shell bypass 403
UnknownSec Shell
:
/
usr
/
share
/
perl5
/
unicore
/ [
drwxr-xr-x
]
upload
mass deface
mass delete
console
info server
name :
Name.pm
# !!!!!!! DO NOT EDIT THIS FILE !!!!!!! # This file is machine-generated by lib/unicore/mktables from the Unicode # database, Version 9.0.0. Any changes made here will be lost! # !!!!!!! INTERNAL PERL USE ONLY !!!!!!! # This file is for internal use by core Perl only. The format and even the # name or existence of this file are subject to change without notice. Don't # use it directly. Use Unicode::UCD to access the Unicode character data # base. package charnames; # This module contains machine-generated tables and code for the # algorithmically-determinable Unicode character names. The following # routines can be used to translate between name and code point and vice versa { # Closure # Matches legal code point. 4-6 hex numbers, If there are 6, the first # two must be 10; if there are 5, the first must not be a 0. Written this # way to decrease backtracking. The first regex allows the code point to # be at the end of a word, but to work properly, the word shouldn't end # with a valid hex character. The second one won't match a code point at # the end of a word, and doesn't have the run-on issue my $run_on_code_point_re = qr/(?^aax: (?: 10[0-9A-F]{4} | [1-9A-F][0-9A-F]{4} | [0-9A-F]{4} ) \b)/; my $code_point_re = qr/(?^aa:\b(?^aax: (?: 10[0-9A-F]{4} | [1-9A-F][0-9A-F]{4} | [0-9A-F]{4} ) \b))/; # In the following hash, the keys are the bases of names which include # the code point in the name, like CJK UNIFIED IDEOGRAPH-4E01. The value # of each key is another hash which is used to get the low and high ends # for each range of code points that apply to the name. my %names_ending_in_code_point = ( 'CJK COMPATIBILITY IDEOGRAPH' => { 'high' => [ 64109, 64217, 195101, ], 'low' => [ 63744, 64112, 194560, ], }, 'CJK UNIFIED IDEOGRAPH' => { 'high' => [ 19893, 40917, 173782, 177972, 178205, 183969, ], 'low' => [ 13312, 19968, 131072, 173824, 177984, 178208, ], }, 'TANGUT IDEOGRAPH' => { 'high' => [ 100332, ], 'low' => [ 94208, ], }, ); # The following hash is a copy of the previous one, except is for loose # matching, so each name has blanks and dashes squeezed out my %loose_names_ending_in_code_point = ( 'CJKCOMPATIBILITYIDEOGRAPH' => { 'high' => [ 64109, 64217, 195101, ], 'low' => [ 63744, 64112, 194560, ], }, 'CJKUNIFIEDIDEOGRAPH' => { 'high' => [ 19893, 40917, 173782, 177972, 178205, 183969, ], 'low' => [ 13312, 19968, 131072, 173824, 177984, 178208, ], }, 'TANGUTIDEOGRAPH' => { 'high' => [ 100332, ], 'low' => [ 94208, ], }, ); # And the following array gives the inverse mapping from code points to # names. Lowest code points are first my @code_points_ending_in_code_point = ( { 'high' => 19893, 'low' => 13312, 'name' => 'CJK UNIFIED IDEOGRAPH', }, { 'high' => 40917, 'low' => 19968, 'name' => 'CJK UNIFIED IDEOGRAPH', }, { 'high' => 64109, 'low' => 63744, 'name' => 'CJK COMPATIBILITY IDEOGRAPH', }, { 'high' => 64217, 'low' => 64112, 'name' => 'CJK COMPATIBILITY IDEOGRAPH', }, { 'high' => 100332, 'low' => 94208, 'name' => 'TANGUT IDEOGRAPH', }, { 'high' => 173782, 'low' => 131072, 'name' => 'CJK UNIFIED IDEOGRAPH', }, { 'high' => 177972, 'low' => 173824, 'name' => 'CJK UNIFIED IDEOGRAPH', }, { 'high' => 178205, 'low' => 177984, 'name' => 'CJK UNIFIED IDEOGRAPH', }, { 'high' => 183969, 'low' => 178208, 'name' => 'CJK UNIFIED IDEOGRAPH', }, { 'high' => 195101, 'low' => 194560, 'name' => 'CJK COMPATIBILITY IDEOGRAPH', }, , ); # Convert from code point to Jamo short name for use in composing Hangul # syllable names my %Jamo = ( 4352 => 'G', 4353 => 'GG', 4354 => 'N', 4355 => 'D', 4356 => 'DD', 4357 => 'R', 4358 => 'M', 4359 => 'B', 4360 => 'BB', 4361 => 'S', 4362 => 'SS', 4363 => '', 4364 => 'J', 4365 => 'JJ', 4366 => 'C', 4367 => 'K', 4368 => 'T', 4369 => 'P', 4370 => 'H', 4449 => 'A', 4450 => 'AE', 4451 => 'YA', 4452 => 'YAE', 4453 => 'EO', 4454 => 'E', 4455 => 'YEO', 4456 => 'YE', 4457 => 'O', 4458 => 'WA', 4459 => 'WAE', 4460 => 'OE', 4461 => 'YO', 4462 => 'U', 4463 => 'WEO', 4464 => 'WE', 4465 => 'WI', 4466 => 'YU', 4467 => 'EU', 4468 => 'YI', 4469 => 'I', 4520 => 'G', 4521 => 'GG', 4522 => 'GS', 4523 => 'N', 4524 => 'NJ', 4525 => 'NH', 4526 => 'D', 4527 => 'L', 4528 => 'LG', 4529 => 'LM', 4530 => 'LB', 4531 => 'LS', 4532 => 'LT', 4533 => 'LP', 4534 => 'LH', 4535 => 'M', 4536 => 'B', 4537 => 'BS', 4538 => 'S', 4539 => 'SS', 4540 => 'NG', 4541 => 'J', 4542 => 'C', 4543 => 'K', 4544 => 'T', 4545 => 'P', 4546 => 'H', ); # Leading consonant (can be null) my %Jamo_L = ( '' => 11, 'B' => 7, 'BB' => 8, 'C' => 14, 'D' => 3, 'DD' => 4, 'G' => 0, 'GG' => 1, 'H' => 18, 'J' => 12, 'JJ' => 13, 'K' => 15, 'M' => 6, 'N' => 2, 'P' => 17, 'R' => 5, 'S' => 9, 'SS' => 10, 'T' => 16, ); # Vowel my %Jamo_V = ( 'A' => 0, 'AE' => 1, 'E' => 5, 'EO' => 4, 'EU' => 18, 'I' => 20, 'O' => 8, 'OE' => 11, 'U' => 13, 'WA' => 9, 'WAE' => 10, 'WE' => 15, 'WEO' => 14, 'WI' => 16, 'YA' => 2, 'YAE' => 3, 'YE' => 7, 'YEO' => 6, 'YI' => 19, 'YO' => 12, 'YU' => 17, ); # Optional trailing consonant my %Jamo_T = ( 'B' => 17, 'BS' => 18, 'C' => 23, 'D' => 7, 'G' => 1, 'GG' => 2, 'GS' => 3, 'H' => 27, 'J' => 22, 'K' => 24, 'L' => 8, 'LB' => 11, 'LG' => 9, 'LH' => 15, 'LM' => 10, 'LP' => 14, 'LS' => 12, 'LT' => 13, 'M' => 16, 'N' => 4, 'NG' => 21, 'NH' => 6, 'NJ' => 5, 'P' => 26, 'S' => 19, 'SS' => 20, 'T' => 25, ); # Computed re that splits up a Hangul name into LVT or LV syllables my $syllable_re = qr/(|B|BB|C|D|DD|G|GG|H|J|JJ|K|M|N|P|R|S|SS|T)(A|AE|E|EO|EU|I|O|OE|U|WA|WAE|WE|WEO|WI|YA|YAE|YE|YEO|YI|YO|YU)(B|BS|C|D|G|GG|GS|H|J|K|L|LB|LG|LH|LM|LP|LS|LT|M|N|NG|NH|NJ|P|S|SS|T)?/; my $HANGUL_SYLLABLE = "HANGUL SYLLABLE "; my $loose_HANGUL_SYLLABLE = "HANGULSYLLABLE"; # These constants names and values were taken from the Unicode standard, # version 5.1, section 3.12. They are used in conjunction with Hangul # syllables my $SBase = 0xAC00; my $LBase = 0x1100; my $VBase = 0x1161; my $TBase = 0x11A7; my $SCount = 11172; my $LCount = 19; my $VCount = 21; my $TCount = 28; my $NCount = $VCount * $TCount; sub name_to_code_point_special { my ($name, $loose) = @_; # Returns undef if not one of the specially handled names; otherwise # returns the code point equivalent to the input name # $loose is non-zero if to use loose matching, 'name' in that case # must be input as upper case with all blanks and dashes squeezed out. if ((! $loose && $name =~ s/$HANGUL_SYLLABLE//) || ($loose && $name =~ s/$loose_HANGUL_SYLLABLE//)) { return if $name !~ qr/^$syllable_re$/; my $L = $Jamo_L{$1}; my $V = $Jamo_V{$2}; my $T = (defined $3) ? $Jamo_T{$3} : 0; return ($L * $VCount + $V) * $TCount + $T + $SBase; } # Name must end in 'code_point' for this to handle. return if (($loose && $name !~ /^ (.*?) ($run_on_code_point_re) $/x) || (! $loose && $name !~ /^ (.*) ($code_point_re) $/x)); my $base = $1; my $code_point = CORE::hex $2; my $names_ref; if ($loose) { $names_ref = \%loose_names_ending_in_code_point; } else { return if $base !~ s/-$//; $names_ref = \%names_ending_in_code_point; } # Name must be one of the ones which has the code point in it. return if ! $names_ref->{$base}; # Look through the list of ranges that apply to this name to see if # the code point is in one of them. for (my $i = 0; $i < scalar @{$names_ref->{$base}{'low'}}; $i++) { return if $names_ref->{$base}{'low'}->[$i] > $code_point; next if $names_ref->{$base}{'high'}->[$i] < $code_point; # Here, the code point is in the range. return $code_point; } # Here, looked like the name had a code point number in it, but # did not match one of the valid ones. return; } sub code_point_to_name_special { my $code_point = shift; # Returns the name of a code point if algorithmically determinable; # undef if not # If in the Hangul range, calculate the name based on Unicode's # algorithm if ($code_point >= $SBase && $code_point <= $SBase + $SCount -1) { use integer; my $SIndex = $code_point - $SBase; my $L = $LBase + $SIndex / $NCount; my $V = $VBase + ($SIndex % $NCount) / $TCount; my $T = $TBase + $SIndex % $TCount; $name = "$HANGUL_SYLLABLE$Jamo{$L}$Jamo{$V}"; $name .= $Jamo{$T} if $T != $TBase; return $name; } # Look through list of these code points for one in range. foreach my $hash (@code_points_ending_in_code_point) { return if $code_point < $hash->{'low'}; if ($code_point <= $hash->{'high'}) { return sprintf("%s-%04X", $hash->{'name'}, $code_point); } } return; # None found } } # End closure 1;
© 2024 UnknownSec