File::Seek

Category:	Text Processing
Author/Contact Info	tachyon aka Dr James Freeman jfreeman@tassie.net.au
Description:	This module implements the halve the difference algorithm to efficiently (and rapidly) find an element(s) in a sorted file. It provides a number of useful methods and can reduce search times in large files by several orders of magnitude as it uses a geometric search rather than the typical linear approach. Logfiles are a typical example where this is useful. I have never written anything I considered worthy of submitting to CPAN but thought this might be worthwhile.
package File::Seek; use strict; # use warnings; # you can use warnings if you have 5.6+ use Time::Local; require Exporter; use vars qw( @ISA @EXPORT @EXPORT_OK %EXPORT_TAGS $VERSION $VERSION ); @ISA = qw( Exporter ); @EXPORT = (); @EXPORT_OK = qw( alphabetic numeric find_time get_between get_last ) +; %EXPORT_TAGS = ( 'all' => \@EXPORT_OK ); $VERSION = '0.01'; my ($count, $exact_match ); my $debug = 0; # set true to watch progression of algorithm my $max_tries = 42; my $descending = 0; my $cuddle = 0; my $line_length = 80; my $error_msg = ''; my $stationary = 0; my $silent = 0; my $NAME = 'File::Seek'; my $EMAIL = 'jfreeman@tassie.net.au'; my %months = ( Jan => 0, Feb => 1, Mar => 2, Apr => 3, May => 4, Jun +=> 5, Jul => 6, Aug => 7, Sep => 8, Oct => 9, Nov => 10, Dec +=> 12); my $default_rec_sep = ($^O =~ m/win32\|vms/i) ? "\015\012" : ( $^O =~ /mac/i ) ? "\015" : "\012"; # some subs to set optional vars OO style sub set_cuddle { $cuddle = 1 }; sub set_no_cuddle { $cuddle = 0 }; sub set_descending { $descending = 1 }; sub set_ascending { $descending = 0 }; sub set_max_tries { $max_tries = shift \|\| 42 }; sub set_line_length { $line_length = shift \|\| 80; $line_length = 80 un +less $line_length >= 1 }; sub set_silent { $silent = 1 }; sub set_verbose { $silent = 0 }; sub set_debug { $debug = 1 }; sub no_debug { $debug = 0 }; sub was_exact { $exact_match }; sub error { $error_msg; }; # basic line munge (just chomp it) sub basic_munge { local $_ = shift \|\| return undef; chomp; return +$_ }; sub alphabetic { local FILE = shift; my $string = shift; my $munge_ref = shift \|\| \&basic_munge; $error_msg = ''; $stationary = 0; _find( FILE, $string, $munge_ref, \&_test_alphabetic ); } sub numeric { local FILE = shift; my $number = shift; my $munge_ref = shift \|\| \&basic_munge; $error_msg = ''; $stationary = 0; _find( FILE, $number, $munge_ref, \&_test_numeric ); } sub find_time { local FILE = shift; my $find = shift; my $not_gmtime = shift; $error_msg = ''; $stationary = 0; my $munge_ref = \&get_epoch_seconds; my $epoch = get_epoch_seconds( $find ); # if $epoch is defined we assume a date string else real epoch sec +s $find = (defined $epoch) ? $epoch : $find; _find( FILE, $find, $munge_ref, \&_test_numeric ); } sub get_epoch_seconds { my $line = shift; return undef unless defined $line; # grab a scalar localtime looking like string from the line my ($wday,$mon,$mday,$hours,$min,$sec,$year) = $line =~ m/(\w\w\w)\s+(\w\w\w)\s+(\d{1,2})\s+(\d\d):(\d\d):(\d +\d)\s+(\d{4})/; unless ($year) { $error_msg = "Unable to find time like string in line:\n$line" +; warn $error_msg unless $silent; return undef; } $mon = $months{$mon}; # convert to numerical months 0 - 11 return timegm($sec,$min,$hours,$mday,$mon,$year); } sub get_between { local FILE = shift; my $begin = shift \|\| 0; my $finish = shift \|\| 0; my $rec_sep = shift \|\| $default_rec_sep; $error_msg = ''; binmode FILE; ($begin , $finish) = ($finish, $begin) if $begin > $finish; my $bytes = $finish - $begin; seek FILE, $begin, 0; my $read = read (FILE, my $buffer, $bytes); if ( $read < $bytes ) { $error_msg = "Short read $NAME\nWanted: $bytes Got: $read\n"; warn $error_msg unless $silent; return undef; } $buffer = substr $buffer, 0, $bytes; my @lines = split $rec_sep, $buffer; return wantarray ? @lines : [ @lines ]; } sub get_last { local FILE = shift; my $num_lines = shift; my $rec_sep = shift \|\| $default_rec_sep; $error_msg = ''; binmode FILE; my $file_size = -s FILE; my $read = $line_length * $num_lines; my @file; GET: { $read = $read << 1; # double our estimate my $position = $file_size - $read; if ($position < 0 ) { seek FILE, 0, 0; $read = read ( FILE, my $buffer, $file_size ); @file = split "$rec_sep", $buffer; chomp (@file); if ( $num_lines > @file ) { $error_msg = "$NAME Wanted $num_lines lines but file +only "; $error_msg .= "contains" . @file . " lines. Whole file + returned\n"; warn $error_msg unless $silent; return wantarray ? @file : [ @file ]; } splice @file, 0, (scalar @file - $num_lines); ; return wantarray ? @file : \@file; } else { seek FILE, $position, 0; $read = read ( FILE, my $buffer, $read ); my $count_lines = $buffer; my $line_count = $count_lines =~ s/$rec_sep//og; my $average_line_length = ($line_count) ? ( $read / $line_ +count ) : $read; if ($average_line_length > $line_length ) { $line_length = $average_line_length; $read = $num_lines * $average_line_length; } redo GET unless $num_lines < $line_count; # first line will +be a partial @file = split "$rec_sep", $buffer; } splice @file, 0, (scalar @file - $num_lines); ; return wantarray ? @file : \@file; } } # this is the main routine that implements the halve the difference se +arch sub _find{ my ( $partial, $line, $next ); local FILE = shift; my $find = shift; my $munge_ref = shift; my $comp_type = shift; my $file_size = -s FILE; my $top = 0; my $bottom = $file_size; $exact_match = 0; $count = 0; # first line is an edge case, so we test it now seek FILE, 0, 0; $line = &$munge_ref( scalar <FILE> ); $next = &$munge_ref( scalar <FILE> ); unless (defined $line and defined $next) { $error_msg = "$NAME Unable to munge valid data from first or s +econd lines\n"; warn $error_msg unless $silent; return undef; } my $ans = &$comp_type($find, $line, $next); if ( $ans == 0 or ($descending and $ans == 1) or (not $descending +and $ans == -1) ) { seek FILE, 0, 0; my $gobble = <FILE> if $exact_match == 2; return tell FILE; } # start the halve the difference loop, we count iterations and # will abort the loop if we exceed the specified $max_tries while ( ++$count ) { my $middle = int(($top+$bottom)/2); seek FILE, $middle , 0; $partial = <FILE>; $line = &$munge_ref( scalar <FILE> ); $next = &$munge_ref( scalar <FILE> ); $ans = &$comp_type($find, $line, $next); print "A:'$ans' C:'$count' T:'$top' B:'$bottom' Find:'$find' L +:'$line' N:'$next'\n" if $debug; unless (defined $ans) { $error_msg = "\n\n\nArk, $NAME got to EOF\n"; $error_msg .= &_debug($find, $line, $next, $file_size, $to +p, $bottom, $descending); warn $error_msg unless $silent; return undef; } if ( $ans ) { if ( $descending ) { ( $ans == 1 ) ? $bottom = $middle : $top = $middle; } else { ( $ans == 1 ) ? $top = $middle : $bottom = $middle; + } } else { seek FILE, $middle, 0; my $partial = <FILE>; if ($exact_match) { my $gobble = <FILE> if $exact_match == 2; } else { my $gobble = <FILE> unless $cuddle; } my $pos = tell FILE; # end of file is an edge case return ( $pos < $file_size ) ? $pos : undef; } if ( $count >= $max_tries ) { $error_msg = "\n\n\nArk, $NAME baling out of infinite loo +p after $max_tries tries\n"; $error_msg .= &_debug($find, $line, $next, $file_size, $to +p, $bottom, $descending); warn $error_msg unless $silent; return undef; } } } # numeric test routine { my $last_line = 0; my $last_next = 0; sub _test_numeric { my ($find, $line, $next) = @_; # EOF if $line is not defined return undef unless defined $line; # check for movement - if repeatedly none we have reached EOF. if ($line eq $last_line and defined $next and defined $last_ne +xt and $next eq $last_next) { $stationary++; if ($stationary > 2) { $stationary = 0; return undef; } } ($last_line, $last_next) = ($line, $next); # check for an exact match $exact_match = 2 if defined $next and $find == $next; $exact_match = 1 if $find == $line; # line must be defined return 0 if $exact_match; # check for between-ness depending on sort order return 0 if !$descending and defined $next and $line < $find an +d $find < $next; return 0 if $descending and defined $next and $line > $find an +d $find > $next; # otherwise indicate which way to jump return +1 if $line < $find; return -1 if $line > $find; } } # alphabetic test routine { my $last_line = ''; my $last_next = ''; sub _test_alphabetic { my ($find, $line, $next) = @_; return undef unless defined $line; # check for movement - if repeatedly none we have reached EOF. + if ($line eq $last_line and defined $next and defined $last_ne +xt and $next eq $last_next) { $stationary++; if ($stationary > 2) { $stationary = 0; return undef; } } ($last_line, $last_next) = ($line, $next); # check for an exact match $exact_match = 2 if defined $next and $find eq $next; $exact_match = 1 if $find eq $line; return 0 if $exact_match; # check for between-ness depending on sort order return 0 if !$descending and defined $next and $line lt $find a +nd $find lt $next; return 0 if $descending and defined $next and $line gt $find a +nd $find gt $next; # otherwise indicate which way to jump return +1 if $line lt $find; return -1 if $line gt $find; } } sub _debug { my ($find, $line, $next, $file_size, $top, $bottom, $mode) = @_; $line = 'undef' unless defined $line; $next = 'undef' unless defined $next; $line = sprintf "0x%x", ord $line unless $line; $next = sprintf "0x%x", ord $next unless $next; $mode = ($mode)? "Descending" : "Ascending"; my $message = "Failed to find: '$find'\n"; $message .= "The search mode for the file was '$mode order'\n"; $message .= "\$line:\t$line\n"; $message .= "\$next:\t$next\n"; $message .= sprintf "File size: %12d Bytes\n", $file_size; $message .= sprintf "\$top: %12d Bytes\n", $top; $message .= sprintf "\$bottom: %12d Bytes\n", $bottom; $message .= "Perhaps try reversing the search mode\n"; $message .= "Are you using the correct method - alhpabetic or nu +meric?\n\n"; $message .= "If you think it is a bug please send a bug report t +o:\n"; $message .= "$EMAIL\n"; $message .= "A sample of the file, the call to this module and\n +"; $message .= "this error message will help to fix the problem\n"; return $message; } "tachyon"; __END__ =head1 NAME Seek.pm - A Perl module providing fast access to large files =head1 SYNOPSIS use File::Seek ':all'; open BIG, $file or die $!; # find a number or the first number greater in a file (ascending ord +er) $tell = numeric( BIG, $number ); # read a line in from where we matched in the file $line = <BIG>; print "Found exact match as $line" if File::Seek:was_exact(); # find a string or the first string greater in a file (alphabetical +order) $tell = alphabetic( BIG, $string ); $line = <BIG>; # find a date in a logfile supplying a scalar localtime type string $tell = find_time( BIG, "Thu Aug 23 22:59:16 2001" ); # or supplying GMT epoch time $tell = find_time( BIG, 998571554 ); # get all the lines after our date @lines = <BIG>; # get the lines between two logfile dates $begin = find_time( LOG, $start ); $end = find_time( LOG, $finish ); # get lines as an array @lines = get_between( LOG, $begin, $end ); # get lines as an array reference $lines = get_between( LOG, $begin, $end ); # use you own sub to munge the file line data before comparison $tell = numeric( BIG, $number, \&epoch ); $tell = alphabetic( BIG, $string, \&munge_line ); # use methods on files in reverse alphabetic or descending numerical + order File::Seek::set_descending(); # for inexact matches set FH so first value read is before and secon +d after File::Seek::set_cuddle(); # get last $n lines of any file as an array @lines = get_last( BIG, $n ) # or an array reference $lines = get_last( BIG, $n ) # change the input record separator from the OS default @lines = get_last( BIG, $n, $rec_sep ) =head1 DESCRIPTION This module provides fast access to data from large files. Three metho +ds numeric() alphabetic() and find_time() depend on the file data being sorted in some way. Logfiles are a typical example of big files +that are sorted (by date stamp). The get_between() method can be used to ge +t a chunk of lines efficiently from anywhere in the file. The required p +ostion(s) for the get_between() method are supplied by the previous methods. The + get_last() method will efficiently get the last N lines of any file, s +orted or not. With sorted data a linear search is not required. Here is a typical li +near search while (<FILE>) { next unless /$some_cond/ # found cond, do stuff } Remember that old game where you try to guess a number between lets sa +y 0 and say 128? Let's choose 101 and now try to guess it. Using a linear search is the same as going 1 higher 2 higher 3 higher +... 100 higher 101 correct! Consider the geometric approach: 64 higher 96 +higher 112 lower 104 lower 100 higher 102 lower - ta da must be 101! This is +the halving the difference search method and can be applied to any data se +t where we can logically say higher or lower. In other words any sorted data s +et can be searched like this. It is a far more efficient method - see the SPE +ED section for a quick analysis. =head2 The two basic methods - numeric() and alphabetic() There are two basic methods - numeric() to do numeric searches and alphabetic() that does alphabetic searches. You call the functions like this: $tell = numeric( BIG, $find ); $tell = alphabetic( BIG, $find ); These methods take two required arguments. BIG is a FILEHANDE to read + from. $find is the item you wish to find. $find must be appropriate to the f +unction as the numeric method will make numeric comparisons ( == < > ). Simila +rly the alphabetic method makes string comparisons ( eq lt gt ). You will get +strange results if you use the wrong method just as you do if you say use == w +hen you actually meant eq =head2 Return values with search success and failure The return value from the numeric() and alphabetic() methods depend on + the result of the search. If the search fails the return value is undefine +d. A search can succeed in two ways. If an exact match is found then the current file position pointer is set to the beginning of the matching +line. The return value is the corresponding response from tell(). This means + that the next read from <FILEHANDLE> will return the matching line. Subsequent reads return the following lines as expected. Alternatively a search will succeed if a point in the file can be foun +d such that $find is cuddled between two adjacent lines. For example consider + searching for the number 42 in a file like this: .. 36 40 <- Before 44 <- After 48 .. The number 42 is not actually there but the search will still succeed +as it is between 40 and 44. By default the file postion pointer is set to th +e beginning of the line '44' so the next read from <FILEHANDLE> will ret +urn this line. If the File::Seek::set_cuddle() function is called then the + file position pointer will be set to the beginning of line '40' so that the + first two reads from <FILEHANDLE> will cuddle the in-between value in +$find. =head2 Adding line munging to make the basic methods more useful Both the numeric and alphabetic subs take an optional third argument. This optional argument is a reference to a subroutine to munge the file lines so that suitable values are extracted for comparison to $fi +nd. $tell = numeric( BIG, $find, \&munge_line ); $tell = alphabetic( BIG, $find, \&munge_line ); A good example of this is the find_time() function. This is just an implementation of the basic numeric algorithm similar to this. $tell = numeric ( BIG, $epoch_seconds, \&get_epoch_seconds ); sub get_epoch_seconds { use Time::Local; my $line = shift; return undef unless defined $line; my %months = ( Jan => 0, Feb => 1, Mar => 2, Apr => 3, May => 4, Jun = +> 5, Jul => 6, Aug => 7, Sep => 8, Oct => 9, Nov => 10, Dec = +> 12); # grab a scalar localtime looking like string from the line my ($wday,$mon,$mday,$hours,$min,$sec,$year) = $line =~ m/(\w\w\w)\s+(\w\w\w)\s+(\d{1,2})\s+(\d\d):(\d\d) +:(\d\d)\s+(\d{4})/; unless ($year) { $error_msg = "Unable to find time like string in line:\n$l +ine"; warn $error_msg unless $silent; return undef; } $mon = $months{$mon}; # convert to numerical months 0 - 11 return timegm($sec,$min,$hours,$mday,$mon,$year); } As the search is made the test lines are passed to the munging sub. Th +is sub needs to return a string or number that we can perform comparison on. +In this case the sub looks for something that looks like a scalar localtime() +string, and assuming this is a date passes it to timegm() for conversion to epoch seconds and returns this number. You can see further examples of this in the test suite test.pl =head2 find_time() The find_time() function is an implementation of the basic numeric met +hod as discussed briefly above. You call it like: $tell = find_time( LOG, 'Thu Jan 1 00:42:00 1970' ); $tell = find_time( LOG, $epoch_seconds ); You may use either a scalar localtime() like string or epoch seconds. +If you use epoch seconds it assumes gmtime. If in doubt use the string as alt +hough it works internally with gmtime the offsets cancel out and the correct + result is returned. =head2 Getting lines from the middle of a file - get_between() Say you have a logfile and you want to get the log between one date an +d another. You can simply use two calls to the find_time() to get the be +ginning and end positions and then use get_between() to get the lines. # get the lines between two logfile dates $begin = find_time( LOG, $start ); $end = find_time( LOG, $finish ); # get lines as an array @lines = get_between( LOG, $begin, $end ); # get lines as an array reference $lines = get_between( LOG, $begin, $end ); The get_between() method returns an array in list context as above and + a reference to an array in scalar context. This function needs to apply binmode so it splits the lines based on a + system specific default record separator. This is derived as below: my $default_rec_sep = ($^O =~ m/win32\|vms/i) ? "\015\012" : ( $^O =~ /mac/i ) ? "\015" : "\012"; You can override this on a per file basis by passing the record separa +tor to the get_between() function. @lines = get_between( LOG, $begin, $end, $rec_sep ); Modifying $/ has no effect. Note that the record separator is not ret +urned* in the array. As a result the returned array has effectively had every + element chomped. Warning - this method will apply binmode to the FH so line endings will possibly not be converted properly if you try to continue to read + from it. As there is no unbinmode() close the FH afterwards and reopen it i +f you want to read from it. You can seek FH, 0, $end if say you want to read + more lines after $end. =head2 Getting the lines from the beginning of a file - get_between() Using the get_between() method you can efficiently get the lines at th +e beginning of a file. Although you can just read in lines sequentially +with a while loop this requires that you test each line. If you can find th +e end point using the find_time() numeric() or alphabetic() methods you can the just get what you need. For large files many thousands of unnecessary tests are avoided saving time. Using the example above you simply set $begin to 0 $begin = 0; $end = find_time( LOG, $finish ); @lines = get_between( LOG, $begin, $end ); =head2 Getting lines from the end of a file - get_between() You can similarly use get between to get all the lines from a specific + point up to the end of the file. The end is just the size of the file so: $begin = find_time( LOG, $start ); $end = -s LOG; @lines = get_between( LOG, $begin, $end ); =head2 Getting lines from the end of a file - get_last() This method does not depend on the file being sorted to work. When you use the get_last() method the module estimates how many bytes + at the end of the file to read in. To make the estimate the module multi +plies the default line length (80 chars) by the number of lines required and + then doubles it. If it does not get sufficient lines on its first attempt it re-estimat +es the line length from the actual data read in, re-calculates the read, doubles it and then tries again. This algorithm is unlikely +to take more than 2 reads but if you have unusually long of short lines y +ou may get a small speed benefit by using the set_line_length() method to set + the average line length. The default is 80 chars per line. Setting the lin +e length close to the actual will also avoid reading a excessive quantity of da +ta into memory. # get last $n lines of any file as an array @lines = get_last( BIG, $n ) # or an array reference $lines = get_last( BIG, $n ) # change the input record separator from the default @lines = get_last( BIG, $n, $rec_sep ) This function needs to apply binmode so it splits the lines based on a + system specific default record separator. This is derived as below: my $default_rec_sep = ($^O =~ m/win32\|vms/i) ? "\015\012" : ( $^O =~ /mac/i ) ? "\015" : "\012"; You can override this on a per file basis by passing the record separa +tor $rec_sep to the get_last() function as shown. Modifying $/ has no effe +ct. Note that the record separator is not returned* in the array. As a result the returned array has effectively had every element chomped. Warning - this method will apply binmode to the FH so line endings will possibly not be converted properly if you try to continue to read + from it. As there is no unbinmode() close the FH afterwards and reopen it i +f you want to read from it. You can seek FH, 0, $end if say you want to read + more lines after $end. =head1 EXPORT Nothing is exported by default. The following 5 methods are available +for import: alphabetic() numeric() find_time() get_between() get_last() You can import just the method you want with a: use File::Seek 'numeric'; or all 5 methods using the ':all' tag. use File::Seek ':all'; =head1 OPTIONS There are some options available via non exported function calls. You will need to fully specify the name if you want to use thes +e. =head2 File::Seek::error() If a function returns undefined there has been an error. error() will contain the text of the last error message or a null string if there was no error. =head2 File::Seek::was_exact() was_exact() will return true if an exact match was found. It will be false if the match was in between or failed. =head2 File::Seek::set_cuddle() File::Seek::set_no_cuddle() set_cuddle() changes the default line returned for in between matches + as discussed above and set_no_cuddle() restores default behaviour =head2 File::Seek::set_descending() File::Seek::set_ascending() By default ascending numerical order and alphabetical order are assume +d. This assumption can be reversed by calling set_descending() and reset by calling set_ascending() We need to know the order to seek within th +e file in the correct direction. =head2 File::Seek::set_max_tries($max) This sets the maximum times that the module will try the halve the difference search before it decides there is a problem and bails out. The default value is 42 which allows files with up to 242 or a bit m +ore than 1012 lines to be processed. A seek in a million line file will +take a mere 20 tries to find the required value. =head2 File::Seek::set_line_length($integer) When you use the get_last() method the module uses its default line length to estimate how many bytes at the end of the file to read +in. You can improve speed slightly and decrease memory usage by setting an + accurate line length. The default is 80 chars per line. The function w +ill work fine regardless of what the line length is, this is just an effic +iency tweak. =head2 File::Seek::set_silent() File::Seek::set_verbose() You can silence or activate error messages by calling these two subs. +The default is verbose. =head2 File::Seek::set_debug() File::Seek::no_debug() Sets debug on or off. Default is of course off. =head1 SPEED Here is a table that demonstrates the advantage of using the halve the + difference algorithm. Num items Lin avg Geom avg Lin:Geom 2 1 1 1 4 2 2 1 8 4 3 1 16 8 4 2 32 16 5 3 64 32 6 5 128 64 7 9 256 128 8 16 512 256 9 28 1024 512 10 51 2048 1024 11 93 4096 2048 12 170 8192 4096 13 315 16384 8192 14 585 32768 16384 15 1092 65536 32768 16 2048 131072 65536 17 3855 262144 131072 18 7281 524288 262144 19 13797 1048576 524288 20 26214 Even though there is an overhead involved with this search this is min +or as the number of tests required is so much less. Speed increases of 10 +0-1000 of times are typical. An OO interface slows things down by > 50% so is not used. =head1 BUGS Bound to be some. The binmoding of the FH by get_between() and get_las +t() can not be easily avoided. =head1 AUTHOR (c) Dr James Freeman 2000-01 E<lt>jfreeman@tassie.net.auE<gt> All rights reserved. This package is free software and is provided ``as is'' without expres +s or implied warranty. It may be used, redistributed and/or modified under +the terms of the Perl Artistic License (see http://www.perl.com/perl/misc/Artist +ic.html) =head1 SEE ALSO For details about the mystical significance of the number 42 and how i +t can be applied to Life the Universe and everything see The Hitch Hiker's G +uide to the Galaxy 'trilogy' by the recently departed Douglas Adams.

Back to Code Catacombs