This manual page documents the format of file name databases for the
GNU version of
locate.
The file name databases contain lists of files that were in
particular directory trees when the databases were last updated.
There can be multiple databases. Users can select which databases
locate searches using an environment variable or command line
option; see locate(1). The system administrator can choose the
file name of the default database, the frequency with which the
databases are updated, and the directories for which they contain
entries. Normally, file name databases are updated by running the
updatedb program periodically, typically nightly; see
updatedb(1).
updatedb runs a program called frcode to compress the list
of file names using front-compression, which reduces
the database size by a factor of 4 to 5. Front-compression (also
known as incremental encoding) works as follows.
The database entries are a sorted list (case-insensitively, for users'
convenience). Since the list is sorted, each entry is likely to share
a prefix (initial string) with the previous entry. Each database
entry begins with an offset-differential count byte, which is the
additional number of characters of prefix of the preceding entry to
use beyond the number that the preceding entry is using of its
predecessor. (The counts can be negative.) Following the count is a
null-terminated ASCII remainder --- the part of the name that follows
the shared prefix.
If the offset-differential count is larger than can be stored in a
byte (+/-127), the byte has the value 0x80 and the count follows in a
2-byte word, with the high byte first (network byte order).
Every database begins with a dummy entry for a file called `LOCATE02',
which locate checks for to ensure that the database file has the
correct format; it ignores the entry in doing the search.
Databases can not be concatenated together, even if the first
(dummy) entry is trimmed from all but the first database. This
is because the offset-differential count in the first entry of the
second and following databases will be wrong.
There is also an old database format, used by Unix
locate
and
find
programs and earlier releases of the GNU ones. updatedb runs
programs called bigram and code to produce old-format
databases. The old format differs from the above description in the
following ways. Instead of each entry starting with an
offset-differential count byte and ending with a null, byte values
from 0 through 28 indicate offset-differential counts from -14 through
14. The byte value indicating that a long offset-differential count
follows is 0x1e (30), not 0x80. The long counts are stored in host
byte order, which is not necessarily network byte order, and host
integer word size, which is usually 4 bytes. They also represent a
count 14 less than their value. The database lines have no
termination byte; the start of the next line is indicated by its first
byte having a value <= 30.
In addition, instead of starting with a dummy entry, the old database
format starts with a 256 byte table containing the 128 most common
bigrams in the file list. A bigram is a pair of adjacent bytes.
Bytes in the database that have the high bit set are indexes (with the
high bit cleared) into the bigram table. The bigram and
offset-differential count coding makes these databases 20-25% smaller
than the new format, but makes them not 8-bit clean. Any byte in a
file name that is in the ranges used for the special codes is replaced
in the database by a question mark, which not coincidentally is the
shell wildcard to match a single character.
EXAMPLE
Input to frcode:
/usr/src
/usr/src/cmd/aardvark.c
/usr/src/cmd/armadillo.c
/usr/tmp/zoo
Length of the longest prefix of the preceding entry to share:
0 /usr/src
8 /cmd/aardvark.c
14 rmadillo.c
5 tmp/zoo
Output from frcode, with trailing nulls changed to newlines
and count bytes made printable:
The best way to report a bug is to use the form at
http://savannah.gnu.org/bugs/?group=findutils.
The reason for this is that you will then be able to track progress in
fixing the problem. Other comments about locate(1) and about
the findutils package in general can be sent to the
bug-findutils
mailing list. To join the list, send email to
bug-findutils-request@gnu.org.
