.\"
.\"	@(#)grok.1	6.5 (CSU) 10/11/91
.\"

.TH GROK 1 "November 11, 1991
.SH NAME
grok \- Match and repair errors in regular expressions
.SH SYNOPSIS
.in +\w'\fBgrok \fR'u
.ti -\w'\fBgrok \fR'u
\fBgrok \fR \fIdefinition-file\fR

.SH DESCRIPTION
.I Grok
matches instances of \fIlex\fR-style regular expressions from the input
and prints them.
In addition, it tries to repair input lines to make them instances of the
regular expression, using single-character insertions, deletions, and 
substitutions.
Different single-character edits can be assigned different costs,
and \fIgrok\fR will find a repair of lowest total cost.

The algorithm used to find lowest-cost repairs takes time proportional
to the product of the length of the input and the cost of the repair.
A maximum repair cost tolerance must be supplied.
\fIgrok\fR ignores input lines that cannot be repaired within the tolerance.
The match is \fIanchored\fR, unlike the matching performed by \fIgrep\fR.

The \fIdefinition-file\fR contains entries to define character
repair costs, followed by a pair of percent signs alone on a line,
followed by a single regular expression.   For example, the following 
defines repair for lines scanned from a bank statement:

.ne 4
.nf
.ft L
% correct anything to anything else for 3
%correct        .        .        3
 
% a number of common numeric misreads.  4 is often read as (+
% but we cannot correct string to string.
%correct        "Z"        "2"        1
%correct        "z"        "2"        1
%correct        "S"        "5"        1
%correct        "s"        "5"        1
%correct        "l"        "1"        1
%correct        "i"        "1"        1
%correct        "I"        "1"        1
%correct        "o"        "0"        1
%correct        "O"        "0"        1
%correct        "r"        "5"        1
%correct        "?"        "9"        1
%correct        "+"        "4"        1

% default insertion cost is 2
%insert         .          2

% charge extra for numbers to avoid hallucinating numbers.
%insert         [0-9]      3

% delete anything for 2.  correction cost <= insert + delete.
%delete         .          2
%delete         [0-9]      3

% the maximum tolerable error cost 
%tolerance      20
%%
([ ]{20,30}[+ ]([ ]{5,7}[0-9]{5}|"NO ITEM # ")[ ]{3,4}[0-9]{9}
[ ]{5,7}[0-1][0-9]-[0-3][0-9][ ]{6,13}(""|[1-9][0-9]{0,2}|[1-9]
[0-9]{0,2}","[0-9]{3})\.[0-9]{2}([ ]{5,10}[+ ]([ ]{5,7}[0-9]{5}|
"NO ITEM # ")[ ]{3,4}[0-9]{9}[ ]{5,7}[0-1][0-9]-[0-3][0-9][ ]{6,13}
(""|[1-9][0-9]{0,2}|[1-9][0-9]{0,2}","[0-9]{3})\.[0-9]{2})?)|
("       ENCLOSED           NUMBER        NUMBER".*)

.ft R
.fi

\fB%correct\fR, \fB%insert\fR, and \fB%delete\fR lines supply repair costs
to replace each character with each other, as well as for insertions
and deletions.  Single-character regular expressions are accepted,
and later costs overwrite earlier ones.

The regular expression has been broken into several lines for display.
It defines the expected
structure of important lines on one style of bank statement.

\fBGrok\fR outputs the repaired lines.

.SH BUGS
No provision is made for breaking long defining expressions across lines.

.SH AUTHOR
Alan Wendt.  Contributors include Eugene Myers (repair algorithm), Vern Paxson (\fBflex\fR code).