If the table or list file is not large (under several hundred rows), it is probably not worth the effort to index it: in that case, use row. search is for high speed or interactive access using indexed keys, while row is slower, but can retrieve rows that match complex logical conditions and regular expressions.

search reads its input in three ways: (1) interactively, from the keyboard; (2) from a keytable, using the "`< file'" form; and (3) from a pipe.

Exactly how search looks for keys depends upon the method specified on the command line. For some methods it first searches another table called a secondary index file that was built by the index program. In one case, sequential, search looks at every row. To determine which method to use, try them all and use the UNIX time command or the timesearch shell script to find out how long each method takes.

-mr

record number. Record numbers are integers which specify the offset in the file of the row to retrieve. For example, if the record number is 5, search -mr will get the fifth row in the file. search uses a secondary file that is fixed length and has the offset for each record in the big table. The record method is a fast way to retrieve a row, but only if its position in the file is known and maintained. Another program, record, does the same thing without needing an index, but it's slower on large files. Because search is looking for rows by position rather than content, the -n, -x, +x, and -i options do not apply.

-ms

sequential or linear. A sequential search looks at each record in the big table in first row to last row order. It is the default method and is the same as using no -m? switch at all. It is provided as a benchmark for timing other methods. If the file is not big enough to make the other access methods worthwhile, use sequential search or row to save the overhead of index. Sequential can search multiple key columns, and any column can be searched. Searching with the sequential method is different from general UNIX pattern matching programs like grep, or bm because it compares chosen columns instead of scanning entire rows.

-mb

binary method. A binary search first looks at the middle record of the file. If the key column value is greater than the key that search is looking for, search will look at the record one quarter of the way through the file. search will continue to divide the part of the file it looks at in half, until the row is found, or not. The binary method requires that the file be sorted on the key column. If tableorlist is to be searched using the -x, +x or -n switches, then those switches must also be used with the index command prior to searching.

The following table shows how many file accesses are required for files with various numbers of rows -- from small to huge.

Accesses   Records              Number
--------   -------              ------
10         thousand              1,000
20         million           1,000,000
30         billion       1,000,000,000
40         trillion  1,000,000,000,000

-mi

inverted method. An inverted search uses a secondary file created by the index program. This index file has two columns. One is the key column from the table which is sorted by the index program so that search can do a binary search for the key. If the key is found, search looks at the adjacent offset column for the address of the row in the big table. With this address, search retrieves the row in a single disk read. The inverted method takes a little longer than binary, but sorting the index file is much faster than sorting the big file. If tableorlist is to be searched using the -x, +x or -n switches, then those switches must also be used with the index command which creates the tableorlist.i secondary file.

-mh

hash method. A hash search computes an address from the key and looks in the secondary file at that address, where it finds an offset in the main file. search retrieves the row at the main file offset and compares the keys. If they match, the row is printed. If the offset is zero, then the keys do not match and the row is not in the file. If another key hashes to the same address (called a collision), then search will keep looking at subsequent entries in the secondary table until it finds a match, or a zero offset. A hash search usually takes two disk accesses when the hash table is half full. This is why hash tables are set up to be twice as large as the number of records we are hashing. As the data table grows, it should be reindexed to keep the hash speed down. Hash is a good method for tables that must remain static: that is, they should not change significantly over time.

It is important that other options used with the index command which creates the hash table must also be used in the search command. This is because the input key is defined by the option prior to hashing: for example, if -x is specified, then the key is stripped of excess blanks and converted to lower case before hashing. If -x is not used in the subsequent search command, then the hash value will be different and the key will not be found, even though it exists. This applies to the +x and -n. options as well. The -i switch is the exception to this rule: it may be specified on the search command line to suppress printing of blank keys even if it was not used on the index command line.

-fc

use c (instead of TAB) as the input column separation character.

-Dn

turn on debugging at level n, or 1 if n is not specified.

-h

suppress printing of the table head line.

-s

only a single row is searched for. This will speed up the search because it quits as soon as the first match is found.

-x

allow partial matching and ignore case (like Berkeley look). By default, non-blank values must match exactly.

+x

exact match. The default mode ignores blanks when comparing keys; using +x forces byte-for-byte matching of key and column contents.

-i

ignore blank key matches.

-l

print the location of the record and also the location of the index record in the index table. This location, in the form: "start end xstart xend," can be used by replace, lock and unlock. By default, search prints this location on the "standard error" to separate it from the output table. seek is a shell program which uses this option to find the location and output the record into another file at the same time. Turning on this option also turns on the -s option because only one row (per user) may be locked and/or replaced at a time.

$ cat inventory
Item    Amount  Cost    Value   Desc
----    ------  ----    -----   -----------
1       4       50.1            rubber gloves
03      23      19      437     plates
001     100     5.23    500     test tubes
2       99      24      2376    cleaning cloth
4.0     89      147     13083   bunsen burners
005     5       175     875     scales
1.000   5       80      400     clamps

$ index -mi -n inventory Item
$ cat inventory.i
   Offset       Item
---------       -------
       63       1
      107       001
      219       1.000
      135       2
       87       03
      166       4.0
      198       005

$ echo 1 | search -mi -n inventory Item
Item    Amount  Cost    Value   Desc
----    ------  ----    -----   -----------
1       4       50.1            rubber gloves
001     100     5.23    500     test tubes
1.000   5       80      400     clamps