[manual index][section index]


Dbm: Dbf, init - data base with hashed indexing


include "dbm.m";

dbm := load Dbm Dbm->PATH;
Datum, Dbf: import dbm;

Datum:  type array of byte;

Dbf: adt {
    create: fn(file: string, perm: int): ref Dbf;
    open:   fn(file: string, flags: int): ref Dbf;

    fetch:  fn(db: self ref Dbf, key: Datum): Datum;
    delete: fn(db: self ref Dbf, key: Datum): int;
    store:  fn(db: self ref Dbf, key: Datum, data: Datum,
                  replace: int): int;

    firstkey: fn(db: self ref Dbf): Datum;
    nextkey:  fn(db: self ref Dbf, key: Datum): Datum;

    flush:    fn(db: self ref Dbf);
    isrdonly: fn(db: self ref Dbf): int;

init:   fn();


Dbm maintains key/content pairs in a data base. The functions will handle very large (a billion blocks) databases and will access a keyed item in one or two filesystem accesses.

Keys and contents are both represented by arrays of bytes (with the synonym Datum), allowing arbitrary binary values.

The data base is stored in two files. One file is a directory containing a bit map and has .dir as its suffix. The second file contains all data and has .pag as its suffix. An application can access several databases at once, but must avoid concurrent operations on any one database (eg, by using a monitor process to control access).

Init must be called before any other operation of the module.

A database is created by Dbf.create, which accepts a file permission parameter perm, as described for Sys->create (see sys-open(2)); it creates the two files file.dir and file.pag. If successful, it returns a Dbf reference describing the database, which is open for reading and writing. (It will truncate an existing database.) It returns nil if it cannot create the database for some reason, and sets the error string.

Dbf.open accepts a mode parameter as described in sys-open(2), and opens the existing database in file.dir and file.pag. If successful, it returns a Dbf reference describing the database, which is open either for reading and writing (ie, Sys->ORDWR), or only for reading (Sys->OREAD) as determined by mode. It returns nil if the database cannot be opened successfully, and sets the error string.

The remaining operations apply to an existing Dbf reference db:

Return the data stored under a key; nil is returned if the key is not in the database.
db.store(key, data, replace)
Store data under the given key. If replace is non-zero, store will simply replace the existing value by the new one if the key is already in the database; if replace is zero store will return 0 if the new item was inserted, but 1 if the key already appears in the database, and the new value will not be stored.
Key and its associated value is removed from the database.
Return the first key in the database; return nil if the database is empty.
Return the key following the given key, or nil if there is none.
Discard any data cached from the file. The cache is write-through, so it is not necessary to flush the file before the application exits.
Return true if db was opened only for reading and writes are not allowed.


A linear pass through all keys in a database may be made, in an (apparently) random order, by use of Dbf.firstkey and Dbf.nextkey. This code will traverse the data base:

for(key := db.firstkey(); key != nil; key = db.nextkey(key)){
	d := db.fetch(key);

The order of keys presented by Dbf.firstkey and Dbf.nextkey depends on a hashing function, not on anything interesting.




All functions that return an int indicate errors with negative values. A zero return indicates success. Routines that return pointers, including values of Datum, return nil values on error. Dbf.create and Dbf.open return nil on failure to access the database, setting the error string to a more detailed diagnostic.


On some systems (notably Plan 9 but also some Unix systems), the .pag file might contain holes where no data block has ever been written so that its apparent size is about four times its actual content. These files cannot be copied by normal means (cp, cat) without filling in the holes.

Except for firstkey and nextkey, Datum values returned by these functions point to storage that is changed by subsequent calls.

The sum of the sizes of a key/content pair must not exceed the internal block size (currently 512 bytes). Moreover all key/content pairs that hash together must fit on a single block. Dbf.store will return an error in the event that a block fills with inseparable data.

DBM(2 ) Rev:  Tue Mar 31 02:42:39 GMT 2015