| DBZ(3) | Library Functions Manual | DBZ(3) |
dbminit, fetch, store, dbmclose - somewhat dbm-compatible database
routines
dbzfresh, dbzagain, dbzfetch, dbzstore - database routines
dbzsync, dbzsize, dbzincore, dbzcancel, dbzdebug - database routines
#include <dbz.h>
dbminit(base) char *base;
datum fetch(key) datum key;
store(key, value) datum key; datum value;
dbmclose()
dbzfresh(base, size, fieldsep, cmap, tagmask) char *base; long size; int fieldsep; int cmap; long tagmask;
dbzagain(base, oldbase) char *base; char *oldbase;
datum dbzfetch(key) datum key;
dbzstore(key, value) datum key; datum value;
dbzsync()
long dbzsize(nentries) long nentries;
dbzincore(newvalue)
dbzcancel()
dbzdebug(newvalue)
These functions provide an indexing system for rapid random access to a text file (the base file). Subject to certain constraints, they are call-compatible with dbm(3), although they also provide some extensions. (Note that they are not file-compatible with dbm or any variant thereof.)
In principle, dbz stores key-value pairs, where both key and value are arbitrary sequences of bytes, specified to the functions by values of type datum, typedefed in the header file to be a structure with members dptr (a value of type char * pointing to the bytes) and dsize (a value of type int indicating how long the byte sequence is).
In practice, dbz is more restricted than dbm. A dbz database must be an index into a base file, with the database values being fseek(3) offsets into the base file. Each such value must ``point to'' a place in the base file where the corresponding key sequence is found. A key can be no longer than DBZMAXKEY (a constant defined in the header file) bytes. No key can be an initial subsequence of another, which in most applications requires that keys be either bracketed or terminated in some way (see the discussion of the fieldsep parameter of dbzfresh, below, for a fine point on terminators).
Dbminit opens a database, an index into the base file base, consisting of files base.dir and base.pag which must already exist. (If the database is new, they should be zero-length files.) Subsequent accesses go to that database until dbmclose is called to close the database. The base file need not exist at the time of the dbminit, but it must exist before accesses are attempted.
Fetch searches the database for the specified key, returning the corresponding value if any. Store stores the key-value pair in the database. Store will fail unless the database files are writeable. See below for a complication arising from case mapping.
Dbzfresh is a variant of dbminit for creating a new database with more control over details. Unlike for dbminit, the database files need not exist: they will be created if necessary, and truncated in any case.
Dbzfresh's size parameter specifies the size of the first hash table within the database, in key-value pairs. Performance will be best if size is a prime number and the number of key-value pairs stored in the database does not exceed about 2/3 of size. (The dbzsize function, given the expected number of key-value pairs, will suggest a database size that meets these criteria.) Assuming that an fseek offset is 4 bytes, the .pag file will be 4*size bytes (the .dir file is tiny and roughly constant in size) until the number of key-value pairs exceeds about 80% of size. (Nothing awful will happen if the database grows beyond 100% of size, but accesses will slow down somewhat and the .pag file will grow somewhat.)
Dbzfresh's fieldsep parameter specifies the field separator in the base file. If this is not NUL (0), and the last character of a key argument is NUL, that NUL compares equal to either a NUL or a fieldsep in the base file. This permits use of NUL to terminate key strings without requiring that NULs appear in the base file. The fieldsep of a database created with dbminit is the horizontal-tab character.
For use in news systems, various forms of case mapping (e.g. uppercase to lowercase) in keys are available. The cmap parameter to dbzfresh is a single character specifying which of several mapping algorithms to use. Available algorithms are:
Mapping algorithm 0 (no mapping) is faster than the others and is overwhelmingly the correct choice for most applications. Unless compatibility constraints interfere, it is more efficient to pre-map the keys, storing mapped keys in the base file, than to have dbz do the mapping on every search.
For historical reasons, fetch and store expect their key arguments to be pre-mapped, but expect unmapped keys in the base file. Dbzfetch and dbzstore do the same jobs but handle all case mapping internally, so the customer need not worry about it.
Dbz stores only the database values in its files, relying on reference to the base file to confirm a hit on a key. References to the base file can be minimized, greatly speeding up searches, if a little bit of information about the keys can be stored in the dbz files. This is ``free'' if there are some unused bits in an fseek offset, so that the offset can be tagged with some information about the key. The tagmask parameter of dbzfresh allows specifying the location of unused bits. Tagmask should be a mask with one group of contiguous 1 bits. The bits in the mask should be unused (0) in most offsets. The bit immediately above the mask (the flag bit) should be unused (0) in all offsets; (dbz)store will reject attempts to store a key-value pair in which the value has the flag bit on. Apart from this restriction, tagging is invisible to the user. As a special case, a tagmask of 1 means ``no tagging'', for use with enormous base files or on systems with unusual offset representations.
A size of 0 given to dbzfresh is synonymous with the local default; the normal default is suitable for tables of 90-100,000 key-value pairs. A cmap of 0 (NUL) is synonymous with the character 0, signifying no case mapping (note that the character ? specifies the local default mapping, normally C). A tagmask of 0 is synonymous with the local default tag mask, normally 0x7f000000 (specifying the top bit in a 32-bit offset as the flag bit, and the next 7 bits as the mask, which is suitable for base files up to circa 24MB). Calling dbminit(name) with the database files empty is equivalent to calling dbzfresh(name,0,'\t','?',0).
When databases are regenerated periodically, as in news, it is simplest to pick the parameters for a new database based on the old one. This also permits some memory of past sizes of the old database, so that a new database size can be chosen to cover expected fluctuations. Dbzagain is a variant of dbminit for creating a new database as a new generation of an old database. The database files for oldbase must exist. Dbzagain is equivalent to calling dbzfresh with the same field separator, case mapping, and tag mask as the old database, and a size equal to the result of applying dbzsize to the largest number of entries in the oldbase database and its previous 10 generations.
When many accesses are being done by the same program, dbz is massively faster if its first hash table is in memory. If an internal flag is 1, an attempt is made to read the table in when the database is opened, and dbmclose writes it out to disk again (if it was read successfully and has been modified). Dbzincore sets the flag to newvalue (which should be 0 or 1) and returns the previous value; this does not affect the status of a database that has already been opened. The default is 0. The attempt to read the table in may fail due to memory shortage; in this case dbz quietly falls back on its default behavior. Stores to an in-memory database are not (in general) written out to the file until dbmclose or dbzsync, so if robustness in the presence of crashes or concurrent accesses is crucial, in-memory databases should probably be avoided.
Dbzsync causes all buffers etc. to be flushed out to the files. It is typically used as a precaution against crashes or concurrent accesses when a dbz-using process will be running for a long time. It is a somewhat expensive operation, especially for an in-memory database.
Dbzcancel cancels any pending writes from buffers. This is typically useful only for in-core databases, since writes are otherwise done immediately. Its main purpose is to let a child process, in the wake of a fork, do a dbmclose without writing its parent's data to disk.
If dbz has been compiled with debugging facilities available (which makes it bigger and a bit slower), dbzdebug alters the value (and returns the previous value) of an internal flag which (when 1; default is 0) causes verbose and cryptic debugging output on standard output.
Concurrent reading of databases is fairly safe, but there is no (inter)locking, so concurrent updating is not.
The database files include a record of the byte order of the processor creating the database, and accesses by processors with different byte order will work, although they will be slightly slower. Byte order is preserved by dbzagain. However, agreement on the size and internal structure of an fseek offset is necessary, as is consensus on the character set.
An open database occupies three stdio streams and their corresponding file descriptors; a fourth is needed for an in-memory database. Memory consumption is negligible (except for stdio buffers) except for in-memory databases.
Functions returning int values return 0 for success, -1 for failure. Functions returning datum values return a value with dptr set to NULL for failure. Dbminit attempts to have errno set plausibly on return, but otherwise this is not guaranteed. An errno of EDOM from dbminit indicates that the database did not appear to be in dbz format.
The original dbz was written by Jon Zeeff (zeeff@b-tech.ann-arbor.mi.us). Later contributions by David Butler and Mark Moraes. Extensive reworking, including this documentation, by Henry Spencer (henry@zoo.toronto.edu) as part of the C News project. Hashing function by Peter Honeyman.
The dptr members of returned datum values point to static storage which is overwritten by later calls.
Unlike dbm, dbz will misbehave if an existing key-value pair is `overwritten' by a new (dbz)store with the same key. The user is responsible for avoiding this by using (dbz)fetch first to check for duplicates; an internal optimization remembers the result of the first search so there is minimal overhead in this.
Waiting until after dbminit to bring the base file into existence will fail if chdir(2) has been used meanwhile.
The RFC822 case mapper implements only a first approximation to the hideously-complex RFC822 case rules.
The prime finder in dbzsize is not particularly quick.
Should implement the dbm functions delete, firstkey, and nextkey.
On C implementations which trap integer overflow, dbz will refuse to (dbz)store an fseek offset equal to the greatest representable positive number, as this would cause overflow in the biased representation used.
Dbzagain perhaps ought to notice when many offsets in the old database were too big for tagging, and shrink the tag mask to match.
Marking dbz's file descriptors close-on-exec would be a better approach to the problem dbzcancel tries to address, but that's harder to do portably.
| 3 Feb 1991 | C News |