Server IP : 85.214.239.14 / Your IP : 3.145.83.149 Web Server : Apache/2.4.62 (Debian) System : Linux h2886529.stratoserver.net 4.9.0 #1 SMP Tue Jan 9 19:45:01 MSK 2024 x86_64 User : www-data ( 33) PHP Version : 7.4.18 Disable Function : pcntl_alarm,pcntl_fork,pcntl_waitpid,pcntl_wait,pcntl_wifexited,pcntl_wifstopped,pcntl_wifsignaled,pcntl_wifcontinued,pcntl_wexitstatus,pcntl_wtermsig,pcntl_wstopsig,pcntl_signal,pcntl_signal_get_handler,pcntl_signal_dispatch,pcntl_get_last_error,pcntl_strerror,pcntl_sigprocmask,pcntl_sigwaitinfo,pcntl_sigtimedwait,pcntl_exec,pcntl_getpriority,pcntl_setpriority,pcntl_async_signals,pcntl_unshare, MySQL : OFF | cURL : OFF | WGET : ON | Perl : ON | Python : ON | Sudo : ON | Pkexec : OFF Directory : /usr/share/doc/libalgorithm-diff-perl/examples/ |
Upload File : |
#!/usr/bin/perl # # `Diff' program in Perl # Copyright 1998 M-J. Dominus. (mjd-perl-diff@plover.com) # # This program is free software; you can redistribute it and/or modify it # under the same terms as Perl itself. # # Altered to output in `context diff' format (but without context) # September 1998 Christian Murphy (cpm@muc.de) # # Context lines feature added # Unified, "Old" (Standard UNIX), Ed diff added September 1998 # Reverse_Ed (-f option) added March 1999 # Amir D. Karger (karger@bead.aecom.yu.edu) # # Modular functions integrated into program # February 1999 M-J. Dominus (mjd-perl-diff@plover.com) # # In this file, "item" usually means "line of text", and "item number" usually # means "line number". But theoretically the code could be used more generally use strict; use Algorithm::Diff qw(diff); # GLOBAL VARIABLES #### # After we've read up to a certain point in each file, the number of items # we've read from each file will differ by $FLD (could be 0) my $File_Length_Difference = 0; #ed diff outputs hunks *backwards*, so we need to save hunks when doing ed diff my @Ed_Hunks = (); ######################## my $usage = << "ENDUSAGE"; Usage: $0 [{-c | -C lines -e | -f | -u | -U lines |-q | -i | -w}] oldfile newfile -c do a context diff with 3 lines of context -C do a context diff with 'lines' lines of context (implies -c) -e create a script for the ed editor to change oldfile to newfile -f like -e but in reverse order -u do a unified diff with 3 lines of context -U do a unified diff with 'lines' lines of context (implies -u) -q report only whether or not the files differ -i ignore differences in Upper/lower-case -w ignore differences in white-space (space and TAB characters) By default it will do an "old-style" diff, with output like UNIX diff ENDUSAGE my $Context_Lines = 0; # lines of context to print. 0 for old-style diff my $Diff_Type = "OLD"; # by default, do standard UNIX diff my ($opt_c, $opt_u, $opt_e, $opt_f, $opt_q, $opt_i, $opt_w); my $compareRoutineRef = undef; while ($ARGV[0] =~ /^-/) { my $opt = shift; last if $opt eq '--'; if ($opt =~ /^-C(.*)/) { $Context_Lines = $1 || shift; $opt_c = 1; $Diff_Type = "CONTEXT"; } elsif ($opt =~ /^-c$/) { $Context_Lines = 3; $opt_c = 1; $Diff_Type = "CONTEXT"; } elsif ($opt =~ /^-e$/) { $opt_e = 1; $Diff_Type = "ED"; } elsif ($opt =~ /^-f$/) { $opt_f = 1; $Diff_Type = "REVERSE_ED"; } elsif ($opt =~ /^-U(.*)$/) { $Context_Lines = $1 || shift; $opt_u = 1; $Diff_Type = "UNIFIED"; } elsif ($opt =~ /^-u$/) { $Context_Lines = 3; $opt_u = 1; $Diff_Type = "UNIFIED"; } elsif ($opt =~ /^-q$/) { $Context_Lines = 0; $opt_q = 1; $opt_e = 1; $Diff_Type = "ED"; } elsif ($opt =~ /^-i$/) { $opt_i = 1; $compareRoutineRef = \&compareRoutine; } elsif ($opt =~ /^-w$/) { $opt_w = 1; $compareRoutineRef = \&compareRoutine; } else { $opt =~ s/^-//; bag("Illegal option -- $opt"); } } if ($opt_q and grep($_,($opt_c, $opt_f, $opt_u)) > 1) { bag("Combining -q with other options is nonsensical"); } if (grep($_,($opt_c, $opt_e, $opt_f, $opt_u)) > 1) { bag("Only one of -c, -u, -f, -e are allowed"); } bag($usage) unless @ARGV == 2; ######## DO THE DIFF! my ($file1, $file2) = @ARGV; my ($char1, $char2); # string to print before file names if ($Diff_Type eq "CONTEXT") { $char1 = '*' x 3; $char2 = '-' x 3; } elsif ($Diff_Type eq "UNIFIED") { $char1 = '-' x 3; $char2 = '+' x 3; } open (F1, $file1) or bag("Couldn't open $file1: $!"); open (F2, $file2) or bag("Couldn't open $file2: $!"); my (@f1, @f2); chomp(@f1 = <F1>); close F1; chomp(@f2 = <F2>); close F2; # diff yields lots of pieces, each of which is basically a Block object my $diffs = diff(\@f1, \@f2, $compareRoutineRef); exit 0 unless @$diffs; if ($opt_q and @$diffs) { print "Files $file1 and $file2 differ\n"; exit 1; } if ($Diff_Type =~ /UNIFIED|CONTEXT/) { my @st = stat($file1); my $MTIME = 9; print "$char1 $file1\t", scalar localtime($st[$MTIME]), "\n"; @st = stat($file2); print "$char2 $file2\t", scalar localtime($st[$MTIME]), "\n"; } my ($hunk,$oldhunk); # Loop over hunks. If a hunk overlaps with the last hunk, join them. # Otherwise, print out the old one. foreach my $piece (@$diffs) { $hunk = new Hunk ($piece, $Context_Lines); next unless $oldhunk; # first time through # Don't need to check for overlap if blocks have no context lines if ($Context_Lines && $hunk->does_overlap($oldhunk)) { $hunk->prepend_hunk($oldhunk); } else { $oldhunk->output_diff(\@f1, \@f2, $Diff_Type); } } continue { $oldhunk = $hunk; } # print the last hunk $oldhunk->output_diff(\@f1, \@f2, $Diff_Type); # Print hunks backwards if we're doing an ed diff map {$_->output_ed_diff(\@f1, \@f2, $Diff_Type)} @Ed_Hunks if @Ed_Hunks; exit 1; # END MAIN PROGRAM sub bag { my $msg = shift; $msg .= "\n"; warn $msg; exit 2; } sub compareRoutine { my $line = shift; $line =~ s/[ \t\r\n]+//g if ($opt_w); $line = uc( $line ) if ($opt_i); return $line; } ######## # Package Hunk. A Hunk is a group of Blocks which overlap because of the # context surrounding each block. (So if we're not using context, every # hunk will contain one block.) { package Hunk; sub new { # Arg1 is output from &LCS::diff (which corresponds to one Block) # Arg2 is the number of items (lines, e.g.,) of context around each block # # This subroutine changes $File_Length_Difference # # Fields in a Hunk: # blocks - a list of Block objects # start - index in file 1 where first block of the hunk starts # end - index in file 1 where last block of the hunk ends # # Variables: # before_diff - how much longer file 2 is than file 1 due to all hunks # until but NOT including this one # after_diff - difference due to all hunks including this one my ($class, $piece, $context_items) = @_; my $block = new Block ($piece); # this modifies $FLD! my $before_diff = $File_Length_Difference; # BEFORE this hunk my $after_diff = $before_diff + $block->{"length_diff"}; $File_Length_Difference += $block->{"length_diff"}; # @remove_array and @insert_array hold the items to insert and remove # Save the start & beginning of each array. If the array doesn't exist # though (e.g., we're only adding items in this block), then figure # out the line number based on the line number of the other file and # the current difference in file lengths my @remove_array = $block->remove; my @insert_array = $block->insert; my ($a1, $a2, $b1, $b2, $start1, $start2, $end1, $end2); $a1 = @remove_array ? $remove_array[0 ]->{"item_no"} : -1; $a2 = @remove_array ? $remove_array[-1]->{"item_no"} : -1; $b1 = @insert_array ? $insert_array[0 ]->{"item_no"} : -1; $b2 = @insert_array ? $insert_array[-1]->{"item_no"} : -1; $start1 = $a1 == -1 ? $b1 - $before_diff : $a1; $end1 = $a2 == -1 ? $b2 - $after_diff : $a2; $start2 = $b1 == -1 ? $a1 + $before_diff : $b1; $end2 = $b2 == -1 ? $a2 + $after_diff : $b2; # At first, a hunk will have just one Block in it my $hunk = { "start1" => $start1, "start2" => $start2, "end1" => $end1, "end2" => $end2, "blocks" => [$block], }; bless $hunk, $class; $hunk->flag_context($context_items); return $hunk; } # Change the "start" and "end" fields to note that context should be added # to this hunk sub flag_context { my ($hunk, $context_items) = @_; return unless $context_items; # no context # add context before my $start1 = $hunk->{"start1"}; my $num_added = $context_items > $start1 ? $start1 : $context_items; $hunk->{"start1"} -= $num_added; $hunk->{"start2"} -= $num_added; # context after my $end1 = $hunk->{"end1"}; $num_added = ($end1+$context_items > $#f1) ? $#f1 - $end1 : $context_items; $hunk->{"end1"} += $num_added; $hunk->{"end2"} += $num_added; } # Is there an overlap between hunk arg0 and old hunk arg1? # Note: if end of old hunk is one less than beginning of second, they overlap sub does_overlap { my ($hunk, $oldhunk) = @_; return "" unless $oldhunk; # first time through, $oldhunk is empty # Do I actually need to test both? return ($hunk->{"start1"} - $oldhunk->{"end1"} <= 1 || $hunk->{"start2"} - $oldhunk->{"end2"} <= 1); } # Prepend hunk arg1 to hunk arg0 # Note that arg1 isn't updated! Only arg0 is. sub prepend_hunk { my ($hunk, $oldhunk) = @_; $hunk->{"start1"} = $oldhunk->{"start1"}; $hunk->{"start2"} = $oldhunk->{"start2"}; unshift (@{$hunk->{"blocks"}}, @{$oldhunk->{"blocks"}}); } # DIFF OUTPUT ROUTINES. THESE ROUTINES CONTAIN DIFF FORMATTING INFO... sub output_diff { # First arg is the current hunk of course # Next args are refs to the files # last arg is type of diff my $diff_type = $_[-1]; my %funchash = ("OLD" => \&output_old_diff, "CONTEXT" => \&output_context_diff, "ED" => \&store_ed_diff, "REVERSE_ED" => \&output_ed_diff, "UNIFIED" => \&output_unified_diff, ); if (exists $funchash{$diff_type}) { &{$funchash{$diff_type}}(@_); # pass in all args } else {die "unknown diff type $diff_type"} } sub output_old_diff { # Note that an old diff can't have any context. Therefore, we know that # there's only one block in the hunk. my ($hunk, $fileref1, $fileref2) = @_; my %op_hash = ('+' => 'a', '-' => 'd', '!' => 'c'); my @blocklist = @{$hunk->{"blocks"}}; warn ("Expecting one block in an old diff hunk!") if scalar @blocklist != 1; my $block = $blocklist[0]; my $op = $block->op; # +, -, or ! # Calculate item number range. # old diff range is just like a context diff range, except the ranges # are on one line with the action between them. my $range1 = $hunk->context_range(1); my $range2 = $hunk->context_range(2); my $action = $op_hash{$op} || warn "unknown op $op"; print "$range1$action$range2\n"; # If removing anything, just print out all the remove lines in the hunk # which is just all the remove lines in the block if ($block->remove) { my @outlist = @$fileref1[$hunk->{"start1"}..$hunk->{"end1"}]; map {$_ = "< $_\n"} @outlist; # all lines will be '< text\n' print @outlist; } print "---\n" if $op eq '!'; # only if inserting and removing if ($block->insert) { my @outlist = @$fileref2[$hunk->{"start2"}..$hunk->{"end2"}]; map {$_ = "> $_\n"} @outlist; # all lines will be '> text\n' print @outlist; } } sub output_unified_diff { my ($hunk, $fileref1, $fileref2) = @_; my @blocklist; # Calculate item number range. my $range1 = $hunk->unified_range(1); my $range2 = $hunk->unified_range(2); print "@@ -$range1 +$range2 @@\n"; # Outlist starts containing the hunk of file 1. # Removing an item just means putting a '-' in front of it. # Inserting an item requires getting it from file2 and splicing it in. # We splice in $num_added items. Remove blocks use $num_added because # splicing changed the length of outlist. # We remove $num_removed items. Insert blocks use $num_removed because # their item numbers---corresponding to positions in file *2*--- don't take # removed items into account. my $low = $hunk->{"start1"}; my $hi = $hunk->{"end1"}; my ($num_added, $num_removed) = (0,0); my @outlist = @$fileref1[$low..$hi]; map {s/^/ /} @outlist; # assume it's just context foreach my $block (@{$hunk->{"blocks"}}) { foreach my $item ($block->remove) { my $op = $item->{"sign"}; # - my $offset = $item->{"item_no"} - $low + $num_added; $outlist[$offset] =~ s/^ /$op/; $num_removed++; } foreach my $item ($block->insert) { my $op = $item->{"sign"}; # + my $i = $item->{"item_no"}; my $offset = $i - $hunk->{"start2"} + $num_removed; splice(@outlist,$offset,0,"$op$$fileref2[$i]"); $num_added++; } } map {s/$/\n/} @outlist; # add \n's print @outlist; } sub output_context_diff { my ($hunk, $fileref1, $fileref2) = @_; my @blocklist; print "***************\n"; # Calculate item number range. my $range1 = $hunk->context_range(1); my $range2 = $hunk->context_range(2); # Print out file 1 part for each block in context diff format if there are # any blocks that remove items print "*** $range1 ****\n"; my $low = $hunk->{"start1"}; my $hi = $hunk->{"end1"}; if (@blocklist = grep {$_->remove} @{$hunk->{"blocks"}}) { my @outlist = @$fileref1[$low..$hi]; map {s/^/ /} @outlist; # assume it's just context foreach my $block (@blocklist) { my $op = $block->op; # - or ! foreach my $item ($block->remove) { $outlist[$item->{"item_no"} - $low] =~ s/^ /$op/; } } map {s/$/\n/} @outlist; # add \n's print @outlist; } print "--- $range2 ----\n"; $low = $hunk->{"start2"}; $hi = $hunk->{"end2"}; if (@blocklist = grep {$_->insert} @{$hunk->{"blocks"}}) { my @outlist = @$fileref2[$low..$hi]; map {s/^/ /} @outlist; # assume it's just context foreach my $block (@blocklist) { my $op = $block->op; # + or ! foreach my $item ($block->insert) { $outlist[$item->{"item_no"} - $low] =~ s/^ /$op/; } } map {s/$/\n/} @outlist; # add \n's print @outlist; } } sub store_ed_diff { # ed diff prints out diffs *backwards*. So save them while we're generating # them, then print them out at the end my $hunk = shift; unshift @Ed_Hunks, $hunk; } sub output_ed_diff { # This sub is used for ed ('diff -e') OR reverse_ed ('diff -f'). # last arg is type of diff my $diff_type = $_[-1]; my ($hunk, $fileref1, $fileref2) = @_; my %op_hash = ('+' => 'a', '-' => 'd', '!' => 'c'); # Can't be any context for this kind of diff, so each hunk has one block my @blocklist = @{$hunk->{"blocks"}}; warn ("Expecting one block in an ed diff hunk!") if scalar @blocklist != 1; my $block = $blocklist[0]; my $op = $block->op; # +, -, or ! # Calculate item number range. # old diff range is just like a context diff range, except the ranges # are on one line with the action between them. my $range1 = $hunk->context_range(1); $range1 =~ s/,/ / if $diff_type eq "REVERSE_ED"; my $action = $op_hash{$op} || warn "unknown op $op"; print ($diff_type eq "ED" ? "$range1$action\n" : "$action$range1\n"); if ($block->insert) { my @outlist = @$fileref2[$hunk->{"start2"}..$hunk->{"end2"}]; map {s/$/\n/} @outlist; # add \n's print @outlist; print ".\n"; # end of ed 'c' or 'a' command } } sub context_range { # Generate a range of item numbers to print. Only print 1 number if the range # has only one item in it. Otherwise, it's 'start,end' # Flag is the number of the file (1 or 2) my ($hunk, $flag) = @_; my ($start, $end) = ($hunk->{"start$flag"},$hunk->{"end$flag"}); $start++; $end++; # index from 1, not zero my $range = ($start < $end) ? "$start,$end" : $end; return $range; } sub unified_range { # Generate a range of item numbers to print for unified diff # Print number where block starts, followed by number of lines in the block # (don't print number of lines if it's 1) my ($hunk, $flag) = @_; my ($start, $end) = ($hunk->{"start$flag"},$hunk->{"end$flag"}); $start++; $end++; # index from 1, not zero my $length = $end - $start + 1; my $first = $length < 2 ? $end : $start; # strange, but correct... my $range = $length== 1 ? $first : "$first,$length"; return $range; } } # end Package Hunk ######## # Package Block. A block is an operation removing, adding, or changing # a group of items. Basically, this is just a list of changes, where each # change adds or deletes a single item. # (Change could be a separate class, but it didn't seem worth it) { package Block; sub new { # Input is a chunk from &Algorithm::LCS::diff # Fields in a block: # length_diff - how much longer file 2 is than file 1 due to this block # Each change has: # sign - '+' for insert, '-' for remove # item_no - number of the item in the file (e.g., line number) # We don't bother storing the text of the item # my ($class,$chunk) = @_; my @changes = (); # This just turns each change into a hash. foreach my $item (@$chunk) { my ($sign, $item_no, $text) = @$item; my $hashref = {"sign" => $sign, "item_no" => $item_no}; push @changes, $hashref; } my $block = { "changes" => \@changes }; bless $block, $class; $block->{"length_diff"} = $block->insert - $block->remove; return $block; } # LOW LEVEL FUNCTIONS sub op { # what kind of block is this? my $block = shift; my $insert = $block->insert; my $remove = $block->remove; $remove && $insert and return '!'; $remove and return '-'; $insert and return '+'; warn "unknown block type"; return '^'; # context block } # Returns a list of the changes in this block that remove items # (or the number of removals if called in scalar context) sub remove { return grep {$_->{"sign"} eq '-'} @{shift->{"changes"}}; } # Returns a list of the changes in this block that insert items sub insert { return grep {$_->{"sign"} eq '+'} @{shift->{"changes"}}; } } # end of package Block