z are CJK characters, then "XY" and "YZ" are indexed as words. The remaining non-CJK parts are tokenized with border-chars and inner-chars, as with the default tokenizer. Finally, the :japanese tokenizer uses a Japanese morphological analyzer to tokenize the Japanese sentences. Using an internal dictionary, it can extract Japanese words more precisely than the simple-cjk tokenizer. The border-chars and inner-chars are ignored.
modify-freetext-index name  &key  db  predicates  index-fields  index-resources  index-literals  minimum-word-size  stop-words  border-chars  inner-chars  tokenizer  word-filters  re-index-p
function

Modify an existing freetext index. When not given, the configuration arguments retain their old value for this index. See create-freetext-index for the meaning of the settings.

If re-index-p is true, the index will be rebuilt, and existing triples will be re-indexed. If it is nil, the new settings will only be used for triples added afterwards.

drop-freetext-index name  &key  db
function
Deletes the named freetext index.
list-freetext-indices &optional  db
function
Returns a list of strings, naming the freetext indices that have been created for this triple-store.

The following functions are index field accessors, which, given the name of an index (as a string), return a configuration setting for that index.

freetext-index-index-fields name  &key  db
function
Retrieve the index-fields configuration option for a freetext-index.
freetext-index-index-literals name  &key  db
function
Retrieve the index-literals configuration option for a freetext-index.
freetext-index-index-resources name  &key  db
function
Retrieve the index-resources configuration option for a freetext-index.
freetext-index-minimum-word-size name  &key  db
function
Retrieve the minimum-word-size configuration option for a freetext-index.
freetext-index-predicates name  &key  db
function
Retrieve the predicates configuration option for a freetext-index.
freetext-index-stop-words name  &key  db
function
Retrieve the stop-words configuration option for a freetext-index.
freetext-index-tokenizer name  &key  db
function
Retrieve the tokenizer configuration option for a freetext-index.
freetext-index-word-filters name  &key  db
function
Retrieve the word-filters configuration option for a freetext-index.

These are functions for querying the freetext indices:

freetext-get-triples expression  &key  index  db  sorted-p
function

Returns a cursor that iterates over the triples that match the expression. The index argument can be given to search only a single index. When not given, all available indices are used. The sorted-p argument, which defaults to nil, can be given a truth value to cause the results to come back sorted by relevance (most relevant triples first).

(iterate-cursor (triple (freetext-get-triples '(match "amsterd*")))  
    (print triple)) 
freetext-get-unique-subjects expression  &key  db  index
function

Returns all the unique subjects in triples whose objects contain expression. This is a useful function in prolog queries.

(select (?person)  
  (lisp ?list  
     (freetext-get-unique-subjects '(and "collection" "people")))  
  (member ?person ?list)  
  (q- ?person !rdfs:subClassOf !c:AsianCitizenOrSubject)) 

From strings to words: Tokenizing

To index freetext, the engine first breaks strings into words (a process called tokenization). The built-in tokenizer defines a word as any run of letters that is at least as long as the minimum-word-size specified for an index. Dashes, underscores and periods do not break words unless they occur at the start or the end. It will also omit the stop words configured for the index. The default list is:

Below are some examples of strings and the words the tokenizer will extract from them:

Freetext Query Expressions

Here is the informal grammar used to build query expressions:

"string"
Matches triples containing all the words in the string.
(and <patter1> <pattern2> ...)
Matches triples that match all of the given patterns.
(or <pattern1> <pattern2> ...)
Matches triples that match any of the given patterns.
(phrase "string")
Matches triples that contain the exact string given.
(match "?string*")
Matches triples that match the given string, supporting wildcards, where ? matches any single character, and * matches any number of characters. Note that, in a big triple-store, matching patterns starting with a * or ? can be very slow.

Here are some examples:

When searching for (string-shaped) user input, the utility function text-index:parse-query might come in handy. It takes a string like "\"common lisp\" (programming | devel*)" and turns it into (and (phrase "common lisp") (or "programming" (match "devel*"))).

Freetext Error Conditions

The following are classes of conditions signaled when an error occurs.

text-index-already-exists-error
class
Signalled when trying to create a freetext-index with a name that is already taken.
text-index-not-found-error
class
Signalled when trying to access a freetext-index that does not exist.

Indexing words in alphabets other than Roman

The :tokenizer keyword argument to create-freetext-index specifies the tokenizer to use. :default works for most European languages. :japanese specifies specifies the Japanese language tokenizer, as the following screenshot shows:

Creating a Japanese freetext index

Data-type and Predicate Mapping

Most triple-stores work internally only with strings. AllegroGraph, however, can store a wide range of datatypes directly in its triples. This ability not only allows for huge reductions in triple-store size but also lets AllegroGraph execute range queries remarkably quickly. Use the supported-types function to see a list of datatypes that AllegroGraph can encode directly. The datatypes are specified as keywords or integers (with the integers being used internally). You can translate in either direction using type-code->type-name and type-name->type-code. You can add encoded-triples (i.e., triples some of whose parts are encoded UPIs rather than references to strings) directly using add-triple or by setting up mappings between particular predicates or specific datatypes and then using one of the bulk loading functions. In the former case, you use value->upi and in the later, you use datatype-mappings and predicate-mappings.

For example, here is how to specify that the datatype xsd:double maps to an AllegroGraph :double-float and the predicate <http://www.example.com/predicate#age> maps to an :unsigned-byte

> (setf (datatype-mapping  
   "http://www.w3.org/2001/XMSchema#double") :double-float)  
  :double-float  
 
 
> (setf (predicate-mapping  
   "http://www.example.com/predicate#age") :unsigned-byte)  
  :unsigned-byte 

When you add a triple (either via add-triple, SPARQL UPDATE or one of the bulk load functions), AllegroGraph will examine it to see if it satisfies any type mappings. If it does, then an encoded-triple will be added to the triple-store instead of one using string literals. If necessary, you can tell AllegroGraph to preserve strings but this is not recommended because of the space and time performance penalties.

automatically-mapped-types
function

Returns a list of the automatically mapped datatypes.

The list is a list of pairs where the first entry in each pair is the short type-name and the second entry is the XSD-type as a future-part. For example:

 (:unsigned-int !<http://www.w3.org/2001/XMLSchema#unsignedInt>) 

UPI type codes +geospatial+ +rdf-blank-node+ +rdf-byte+ +rdf-date+ +rdf-date-time+ +rdf-default-graph+ +rdf-double-float+ +rdf-gyear+ +rdf-int+ +rdf-latitude+ +rdf-literal+ +rdf-literal-language+ +rdf-literal-short+ +rdf-literal-typed+ +rdf-long+ +long-88+ +rdf-longitude+ +rdf-resource+ +rdf-short+ +rdf-single-float+ +telephone-number+ +rdf-time+ +rdf-unsigned-byte+ +rdf-unsigned-int+ +rdf-unsigned-long+ +unsigned-long-88+ +rdf-unsigned-short+ +subscript+ +triple-id-tag+
group

These constants can be used to specify a UPI's type-code. See upi->value and value->upi for more details.

The types +long-88+ and +unsigned-long-88+ are similar to the standard RDF long integer types but extend the range to 88 bits.

datatype-mapping part  &optional  db
function

Returns the type encoding for part, an XML-Schema type designator. This can be a string, a UPI or a future-part. For example:

> (datatype-mapping "<http://www.w3.org/2001/XMLSchema#unsignedByte>")  
!xs:unsignedByte 

The scope of the mapping is the db (which defaults to
*db*). Use (setf datatype-mapping) to add an association between a datatype and an encoded UPI type

encoded-types
function

Returns a list of type names that can be used in value->upi to encode a string into an encoded UPI. To see the corresponding type code for a name, use the type-name->type-code function.

See the section on type mapping for more information.

hashed-upi-p upi
function
Returns true if UPI is hashed. I.e., if its associated data is stored in the triple-store's string-table rather than being directly encoded.
predicate-mapping part  &optional  db
function

Returns the type encoding for part. Part should be the URIref of a predicate. For example:

> (predicate-mapping !<http://www.example.org/property/height>)  
!xs:double 

The scope of the mapping is the db (which defaults to *db*). Use (setf predicate-mapping) to create a mapping between a URIref of a predicate and one of AllegroGraph's supported-types.

> (setf (predicate-mapping !<http://www.example.org/property/height>)  
        :double-float)  
:double-float 
supported-types
function

Returns a list of all available type names. Use encoded-types to restrict the result to only those types that produce encoded UPIs. To see the corresponding type code for a name, use the type-name->type-code function.

See the section on type mapping for more information.

type-code->type-name code
function
Returns the type name associated with code. See type-name->type-code.
type-name->type-code name
function

Returns the type code associated with the name. See type-code->type-name and supported-types. The name can be one of the supported-types or one of:

  • :blank-node
  • :resource
  • :literal
  • :literal-language
  • :literal-short
  • :literal-typed

The function returns nil if there is no type-code corresponding to name.

upi->value upi  &key  db  errorp
function

Decodes UPI and returns the value, the type-code and any extra information as multiple values.

  • upi - the UPI to decode

  • :db - specifies the triple-store from which the UPI originates.

The value, type-code, and extra information are interpreted as follows:

  • value - a string, integer, or float representing the contents of the UPI
  • type-code - an integer corresponding to one of the defined UPI types (see supported-types for more information). You can use type-code->type-name to see the English name of the type.
  • extra - There are three disjoint sets of literals in RDF:
    1. "simple literals" -- a string.
    2. literals with datatypes
    3. literals with language annotations.
  • A literal cannot have both a language and a datatype; if a literal has type-name :literal-typed (i.e., its type-code is 2), then the extra value is a datatype URI; if its type-code is 3, then extra will be the language code.

Examples:

> (upi->value (value->upi 22 :byte))  
22  
18  
nil  
 
> (upi->value (upi (literal "hello" :language "en")))  
"hello"  
3  
"en" 

See value->upi for additional information.

value->upi value  encode-as  &optional  upi
function

Returns a UPI that encodes value using type encode-as.

encode-as can be a type-code or a type-name (see supported-types and type-name->type-code for details. If a upi keyword argument is not supplied, then a new UPI will be created. See upi->value for information on retrieving the original value back from an encoded UPI.

Note that value->upi can not create hashed UPIs (i.e., UPIs whose value must be stored in the string dictionary. To create these, use intern-resource and intern-literal.

clear-type-mappings &key  db  keep-automatic-type-mappings
function

Remove all type mappings from the triple store db.

Remove type mappings added with datatype-mapping and predicate-mapping from the triple-store db. If db is not specified, it defaults to the value of *db*.

keep-automatic-type-mappings controls whether or not the automatically mapped types stay mapped or if they are also cleared. This defaults to true. Use automatically-mapped-types to see which types are mapped automatically.

collect-all-type-mappings &optional  db
function

Return a list of the datatype and predicate mappings of db.

  • :db - defaults to *db*.

Each item in the list returned consists of a list of three elements:

  1. the URIref used for the mapping (predicate or datatype)
  2. the type-code of the mapping
  3. a keyword that indicates if the mapping is a :datatype or :predicate mapping.
map-all-type-mappings fn  &optional  db
function

Applies fn to each datatype and predicate mapping of db.

  • :db - defaults to *db*.

fn should be a function of three arguments. For each mapping, it will called with

  1. the URIref used for the mapping (predicate or datatype)
  2. the type-code of the mapping (if simple) or the mapping identifier (if extended)
  3. a keyword that indicates if the mapping is a :datatype or :predicate mapping.

(See collect-all-type-mappings if you just want a list of the mappings.).

print-type-mappings &optional  db
function

Print the type mappings of db to *standard-output*.

If not specified, db defaults to *db*.

Federation

Managing Massive Data - Federation

The block diagram we saw above is abstract: it can be implemented in many different ways. AllegroGraph 4.7 uses the same programming API to connect to local triple-stores (either on-disk or in-memory), remote-triple-stores and federated triple-stores. A federated store collects multiple triple-stores of any kind into a single virtual store that can be manipulated as if it were a simple local-store. Federation provides three big benefits:

  1. it scales well,
  2. it makes triple-stores more manageable, and
  3. it makes data archive extremely simple.

A high-level description of these benefits is provided in the AllegroGraph introduction. This guide focuses on the classes that are used to build a federation.

AllegroGraph's internal architecture

Internally, an open AllegroGraph triple-store is an instance of one of the classes depicted below. Most of the time, you won't need to be concerned with this class implementation because AllegroGraph will manage it transparently. We're depicting them here because they also serve to illustrate many of AllegroGraph's capabilities.

AllegroGraph Triple-store Classes

Let's look at each of these in turn.

An Abstract-triple-store defines the main interfaces a triple-store must implement. This class has four main subclasses:

  1. concrete-triple-stores manage actual triples whereas the other three function as wrappers between real triples and the store. These can be instances of triple-db or in-memory-triple-store.

  2. federated-triple-stores provide mechanisms to group and structure arbitrary collections of other triple-stores.

  3. Encapsulated-triple-stores let us add new behaviors to existing stores in a controlled and easily optimized fashion. The best example of an encapsulated-store is a reasoning-triple-store which endow triple-stores with RDFS++, rule based or other reasoning engines. Other encapsulated-store examples are graph-filtered-triple-store and ground-only-triple-store.

  4. Finally, a remote-triple-store lets AllegroGraph use triple-stores being served by other processes either locally or anywhere on the network. These triple-stores can be other AllegroGraph stores or connections to Oracle and Sesame ones. 8

By combining these four classes, you can build a triple-store composed of leaf stores from anywhere, implementing differing reasoning rules, from entirely different architectures and treat them as if they comprise a single unified store living on your desktop.

Managing triple-stores

federate-triple-stores name  stores  &key
function
Create a new federated triple-store named name which federates each of the triple-stores listed in stores. This also sets *db* to the value of the federated store.
encapsulate-triple-store db  name  wrapper-class  &rest  wrapper-initargs
function

Create a new encapsulated-triple-store wrapping db and and return it.

The new store will be named name and will be of type wrapper-class (which must be a subclass of encapsulated-triple-store. The additional wrapper-initargs are passed along to the newly created store.

make-graph-filtered-store &key  name  db  graphs  cached-wrapper-key
function

Create an encapsulated triple-store that allows only triples in specified graphs to be seen.

The encapsulated triple-store

  • db - the store to be encapsulated. The default is taken from *db*.

  • name - a unique name for the new store. The default is to prefix the name of the inner store with graph-filtered- and add a gensym suffix.

  • graphs - A non-empty list of triple part specifiers. Only the triples in the graphs specified by this list are visible in the encapsulated store.

  • cached-wrapper-key - If this argument is nil, the call creates a new instance of a filtering wrapper. If the argument is t, a new key is generated from the list of graphs. This key will be equalp to any other key generated from the same set of graphs. If the argument is any other non-nil value, it is used as a key in an equalp hash-table; if a previously created wrapper is found, the previous instance is returned. If no entry is found, a new wrapper is created and saved in the table. The default value is taken from the variable *cached-wrapper-key*. The initial value of this variable is nil.
inner-triple-store store
macro

Removes one layer of encapsulation from an encapsulated-triple-store and returns the triple-store that was encapsulated.

If store is not encapsulated, then this returns store unchanged. See encapsulate-triple-store for details.

ground-triple-store &optional  db
function

This function returns a triple store that does not contain any inferred triples.

If the given triple store is not a reasoning store, then it is returned unchanged.

If the triple store is a reasoning encapsulation, then the inner triple store is returned.

In some cases, the function returns a new encapsulated store instance that represents the desired intent.

Triple-store classes

abstract-triple-store
class
The base class of all AllegroGraph triple-store instances. It is responsible for managing resources local to the store (e.g., threads, files and the like) and for keeping track of the triple-store's ID.
concrete-triple-store
class
A concrete-triple-store is a container for real, honest-to-goodness (i.e., concrete) triples.
encapsulated-triple-store
class
A triple-store that wraps its inner-triple-store in order to provide additional processing or override the usual processing. A reasoning-triple-store is a kind of encapsulated-triple-store.
federated-triple-store
class
A federated-triple-store composes one or more leaf stores so that they can be treated as a single store. See the reference guide for details.
reasoning-triple-store
class
An encapsulated-triple-store that adds reasoning to a triple-store. This store contains a reasoner slot that controls the sort of reasoning that the triple-store supplies. The rdfs++-reasoner is an example of the sort of reasoning that AllegroGraph provides.
filtered-triple-store
class
An encapsulated-triple-store that presents a subset of the triples in the inner store.
graph-filtered-triple-store
class
An encapsulated-triple-store that presents a subset of the triples in the inner store based on a fixed set of graph labels.
ground-only-triple-store
class
An encapsulated-triple-store that excludes inferred triples from any search results.
in-memory-triple-store
class

An in-memory-triple-store implements the AllegroGraph API for a triple-store contained entirely in memory.

Note that in-memory-triple-stores work well only for small triple-stores, do not implement fast range-queries and may be missing full support for other AllegroGraph features.

remote-triple-store
class
A remote-triple-store acts as a local proxy for a triple-store located in another AllegroGraph process. This process can be on the same machine or somewhere out on the network. A connection to a remote-triple-store requires a port, username and password and accepts an optional scheme.
triple-db
class

The default triple-store class used by AllegroGraph in calls to create-triple-store, open-triple-store and so on.

triple-db implements the full AllegroGraph API and represents a triple-store accessed directly on a server. They should be compared with remote-triple-stores which function via client API.

Note AllegroGraph's federated triple-stores help solve problems associated with the management and understanding of massive datasets -- AllegroGraph does not keep track of where those datasets live or how to access them. You are responsible for finding each leaf triple-store and configuring them as best suits your particular problem space. You are also responsible for ensuring that leaf triple-stores remain open for as long as the federation is in use. Franz will continue to enhance AllegroGraph's federation support and the persistence of complex relationships between triple-stores. Please be sure you are using the most recent release of AllegroGraph and to contact us ([email protected]) if you have any questions or issues.

Modifying data in federated triple-stores

With a few exceptions that will be mentioned below, AllegroGraph treats federated triple-stores as if they were read-only. The primary reason for this is that AllegroGraph cannot know which of the leaf triple-stores you wish to modify. You can still make modifications to a federated triple-store, but to do so you must select one of the leaf stores and modify it.

Federation, triple-ids, default-graphs and blank nodes

Federation adds a new ambiguity to triples, the default-graph and blank node identifiers: from which leaf store did they come? To see the sort of problem that can arise, imagine that we federate two small stores into one as in the figure below:

Sample Federation

Now let's try a simple query: What is the rdf:type of the thing whose ex:isNamed is <Gary>? To answer this, we would first ask for the triples whose predicate is ex:isNamed and whose object is <Gary> and find this:

_:anon1 ex:isNamed "Gary" 

then we would ask for the triples whose subject is _:anon1 and whose predicate is rdf:type. If the federation could not tell blank nodes from different stores apart, then it would give us back two answers:

_anon1 rdf:type <Human>  
_anon1 rdf:type <Dog> 

and that would be wrong!

To prevent this, AllegroGraph keeps track of the originating triple-store for each blank node, triple-id and default-graph UPI. If you look at any of these UPIs or triples in its raw form (i.e., if enable-print-decoded is nil), then you will see that some of the bytes in these data-structures encode the triple-store's ID.

Franz is still exploring the semantics surrounding some of these issues (for example, sometimes if may make more sense to allow multiple leaf stores to use the same default-graph UPI) and will continue to improve federation support by making it more flexible without giving up its speed or ease of use.

Simple Reification

Suppose I have triple t that states s p o and I want to further claim that I believe it. I'd like to say:

This ability to talk about an assertion is called reification. In RDF, reifying a triple requires creating a new blank node (_:anon) so that we can say

and then making (at least) four assertions:

  1. _:anon is rdf:type rdfs:Statement,
  2. _:anon has rdfs:subject s,
  3. _:anon has rdfs:predicate p , and
  4. _:anon has rdf:object o

This is a lot of work when all we really want is to say something about an assertion. Because AllegroGraph provides both persistent triple-ids and encoded UPIs, we can reify another triple by saying simply:

In Java, the code for this would look like:

AllegroGraph ag;  
Triple t;  
 
...  
 
ag.addStatement("<name:Gary>", "<some:believes>",  
                ag.createEncodedLiteral(t.queryAGId(), "triple-id")); 

and in Lisp, we would use value->upi with a type-code of +triple-id-tag+.

SPARQL Query Language

parse-sparql string  &optional  default-prefixes  default-base
function

Parse a SPARQL query string into an s-expression.

This function is useful for three reasons: validation and inspection of queries, manual manipulation of query expressions without text processing, and performing parsing at a more convenient time than during query execution.

You do not need an open triple-store in order to parse a query. Any parse errors will signal a sparql-parse-error.

The optional arguments provide BASE and PREFIX arguments to the parser without inserting them textually into the query.

default-base
A string to use as the BASE for the SPARQL query.
default-prefixes
A hash-table mapping string prefixes to their expansions or a list of two element lists where each sublist contains the prefix and its expansion. For example:
(("rdf" "http://www.w3.org/1999/02/22-rdf-syntax-ns#")  
 ("rdfs" "http://www.w3.org/2000/01/rdf-schema#")  
 ("owl" "http://www.w3.org/2002/07/owl#")  
 ("xsd" "http://www.w3.org/2001/XMLSchema#")  
 ("xs" "http://www.w3.org/2001/XMLSchema#")  
 ("fn" "http://www.w3.org/2005/xpath-functions#")  
 ("err" "http://www.w3.org/2005/xqt-errors#")) 

This list uses the same format as db.agraph:standard-namespaces.

parse-sparql returns the s-expression representation of the query string.

run-sparql query  &rest  args  &key  engine  db  output-stream  rdf-format  results-format  host  &allow-other-keys
function

run-sparql takes a SPARQL query as input and returns bindings or new triples as output.

Since AllegroGraph 3.0 it is a convenient wrapper for the db-run-sparql methods specialized on particular database classes and query engines. You might consider using those methods directly to gain more control over the execution of your queries.

You should consider specifying an engine argument in your invocations of run-sparql; the choice of default execution engine is not guaranteed to remain the same in future releases.

Allowable values for engine are keyword symbols returned by valid-query-engines.

The precise arguments supplied to run-sparql vary according to the query engine. These are the typical arguments expected by the default engines.

SELECT and ASK query results will be presented according to the value provided for results-format, whilst the RDF output of DESCRIBE and CONSTRUCT will be serialized according to rdf-format. Both of these arguments take keyword values.

If the format is programmatic (that is, it is intended to return values rather than print a representation; :arrays is an example) then any results will be returned as the first value, and nothing will be printed on output-stream.

  • The query can be a string, which will be parsed by parse-sparql, or an s-expression as produced by parse-sparql. If you expect to run a query many times, you can avoid some parser overhead by parsing your query once and calling run-sparql with the parsed representation.
    If `query` is a string, then `default-base` and  
    `default-prefixes` are provided to [parse-sparql][]  
    to use when parsing the query. See the documentation  
    for that function for details. Parser errors signaled  
    within [parse-sparql][] will be propagated onwards by  
    `run-sparql`. 
  • default-base A string to use as the BASE for the SPARQL query (only used when query is a string).

  • default-prefixes A hash-table mapping string prefixes to their expansions or a list of two element lists where each sublist contains the prefix and its expansion (only used when query is a string; see parse-sparql for details).

  • Results or new triples will be serialized to output-stream. If a programmatic format is chosen for output, the stream is irrelevant. An error will be signaled if output-stream is not a stream, t, or nil.

  • If limit, offset, from, or from-named are provided, they override the corresponding values specified in the query string itself. As FROM and FROM NAMED together define a dataset, and the SPARQL Protocol specification states that a dataset specified in the protocol (in this case, the programmatic API) overrides that in the query, if either from or from-named are non-nil then any dataset specifications in the query are ignored. You can specify that the contents of the query are to be partially overridden by providing t as the value of one of these arguments. This is interpreted as 'use the contents of the query'. from and from-named should be lists of URIs: future-parts, UPIs, or strings.

  • default-dataset-behavior controls how the query engine builds the dataset environment if FROM or FROM NAMED are not provided. Valid options are :all (ignore graphs; include all triples) and :default (include only the store's default graph).

  • default-graph-uris allows you to specify a list of resources which, when encountered in the SPARQL dataset specification, are to be treated as the default graph of the store. Each resource can be a resource UPI, resource future-part, or a URI string. For example, specifying '("http://example.com/default") will cause a query featuring

    FROM <http://example.com/default>  
    FROM <http://example.com/baz> 
  • to execute against the union of the contents of the named graph <http://example.com/baz> and the store's default graph, as determined by (default-graph-upi db).

  • with-variables should be an alist of variable names and values. The variable names can be strings (which will be interned in the package in which the query is parsed) or symbols (which should be interned in the package in which the query is to be, or was, parsed). The variable names can include or omit a leading '?'. Note that a query literal in code might be parsed at compile time. Using strings is the most reliable method for naming variables.
  • Before the query is executed, the variables named after symbols will be bound to the provided values.

    This allows you to use variables in your query which are externally imposed, or generated by other queries. The format expected by with-variables is the same as that used for each element of the list returned by the :alists results-format.

  • db (*db* by default) specifies the triple store against which queries should run.

  • destination-db (db by default) specifies the triple store against which Update modifications should take place. This is primarily of use when db is a read-only wrapper around a writable store, such as when reasoning has been applied.

  • If verbosep is non-nil, status information is written to *sparql-log-stream* (*standard-output* by default).

Three additional extensions are provided for your use.

  • If extendedp is true (or *use-extended-sparql-verbs-p* is true, and the argument omitted) then extensions like AllegroGraph's GEO syntax are enabled. Extensions are enabled by default in all versions of AllegroGraph after 3.2.

  • If memoizep is true (or *build-filter-memoizes-p* is true, and the argument omitted) calls to SPARQL query functions (such as STR, fn:matches, and extension functions) will be memoized for the duration of the query. For most queries this will yield speed increases when FILTER or ORDER BY are used, at the cost of additional memory consumption (and consequent GC activity). For some queries (those where repetition of function calls is rare) the cost of memoization will outweigh the benefits. In large queries which call SPARQL functions on many values, the size of the memos can grow large.

  • Memoization also requires that your extension functions do not depend on side-effects. The standard library is correct in this regard.

  • In some circumstances you can achieve substantial speed increases by sharing your memos between queries. Create a normal eql hash-table with (make-hash-table), passing it as the value of the memos argument to run-sparql. This hash-table will gradually fill with memos for each used query function.

If you wish to globally enable memoization, set the variables as follows:

(progn  
  (setf *build-filter-memoizes-p* t)  
  (setf *sparql-sop-memos* (make-hash-table))) 

Be aware that the size of *sparql-sop-memos* could grow very large indeed. You might consider using a weak hash-table, or periodically discarding the contents of the hash-table.

  • load-function is a function with signature (uri db &optional type) or nil. If it is a function, it is called once for each FROM and FROM NAMED parameter making up the dataset of the query. The execution of the query commences once each parameter has been processed. The type argument is either :from or :from-named, and the uri argument is a part (ordinarily a future-part) naming a URI. The default value is taken from *dataset-load-function*. You can use this hook function to implement loading of RDF before the query is executed.

  • permitted-verbs is a keyword, either :all or :read-only. This defaults to :all, and will permit any kind of SPARQL or SPARQL/Update query. Use :read-only to allow only SELECT, ASK, DESCRIBE, and CONSTRUCT queries. Note that you must also enable extended mode (using :extendedp :update) to use SPARQL/Update operations.

The values returned by run-sparql are dependent on the verb used. The first value is typically disregarded in the case of results being written to output-stream. If output-stream is nil, the first value will be the results collected into a string (similar to the way in which cl:format operates).

The second value is the query verb: one of :select, :ask, :construct, or :describe. Other values are possible in extended mode.

The third value, for SELECT queries only, is a list of variables. This list can be used as a key into the values returned by the :arrays and lists results formats, amongst other things.

Individual results formats are permitted to return additional values.

sparql->algebra sparql  &key  fixed-places  simplifyp  optimizep  extendedp
function

Take a SPARQL query and return the corresponding algebra expression.

For example:

(sparql.algebra:sparql->algebra  
 "PREFIX fti: <http://franz.com/ns/allegrograph/2.2/textindex/>  
  SELECT ?g ?x { GRAPH ?g { ?x fti:match 'world' } FILTER ( <urn:x> = ?x ) }")  
==>  
(:filter  
 (:graph ?g  
  (:bgp  
   #(?x !<http://franz.com/ns/allegrograph/2.2/textindex/match> !"world")))  
 (= !urn:x ?x)) 

See the SPARQL documentation for more details on the SPARQL algebra.

If simplifyp is true (the default), the algebra expression is simplified according to the SPARQL algebra simplification rules. There should ordinarily be no reason to do otherwise.

If optimizep is true (the default), the query is optimized before being returned. This implies simplifyp.

If extendedp is true (the default), extended SPARQL syntax understood by AllegroGraph is accepted. If nil, only standard SPARQL is accepted.

Prolog Select

The q and q- functors

The main interface to Prolog is the function q- (for simple query). The q- functor is analogous to get-triples. There are many versions of the functor but the simplest one has three arguments (for subject, predicate and object). You can supply any combination of these arguments and the functor will iterate over all of the possible matches in the triple-store, unifying its logic variables to each result in turn. The example below assumes that enable-print-decoded is active.

> (?- (q- ?x !rdf:type ?z))  
?x = {http://example/ok#ViewDerivedObject}  
?z = {owl:Class}  
                   ;; type Enter  
 
?x = {http://example/ok#UserActions}  
?z = {owl:Class}  
                   ;; type period ( . ) 

This documentation uses the usual Prolog argument notation. A + prefix on an argument indicates that is must be supplied (it is input to the predicate), a - prefix indicates the argument is output and must not be supplied, and a ± prefix indicates the argument may be either.

As is usual with Prolog, you can stop the iteration by typing a period (' . '). The q- variant of the query functor even supports AllegroGraph range queries.

q
(q ±s ±p ±o) (q ±s ±p ±o ±g) (q ±s ±p ±o ±g ±id) (q ±s ±p ±o ±g ±id -triple)
functor

Query the triple-store. If the store is a simple store, then q is the same as q-. If the store is an encapsulated-triple-store (for example, a reasoning-triple-store) then q and q- will differ. Unify s, p, and o with the triples in the store.

The third argument may be of the form (-o +o-min +o-max) to specify a range of values for the object component of the triples. Range queries are always bypass any reasoning wrappers on the store.

Unify s, p, o, and g with the triples in the store.

The third argument may be of the form (-o +o1 +o2) to specify a range of values for the object component of the triples. Range queries always bypass any reasoning wrappers on the store.

The fourth argument may be of the form (-g +g1 +g2) to specify a range of values for the graph component of the triples. Only one of the object or graph positions may specify a range.

The fourth argument may also be of the form (-g (:member +g1 +g2 ...)) or (-g (:member . +glist)) to specify a finite set of allowed graph values.

Unify s, p, o, g, and id with the triples in the store.

Unify s, p, o, g, id and triple.

q-
(q- ±s ±p ±o) (q- ±s ±p (-o +o-min +o-max)) (q- ±s ±p ±o ±g) (q- ±s ±p ±o (-g +g-min +g-max)) (q- ±s ±p ±o ±g ±id -triple)
functor

Query a triple-store's ground store. If the store is not a reasoning store, then q- is equivalent to q Unify s, p, and o with the triples in the store.

The third argument may be of the form (-o +o1 +o2) to specify a range of values for the object component of the triples. Range queries always bypass any reasoning wrappers on the store.

Execute a range query for triples whose object is between o-min and o-max. Unify s and p.

Unify s, p, o, and g with the triples in the store. See (q ?s ?p ?o ?g) for a detailed description of the arguments

Execute a range query for triples whose graph is between g-min and g-max. Unify s, p, and o.

Unify s, p, o, g, id and triple.

qd
(qd +db ±s ±p ±o) (qd +db ±s ±p ±o ±g) (qd +db ±s ±p ±o ±g ±id) (qd +db ±s ±p ±o ±g ±id -triple)
functor

Query the triple store specified by the first argument. The remaining arguments specify s, p, o, and/or g` components as in q and q-.

The first argument must be a symbol that names a store defined in a :named-store planner option or it must be a Prolog expression that evaluates to a triple store. The symbol *db* can appear as a name; in that case the qd form is equivalent to a q form with the same triple parts. Unify s, p, and o with the triples in the store.

The fourth argument may be of the form (-o +o1 +o2) to specify a range of values for the object component of the triples. Range queries always bypass any reasoning wrappers on the store.

Unify s, p, o, and g with the triples in the store. See (q ?s ?p ?o ?g) for a detailed description of the arguments

Unify s, p, o, g, and id with the triples in the store.

Unify s, p, o, g, id and triple.

The Select Macros

You can combine multiple q clauses into a query using one of the family of select macros. These evaluate a list of Prolog clauses and then do something with the solutions found 9 . There are three dimensions in which this family of macros differ:

Each macro takes as its first argument a template. Often the template is a a single variable or a simple list of variables. But in general the template may be a tree composed of Lisp data and Prolog variables

The select macros will find each solution to the list of supplied Prolog clauses. It will then build up a fresh instance of the template tree with each Prolog var in the tree replaced by the variable in the solution. For example:

> (select ((:name ?name) (:age ?age))  
    (q- ?x ?isa !ex:person)  
    (q- ?x ?hasname ?name)  
    (q- ?x ?hasage ?age)) 

would return a list of results like

(((:name "Gary") (:age 12))  
 ((:name "Xaviar") (:age 56))  
 ...  
 ) 

Select planner options described below can also be used to control the behavior of the select expression and the format of the result. If planner options are used for this purpose, it is recommended to use the basic select macro instead of the variants. Using a variant with a conflicting option may yield unexpected results.

You can find other examples of using select in the Using Prolog with AllegroGraph tutorial.

select template  &body  clauses
macro

Evaluate the Prolog clauses and return a list of all solutions. Each solution is a copy of the template (which is a tree) where each Prolog variable is replaced by its string value. See select0 if you want UPIs rather than strings.

The code generated by all the variants of this macro always binds *db*. Consequently, any modification of this variable in the body clauses will disappear when control exist the code.

select0 template  &body  clauses
macro
Evaluate the clauses and return a list of all solutions. Solutions consist fresh copies of the template (which is a tree) where each Prolog variable in the tree is replaced by its UPI value in the solution. If you are interested in part-names rather than raw UPIs, see select.
select/callback (&rest  vars)  callback  &body  clauses
macro
Evaluate the Prolog clauses and call callback once for each solution. Solutions consist of a fresh copy of the template with each var replaced by the string value of the Prolog variable in the solution. The callback argument is evaluated. See select0/callback if you are want UPIs rather than strings.
select0/callback (&rest  vars)  callback  &body  clauses
macro
Evaluate the Prolog clauses and call callback once for each solution. Each solution is a copy of the template with each Prolog variable replaced by its UPI. The callback argument is evaluated. See select/callback if you want strings rather than UPIs.
select-distinct template  &body  clauses
macro
Evaluate the clauses and return a list of all distinct solutions. Each solution is a copy of the template where each Prolog variable in the tree is replaced by its string value See select0-distinct if you want UPIs rather than strings.
select0-distinct template  &body  clauses
macro
Evaluate the Prolog clauses and return a list of all distinct solutions. Each solution is a copy of the template (which is a tree) where each Prolog variable is replaced by its UPI value. See select-distinct if you want strings rather than UPIs,
select-distinct/callback (&rest  vars)  callback  &body  clauses
macro
Evaluate the Prolog clauses and call callback once for each distinct solution. Each solution is a fresh copy of the template with each var replaced by the string value of the Prolog variable in the solution. The callback argument is evaluated. See select0-distinct/callback if you want UPIs rather than strings.
select0-distinct/callback (&rest  vars)  callback  &body  clauses
macro
Evaluate the Prolog clauses and call callback once for each distinct solution. Each solution is a fresh copy of the template with each var replaced by the UPI value of the Prolog variable in the solution. The callback argument is evaluated. See select-distinct/callback if you want strings rather than UPIs.

Prolog and range queries

You can make range queries using the same prolog q functor as you do to make non-range queries. We support range queries on either the object or the graph fields (but not on both simultaneously in the same clause). What follows is a list of all the 9 possibilities for q. Each token in the functor may be a UPI literal, a Prolog variable (possibly anonymous) or nil. We use a - prefix on a token to indicate that it is a placeholder and it cannot be used as a variable; a + prefix to indicate that the token must be supplied and a ? to indicate that the token is either.

Suppose, for example, that we use AllegroGraph to record the velocities during a mythical 1000-km trip:

;;;;;;;;;;;;;;;;;;;; A 1000 KM trip.  
 
(in-package :triple-store-user)  
 
(create-triple-store "My1000KmTrip" :if-exists :supersede)  
 
(register-namespace "t" "http://www.me.disorg#"  
                    :errorp nil)  
 
;; add triples describing a 1000-km trip  
(loop with km = 0           ; distance in km  
    with time = 0           ; time in seconds  
    as kmh = (random 80.0)  ; velocity in km/hr  
    do (add-triple  
        !t:me  
        !t:distance  
        (value->upi km :double-float)  
        :g (value->upi time :unsigned-long))  
       (add-triple  
        !t:me  
        !t:velocity  
        (value->upi kmh :double-float)  
        :g (value->upi time :unsigned-long))  
       (incf km (/ kmh 60))  
       (incf time 60)  
    count 1             ; Number of data points, which is  
                        ; also the duration in minutes.  
    while (< km 1000)) 

We can then use the following select query to find triples that represent the 1-minute periods where average speed is 75..80 km/h and the distance at which it occurred.

(loop for (time speed distance)  
    in (sort (select (?time ?v ?distance)  
           (lisp ?v75 (value->upi 75 :double-float))  
           (lisp ?v80 (value->upi 80 :double-float))  
           (q- !t:me !t:velocity (?v ?v75 ?v80) ?time)  
           (q- !t:me !t:distance ?distance ?time))  
         (lambda (x y) (< (first x) (first y))))  
    do (format t "~7d ~5f ~5f~%" (/ time 60) speed distance)) 

This last form could be written more succinctly using the ++ syntax explained in the tutorial:

(loop for (time speed distance)  
    in (sort (select (?time ?v ?distance)  
               (q- !t:me  
                   !t:velocity  
                   (?v (value->upi 75 :double-float)  
                       (value->upi 80 :double-float))  
                   ?time)  
               (q- !t:me !t:distance ?distance ?time))  
             (lambda (x y) (< (first x) (first y))))  
    do (format t "~7d ~5f ~5f~%" (/ time 60) speed distance)) 

Other AllegroGraph Prolog functors

subject
(subject ±subject +triple)
functor
Unify subject with the subject field of triple.
predicate
(predicate ±predicate +triple)
functor
Unify predicate with the predicate field of triple.
object
(object ±object +triple)
functor
Unify object with the object field of triple.
graph
(graph ±graph +triple)
functor
Unify graph with the graph field of triple.
triple-id
(triple-id ±triple-id +triple)
functor
Unify triple-id with the triple-id field of triple.
part=
(part= +part1 +part2)
functor
Succeed if part1 is part= to part2. Each of part1 and part2 should be UPIs, future-parts or strings in N-Triples format that can be converted into resources or literals.
upi<
(upi< +upi1 +upi2)
functor
Succeed if upi1 is upi< to upi2. Each of upi1 and upi2 must be UPIs.
upi<=
(upi<= +upi1 +upi2)
functor
Succeed if upi1 is upi<= to upi2. Each of upi1 and upi2 must be UPIs.
upi=
(upi= +upi1 +upi2)
functor
Succeed if upi1 is upi= to upi2. Each of upi1 and upi2 must be UPIs.

Select Query Planner

AllegroGraph's Prolog query language (select) includes an optimizer that speeds up complex queries via:

As part of the transformation process, the query-optimizer works with AllegroGraph's dynamic RDFS++ reasoner to perform some inferences at query planning-time (rather than repeatedly during query run-time).

The optimizer works automatically and in most situations you do not need to do anything special to make use of it: simply ask your Prolog query and let the optimizer run. Query optimization is a black art, however, so sometimes you may want to see what is happening inside AllegroGraph. Furthermore, there are some additional, non-automatic tuning options that you may want to consider.

AllegroGraph's query optimizer and dynamic-reasoner are both areas of active research and development. We value your feedback on both the positives and the negatives so that we can continue to make AllegroGraph the best RDF graph database for your projects.

Note that there is a problem in the query optimizer's handling of uncommitted triples and RDFS reasoning. In certain cases involving uncommitted triples, the optimizer will produce incorrect plans that may return fewer results than expected. The two possible work-arounds are to either call commit-triple-store before executing the query or to add the clause (:use-maps nil) to the query plan. This problem will be corrected in the near future.

Query plan options

Query plan options can be used to control the planning and transformation process. You can specify these options by including keyword clauses in the body of your query. A keyword clause is a two item list consisting of the option (a keyword) and its value. For example, this query uses the :display-plan option so that the generated plan will be printed to *standard-output*:

(select0 (?x)  
  (:display-plan t)  
  (q ?x !ub:memberOf !<http://www.Department0.University0.edu>)  
  (q ?x !rdf:type !ub:Person)) 

There are four kinds of plan options: store management, introspection, planner control, and result manipulation. The current list of options is:

More options may be added in the future. See below for details on how you can determine which options exist and their current settings.

Query Planner Examples

The following example queries are taken from the LUBM Benchmark. We will cover several examples showing the different sorts of transformations that the query planner performs.

Type Filter Removal and Transformations

Q3 asks for all of the publications of a particular professor:

(select0 (?x)  
    (:display-plan t)  
    (q ?x !ub:publicationAuthor !u0d0:AssistantProfessor0)  
    (q ?x !rdf:type !ub:Publication)) 

We use the :display-plan option so that we can see how AllegroGraph interprets the query. Note that the rdfs:domain of !ub:publicationAuthor is !ub:Publication so the second clause is redundant and can be removed. 10 Secondly, !ub:publicationAuthor does not participate in any RDFS++ reasoning for this store. Thus the planner transforms the q clause into an internal q/upis- one:

;; Return 1 variables: ?x  
;; estimate:          256 results per binding  
(q/upis- ?x {publicationAuthor} {AssistantProfessor0}) 

The q/upis- functor performs certain extra though minor optimizations over and above the more generic q-.

Clause Reordering and More Transformations

Query 2 asks for a listing of students who take a course from their advisors:

(select0 (?x ?y ?z)  
    (:verbose t)  
    (:display-plan t)  
    (q ?x !ub:advisor ?y)        
    (q ?x !ub:takesCourse ?z)  
    (q ?y !ub:teacherOf ?z)  
    (q ?x !rdf:type !ub:Student)  
    (q ?y !rdf:type !ub:Faculty)  
    (q ?z !rdf:type !ub:Course)) 

Note that we've used the :verbose query plan option to display additional information during the query process. The information display below is unlikely to match what you see since it is dependent on the inner workings of the planner and the particular store being used. Here is the information from the planning process.

;; Compute estimate: nil {advisor} nil nil :                             204,997  
;; Compute estimate: nil {takesCourse} nil nil :                       1,433,856  
;; Compute estimate: nil {teacherOf} nil nil :                           108,288  
;; Compute estimate: nil {type} {ResearchAssistant} nil :                 36,608  
;; Compute estimate: nil {type} {GraduateStudent} nil :                  126,208  
;; Compute estimate: nil {type} {UndergraduateStudent} nil :             393,984  
;; Compute estimate: nil {type} {Student} nil :                              256  
;;   Choose q ?y {teacherOf} ?z: approximately 108,288 bindings for ?y, ?z.  
 
;; Compute estimate: nil {advisor} 108288 nil :                                9  
;; Compute estimate: nil {takesCourse} 108288 nil :                           14  
;; Compute estimate: nil {type} {ResearchAssistant} nil :                 36,608  
;; Compute estimate: nil {type} {GraduateStudent} nil :                  126,208  
;; Compute estimate: nil {type} {UndergraduateStudent} nil :             393,984  
;; Compute estimate: nil {type} {Student} nil :                              256  
;;   Choose q ?x {advisor} ?y: approximately 9 bindings for ?x.  
 
;; Compute estimate: 8.177778 {takesCourse} 108288 nil :                       1  
;; Compute estimate: 8.177778 {type} {ResearchAssistant} nil :                 1  
;; Compute estimate: 8.177778 {type} {GraduateStudent} nil :                   1  
;; Compute estimate: 8.177778 {type} {UndergraduateStudent} nil :              1  
;; Compute estimate: 8.177778 {type} {Student} nil :                           1  
;;   Choose q ?x {takesCourse} ?z.  
 
;; Compute estimate: 1 {type} {ResearchAssistant} nil :                        1  
;; Compute estimate: 1 {type} {GraduateStudent} nil :                          1  
;; Compute estimate: 1 {type} {UndergraduateStudent} nil :                     1  
;; Compute estimate: 1 {type} {Student} nil :                                  1  
;;   Choose q ?x {type} {Student}. 

And now the generated plan:

;; Return 3 variables: ?x, ?y, ?z  
(db.agraph::find-or-create-map #:?map13983 {advisor} :object :subject)  
(db.agraph::find-or-create-map #:?map13984 {takesCourse} :object  
 :subject)  
(db.agraph::find-or-create-map #:?map13985 (?? +rdf-type-uri+) :object  
 :subject)  
!  
;; estimate:      108,288 results per binding  
(q/upis- ?y {teacherOf} ?z)  
 
;; estimate:            9 results per binding  
(db.agraph::q-upi-table #:?map13983 ?y ?x)  
 
;; estimate:            1 results per binding  
(db.agraph::q-upi-table #:?map13984 ?z ?x)  
 
;; estimate:            4 results per binding  
(lispp*  
 (let ((reasoner (reasoner *db*)))  
   (declare (ignorable reasoner))  
   (or (db.agraph.upi-maps:upi-pair-present-p #:?map13985  
         {UndergraduateStudent} ?x)  
       (db.agraph.upi-maps:upi-pair-present-p #:?map13985  
         {GraduateStudent} ?x)  
       (db.agraph.upi-maps:upi-pair-present-p #:?map13985  
         {ResearchAssistant} ?x)  
       (db.agraph.upi-maps:upi-pair-present-p #:?map13985 {Student}  
         ?x)))) 

The compute estimates shows how the query planner is attempting to reorder the clauses more favorably. The planner currently uses a simple greedy heuristic but this will change as the planner evolves. Once the ordering is determined, the transformation process begins. There are two main transformations used in this query: UPI maps and reasoner simplification.

Transformations

The query-planner and dynamic-reasoner use the following broad categories of transformation: reordering, UPI maps, reasoner simplifications, and clause simplifications. Generally speaking, you should not worry about the exact nature and syntax of the transformations as these are both likely to change.

Reordering

When a query has many clauses, the order in which they are executed can have a huge effect on total query speed. The query-planner uses a set of heuristics and statistical information about the triple-store in order to try and determine the best order in which to run the query. The heuristics may fail especially for complex queries. If you suspect that the planner is ordering your query incorrectly, use the :display-plan and :reorder query-plan-options to diagnose the problem.

UPI maps transformation

A UPI map is a persistent index mapping one column of a triple to another. Typically, the index is over only a portion of all of the triples in a store. For example, a UPI map might map from object to subject for all rdf:type triples in the store. This map would make determining whether or not some x was of type C very fast. UPI maps provide a cache for certain kinds of queries and can dramatically improve query speed.

The query-planner and dynamic-reasoner determine when building UPI maps makes sense for a query and will build them as necessary the first time that they are needed. Once built, maps can be reused until the store changes. See the reference section below for some variables that help control how the query planner uses UPI maps.

The UPI map transformation appears in a query-plan in two places:

  1. The finding or building of the map. This looks like
  2. (find-or-create-map #:?map13983 {advisor} :object :subject) 
  3. The use of the map:
  4. (db.agraph::q-upi-table #:?map13983 ?y ?x) 

The :use-maps query-plan-option controls whether or not a query will use any UPI maps.

reasoner simplification

The dynamic-reasoner examines each clause in the context of the store when the query is run. Depending on the store's ontology, many clauses can be re-written into faster executing forms at query-plan time. Most of the transformations let AllegroGraph move computation out of the inner-loop of the reasoner; they can help greatly increase query speed. As an example, consider how AllegroGraph can transform (q ?x !rdf:type !ub:Student). The type information in the store will not change while the query is running, so AllegroGraph can consult the ontology once and determine that ?x can be inferred to be a student if there is a ground triple that says ?x is of rdf:type ub:UndergraduateStudent, ub:GraduateStudent, ub:ResearchAssistant, or ub:Student. Adding UPI maps into the mix, the finished transformation looks like:

(lispp*  
 (or (db.agraph.upi-maps:upi-pair-present-p #:?map13985  
       {UndergraduateStudent} ?x)  
     (db.agraph.upi-maps:upi-pair-present-p #:?map13985  
       {GraduateStudent} ?x)  
     (db.agraph.upi-maps:upi-pair-present-p #:?map13985  
       {ResearchAssistant} ?x)  
     (db.agraph.upi-maps:upi-pair-present-p #:?map13985 {Student}  
       ?x))) 

The set of transformation and their actual syntax will change over time as the query-planner and dynamic-reasoner become more capable.

Clause simplification

Though reasoning transformations provide the most dramatic speedups, there are some other transformation that can also make queries more efficient. Most of these have more to do with reducing consing and shaving small amounts off of clause execution. An example is the transformation of q- clauses into asserted-1, asserted-po, asserted-sp, or asserted-p clauses. Because the latter four have slightly simpler execution paths, they provide measurable increases in speed.

Note that these asserted-* functors may change or disappear in future revisions and should not be relied in outside of the query planner itself.

Store Management Options

The select options :graph-filter and :reasoner allow a program to modify the store used in the select query without affecting the current default store in the surrounding program.

A query such as

(select (?x)  
   (:graph-filter !ex:g1)  
   (q ?x !ex:pred ?y)) 

is applied to the subset of triples where the graph component is !ex:g1.

And the query

(select (?x)  
   (:graph-filter !ex:g1)  
   (:reasoner)  
   (q ?x !ex:pred ?y)) 

is applied to all the triples that can be inferred from the set of triples where the graph component is !ex:g1.

In both cases the triple store (bound to db) in the surrounding context is not affected.

Note that a query such as

(select (?x)  
   (:reasoner)  
   (:graph-filter !ex:g1)  
   (q ?x !ex:pred ?y)) 

is likely to yield surprising or unexpected results. In this case, the graph filter is applied to the triples generated by the reasoner. But in the default case, these inferred triples use the default graph of the base store as the graph component of the inferred triples. It would be more constructive to pose the query

(select (?x)  
   (:reasoner :inferred-graph !ex:g2)  
   (:graph-filter !ex:g1 !ex:g2)  
   (q ?x !ex:pred ?y))  
 
 
 
 
 

Reference

default-query-plan-option name
function
Return the default value of query-plan-option name.
defined-query-plan-options &optional  include-values-p
function
Return a list of the defined query plan options and, optionally, their settings.
*expand-type-clause-with-map-p*
variable

Controls whether or not UPI maps are used with type inference.

If true, then the dynamic-reasoner will build a UPI map on rdf:type and use this map for subsequent type inference.

*last-query-plan*
variable

This variable will be nil or contain an instance of the type query-plan.

query-plan is an opaque data structure. It should not generally be used other than internally.

*use-maps-for-subproperties-s-o-p*
variable

Controls whether or not UPI maps may be used for sub-property reasoning.

If true, then the dynamic-reasoner may decide to use UPI maps when a predicate is involved in sub-property reasoning.

RDFS++ Reasoning

AllegroGraph's reasoner supports the following predicates:

and OWL restriction reasoning like owl:hasValue, owl:someValuesFrom and owl:allValuesFrom (see the reasoner tutorial and the hasValue reasoning tutorials for more details).

Note that the reasoner is designed to efficiently answer questions like "What are properties of this subject?" or "Tell me all the subjects that have this type?" It is not designed for efficiently determining all of the inferred triples in a store (e.g., "Tell me everything you know."). You should expect fast answers for queries whose predicate is specified but you will not see such good performance for more open-ended calls to get-triples. We will continue to enhance the reasoner as AllegroGraph develops.

The following additional functions and variables are also part of the reasoner:

apply-rdfs++-reasoner &rest  args  &key  db  name  restriction-reasoning-enabled  remote-reasoning  inferred-graph  cached-wrapper-key
function

Return a new reasoning-triple-store that encapsulates db and adds an rdfs++-reasoner.

  • db - the triple-store to encapsulate. The defaults to the value of *db*.

  • name - the name to give the encapsulated-triple-store. Defaults to the current db-name of the db prefixed with rdfs++- and with a generated suffix .

  • restriction-reasoning-enabled - when true, enable OWL hasValue reasoning. The default is nil.

  • inferred-graph - This argument specifies the graph component of all inferred triples. The default is to use the default graph of the specified triple store.

  • cached-wrapper-key - If this argument is nil, the call creates a new instance of a reasoning wrapper; this instance will cause a new invocation of prepare-reasoning when the first query is issued. If the argument is not nil, it is used as a key in an equalp hash-table; if a previously created wrapper is found, the previous instance is returned. If no entry is found, a new wrapper is created and saved in the table. The default value is taken from the variable *cached-wrapper-key*. The initial value of this variable is nil.

  • remote-reasoning - This argument is meaningful only when the db (or *db*) argument is a remote-triple-store instance. A non-nil value specified that the reasoning wrapper is created on the server side. The value returned by apply-rdfs++-reasoner is a new remote-triple-store instance pointing to the reasoner on the server side.
inferred-triple-p triple
function
Returns true if triple is the result of inference (as opposed to be physically present in the triple-store).
rdfs++-reasoner
class
Though not a formal definition, an RDFS++ reasoner provides RDFS entailment along with owl:sameAs and owl:InverseFunctionalProperty. The AllegroGraph reasoner also supports inverse properties.
reasoner g27570
function
Returns the reasoner of a reasoning-triple-store.
reasoning-triple-store
class
An encapsulated-triple-store that adds reasoning to a triple-store. This store contains a reasoner slot that controls the sort of reasoning that the triple-store supplies. The rdfs++-reasoner is an example of the sort of reasoning that AllegroGraph provides.

AllegroGraph also supports restriction reasoning. 11

restriction-reasoning-enabled-p rdfs++-reasoner
function

Returns whether or not restriction reasoning is enabled in the RDFS++ reasoner.

This includes reasoning about owl:hasValue, owl:someValuesFrom and owl:allValuesFrom.

You can control whether or not has-value reasoning is enabled when you use apply-rdfs++-reasoner.

Prepare-reasoning is a function called internally by the AllegroGraph reasoner. It is unlikely that you will ever need to call it yourself. Its purpose is to update the internal data structures used by the reasoner.

prepare-reasoning reasoner  &key  verbose  force  show-progress
function

This function has to be called before any inferences are made. It creates internal hash-tables to speed up the reasoner. In normal operation, AllegroGraph will call prepare-reasoning as necessary. You can see diagnostic messages by using the parameter verbose which defaults to the value of nil. You can also force the hash-tables to be regenerated using the force parameter. Finally, the show-progress keyword argument can be used to cause prepare-reasoning to print a message for each of the hash-tables it builds.

The function prepare-reasoning returns no value.

Specialized Datatypes

In addition to the "encoded UPIs" for the various computer numeric datatypes, AllegroGraph supports several specialized datatypes for efficient storage, manipulation, and search of Social Network, Geospatial and Temporal information.

Social Network Analysis

By viewing interactions as connections in a graph, we can treat a multitude of different situations using the tools of Social Network Analysis (SNA). SNA lets us answer questions like:

A Example Social Graph

The field is full of rich mathematical techniques and powerful algorithms. AllegroGraph's SNA toolkit includes an array of search methods, tools for measuring centrality and importance, and the building blocks for creating more specialized measures. These tools can be used with any "network" data set whether its connections between companies and directors, predator/prey food webs, chemical interactions, or links between web sites.

Navigating Through a Graph - Generators

The graph functions described below all use generators in their definitions. A generator is a function of one-argument, a vertex, that returns a list of nodes that are connected to the vertex.

We provide several helper utilities and a generator definition language to define both simple ad complex generator functions. When the graph operations below take a generator as argument, you may supply either a function of one argument or the name of a generator that has been defined using the generator definition language or a neighbor-matrix (see below).

defsna-generator name  (&optional  var)  &body  body
macro

Define a new SNA generator function and associate it with `name.

A generator is a function of one-argument, a vertex, that returns a list of nodes that are connected to the vertex. In a directed graph, connected may refer to the nodes that point to a specific vertex (the subjects-of) or it may refer to the nodes at which the vertex points (the objects-of). Generators can also ignore edge direction in a graph and return a list of nodes that are adjacent to the specific vertex.

The body should be a list of clauses where each clause is one of:

  • (:objects-of &key node p)
  • (:subjects-of &key node p)
  • (:undirected &key p)
  • (select var select-clause*)

When called, the generator returns the distinct union of the nodes found by each clause.

Some simple examples:

(defsna-generator involved-with (node)  
  (:objects-of :node node :p (!ex:kisses !ex:sleepsWith)))  
 
;; we can leave off `node`  
(defsna-generator involved-with ()  
  (:objects-of :p (!ex:kisses !ex:sleepsWith)))  
 
(defsna-generator involved-with ()  
  (:undirected :p (!ex:kisses !ex:sleepsWith)))  
 
(defsna-generator in-movie-with (node)  
  (select (?x)  
    (q- (?? ?x) !franz:stars node))) 

defsna-generator returns the new function.

Note that:

  • In both :objects-of and :subjects-of, the node parameter is syntactic sugar and can be left off.

  • In all three of :objects-of, :subjects-of, and :undirected, the p variable can be a single property or a list of properties.
find-sna-generator name
function

Lookup a registered sna-generator by name and return its associated function.

name can be:

  • a generator function itself (in which case it is simply returned),

  • a neighbor-matrix created via build-neighbor-matrix

  • the name of a generator previously defined with the defsna-generator macro,

  • the name of a neighbor-matrix previously defined with defneighbor-matrix, or

  • a symbol naming an existing function which is assumed to act as a generator.
register-sna-generator name  function
function
Associate name with an SNA generator function. Returns the new function. If the function is nil, then the association is removed.
Examples
(register-namespace "ex" "http://namespace.franz.com/namespace#")  
 
(defsna-generator involved-with ()  
  (:objects-of :p (!ex:kisses !ex:intimateWith)))  
 
(defsna-generator involved-with (node)  
  (:objects-of :node node :p (!ex:kisses !ex:intimateWith)))  
 
(defsna-generator involved-with ()  
  (:undirected :p (!ex:kisses !ex:intimateWith)))  
 
(defsna-generator in-movie-with (node)  
  (select (?x)  
    (q- ?x !franz:stars (?? node)))) 

(Note the use of Allegro Prolog ?? idiom that we use to treat the symbol node as a variable in the body of the macro.)

Saving time - the Neighbor-Matrix

Generators provide a fast and flexible mechanism to navigate though arbitrary graphs. AllegroGraph automatically saves the results of recent node expansions but generators still require repeated queries against the triple-store. If you know that you will be operating on a particular group of nodes in your graph, you can greatly increase the speed of some of the SNA algorithms by building a neighbor-matrix (which is essentially an in-memory sparse adjacency matrix) and using it rather than the generator in your queries. You can build a neighbor-matrix with build-neighbor-matrix.

defneighbor-matrix name  group  generator  max-depth
macro

Defines a Neighbor-Matrix structure for group using generator including all nodes reachable within max-depth steps from group and associates it with name.

If max-depth is zero, then only the group will be queried; if it is one, then the cache will include the group and it neighbors; etc.

build-neighbor-matrix group  generator  max-depth
function
Return an Neighbor-Matrix structure for group using generator including all nodes reachable within max-depth steps from group. I.e., If max-depth is zero, then only the group will be queried; if it is one, then the cache will include the group and it neighbors; etc.
find-neighbor-matrix name  &optional  errorp
function

Lookup a registered neighbor-matrix by name and return it.

name should be a neighbor-matrix (in which case it is returned) or a symbol naming a previously registered neighbor-matrix. See defneighbor-matrix macro.

register-neighbor-matrix name  matrix
function

Associate the neighbor-matrix matrix with name.

To remove a matrix use register-neighbor-matrix with nil.

The neighbor-matrix for a particular group and generator saves all of the information needed to use the SNA algorithms (but only for that group and that generator). To use it, call the algorithm you want but pass in the neighbor-matrix instead of the generator. Secondly, the group argument to the function must be set to nil and any actor arguments must be members of the group used to create the matrix. The following examples compute the same result:

Using just the generator
> (defsna-generator knows ()  
    (:undirected :p !f:knows))  
#<Function (:anonymous-lambda 35) @ #x11e18772>  
 
> (setf group (ego-group !<http://www.franz.com/ont/person-1> 2 'knows))  
({person-898} {person-47649} {person-91744} {person-90811}  
 {person-17690} {person-59736} {person-84513} {person-63689}  
 {person-91562} {person-62312} ...)  
 
> (length group)  
147  
 
> (time (group-degree-centrality group 'knows))  
; cpu time (non-gc) 100 msec user, 120 msec system  
; cpu time (gc)     20 msec user, 0 msec system  
; cpu time (total)  120 msec user, 120 msec system  
; real time  251 msec  
; space allocation:  
;  4,795 cons cells, 1,629,912 other bytes, 0 static bytes  
7.1275467e-4 
Using the Neighbor-Matrix
> (time (setf neighbor-matrix  
	    (build-neighbor-matrix group 'knows 0)))  
; cpu time (non-gc) 130 msec user, 170 msec system  
; cpu time (gc)     30 msec user, 0 msec system  
; cpu time (total)  160 msec user, 170 msec system  
; real time  1,263 msec  
; space allocation:  
;  5,577 cons cells, 965,136 other bytes, 0 static bytes  
#<neighbor-matrix 147 members @ #x11bbbe12>  
 
> (time (group-degree-centrality group neighbor-matrix))  
; cpu time (non-gc) 10 msec user, 0 msec system  
; cpu time (gc)     0 msec user, 0 msec system  
; cpu time (total)  10 msec user, 0 msec system  
; real time  9 msec  
; space allocation:  
;  594 cons cells, 684,728 other bytes, 0 static bytes  
7.1275467e-4 

It's clear that the in-memory neighbor-matrix gives a tremendous speed boost (and is much more memory efficient). On the down side, it took some time to build so it would not make sense to use it unless you need to expand nodes many times. One area where this occurs is whenever an SNA algorithm relies on path-finding. For example, the actor-closeness-centrality algorithm requires finding the shortest path between an actor and every other member of the group. Lets compare:

> (time (actor-closeness-centrality  
         !<http://www.franz.com/ont/person-1> group 'knows))  
; cpu time (non-gc) 1,750 msec user, 2,140 msec system  
; cpu time (gc)     100 msec user, 10 msec system  
; cpu time (total)  1,850 msec user, 2,150 msec system  
; real time  20,200 msec  
; space allocation:  
;  84,703 cons cells, 13,531,840 other bytes, 0 static bytes  
0.5214286  
 
> (time (actor-closeness-centrality  
         !<http://www.franz.com/ont/person-1> group neighbor-matrix))  
; cpu time (non-gc) 10 msec user, 0 msec system  
; cpu time (gc)     0 msec user, 0 msec system  
; cpu time (total)  10 msec user, 0 msec system  
; real time  12 msec  
; space allocation:  
;  13,273 cons cells, 1,803,600 other bytes, 0 static bytes  
0.5214286 

As the groups get larger, path-finding algorithms do not scale well but this poor performance can be offset for a while using the neighbor-matrices.

Search

There are many ways to search in a graph and the one you choose will depend both on the application and on the kind of graph data you have collected. The SNA library provides multiple search functions whose operation is controlled by generators as defined above. Most operations come in several flavors. For example,

We also provide an additional set of search functions specialized for bipartite graphs.

Breadth First Search
breadth-first-search start  end  generator  &key  maximum-depth  db
function

Search db in breadth first fashion starting at start and looking for end, using generator to expand the search space. Returns nil if no path can be found (within depth) and the first such path found otherwise. The arguments are:

  • start - a UPI or future-part or string that can be converted into a UPI.
  • end - a UPI, future-part or string that can be converted into a UPI.
  • generator - a function of one argument or a generator name. See the description of generators for additional details.
  • maximum-depth - If supplied, this restricts the length of the paths searched. Only paths (from start) less than or equal to this will be checked. If maximum-depth is not specified, it defaults to nil which means to search all paths. Warning in large graphs this can be very expensive.
breadth-first-distance start  end  generator  &key  maximum-depth  db
function
Return the length of the first path found between start and end using breadth-first-search to find the path and generator to determine when two nodes are adjacent. Returns nil if no path exists.
map-breadth-first-search-paths start  end  generator  fn  &key  maximum-depth  db
function
Apply fn to each path between start and end using generator to determine adjacent nodes and breadth-first-search to find paths.
all-breadth-first-search-paths start  end  generator  &key  maximum-depth  db
function

Return a list of all (the shortest) paths between start and end using generator to determine adjacent nodes and breadth-first-search to find paths.

all-breadth-first-search-paths will search for a path no longer than maximum-depth. If maximum-depth is left unspecified, then the entire graph will be searched (obviously, this can be very expensive in large graphs). Note that once a path is found, no longer paths will be found (i.e., the maximum-depth becomes the length of the first path found.) Since breadth-first search is guaranteed to find the shortest path between start and end, this function actually returns a list of the shortest paths.

Please contact [email protected] if your path finding needs are not met by this algorithm so that we can work to make our SNA module more flexible.

Depth First Search
depth-first-search start  end  generator  &key  maximum-depth  db
function

Search db in depth first fashion starting at start and looking for end, using generator to expand the search space. Returns nil if no path can be found (within depth) and the first path found otherwise. The arguments are:

  • start - a UPI, future-part or string that can be converted into a UPI.
  • end - a UPI, future-part or string that can be converted into a UPI.
  • generator - a function of one argument or a generator name. See the description of generators for additional details.
  • maximum-depth - If supplied, this restricts the length of the paths searched. Only paths (from start) less than or equal to this will be checked. If maximum-depth is not specified, it defaults to nil which means to search all paths. Warning in large graphs this can be very expensive.
depth-first-distance start  end  generator  &key  maximum-depth  db
function
Return the length of the first path found between start and end using depth-first-search to find the path and generator to determine when two nodes are adjacent. Returns nil if no path exists.
depth-first-search-path
(depth-first-search-path +start +end +generator -path) (depth-first-search-path +start +end +generator +max-depth -path)
functor

Unifies path with the first path found from start to end using a depth first search strategy.

  • start - the node at which to start
  • end - the node at which to stop
  • generator - a function that specifies how to find the neighbors of an actor in the graph (see generators for more details).
  • max-depth - the maximum depth to search.
  • path - bound to a list of UPIs representing a path from start to end. Each element in the path will be a UPI U such that there is a triple whose subject is U; whose object is the next element in the list; and which is returned by a call to generator. Finds the first path between start and end of any length. Warning Searching for a path of arbitrary length in a large triple-store can take a long time.

Finds the first path between start and end that is no longer than max-depth.

map-depth-first-search-paths start  end  generator  fn  &key  maximum-depth  db
function
Apply fn to each path between start and end using generator to determine adjacent nodes and depth-first-search to find paths.
all-depth-first-search-paths start  end  generator  &key  maximum-depth  db
function

Return a list of all paths between start and end using generator to determine adjacent nodes and depth-first-search to find paths.

all-depth-first-search-paths will search for a path no longer than maximum-depth. If maximum-depth is left unspecified, then the entire graph will be searched (obviously, this can be very expensive in large graphs). Note that once a node appears in a path, it will not appear in any longer paths (though it may appear in paths of the same or shorter length depending on the order in which nodes are discovered.

Please contact [email protected] if your path finding needs are not met by this algorithm so that we can work to make our SNA module more flexible.

Bidirectional Search
bidirectional-search start  end  generator  &key  maximum-depth  db
function
Search to see if a path exists between start and end (using generator to determine when two nodes are adjacent.
bidirectional-distance start  end  generator  &key  maximum-depth  db
function
Return the length of the first path found between start and end using bidirectional-search to find the path and generator to determine when two nodes are adjacent. Returns nil if no path exists.
map-bidirectional-search-paths start  end  generator  fn  &key  maximum-depth  db
function
Apply fn to each path between start and end using generator to determine adjacent nodes and bidirectional-search to find paths.
all-bidirectional-search-paths start  end  generator  &key  maximum-depth  db
function

Return a list of all (shorter) paths between start and end using generator to determine adjacent nodes and bidirectional-search to find paths.

all-bidirectional-search-paths will search for a path no longer than maximum-depth. If maximum-depth is left unspecified, then the entire graph will be searched. Note that once a path is found, no longer paths will be searched. (Though some longer paths may appear in the results after a shorter one because of the recursive algorithm employed.)

Please contact [email protected] if your path finding needs are not met by this algorithm so that we can work to make our SNA module more flexible.

Bipartite graphs
map-breadth-first-search-paths-bipartite start  end  generator-1  generator-2  fn  &key  maximum-depth  db
function
Apply fn to each path between start and end using generator-1 and generator-2 to determine adjacent nodes and breadth-first-search to find paths. The function assumes that generator-1 can be used on the start node; the node type of the end node does not matter.
map-depth-first-search-paths-bipartite start  end  generator-1  generator-2  fn  &key  maximum-depth  db
function
Apply fn to each path between start and end using generator-1 and generator-2 to determine adjacent nodes and depth-first-search to find paths. The function assumes that generator-1 can be used on the start node; the node type of the end node does not matter.
map-bidirectional-search-paths-bipartite start  end  generator-1  generator-2  fn  &key  maximum-depth  db
function
Apply fn to each path between start and end using generator-1 and generator-2 to determine adjacent nodes and bidirectional-search to find paths.

Graphs Measures

There are many ways to measure information about a graph and its structure. Some of these focus on the graph as a whole while others target particular actors (vertices).

Who is nearby
nodal-neighbors node  generator  &key  db
function

nodal-neighbors returns a list of other nodes that can be reached from node by following an edge. Node should be a vertex in the graph represented by the triple-store; See the description of generators for additional details.

Note that the nodal-degree of a node is the same as `(length (nodal-neighbors node)). See nodal-degree for additional details.

nodal-degree node  generator
function

The nodal-degree of node is the total number of edges that connect it to other nodes in the graph. Edges are found using the generator. See the description of generators for additional details.

Note that the interpretation of nodal-degree depends entirely on the generator being using in the calculation. Suppose, for example, that we have a graph with three nodes and two edges:

 A --> B <-- C 

and that these are encoded in AllegroGraph as

!ex:A !ex:linksTo !ex:B  
!ex:C !ex:linksTo !ex:B 

If we define three generators:

(defsna-generator linksTo (node)  
  (:objects-of :p !ex:linksTo))  
 
(defsna-generator linksFrom (node)  
  (:subjects-of :p !ex:linksTo))  
 
(defsna-generator links (node)  
  (:undirected :p !ex:linksTo)) 

then we will find that

> (nodal-degree !ex:B 'linksTo)  
0  
 
> (nodal-degree !ex:B 'linksFrom)  
2  
 
> (nodal-degree !ex:B 'links)  
2 

Finally, note that nodal-degree is the same as the length of the list of nodes returned by nodal-neighbors.

ego-group node  depth  generator  &key  db
function

The ego-group function returns a list of nodes in nodes ego group.

These are the other nodes in the graph (represented by the triple-store) that can be reached by following paths of at most length depth starting at node. The paths are determined using the generator to select neighbors of each node in the ego-group. See the description of generators for additional details.

Centrality Measures

The centrality of an actor in a network is a measure of its prestige or importance or power. There are many ways to measure it and the best way varies with the application.

graph-density subgraph  generator  &key  db
function
This is the normalized average degree of the actors in the graph. Roughly, how many of the possible connections between actors are realized. Subgraph is a list of nodes and generator is a generator function or name. See the description of generators for additional details.
actor-degree-centrality actor  group  generator  &key  db
function
An actor centrality measure provides insight into how central or important an actor is in the group. Degree-centrality uses the number of actor connections (the degree) as a proxy for importance. Actor is a node in the graph; group is a list of nodes (e.g., the actor's ego group) and generator is either a function or name. See the description of generators for additional details.
group-degree-centrality group  generator  &key  db
function
Group centrality measures the cohesion a group relative to some measure of actor-centrality. group-degree-centrality measures group cohesion by finding the maximum actor centrality in the group, summing the difference between this and each other actor's degree centrality and then normalizing. It ranges from 0 (when all actors have equal degree) to 1 (when one actor is connected to every other and no other actors have connections).
actor-closeness-centrality actor  group  generator  &key  db
function

An actor centrality measure provides insight into how central or important an actor is in the group. Closeness-centrality is the (normalized) inverse average path length of all the shortest paths between the actor and every other member of the group. (The inverse so that higher values indicate more central actors).

Actor is a node in the graph; group is a list of nodes (e.g., the actor's ego group) and generator is either a function or name. See the description of generators for additional details.

group-closeness-centrality group  generator  &key  db
function
Group centrality measures the cohesion a group relative to some measure of actor-centrality. group-closeness-centrality is measured by first finding the actor whose closeness-centrality is maximized and then summing the difference between this maximum value and the actor-closeness-centrality of all other actors. This value is then normalized so that it ranges between 0 and 1.
actor-betweenness-centrality actor  group  generator  &key  db
function

An actor centrality measure provides insight into how central or important an actor is in the group. Betweenness-centrality measures how much control an actor has over communication in the group.

The actor-betweenness-centrality of actor i is computed by counting the number of shortest paths between all pairs of actors (not including i) that pass through actor i. The assumption being that this is the chance that actor i can control the interaction between j and k.

Actor is a node in the graph; group is a list of nodes (e.g., the actor's ego group) and generator is either a function or name. See the description of generators for additional details.

group-betweenness-centrality group  generator  &key  db
function
Group centrality measures the cohesion a group relative to some measure of actor-centrality. group-betweenness-centrality is measured by first finding the actor whose betweenness-centrality is maximized and then summing the difference between this maximum value and the actor-betweenness-centrality of all other actors. This value is then normalized so that it ranges between 0 and 1.
page-rank-centrality group  generator  &key  dampening-factor  db
function

Page-rank centrality is a specific implementation of a more general eigenvector centrality measure. Given a set of nodes in a graph, the page-rank-centrality function returns a vector representing the importance or influence of each node. Though it is not actually computed like this, you can imagine that we repeatedly pick a node at random and then execute a random walk following the edges of the graph while noticing which nodes our walker visits. The page-rank centrality is the probability that the walker will be on a particular node when it stops.

  • group - a list of nodes in the triple-store
  • generator - a generator describing how to get from one to another
  • :dampening-factor - the effects how quickly the iterate algorithm stabilizes

Returns a vector of (unnormalized) page-ranks in the same order as the nodes in group.

Cliques

In graph theory, a clique is a fully connected subgraph in a larger graph. AllegroGraph supplies function to determine is a subgraph is a clique and to find all of the cliques in a graph.

cliquep subgraph  generator  &key  db
function
Return true if every member of subgraph is linked to every other member. As usual, the generator is used to determine whether or not a link exists between two nodes. See the description of generators for additional details.
cliques node  generator  &key  minimum-size  maximum-size  maximal-only
function
Return a list of the cliques of which node is a member (using generator to find adjacent nodes). Each clique returned will contain at least minimum-size and no more than maximum-size members. If `maximal-only is true, then only the largest cliques will be returned.
map-cliques node  generator  fn  &key  minimum-size  maximum-size  maximal-only
function
Apply fn to each clique in which node is a member and whose size is at least minimum-size and less than maximum-size. The generator is used to find adjacent nodes. If maximal-only` is true, then only maximal-cliques are mapped.

Prolog and Social Network Analysis

The functions described above have analogues in the Prolog world.

Prolog Graph Search
breadth-first-search-paths
(breadth-first-search-paths +start +end +generator -path) (breadth-first-search-paths +start +end +generator +max-depth -path)
functor

Unifies path with each path found from start to end using a breadth first search strategy.

  • start - the node at which to start
  • end - the node at which to stop
  • generator - a function that specifies how to find the neighbors of an actor in the graph (see generators for more details).
  • max-depth - the maximum depth to search.
  • path - bound to a list of UPIs representing a path from start to end. Each element in the path will be a UPI U such that there is a triple whose subject is U; whose object is the next element in the list; and which is returned by a call to generator. Iterates over paths between start and end of all lengths. Warning Searching for all possible paths in a large triple-store can take a long amount of time.

Iterates over paths between start and end that are no longer than max-depth.

breadth-first-search-path
(breadth-first-search-path +start +end +generator -path) (breadth-first-search-path +start +end +generator +max-depth -path)
functor

Unifies path to a list of UPIs connecting start and end using generator using a breadth first search strategy.

  • start - the node at which to start
  • end - the node at which to stop
  • generator - a function that specifies how to find the neighbors of an actor in the graph (see generators for more details).
  • max-depth - the maximum depth to search.
  • path - bound to a list of UPIs representing a path from start to end. Each element in the path will be a UPI U such that there is a triple whose subject is U and whose object is the next element in the list and which is returned by a call to generator. Binds path to a list of UPIs connecting start and end.

The path will use generator to find the links between nodes. Since a breadth-first search strategy is used, the shortest path will be found.

Binds path to a list of UPIs connecting start and end.

The path will be no longer than max-depth and will use generator to find the links between nodes. Since a breadth-first search strategy is used, the shortest path will be found.

depth-first-search-paths
(depth-first-search-paths +start +end +generator -path) (depth-first-search-paths +start +end +generator +max-depth -path)
functor

Unifies path with each path found from start to end using a depth first search strategy.

  • start - the node at which to start
  • end - the node at which to stop
  • generator - a function that specifies how to find the neighbors of an actor in the graph (see generators for more details).
  • max-depth - the maximum depth to search.
  • path - bound to a list of UPIs representing a path from start to end. Each element in the path will be a UPI U such that there is a triple whose subject is U; whose object is the next element in the list; and which is returned by a call to generator. Iterates over paths between start and end of all lengths. Warning Searching for all possible paths in a large triple-store can take a long amount of time.

Iterates over paths between start and end that are no longer than max-depth.

bidirectional-search-paths
(bidirectional-search-paths +start +end +generator -path) (bidirectional-search-paths +start +end +generator +max-depth -path)
functor

Unifies path with each path found from start to end using a bidirectional search strategy (i.e., preceding from start and end simultaneously).

  • start - the node at which to start
  • end - the node at which to stop
  • generator - a function that specifies how to find the neighbors of an actor in the graph (see generators for more details).
  • max-depth - the maximum depth to search.
  • path - bound to a list of UPIs representing a path from start to end. Each element in the path will be a UPI U such that there is a triple whose subject is U; whose object is the next element in the list; and which is returned by a call to generator. Iterates over paths between start and end of all lengths. Warning Searching for all possible paths in a large triple-store can take a long amount of time.

Iterates over paths between start and end that are no longer than max-depth.

bidirectional-search-path
(bidirectional-search-path +start +end +generator -path) (bidirectional-search-path +start +end +generator +max-depth -path)
functor

Unifies path with the first path found from start to end using a bidirectional search strategy (i.e., preceding from start and end simultaneously).

  • start - the node at which to start
  • end - the node at which to stop
  • generator - a function that specifies how to find the neighbors of an actor in the graph (see generators for more details).
  • max-depth - the maximum depth to search.
  • path - bound to a list of UPIs representing a path from start to end. Each element in the path will be a UPI U such that there is a triple whose subject is U; whose object is the next element in the list; and which is returned by a call to generator. Binds path to a list of UPIs connecting start and end.

Binds path to a list of UPIs start and end that is no longer than max-depth.

Prolog Graph Measures
nodal-degree
(nodal-degree +actor +generator -degree)
functor

Unifies degree to be the nodal-degree of actor in the current triple-store (i.e., the current value of *db*).

  • actor - a node in the graph
  • generator - the name of the generator function used to compute adjacent nodes in the graph (see the section on generators for more details).
nodal-neighbors
(nodal-neighbors +actor +generator -neighbor)
functor

Unifies neighbor to each vertex that has an edge pointing away from or towards actor in the graph defined by generator in the current triple-store (i.e., the current value of *db*).

  • actor - a node in the graph
  • generator - the name of the function used to determine adjacent nodes. (See generators for more details).
graph-density
(graph-density +group +generator -density)
functor

The density of group is the normalized average degree of the actors in the group. Roughly, this is a measure of how many of the possible connections between actors are actually realized.

  • group - a list of nodes in the graph (these can be UPIs, future-parts, or strings that can be coerced into UPIs).

  • generator - a function that specifies how to find the neighbors of an actor in the graph (see generators for more details).
ego-group
(ego-group +actor +depth +generator -group)
functor

Unify group to the list of UPIs representing the ego group of actor in the graph.

  • actor - a node in the graph
  • depth - how many links to travel when fining the ego group.
  • generator - a function that specifies how to find the neighbors of an actor in the graph (see generators for more details).

See ego-group-member to iterate over each member instead of collecting them all into a list.

ego-group-member
(ego-group-member +actor +depth +generator +member) (ego-group-member +actor +depth +generator -member)
functor

If member is supplied, then ego-group-member succeeds if and only if member is in the ego-group of actor.

Unify member with each of the other actors in actor's ego-group.

  • actor - a node in the graph
  • depth - how many links to travel when fining the ego group.
  • generator - a function that specifies how to find the neighbors of an actor in the graph (see generators for more details).

See ego-group to directly collect all of the members of an actor's ego-group.

clique
(clique +actor +generator -clique) (clique +actor +minimum-size +generator -clique)
functor

Unify clique with each of the graph cliques (i.e., completely connected sub-graphs) in graph in which actor plays a member.

  • actor - a node in the graph
  • generator - a function that specifies how to find the neighbors of an actor in the graph (see generators for more details). The arity three version of clique iterates over all cliques that have three or more members

The arity four version of clique iterates over all cliques that have minimum-size or more members

cliquep
(cliquep +group +generator)
functor

Succeeds if group is a clique in graph (where edges between nodes are defined by generator).

  • group - a list of items that can be coerced into UPIs in the graph
  • generator - a function that specifies how to find the neighbors of an actor in the graph (see generators for more details).
actor-degree-centrality
(actor-degree-centrality +actor +group +generator -centrality)
functor

Unify centrality with the actor-degree-centrality of actor.

  • actor - a UPI in the graph (or something that can be coerced into a UPI).
  • group - a list of items that can be coerced into UPIs in the graph
  • generator - a function that specifies how to find the neighbors of an actor in the graph (see generators for more details).
group-degree-centrality
(group-degree-centrality +group +generator -centrality)
functor

Unify centrality with the group-degree-centrality of group.

  • group - a list of items that can be coerced into UPIs in the graph
  • generator - a function that specifies how to find the neighbors of an actor in the graph (see generators for more details).
actor-closeness-centrality
(actor-closeness-centrality +actor +group +generator -centrality)
functor

Unify centrality with the actor-closeness-centrality of actor.

  • actor - a UPI in the graph (or something that can be coerced into a UPI).
  • group - a list of items that can be coerced into UPIs in the graph
  • generator - a function that specifies how to find the neighbors of an actor in the graph (see generators for more details).
group-closeness-centrality
(group-closeness-centrality +group +generator -centrality)
functor

Unify centrality with the group-closeness-centrality of group.

  • group - a list of items that can be coerced into UPIs in the graph
  • generator - a function that specifies how to find the neighbors of an actor in the graph (see generators for more details).
actor-betweenness-centrality
(actor-betweenness-centrality +actor +group +generator -centrality)
functor

Unify centrality with the actor-betweenness-centrality of actor.

  • actor - a UPI in the graph (or something that can be coerced into a UPI).
  • group - a list of items that can be coerced into UPIs in the graph
  • generator - a function that specifies how to find the neighbors of an actor in the graph (see generators for more details).
group-betweenness-centrality
(group-betweenness-centrality +group +generator -centrality)
functor

Unify centrality with the group-betweenness-centrality of group.

  • group - a list of items that can be coerced into UPIs in the graph
  • generator - a function that specifies how to find the neighbors of an actor in the graph (see generators for more details).

Geospatial Primitives

Introduction

AllegroGraph provides a novel mechanism for efficient storage and retrieval of data elements which are coordinates in a two-dimensional region. We often refer to this as "geospatial" although that term refers more specifically to positions on or around the earth's surface. AllegroGraph supports a more-general notion of two-dimensional coordinates, both on a flat plane such as the surface of a silicon chip, a sphere such as the earth's surface, or any other two-dimensional geometry system a user can define by writing some additional Lisp code. However, we will use the term geospatial to include any such data when there is no danger of ambiguity

AllegroGraph supports a :geospatial UPI type. The geospatial UPI type supports up to 256 subtypes, some predefined and others available for the application programmer to custom design. A geospatial UPI always encodes both an X and Y ordinate inside a single UPI in a way that allows efficient searching. The subtype specifies how external numeric data (e.g. degrees, radians, meters, or nanometers) is converted to and from the geospatial UPI internal representation.

Geospatial Subtypes

It is the job of a geospatial subtype to convert coordinates in some external representation to and from UPI representation. There are helper functions for defining certain important kinds of subtypes, for example, latitude and longitude on the earth's surface, and simple X:Y coordinates over a finite range of the infinite Cartesian plane.

The precision of each ordinate in the internal UPI representation is 32-bits. Dividing the circumference of the earth by 2^32 reveals that the resolution of a 32-bit integer longitude is slightly better than 1 cm.

Suppose we want to search a triple store for geospatial UPIs that are close to a certain XY, i.e., within a bounding box [X1,Y1,X2,Y2]. In a traditional RDF store (perhaps using :latitude and :longitude types) the X and Y would be in separate triples. If the two ordinates were combined a single part, it would still only be possible to search over a range of {X1,X2}, but within that range every Y value must be retrieved and range tested. As the size of the data increases, and therefore the total number of UPIs within the X range {X1,X2} increases, this scales badly.

A geospatial UPI does combine the two ordinates into a single value, but it adds a third component to the UPI. The Y dimension (or latitude) is decimated into strips of a fixed width, and UPIs are sorted first on the strip, then on the X, and finally on the actual Y value.

How does the Y strip help? In any of the AllegroGraph indexes, UPIs in any particular one of the SPOG are sorted as 12-byte unsigned integers. Within any index, all geospatial UPIs of a given subtype will sort together, first sorted on Y strip, then on the unsigned 32-bit integer X, then finally on the unsigned 32-bit integer Y. The Y strip is a decimation of the Y value. In other words, the Y values are divided into narrow strips. UPIs (within a given subtype) are sorted first on the strip, then the X, then the Y. If we expect searches to be concerned with regions (or radii) about R in size, and if the strip width is chosen to be within an order of magnitude of the search size, it helps the search tremendously.

Suppose we have a large number of GPS latitude and longitude information of cell phone locations, and that the data points have a resolution around 10 meters. Suppose we are interested in searching over a radius of 100 meters. If we store the latitude-longitude UPIs using a strip width of 100 meters, then to find all candidates within the bounding box [X-100,Y-100,X+100,Y+100] requires scanning short regions of two strips. Each candidate within those regions can be further filtered for exact distance. The proportion of entries that must be filtered out is 1-(pi/4).

Suppose our strips are larger, 500 meters. In that case our 100 meter radius search needs scan at most only two strips, and frequently only one, but the scan overall is somewhat less efficient because number of entries that must be filtered is greater. Even if the Y region lies completely within a single strip, the proportion of undesired entries that must be filtered is approximately 1-(pi/(4*5)). If the Y region of interest spans two strips, the proportion of entries that must be examined and rejected by the filter is 10-(pi/(4*10)).

If we err in the other direction and make the strips significantly smaller than the typical search radius, the proportion of entries that must be examined and filtered is not much affected, but the number of Y strips that must be scanned over the range [X1,X2] increases, causing more cursor creation and scans of disjoint sections of the index file. For example, if the strip width were 10 meters, the 100-meter radius query would have to examine 10 or 11 strips. This requires more cursor initialization and accessing multiple separate regions of the index.

In practice, even if the strip size for a particular geospatial subtype is as much as an order of magnitude away from the Y range used in a particular search, the strip approach is tremendously more efficient than direct X-Y search even for huge databases. If it is anticipated that data will be searched using widely different ranges at different times, then the data can be duplicated in the database using multiple subtypes with different strip sizes. For example, if we sometimes wanted to find cell phones records within a 100 meter radius of a location, and other times wanted to find calls within a 10 kilometer radius of a location, then it might make sense to record the data with multiple separate triples (or perhaps use both the object and graph UPIs of a single triple) with strip widths appropriate for each of the expected types of searches.

This points out the important limitation of the geospatial UPI strip concept. It requires that the program creating the data know in advance something about how the data will be used, i.e., the typical width of regions of interest. Fortunately, this estimate does not need to be exact. Even an order of magnitude error will not make the data unusable.

Defining and Managing Geospatial Subtypes

AllegroGraph cannot anticipate all of the geospatial subtypes that future applications might find necessary. An application must define the subtypes it wants to use, and there are some helper functions to assist for obvious popular requirements such as Earth-surface latitude/longitude with any desired strip width.

A geospatial subtype needs to define a number of functions to translate to and from the internal fields of the UPI and the external data. It is important to realize that an X-Y location most-intuitively consists of two separate numerical values. Other UPI types once way or another contain only a single value -- a string for resources and string literals, and a single number for the various numeric types, including time etc. This means that the familiar functions upi->value and value->upi may not immediately be applicable. Although the application program rarely needs be concerned with the strip component, it does need communicate back and forth the two values for X and Y.

One way to work around this would be to use Common Lisp's built in complex numbers. Another way is to use Common Lisp's multiple values. The advantage of a complex is that it is a single quantity that contains two numeric values, either integer or float. This is supported, but since manipulating complex number is a a little less efficient and computationally involved, it is more usual to employ conversion functions that deal with the two values as separate numbers.

(A geospatial subtype is associated with a unique UUID. In the future there will be a facility, not yet implemented, which will search for or otherwise reconstruct a subtype from a UUID. It will also be possible to use geospatial UPIs with federation, but currently geospatial UPIs are not supported in federated stores.)

An application can control all aspects of encoding and decoding x/y data to and from geospatial UPIs but this is rarely necessary and only so for exceptional kinds of models. Simple defining macros are provided for define geospatial subtypes for a region on a flat Cartesian plane and for the surface of a sphere (as for longitude/latitude).

register-cartesian-striping xmin  xmax  ymin  ymax  strip-width
function
Creates and returns a geospatial-subtype object implementing Cartesian XY coordinates over the X range xmin..xmax and Y range ymin..ymax and with the strip-width specified.
register-latitude-striping-in-degrees latitude-strip-width-in-degrees  &key  lat-min  lat-max  lon-min  lon-max
function
Like register-latitude-striping-in-miles but the strip width latitude-strip-width-in-km is specified in degrees.
register-latitude-striping-in-miles latitude-strip-width-in-miles  &key  lat-min  lat-max  lon-min  lon-max
function
Creates and returns a geospatial-subtype object implementing spherical longitude-latitude coordinates with the specified latitude-strip-width-in-miles. The range of latitudes and longitudes are given by these keyword arguments with these defaults: lat-min -90.0, lat-max +90.0, lon-min -180.0, lon-max +180.0.
register-latitude-striping-in-km latitude-strip-width-in-km  &key  lat-min  lat-max  lon-min  lon-max
function
Like register-latitude-striping-in-miles but the strip width latitude-strip-width-in-km is specified in kilometers.
register-latitude-striping-in-radians latitude-strip-width-in-radians  &key  lat-min  lat-max  lon-min  lon-max
function
Like register-latitude-striping-in-miles but intended to support spherical coordinates in radians. Not completely implemented in the 3.0 release. Do not use.

Execution of these macros returns an opaque geospatial subtype object that implements the subtype. It is typical to save it somewhere in order to use it later. See the defparameter in the example below.

The first defines a Cartesian subtype covering the specified region of the Cartesian plane. Units of the four arguments are at the discretion of the programmer and may be any type of real numbers.

The other four macros define a spherical subtype. A subtype can be defined to use either longitude/latitude in decimal degrees or radians. For longitude/latitude, the strip width may be given in miles (relative to the earth's average radius), kilometers, or degrees in latitude.

Before using a geospatial subtype it must be defined, usually by executing one of the above operators. Before storing a triple with a part with a geospatial subtype into a triple store, it is necessary to add that subtype to the set of subtypes known to that store. This is done with the add-geospatial-subtype-to-db function.

When opening a triple store containing one or more geospatial subtypes, geospatial subtypes it uses will be recreated in the running AllegroGraph program if they are not already registered in that image.

When working in spherical coordinates it is idiomatic to speak of "latitude and longitude" and to present the latitude before the longitude, as in "37.5N127.3W". Latitude, i.e. the North/South axis, corresponds intuitively to Cartesian Y, and longitude corresponds intuitively to Cartesian X. However, in Cartesian coordinates it is typical to present X before Y, as in "the point {3,-5}". Since Cartesian and spherical coordinates mostly share a common API, we will in AllegroGraph uniformly speak of longitude/latitude and present the longitude before the latitude. The only exception will be in parsing and printing ISO6709 latitude-longitude strings, where the standard defines latitude to appear before longitude.

Internally, the several varieties of latitude-striping subtypes all represent latitude and longitude ordinates as 32-bit unsigned integers. The ranges -90..+90 and -180..+180 are scaled so that any coordinate in ISO6709 format ±DDMMSS±DDDMMSS will round-trip between internal and external string representations without change or loss of precision. Each second in latitude corresponds to 3300 units in the integer encoding, and each second in longitude corresponds to 6600 integer units, so fractional seconds that are exact submultiples will be reproducible. These rather strange numbers were chosen as being close to the maximum resolution possible within 32 bits, and having a large number of prime divisors:

3300:   2 2 3 5 5 11  
6600: 2 2 2 3 5 5 11 

The geospatial programming API generally converts all numeric ordinate values to floats, and returns them similarly. The need to scale ordinates to and from 32-bit unsigned integer quantities, as they are represented in a geospatial UPI, imposes rounding errors. This is even more evident in Cartesian subtypes, since Cartesian ranges are arbitrary, unlike sperical ordinates, so no simple prime divisor selection can be applied. Ordinate numbers that were encoded from pleasantly round decimal fractions, and even some integers, will when externalized show annoying lengthy decimal fractions. This is unpleasant, but the intended application of Allegrograph geospatial encoding is efficient and accurate calculation, not the exact preservation of textual presentation. It is more important that encoded geospatial data can be externalized and reloaded without loss of resolution or accuracy. If you need reliable preservation of RDF lexical form, you might want to store duplicate data as string literals. AllegroGraph obviously always preserves literals accurately.

add-geospatial-subtype-to-db subtype  &optional  db
function
Once a geospatial subtype has been defined in the executing AllegroGraph image, the subtype object must be added to the set of subtypes known to a specific triple-store before triples containing that subtype can be added to the store.
report-loaded-geospatial-subtypes
function
Reports to *standard-output* the (mod 256) subtype index UUIDs of each the geospatial subtype that has been registered in the running AllegroGraph image.
drop-geospatial-subtype subtype
function
Undoes the effect previous calls to the register-geospatial-subtype functions. The subtype argument may be a geospatial-subtype object returned by one of the register functions, or the mod 256 index of the subtype in this executing AllegroGraph image. An error is signaled if the subtype is still in use (as in add-geospatial-subtype-to-db) by any open database.
register-geospatial-subtype &key  documentation  base-uuid  uuid  type  2v-encoder  2v-decoder  2v-writer  2v-reader  x->usb32  usb32->x  y->usb32  usb32->y  usb32->strip  xsd-type  x-min  x-max  y-min  y-max
function

Define a geospatial subtype. It returns two values: an opaque object representing the subtype and an mod 256 integer which is the subtype index within this running AllegroGraph image.

The arguments for this function are complicated. It is usually called from higher-level geospatial subtype definers such as register-cartesian-striping and the several spherical subtype definers, and would only be called in applications that need to define custom geospatial subtypes not covered by those existing functions. If your application requires such definitions, please contact Franz support.

AllegroGraph takes care of the mapping between the subtype index in the local AllegroGraph and the subtype index in an open triple store. In general these will not be the same.

The next three functions are not normally needed by Allegrograph applications but are included for completeness and introspection.

geospatial-subtypes-in-db db
function
Returns a list of the geospatial subtypes that have ever been added to the db. See add-geospatial-subtype-to-db. Some geospatial queries require the subtype as an argument, and since opening a database with geospatial data reconstructs any unregistered subtypes automatically, it may be more convenient when opening an existing store to obtain a subtype object this way rather than by reregistering it manually.
geospatial-subtype-index-in-db geospatial-subtype  db
function
Returns the (mod 256) index of this geospatial subtype in argument db which must be an open triple store. This information is not usually needed by application programs.
geospatial-subtype-index-in-this-image struct
function
Returns the (mod 256) index of this argument geospatial-subtype in the running AllegroGraph image. Not needed in normal programming.

Geospatial UPI accessors geospatial-x geospatial-y geospatial-y-strip geospatial-subtype
group
These functions extract the raw component of a geospatial upi. The return types of geospatial-x and geospatial-y are (unsigned-byte 32), the type of geospatial-y-strip is (unsigned-byte 16 ), and the type of geospatial-subtype is (unsigned-byte 8). Each is setfable. They are needed only in the rare case that an application programmer needs define a subtype other than the built in Cartesian and spherical.

Converting between Geospatial Data and UPIs

geo-parts->upi subtype-index  strip-usb16  x-usb32  y-usb32  &optional  upi
function
Creates a UPI using the specified subtype-index, strip and x and y coordinates.
geospatial->upi subtype  x  y  &optional  upi
function
This creates a UPI with the given geospatial subtype and x and y coordinates.
upi->geospatial upi
function
Returns the x and y coordinates of a geospatial UPI as multiple values.
longitude-latitude->upi subtype  longitude  latitude  &optional  upi
function
Analogous to value->upi, takes a geospatial subtype and a longitude/latitude coordinate and encodes them as a UPI.
upi->longitude-latitude upi
function
Decodes a longitude-latitude geospatial UPI and returns two values, the longitude and latitude, as floating point numbers.

Scaling functions scale-latitude unscale-latitude scale-longitude unscale-longitude
group
These functions convert back and forth between floating-point latitude and longitude values and the internal unsigned-byte 32 representation in a UPI. They are not normally needed by application code.

Using Automatic Predicate and Type Mapping with Geospatial Data

These functions create type or predicate mappings for geospatial subtypes. subtype os a geospatial-subtype object. The predicate-part or datatype arguments are the same as could be passed to setf of datatype-mapping or predicate-mapping.

set-spherical-predicate-mapping predicate-part  subtype  &key  db
function
Similar to (setf predicate-mapping) but is used for spherical geospatial subtypes. Automatic mapping needs to know the intended subtype for geospatial data predicates.
set-cartesian-predicate-mapping predicate-part  subtype  &key  db
function
Similar to (setf predicate-mapping) but is used for Cartesian geospatial subtypes. Automatic mapping needs to know the intended subtype for geospatial data predicates.
set-spherical-datatype-mapping datatype  subtype  &key  db
function
Similar to (setf datatype-mapping) but is used for spherical geospatial subtypes. Automatic mapping needs to know the intended subtype for geospatial datatypes.
set-cartesian-datatype-mapping datatype  subtype  &key  db
function
Similar to (setf datatype-mapping) but is used for Cartesian geospatial subtypes. Automatic mapping needs to know the intended subtype for geospatial datatypes.

Retrieving Geospatial Triples

get-triples-geospatial-bounding-box subtype  p  x-min  x-max  y-min  y-max  &key  db  use-g
function
Returns a cursor that iterates through all the triples with predicate pred and an object UPI that is a geospatial UPI with the given subtype and within the specified Cartesian bounding box. If :use-g is true, the triple's graph slot is searched instead of the object. The units are implicit in the data.
get-triples-geospatial-radius subtype  pred  x  y  radius  &key  db  use-g
function

Returns a cursor that will iterate over the specified triples.

subtype is a geospatial-subtype object such as is returned by register-cartesian-striping.

pred is a UPI or future part specifying the predicate of triples to be matched. It must be supplied.

x and y are the coordinates around which to search, and radius is the search radius. The units of radius are implicit in the subtype. All matching triples with an object of the specified geospatial subtype and within the search radius will be returned by the cursor.

If the use-g argument is specified and is true, the graph part of triples is searched instead of the object. The other arguments are as for get-triples.

The Haversine formula is used to compute the "Great Circle" distance on a sphere. See for background. Each of these functions returns a cursor that will return all the triples with predicate pred and an object UPI that is a geospatial UPI with the given subtype and within the specified spherical radius. If :use-g is true, the graph is searched instead of the object. This is the spherical analog to the bounding box functions above.

get-triples-haversine-miles subtype  pred  lon  lat  radius-in-miles  &key  db  use-g
function

Returns a cursor that will return the specified triples.

subtype is a geospatial-subtype designator. It can be an object such as is returned by register-latitude-striping-in-miles or a string, future-part or UPI describing such a subtype

pred is a UPI or future part specifying the predicate of triples to be matched. It must be supplied.

lon and lat are the coordinates around which to search, and radius-in-miles is the search radius. All matching triples with an object of the specified geospatial subtype and within the search radius will be returned by the cursor.

If the use-g argument is specified and is true, the graph part of triples is searched instead of the object. The other arguments are as for get-triples.

get-triples-haversine-km subtype  pred  lon  lat  radius-in-km  &key  db  use-g
function
Like get-triples-haversine-miles but the radius-in-km is given in kilometers.
get-triples-haversine-radians subtype  pred  lon  lat  radius-in-radians  &key  db  use-g
function

Like get-triples-haversine-miles but the radius-in-radians is given in radians.

This is incompletely implemented and nonfunctional in release 3.0.

Geospatial and Prolog

There are Prolog analogs to the functions above. The triple argument is unified to successive triples that have predicate pred and an object within the bounding box or radius.

triple-inside-bounding-box
(triple-inside-bounding-box -triple +subtype +pred +x-min +x-max +y-min +y-max)
functor
Unify -triple against each triple with predicate +pred with an object of geospatial subtype +subtype within the bounding box x-min x-max y-min y-max. The +subtype is the object returned by register-cartesian-striping and similar functions.
triple-inside-radius
(triple-inside-radius -triple +subtype +pred ±x ±y ±radius)
functor
Unify -triple against each triple with predicate +pred with an object of geospatial subtype +subtype within radius distance of coordinate x y. The +subtype is the object returned by register-cartesian-striping and similar functions.
triple-inside-bounding-box-g
(triple-inside-bounding-box-g -triple +subtype +pred +x-min +x-max +y-min +y-max)
functor
Like triple-inside-bounding-box except it searches against the graph part or each triple instead of the object part.
triple-inside-radius-g
(triple-inside-radius-g -triple +subtype +pred ±x ±y ±radius)
functor
Like triple-inside-radius except it searches against the graph part or each triple instead of the object part.
triple-inside-haversine-miles
(triple-inside-haversine-miles -triple +subtype +pred +lon +lat +radius)
functor
Unify -triple against each triple with predicate +pred with an object of geospatial subtype +subtype within +radius miles of +longitude and +latitude. These last three arguments are typically floating-point numbers. The +subtype is the object returned by register-latitude-striping-in-miles and similar functions.
triple-inside-haversine-km
(triple-inside-haversine-km -triple +subtype +pred +lon +lat +radius)
functor
Like triple-inside-haversine-miles except +radius is in kilometers.
triple-inside-haversine-radians
(triple-inside-haversine-radians -triple +subtype +pred +lon +lat +radius)
functor
Like triple-inside-haversine-miles except it searches against the graph part or each triple instead of the object part.
triple-inside-haversine-miles-g
(triple-inside-haversine-miles-g -triple +subtype +pred +lon +lat +radius)
functor
Like triple-inside-haversine-miles except it searches against the graph part or each triple instead of the object part.
triple-inside-haversine-km-g
(triple-inside-haversine-km-g -triple +subtype +pred +lon +lat +radius)
functor
Like triple-inside-haversine-km except it searches against the graph part or each triple instead of the object part.
triple-inside-haversine-radians-g
(triple-inside-haversine-radians-g -triple +subtype +pred +lon +lat +radius)
functor
Like triple-inside-haversine-radians except it searches against the graph part or each triple instead of the object part.

Haversine Conversions

These compute the haversine distance between two spherical coordinates. The first returns a value in radians while the other two return an actual distance on the earth's surface in kilometers or miles.

haversine-r lon1r  lat1r  lon2r  lat2r
function
Compute the great circle distance between the points on a unit radius sphere specified by lon1r, lat1r, lon2r, and lat2r, all in radians.
haversine-miles lon1  lat1  lon2  lat2
function
Compute the great circle distance in miles between the points on a the Earth's surface specified by lon1, lat1, lon2, and lat2, all in degrees.
haversine-km lon1  lat1  lon2  lat2
function
Compute the great circle distance in kilometers between the points on a the Earth's surface specified by lon1, lat1, lon2, and lat2, all in degrees.

Polygon Support

There is a facility for dealing with polygon regions which can be used to determine if a pair of geospatial ordinates lies within a region defined by a polygon. It is available in Lisp, Java, and Python.

All computations on a polygon are performed on a Lisp data representation, a proper list of conses of X and Y ordinates which are the vertexes of the polygon. Successive vertexes define the edges with an implied edge between the last and first element. There is no requirement that a polygon be convex, or even non-self-intersecting. Polygons work with any geospatial subtype, including both Cartesian and spherical ordinates, although the predicates for the interior of a polygon compute in the Cartesian plane and do not account for spherical ordinates. This is a reasonable optimization for small regions of the sphere, except near the poles.

When a polygon must be stored in a triple store (e.g. the boundaries of a city) by convention it can be done with an ordered set of triples, one for each vertex. The Subject is any convenient identifier for the polygon, the Object is a geospatial UPI for the vertex, and the Graph is available for any other purpose. The Predicate is a nonnegative integer UPI of type +rdf-subscript+ which orders the vertexes. These can be extracted from the triple store by the polygon-vertexes function.

polygon-vertexes polygon-upi
function
Returns an ordered list of conses that are the (longitude . latitude) ordinates for each vertex of the polygon identified by polygon-upi.

The polygon-subject-upi is the subject identifying the desired polygon. This function returns a fresh list of the vertexes of the polygon. A polygon could have been stored in a triple store with code like the following:

(defparameter *cartesian-10*  
 (register-cartesian-striping -1000 1000 -1000 1000 10))  
 
(add-geospatial-subtype-to-db *cartesian-10*)  
 
(defparameter *my-pentagon*  
'((234.56 . 347.31)  
  (243.07 . 348.40)  
  (247.85 . 357.02)  
  (230.68 . 349.52)  
  (231.37 . 348.62)))  
 
(defparameter *penta*  
(with-temp-upi (u)  
  (with-temp-upi (s)  
    (loop with id = (new-blank-node)  
	for vertex-number from 1  
	for (x . y) in *my-pentagon*  
	do (add-triple id  
		       (value->upi vertex-number :subscript)  
		       (geospatial->upi *cartesian-10* x y u))  
	finally (return id))))) 

penta will be set to the new blank node that names this polygon. Its vertexes can now be retrieved given it's UPI as subject.

(polygon-vertexes *penta*) =>  
((234.56006 . 347.31006) (243.07007 . 348.40002)  
 (247.84998 . 357.02002) (230.67993 . 349.52002)  
 (231.37 . 348.62)) 
point-inside-polygon-p polygon  x  y
function
polygon is a list of vertexes such as is returned by polygon-vertexes. Returns true if the point [x,y] is inside the polygon. Boundary cases for vertexes of points on the edges are indeterminate.

This predicate returns true if the given point is inside the polygon. The polygon is a list of coordinates, as above. Points on an edge or vertex will return an indeterminate result since they are subject to float rounding.

(point-inside-polygon-p (polygon-vertexes *penta*) 235.5 350.0) => t  
(point-inside-polygon-p *my-pentagon* 235.5 350.0)              => t  
(point-inside-polygon-p *my-pentagon* 235.5 356.0)              => nil 

See the polygon section in the Geospatial tutorial for more examples how polygons can be used.

polygon-inside-polygon-p inner  outer  &key  outer-convex-p
function
For two lists representing polygons such as is returned by polygon-vertexes, returns true if and only if the inner polygon is contained entirely within the outer polygon. Boundary cases for points on vertexes or edges are indeterminate. The outer-convex-p boolean may be passed, and if true allows a more-efficient algorithm to be used.

This function tests whether the inner polygon is entirely contained by the outer polygon. Again, results are indeterminate if any vertexes of the inner polygon fall on edges or vertexes of the outer polygon, or if any edges of the two polygons are collinear. The outer-convex-p boolean may be passed, and if true allows a more-efficient algorithm to be used.

get-triples-inside-polygon subtype  predicate  polygon  &key  db  use-g
function
polygon must be a list representing a polygon such as is returned by polygon-vertexes. Returns a cursor that will return all triples with predicate predicate and an object that is a geospatial UPI of subtype subtype within the polygon. If use-g is supplied and true, the graph is used instead. Boundary cases for vertexes of points on the edges are indeterminate. The other arguments are as for get-triples-geospatial-bounding-box

This function returns a cursor that return all triples with the specified predicate that have an Object that is a geospatial UPI of the specified subtype and which is within the given polygon. If :use-g is true, the graph is searched instead of the object.

Polygon Support and Prolog

These are the Prolog analogs to get-triples-inside-polygon. They successively unify the -triple argument to each triple inside the given polygon.

triple-inside-polygon
(triple-inside-polygon -triple +subtype +predicate +polygon)
functor
This is the Prolog analog of get-triples-inside-polygon Unify -triple to each triple with predicate +predicate and an object part of geospatial subtype +subtype within the polygon denoted by the list polygon. Boundary cases on the vertexes and edges are not specified.
triple-inside-polygon-g
(triple-inside-polygon-g -triple +subtype +predicate +polygon)
functor
Like triple-inside-polygon except it searches against the graph part or each triple instead of the object part.

Temporal Primitives

AllegroGraph's encoded UPIs support efficient storage and retrieval of temporal data including datetimes, time points, and time intervals:

Once data has been encoded, applications can perform queries involving a broad range of temporal constraints on data in the store, including:

The temporal reasoning tutorial describes all of these capabilities in detail and also functions as a reference guide.

REPL Interface

In addition to the !-reader macro, AllegroGraph provides a handful of functions to make it easier to interact with the triple-store from the REPL.

enable-print-decoded boolean  &key  pprint-dispatch
function

Triples and parts are represented by (simple-array (unsigned-byte 8) 56) and (simple-array (unsigned-byte 8) 12) respectively. By default the Lisp printer prints these as vectors of bytes which is not very informative when using AllegroGraph interactively. This function modifies a pprint-dispatch table to print triples and parts interpretively if this can be done in the current environment. Specifically, the value of *db* must be an open triple store, and the vector being printed must print without error in the manner of print-triple or upi->value. If any of these conditions don't hold the vector is printed normally.

If the boolean argument is true, the informative printing is enabled, otherwise disabled. For example:

> (enable-print-decoded t)  
t  
> (upi !"Dürst"^^xsd:string)  
{Dürst}  
> (enable-print-decoded nil)  
nil  
> (upi !"Dürst"^^xsd:string)  
#(48 61 125 164 117 231 137 163 64 100 17 2) 

The :pprint-dispatch argument specifies the dispatch table to modify, by default the value of print-pprint-dispatch. Once enable-print-decoded has been turned on, you can also use the special-variable *print-decoded* to enable fine-grained control of triple and UPI printing. See the AllegroGraph tutorial for more information on the various notations AllegoGraph uses.

*print-decoded*
variable
When true and enable-print-decoded has been called with true, then triples and UPIs are printed with interpretation.
print-triple triple  &key  format  stream  db
function

Print a triple returned by cursor-next-row or get-triples-list.

Returns the triple printed.

The keyword argument :format, which defaults to :ntriples, specifies how the triple should be printed. The value :ntriples specifies that it should be printed in N-Triples syntax. The value :long indicates that the string value of the part should be used. And the value :concise causes it to use a more concise, but possibly ambiguous, human-readable format.

print-triples triple-container  &key  limit  format  stream  db
function
Display the triples in triple-container which can be either a triple store object, a list of triples such as is returned by get-triples-list, or a cursor such as is returned by get-triples. If the keyword argument :limit is supplied, then at most that many triples will be displayed. The :format keyword argument controls how the triples will be displayed, in either :ntriples, :long, or :concise format. It defaults to :ntriples. The stream` argument can be used to send output to the stream of your choice. If left unspecified, then output will go to standard-output.
pprint-subject part  &key  maximum-depth  format  db
function

Print information about part down to a maximum depth of maximum-depth using the format format. Triples for which part is a subject and their children will be printed. See part->string for information about part printing. See pprint-object to display information based on subjects rather than objects.

Note that pprint-subject is designed to show the actual triples in a store. It uses ground-triple-store to remove one layer of reasoning from db. db defaults to *db*.

pprint-object part  &key  maximum-depth  format  db
function

Print information about part down to a maximum depth of maximum-depth using the format format. Triples for which part is an object and their children will be printed. See part->string for information about part printing. See pprint-subject to display information based on objects rather than subjects.

Note that pprint-object is designed to show the actual triples in a store. It uses ground-triple-store to remove one layer of reasoning from db. db defaults to *db*.

From Parts and UPIs to Values

See upi->value and upi->values for additional information.

part->concise part  &optional  stream  db
function

Return a string representation of part in a short human-readable form

syntax. Part can be either a UPI or a future-part.

part->long part  &optional  stream  db
function

Return a string representation of part in long

syntax. Part can be either a UPI or a future-part.

part->ntriples part  &optional  stream  db
function

Return a string representation of part in N-Triples

syntax. Part can be either a UPI or a future-part.

part->string part  &key  format  db
function
Return a string representation of part (which can be a UPI or a future-part). The :format keyword argument controls the format and can be one of :ntriples, :long, :concise, or :terse, as with the :format argument to print-triple. The default is :ntriples.
part->terse part  &optional  stream  db
function

Return a string representation of part in a very-short human-readable form

syntax. Part can be either a UPI or a future-part.

part->value part  &key  db
function
Return the value of the part. Part can be UPI or a future-part.

Miscellaneous Utilities

UPI hash-tables

gethash/upi upi  table  &optional  default
function
Returns (as multiple values) the value associated with upi in the upi-hash-table table and whether or not the value was found. setf may be used with gethash/upi to modify the value associated with a upi or to add a new association to the hash-table.
hash-table-count/upi table
function
Returns the number of UPI/value associations in the upi-hash-table table.
make-upi-hash-table &rest  rest  &key  size
function
Returns a hash-table optimized for storing UPIs.
maphash/upi fn  table
function
Apple fn to each UPI/value association in the upi-hash-table table. fn should be a function of two arguments. It will be called with a UPI as the first argument and the value associated with the UPI as the second.
remhash/upi upi  table
function
Remove the association (if any) between upi and its value in the upi-hash-table table.

Other

with-temp-triple (var)  &body  body
macro
Binds var to a fresh stack-allocated triple which can be used within the dynamic extent of the code body. Corruption may result if the triple is referenced after exiting the body.
with-temp-upi (var)  &body  body
macro
Binds var to a fresh stack-allocated upi which can be used within the dynamic extent of the code body. Corruption may result if the upi is referenced after exiting the body.
with-temp-upis vars  &body  body
macro
Binds vars to fresh stack-allocated upis which can be used within the dynamic extent of the code body. Corruption may result if the upis are referenced after exiting the body.

string+ &rest  args
function
string+ returns a fresh simple string that contains the printed representation of all the individual objects passed as arguments in the order that they are supplied. Actual strings among objects are copied into the result string. Other objects are made into strings as if they were passed to princ-to-string (although the actual mechanism may be different for some types of objects), with print-pretty bound to nil. This function is designed to allow string to be used as identifiers or labels to be created quickly and efficiently. See the Franz documentation on string+ for more details.
triple-part
type
A triple-part is either a UPI or a future-part.

Indices

All of the functions and variables in AllegroGraph's public interface are indexed below.

Function and Macro index

Functor index

Variable index


  1. While completely unrelated to AllegroGraph, there is another symbol with the same name exported from the dbi.oracle package. If you try to have a package use both the dbi.oracle and the db.agraph packages, these two symbols will signal a package conflict error. Assuming you want to have the db.agraph symbol to be available without a package qualifier, you should define your package something like this

        (defpackage :my-package  
          (:use :cl :dbi.oracle :triple-store ...)  
          (:shadowing-import-from :triple-store :*db*)  
          ...) 

    or else execute this form

        (shadowing-import 'db.agraph:*db*) 

    before using either package.

  2. For information on the N-Triples format, see both the W3C description and the information in the RDF test cases. Additionally the file w3c-ntriples-tests.nt in the sample-inputs directory of the AllegroGraph distribution contains a set of test cases and provides a good overview of what can be represented.
  3. Details of the RDF/XML syntax can be found in the RDF/XML-primer and related documentation.
  4. Language tags are defined in RFC-3066
  5. The function do-cursor has been removed from AllegroGraph as of version 3.0. Please use the function map-cursor and the macro iterate-cursor instead.
  6. The Prolog q- functor provides an interface to range queries for only the object and graph slots of a triple.
  7. It is possible to use range queries with UPIs of type :short-literal since these are stored directly in the UPI and not in the string dictionary. The only caveat is that you must make sure that all of your data really does fit into a short literal UPI. This is equivalent to saying that all of your strings must be less than or equal to 10 bytes in length after they have been UTF-8 encoded.
  8. The Oracle and Sesame connections are in progress. Please contact Franz for more information.
  9. Remember that select is a macro and that using it in the REPL will produce interpreted code. This means that selects run in the REPL will be significantly slower than those that you write inside of compiled functions. (Note also that both the HTTP and the Java interfaces to AllegroGraph ensure that any select calls get compiled before they run). Whether compiled queries execute significantly faster depends on whether the generated code performs a lot of backtracking or other repeated computation.
  10. In this case, there are fewer ub:publicationAuthor triples than rdf:type ones so the type clause provides no help at all. If the type clause was more restrictive, then the planner will keep the type clause and move it first.
  11. Previously, this was called has-value-reasoning-enabled because AllegroGraph only supported owl:hasValue. Now that AllegroGraph supports other restrictions, the name has been changed.