persistency.pl -- Provide persistent dynamic predicates

This module provides simple persistent storage for one or more dynamic predicates. A database is always associated with a module. A module that wishes to maintain a database must declare the terms that can be placed in the database using the directive persistent/1.

The persistent/1 expands each declaration into four predicates:

• name(Arg, ...)
• assert_name(Arg, ...)
• retract_name(Arg, ...)
• retractall_name(Arg, ...)

As mentioned, a database can only be accessed from within a single module. This limitation is on purpose, forcing the user to provide a proper API for accessing the shared persistent data.

Below is a simple example:

:- module(user_db,
          [ attach_user_db/1,           % +File
            current_user_role/2,        % ?User, ?Role
            add_user/2,                 % +User, +Role
            set_user_role/2             % +User, +Role
          ]).
:- use_module(library(persistency)).

:- persistent
        user_role(name:atom, role:oneof([user,administrator])).

attach_user_db(File) :-
        db_attach(File, []).

%%      current_user_role(+Name, -Role) is semidet.

current_user_role(Name, Role) :-
        with_mutex(user_db, user_role(Name, Role)).

add_user(Name, Role) :-
        assert_user_role(Name, Role).

set_user_role(Name, Role) :-
        user_role(Name, Role), !.
set_user_role(Name, Role) :-
        with_mutex(user_db,
                   (  retractall_user_role(Name, _),
                      assert_user_role(Name, Role))).

To be done

- Provide type safety while loading

- Thread safety must now be provided at the user-level. Can we provide generic thread safety? Basically, this means that we must wrap all exported predicates. That might better be done outside this library.

- Transaction management?

- Should assert_<name> only assert if the database does not contain a variant?

 persistent(+Spec)

Declare dynamic database terms. Declarations appear in a directive and have the following format:

:- persistent
        <callable>,
        <callable>,
        ...

Each specification is a callable term, following the conventions of library(record), where each argument is of the form

name:type

Types are defined by library(error).

 current_persistent_predicate(:PI) is nondet

True if PI is a predicate that provides access to the persistent database DB.

 db_attach(:File, +Options)

Use File as persistent database for the calling module. The calling module must defined persistent/1 to declare the database terms. Defined options:

sync(+Sync)

One of close (close journal after write), flush (default, flush journal after write) or none (handle as fully buffered stream).

If File is already attached this operation may change the sync behaviour.

 db_size(+Module, -Terms) is det[private]

Terms is the total number of terms in the DB for Module.

 db_attached(:File) is semidet

True if the context module attached to the persistent database File.

 db_assert(:Term) is det[private]

Assert Term into the database and record it for persistency. Note that if the on-disk file has been modified it is first reloaded.

 db_detach is det

Detach persistency from the calling module and delete all persistent clauses from the Prolog database. Note that the file is not affected. After this operation another file may be attached, providing it satisfies the same persistency declaration.

 sync(+Module, +Stream) is det[private]

Synchronise journal after a write. Using close, the journal file is closed, making it easier to edit the file externally. Using flush flushes the stream but does not close it. This provides better performance. Using none, the stream is not even flushed. This makes the journal sensitive to crashes, but much faster.

 db_retractall(:Term) is det[private]

Retract all matching facts and do the same in the database. If Term is unbound, persistent/1 from the calling module is used as generator.

 db_retract(:Term) is nondet[private]

Retract terms from the database one-by-one.

 db_sync(:What)

Synchronise database with the associated file. What is one of:

reload

Database is reloaded from file if the file was modified since loaded.

update

As reload, but use incremental loading if possible. This allows for two processes to examine the same database file, where one writes the database and the other periodycally calls db_sync(update) to follow the modified data.

gc

Database was re-written, deleting all retractall statements. This is the same as gc(50).

gc(Percentage)

GC DB if the number of deleted terms is greater than the given percentage of the total number of terms.

gc(always)

GC DB without checking the percentage.

close

Database stream was closed

detach

Remove all registered persistency for the calling module

nop

No-operation performed

With unbound What, db_sync/1 reloads the database if it was modified on disk, gc it if it is dirty and close it if it is opened.

 db_sync_all(+What)

Sync all registered databases.