Server IP : 85.214.239.14 / Your IP : 18.191.195.180 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 : /proc/3/root/usr/share/perl/5.36.0/ |
Upload File : |
# !!!!!!! INTERNAL PERL USE ONLY !!!!!!! # This helper module is for internal use by core Perl only. This module is # subject to change or removal at any time without notice. Don't use it # directly. Use the public <charnames> module instead. package _charnames; use strict; use warnings; our $VERSION = '1.50'; use unicore::Name; # mktables-generated algorithmically-defined names use bytes (); # for $bytes::hint_bits use re "/aa"; # Everything in here should be ASCII $Carp::Internal{ (__PACKAGE__) } = 1; # Translate between Unicode character names and their code points. This is a # submodule of package <charnames>, used to allow \N{...} to be autoloaded, # but it was decided not to autoload the various functions in charnames; the # splitting allows this behavior. # # The official names with their code points are stored in a table in # lib/unicore/Name.pl which is read in as a large string (almost 3/4 Mb in # Unicode 6.0). Each code point sequence appears on a line by itself, with # its corresponding name occupying the next line in the string. (Some of the # CJK and the Hangul syllable names are instead determined algorithmically via # subroutines stored instead in lib/unicore/Name.pm). Because of the large # size of this table, it isn't converted into hashes for faster lookup. # # But, user defined aliases are stored in their own hashes, as are Perl # extensions to the official names. These are checked first before looking at # the official table. # # Basically, the table is grepped for the input code point (viacode()) or # name (the other functions), and the corresponding value on the next or # previous line is returned. The grepping is done by turning the input into a # regular expression. Thus, the same table does double duty, used by both # name and code point lookup. (If we were to have hashes, we would need two, # one for each lookup direction.) # # For loose name matching, the logical thing would be to have a table # with all the ignorable characters squeezed out, and then grep it with the # similiarly-squeezed input name. (And this is in fact how the lookups are # done with the small Perl extension hashes.) But since we need to be able to # go from code point to official name, the original table would still need to # exist. Due to the large size of the table, it was decided to not read # another very large string into memory for a second table. Instead, the # regular expression of the input name is modified to have optional spaces and # dashes between characters. For example, in strict matching, the regular # expression would be: # qr/^DIGIT ONE$/m # Under loose matching, the blank would be squeezed out, and the re would be: # qr/^D[- ]?I[- ]?G[- ]?I[- ]?T[- ]?O[- ]?N[- ]?E$/m # which matches a blank or dash between any characters in the official table. # # This is also how script lookup is done. Basically the re looks like # qr/ (?:LATIN|GREEK|CYRILLIC) (?:SMALL )?LETTER $name/ # where $name is the loose or strict regex for the remainder of the name. # The hashes are stored as utf8 strings. This makes it easier to deal with # sequences. I (khw) also tried making Name.pl utf8, but it slowed things # down by a factor of 7. I then tried making Name.pl store the utf8 # equivalents but not calling them utf8. That led to similar speed as leaving # it alone, but since that is harder for a human to parse, I left it as-is. my %system_aliases = ( 'SINGLE-SHIFT 2' => chr utf8::unicode_to_native(0x8E), 'SINGLE-SHIFT 3' => chr utf8::unicode_to_native(0x8F), 'PRIVATE USE 1' => chr utf8::unicode_to_native(0x91), 'PRIVATE USE 2' => chr utf8::unicode_to_native(0x92), ); # These are the aliases above that differ under :loose and :full matching # because the :full versions have blanks or hyphens in them. #my %loose_system_aliases = ( #); #my %deprecated_aliases; #$deprecated_aliases{'BELL'} = chr utf8::unicode_to_native(0x07) if $^V lt v5.17.0; #my %loose_deprecated_aliases = ( #); # These are special cased in :loose matching, differing only in a medial # hyphen my $HANGUL_JUNGSEONG_O_E_utf8 = chr 0x1180; my $HANGUL_JUNGSEONG_OE_utf8 = chr 0x116C; my $txt; # The table of official character names my %full_names_cache; # Holds already-looked-up names, so don't have to # re-look them up again. The previous versions of charnames had scoping # bugs. For example if we use script A in one scope and find and cache # what Z resolves to, we can't use that cache in a different scope that # uses script B instead of A, as Z might be an entirely different letter # there; or there might be different aliases in effect in different # scopes, or :short may be in effect or not effect in different scopes, # or various combinations thereof. This was solved in this version # mostly by moving things to %^H. But some things couldn't be moved # there. One of them was the cache of runtime looked-up names, in part # because %^H is read-only at runtime. I (khw) don't know why the cache # was run-time only in the previous versions: perhaps oversight; perhaps # that compile time looking doesn't happen in a loop so didn't think it # was worthwhile; perhaps not wanting to make the cache too large. But # I decided to make it compile time as well; this could easily be # changed. # Anyway, this hash is not scoped, and is added to at runtime. It # doesn't have scoping problems because the data in it is restricted to # official names, which are always invariant, and we only set it and # look at it at during :full lookups, so is unaffected by any other # scoped options. I put this in to maintain parity with the older # version. If desired, a %short_names cache could also be made, as well # as one for each script, say in %script_names_cache, with each key # being a hash for a script named in a 'use charnames' statement. I # decided not to do that for now, just because it's added complication, # and because I'm just trying to maintain parity, not extend it. # Like %full_names_cache, but for use when :loose is in effect. There needs # to be two caches because :loose may not be in effect for a scope, and a # loose name could inappropriately be returned when only exact matching is # called for. my %loose_names_cache; # Designed so that test decimal first, and then hex. Leading zeros # imply non-decimal, as do non-[0-9] my $decimal_qr = qr/^[1-9]\d*$/; # Returns the hex number in $1. my $hex_qr = qr/^(?:[Uu]\+|0[xX])?([[:xdigit:]]+)$/; sub croak { require Carp; goto &Carp::croak; } # croak sub carp { require Carp; goto &Carp::carp; } # carp sub populate_txt() { return if $txt; $txt = do "unicore/Name.pl"; Internals::SvREADONLY($txt, 1); } sub alias (@) # Set up a single alias { my @errors; my $nbsp = chr utf8::unicode_to_native(0xA0); my $alias = ref $_[0] ? $_[0] : { @_ }; foreach my $name (sort keys %$alias) { # Sort only because it helps having # deterministic output for # t/lib/charnames/alias my $value = $alias->{$name}; next unless defined $value; # Omit if screwed up. # Is slightly slower to just after this statement see if it is # decimal, since we already know it is after having converted from # hex, but makes the code easier to maintain, and is called # infrequently, only at compile-time if ($value !~ $decimal_qr && $value =~ $hex_qr) { my $temp = CORE::hex $1; $temp = utf8::unicode_to_native($temp) if $value =~ /^[Uu]\+/; $value = $temp; } if ($value =~ $decimal_qr) { no warnings qw(non_unicode surrogate nonchar); # Allow any of these $^H{charnames_ord_aliases}{$name} = chr $value; # Use a canonical form. $^H{charnames_inverse_ords}{sprintf("%05X", $value)} = $name; } else { my $ok_portion = ""; $ok_portion = $1 if $name =~ / ^ ( \p{_Perl_Charname_Begin} \p{_Perl_Charname_Continue}* ) /x; # If the name was fully correct, the above should have matched all of # it. if (length $ok_portion < length $name) { my $first_bad = substr($name, length($ok_portion), 1); push @errors, "Invalid character in charnames alias definition; " . "marked by <-- HERE in '$ok_portion$first_bad<-- HERE " . substr($name, length($ok_portion) + 1) . "'"; } else { if ($name =~ / ( .* \s ) ( \s* ) $ /x) { push @errors, "charnames alias definitions may not contain " . "trailing white-space; marked by <-- HERE in " . "'$1 <-- HERE " . $2 . "'"; next; } # Use '+' instead of '*' in this regex, because any trailing # blanks have already been found if ($name =~ / ( .*? \s{2} ) ( .+ ) /x) { push @errors, "charnames alias definitions may not contain a " . "sequence of multiple spaces; marked by <-- HERE " . "in '$1 <-- HERE " . $2 . "'"; next; } $^H{charnames_name_aliases}{$name} = $value; } } } # We find and output all errors from this :alias definition, rather than # failing on the first one, so fewer runs are needed to get it to compile if (@errors) { croak join "\n", @errors; } return; } # alias sub not_legal_use_bytes_msg { my ($name, $utf8) = @_; my $return; if (length($utf8) == 1) { $return = sprintf("Character 0x%04x with name '%s' is", ord $utf8, $name); } else { $return = sprintf("String with name '%s' (and ordinals %s) contains character(s)", $name, join(" ", map { sprintf "0x%04X", ord $_ } split(//, $utf8))); } return $return . " above 0xFF with 'use bytes' in effect"; } sub alias_file ($) # Reads a file containing alias definitions { require File::Spec; my ($arg, $file) = @_; if (-f $arg && File::Spec->file_name_is_absolute ($arg)) { $file = $arg; } elsif ($arg =~ m/ ^ \p{_Perl_IDStart} \p{_Perl_IDCont}* $/x) { $file = "unicore/${arg}_alias.pl"; } else { croak "Charnames alias file names can only have identifier characters"; } if (my @alias = do $file) { @alias == 1 && !defined $alias[0] and croak "$file cannot be used as alias file for charnames"; @alias % 2 and croak "$file did not return a (valid) list of alias pairs"; alias (@alias); return (1); } 0; } # alias_file # For use when don't import anything. This structure must be kept in # sync with the one that import() fills up. my %dummy_H = ( charnames_stringified_names => "", charnames_stringified_ords => "", charnames_scripts => "", charnames_full => 1, charnames_loose => 0, charnames_short => 0, ); sub lookup_name ($$$;$) { my ($name, $wants_ord, $runtime, $regex_loose) = @_; $regex_loose //= 0; # Lookup the name or sequence $name in the tables. If $wants_ord is false, # returns the string equivalent of $name; if true, returns the ordinal value # instead, but in this case $name must not be a sequence; otherwise undef is # returned and a warning raised. $runtime is 0 if compiletime, otherwise # gives the number of stack frames to go back to get the application caller # info. # If $name is not found, returns undef in runtime with no warning; and in # compiletime, the Unicode replacement character, with a warning. # It looks first in the aliases, then in the large table of official Unicode # names. my $result; # The string result my $save_input; if ($runtime && ! $regex_loose) { my $hints_ref = (caller($runtime))[10]; # If we didn't import anything (which happens with 'use charnames ()', # substitute a dummy structure. $hints_ref = \%dummy_H if ! defined $hints_ref || (! defined $hints_ref->{charnames_full} && ! defined $hints_ref->{charnames_loose}); # At runtime, but currently not at compile time, %^H gets # stringified, so un-stringify back to the original data structures. # These get thrown away by perl before the next invocation # Also fill in the hash with the non-stringified data. # N.B. New fields must be also added to %dummy_H %{$^H{charnames_name_aliases}} = split ',', $hints_ref->{charnames_stringified_names}; %{$^H{charnames_ord_aliases}} = split ',', $hints_ref->{charnames_stringified_ords}; $^H{charnames_scripts} = $hints_ref->{charnames_scripts}; $^H{charnames_full} = $hints_ref->{charnames_full}; $^H{charnames_loose} = $hints_ref->{charnames_loose}; $^H{charnames_short} = $hints_ref->{charnames_short}; } my $loose = $regex_loose || $^H{charnames_loose}; my $lookup_name; # Input name suitably modified for grepping for in the # table # User alias should be checked first or else can't override ours, and if we # were to add any, could conflict with theirs. if (! $regex_loose && exists $^H{charnames_ord_aliases}{$name}) { $result = $^H{charnames_ord_aliases}{$name}; } elsif (! $regex_loose && exists $^H{charnames_name_aliases}{$name}) { $name = $^H{charnames_name_aliases}{$name}; $save_input = $lookup_name = $name; # Cache the result for any error # message # The aliases are documented to not match loosely, so change loose match # into full. if ($loose) { $loose = 0; $^H{charnames_full} = 1; } } else { # Here, not a user alias. That means that loose matching may be in # effect; will have to modify the input name. $lookup_name = $name; if ($loose) { $lookup_name = uc $lookup_name; # Squeeze out all underscores $lookup_name =~ s/_//g; # Remove all medial hyphens $lookup_name =~ s/ (?<= \S ) - (?= \S )//gx; # Squeeze out all spaces $lookup_name =~ s/\s//g; } # Here, $lookup_name has been modified as necessary for looking in the # hashes. Check the system alias files next. Most of these aliases are # the same for both strict and loose matching. To save space, the ones # which differ are in their own separate hash, which is checked if loose # matching is selected and the regular match fails. To save time, the # loose hashes could be expanded to include all aliases, and there would # only have to be one check. But if someone specifies :loose, they are # interested in convenience over speed, and the time for this second check # is miniscule compared to the rest of the routine. if (exists $system_aliases{$lookup_name}) { $result = $system_aliases{$lookup_name}; } # There are currently no entries in this hash, so don't waste time looking # for them. But the code is retained for the unlikely possibility that # some will be added in the future. # elsif ($loose && exists $loose_system_aliases{$lookup_name}) { # $result = $loose_system_aliases{$lookup_name}; # } # if (exists $deprecated_aliases{$lookup_name}) { # require warnings; # warnings::warnif('deprecated', # "Unicode character name \"$name\" is deprecated, use \"" # . viacode(ord $deprecated_aliases{$lookup_name}) # . "\" instead"); # $result = $deprecated_aliases{$lookup_name}; # } # There are currently no entries in this hash, so don't waste time looking # for them. But the code is retained for the unlikely possibility that # some will be added in the future. # elsif ($loose && exists $loose_deprecated_aliases{$lookup_name}) { # require warnings; # warnings::warnif('deprecated', # "Unicode character name \"$name\" is deprecated, use \"" # . viacode(ord $loose_deprecated_aliases{$lookup_name}) # . "\" instead"); # $result = $loose_deprecated_aliases{$lookup_name}; # } } my @off; # Offsets into table of pattern match begin and end # If haven't found it yet... if (! defined $result) { # See if has looked this input up earlier. if (! $loose && $^H{charnames_full} && exists $full_names_cache{$name}) { $result = $full_names_cache{$name}; } elsif ($loose && exists $loose_names_cache{$name}) { $result = $loose_names_cache{$name}; } else { # Here, must do a look-up # If full or loose matching succeeded, points to where to cache the # result my $cache_ref; ## Suck in the code/name list as a big string. ## Entries look like: ## "00052\nLATIN CAPITAL LETTER R\n\n" # or # "0052 0303\nLATIN CAPITAL LETTER R WITH TILDE\n\n" populate_txt() unless $txt; ## @off will hold the index into the code/name string of the start and ## end of the name as we find it. ## If :loose, look for a loose match; if :full, look for the name ## exactly # First, see if the name is one which is algorithmically determinable. # The subroutine is included in Name.pl. The table contained in # $txt doesn't contain these. Experiments show that checking # for these before checking for the regular names has no # noticeable impact on performance for the regular names, but # the other way around slows down finding these immensely. # Algorithmically determinables are not placed in the cache because # that uses up memory, and finding these again is fast. if ( ($loose || $^H{charnames_full}) && (defined (my $ord = charnames::name_to_code_point_special($lookup_name, $loose)))) { $result = chr $ord; } else { # Not algorithmically determinable; look up in the table. The name # will be turned into a regex, so quote any meta characters. $lookup_name = quotemeta $lookup_name; if ($loose) { # For loose matches, $lookup_name has already squeezed out the # non-essential characters. We have to add in code to make the # squeezed version match the non-squeezed equivalent in the table. # The only remaining hyphens are ones that start or end a word in # the original. They have been quoted in $lookup_name so they look # like "\-". Change all other characters except the backslash # quotes for any metacharacters, and the final character, so that # e.g., COLON gets transformed into: /C[- ]?O[- ]?L[- ]?O[- ]?N/ $lookup_name =~ s/ (?! \\ -) # Don't do this to the \- sequence ( [^-\\] ) # Nor the "-" within that sequence, # nor the "\" that quotes metachars, # but otherwise put the char into $1 (?=.) # And don't do it for the final char /$1\[- \]?/gx; # And add an optional blank or # '-' after each $1 char # Those remaining hyphens were originally at the beginning or end of # a word, so they can match either a blank before or after, but not # both. (Keep in mind that they have been quoted, so are a '\-' # sequence) $lookup_name =~ s/\\ -/(?:- | -)/xg; } # Do the lookup in the full table if asked for, and if succeeds # save the offsets and set where to cache the result. if (($loose || $^H{charnames_full}) && $txt =~ /^$lookup_name$/m) { @off = ($-[0], $+[0]); $cache_ref = ($loose) ? \%loose_names_cache : \%full_names_cache; } elsif ($regex_loose) { # Currently don't allow :short when this is set return; } else { # Here, didn't look for, or didn't find the name. # If :short is allowed, see if input is like "greek:Sigma". # Keep in mind that $lookup_name has had the metas quoted. my $scripts_trie = ""; my $name_has_uppercase; my @scripts; if (($^H{charnames_short}) && $lookup_name =~ /^ (?: \\ \s)* # Quoted space (.+?) # $1 = the script (?: \\ \s)* \\ : # Quoted colon (?: \\ \s)* (.+?) # $2 = the name (?: \\ \s)* $ /xs) { # Even in non-loose matching, the script traditionally has been # case insensitive $scripts_trie = "\U$1"; $lookup_name = $2; # Use original name to find its input casing, but ignore the # script part of that to make the determination. $save_input = $name if ! defined $save_input; $name =~ s/.*?://; $name_has_uppercase = $name =~ /[[:upper:]]/; } else { # Otherwise look in allowed scripts # We want to search first by script name then by letter name, so that # if the user imported `use charnames qw(arabic hebrew)` and asked for # \N{alef} they get ARABIC LETTER ALEF, and if they imported # `... (hebrew arabic)` and ask for \N{alef} they get HEBREW LETTER ALEF. # We can't rely on the regex engine to preserve ordering like that, so # pick the pipe-seperated string apart so we can iterate over it. @scripts = split(/\|/, $^H{charnames_scripts}); # Use original name to find its input casing $name_has_uppercase = $name =~ /[[:upper:]]/; } my $case = $name_has_uppercase ? "CAPITAL" : "SMALL"; if(@scripts) { SCRIPTS: foreach my $script (@scripts) { if($txt =~ /^ (?: $script ) \ (?:$case\ )? LETTER \ \U$lookup_name $/xm) { @off = ($-[0], $+[0]); last SCRIPTS; } } return unless(@off); } else { return if (! $scripts_trie || $txt !~ /^ (?: $scripts_trie ) \ (?:$case\ )? LETTER \ \U$lookup_name $/xm); @off = ($-[0], $+[0]); } } # Here, the input name has been found; we haven't set up the output, # but we know where in the string # the name starts. The string is set up so that for single characters # (and not named sequences), the name is on a line by itself, and the # previous line contains precisely 5 hex digits for its code point. # Named sequences won't have the 7th preceding character be a \n. # (Actually, for the very first entry in the table this isn't strictly # true: subtracting 7 will yield -1, and the substr below will # therefore yield the very last character in the table, which should # also be a \n, so the statement works anyway.) if (substr($txt, $off[0] - 7, 1) eq "\n") { $result = chr CORE::hex substr($txt, $off[0] - 6, 5); # Handle the single loose matching special case, in which two names # differ only by a single medial hyphen. If the original had a # hyphen (or more) in the right place, then it is that one. $result = $HANGUL_JUNGSEONG_O_E_utf8 if $loose && $result eq $HANGUL_JUNGSEONG_OE_utf8 && $name =~ m/O \s* - [-\s]* E/ix; # Note that this wouldn't work if there were a 2nd # OE in the name } else { # Here, is a named sequence. Need to go looking for the beginning, # which is just after the \n from the previous entry in the table. # The +1 skips past that newline, or, if the rindex() fails, to put # us to an offset of zero. my $charstart = rindex($txt, "\n", $off[0] - 7) + 1; $result = pack("W*", map { CORE::hex } split " ", substr($txt, $charstart, $off[0] - $charstart - 1)); } } # Cache the input so as to not have to search the large table # again, but only if it came from the one search that we cache. # (Haven't bothered with the pain of sorting out scoping issues for the # scripts searches.) $cache_ref->{$name} = $result if defined $cache_ref; } } # Here, have the result character. If the return is to be an ord, must be # any single character. if ($wants_ord) { return ord($result) if length $result == 1; } elsif (! utf8::is_utf8($result)) { # Here isn't UTF-8. That's OK if it is all ASCII, or we are being called # at compile time where we know we can guarantee that Unicode rules are # correctly imposed on the result, or under 'bytes' where we don't want # those rules. But otherwise we have to make it UTF8 to guarantee Unicode # rules on the returned string. return $result if ! $runtime || (caller $runtime)[8] & $bytes::hint_bits || $result !~ /[[:^ascii:]]/; utf8::upgrade($result); return $result; } else { # Here, wants string output. If utf8 is acceptable, just return what # we've got; otherwise attempt to convert it to non-utf8 and return that. my $in_bytes = ! $regex_loose # \p{name=} doesn't currently care if # in bytes or not && (($runtime) ? (caller $runtime)[8] & $bytes::hint_bits : $^H & $bytes::hint_bits); return $result if (! $in_bytes || utf8::downgrade($result, 1)) # The 1 arg # means don't die on failure } # Here, there is an error: either there are too many characters, or the # result string needs to be non-utf8, and at least one character requires # utf8. Prefer any official name over the input one for the error message. if (@off) { $name = substr($txt, $off[0], $off[1] - $off[0]) if @off; } else { $name = (defined $save_input) ? $save_input : $_[0]; } if ($wants_ord) { # Only way to get here in this case is if result too long. Message # assumes that our only caller that requires single char result is # vianame. carp "charnames::vianame() doesn't handle named sequences ($name). Use charnames::string_vianame() instead"; return; } # Only other possible failure here is from use bytes. if ($runtime) { carp not_legal_use_bytes_msg($name, $result); return; } else { croak not_legal_use_bytes_msg($name, $result); } } # lookup_name sub charnames { # For \N{...}. Looks up the character name and returns the string # representation of it. # The first 0 arg means wants a string returned; the second that we are in # compile time return lookup_name($_[0], 0, 0); } sub _loose_regcomp_lookup { # For use only by regcomp.c to compile \p{name=...} # khw thinks it best to not do :short matching, and only official names. # But that is only a guess, and if demand warrants, could be changed return lookup_name($_[0], 0, 1, 1 # Always use :loose matching ); } sub _get_names_info { # For use only by regcomp.c to compile \p{name=/.../} populate_txt() unless $txt; return ( \$txt, \@charnames::code_points_ending_in_code_point ); } sub import { shift; ## ignore class name populate_txt() unless $txt; if (not @_) { carp("'use charnames' needs explicit imports list"); } $^H{charnames} = \&charnames ; $^H{charnames_ord_aliases} = {}; $^H{charnames_name_aliases} = {}; $^H{charnames_inverse_ords} = {}; # New fields must be added to %dummy_H, and the code in lookup_name() # that copies fields from the runtime structure ## ## fill %h keys with our @_ args. ## my ($promote, %h, @args) = (0); while (my $arg = shift) { if ($arg eq ":alias") { @_ or croak ":alias needs an argument in charnames"; my $alias = shift; if (ref $alias) { ref $alias eq "HASH" or croak "Only HASH reference supported as argument to :alias"; alias ($alias); $promote = 1; next; } if ($alias =~ m{:(\w+)$}) { $1 eq "full" || $1 eq "loose" || $1 eq "short" and croak ":alias cannot use existing pragma :$1 (reversed order?)"; alias_file ($1) and $promote = 1; next; } alias_file ($alias) and $promote = 1; next; } if (substr($arg, 0, 1) eq ':' and ! ($arg eq ":full" || $arg eq ":short" || $arg eq ":loose")) { warn "unsupported special '$arg' in charnames"; next; } push @args, $arg; } @args == 0 && $promote and @args = (":full"); @h{@args} = (1) x @args; # Don't leave these undefined as are tested for in lookup_names $^H{charnames_full} = delete $h{':full'} || 0; $^H{charnames_loose} = delete $h{':loose'} || 0; $^H{charnames_short} = delete $h{':short'} || 0; my @scripts = map { uc quotemeta } grep { /^[^:]/ } @args; ## ## If utf8? warnings are enabled, and some scripts were given, ## see if at least we can find one letter from each script. ## if (warnings::enabled('utf8') && @scripts) { for my $script (@scripts) { if (not $txt =~ m/^$script (?:CAPITAL |SMALL )?LETTER /m) { warnings::warn('utf8', "No such script: '$script'"); $script = quotemeta $script; # Escape it, for use in the re. } } } # %^H gets stringified, so serialize it ourselves so can extract the # real data back later. $^H{charnames_stringified_ords} = join ",", %{$^H{charnames_ord_aliases}}; $^H{charnames_stringified_names} = join ",", %{$^H{charnames_name_aliases}}; $^H{charnames_stringified_inverse_ords} = join ",", %{$^H{charnames_inverse_ords}}; # Modify the input script names for loose name matching if that is also # specified, similar to the way the base character name is prepared. They # don't (currently, and hopefully never will) have dashes. These go into a # regex, and have already been uppercased and quotemeta'd. Squeeze out all # input underscores, blanks, and dashes. Then convert so will match a blank # between any characters. if ($^H{charnames_loose}) { for (my $i = 0; $i < @scripts; $i++) { $scripts[$i] =~ s/[_ -]//g; $scripts[$i] =~ s/ ( [^\\] ) (?= . ) /$1\\ ?/gx; } } my %letters_by_script = map { $_ => [ ($txt =~ m/$_(?: (?:small|capital))? letter (.*)/ig) ] } @scripts; SCRIPTS: foreach my $this_script (@scripts) { my @other_scripts = grep { $_ ne $this_script } @scripts; my @this_script_letters = @{$letters_by_script{$this_script}}; my @other_script_letters = map { @{$letters_by_script{$_}} } @other_scripts; foreach my $this_letter (@this_script_letters) { if(grep { $_ eq $this_letter } @other_script_letters) { warn "charnames: some short character names may clash in [".join(', ', sort @scripts)."], for example $this_letter\n"; last SCRIPTS; } } } $^H{charnames_scripts} = join "|", @scripts; # Stringifiy them as a trie } # import # Cache of already looked-up values. This is set to only contain # official values, and user aliases can't override them, so scoping is # not an issue. my %viacode; my $no_name_code_points_re = join "|", map { sprintf("%05X", utf8::unicode_to_native($_)) } 0x80, 0x81, 0x84, 0x99; $no_name_code_points_re = qr/$no_name_code_points_re/; sub viacode { # Returns the name of the code point argument if (@_ != 1) { carp "charnames::viacode() expects one argument"; return; } my $arg = shift; # This is derived from Unicode::UCD, where it is nearly the same as the # function _getcode(), but here it makes sure that even a hex argument # has the proper number of leading zeros, which is critical in # matching against $txt below # Must check if decimal first; see comments at that definition my $hex; if ($arg =~ $decimal_qr) { $hex = sprintf "%05X", $arg; } elsif ($arg =~ $hex_qr) { $hex = CORE::hex $1; $hex = utf8::unicode_to_native($hex) if $arg =~ /^[Uu]\+/; # Below is the line that differs from the _getcode() source $hex = sprintf "%05X", $hex; } else { carp("unexpected arg \"$arg\" to charnames::viacode()"); return; } return $viacode{$hex} if exists $viacode{$hex}; my $return; # If the code point is above the max in the table, there's no point # looking through it. Checking the length first is slightly faster if (length($hex) <= 5 || CORE::hex($hex) <= 0x10FFFF) { populate_txt() unless $txt; # See if the name is algorithmically determinable. my $algorithmic = charnames::code_point_to_name_special(CORE::hex $hex); if (defined $algorithmic) { $viacode{$hex} = $algorithmic; return $algorithmic; } # Return the official name, if exists. It's unclear to me (khw) at # this juncture if it is better to return a user-defined override, so # leaving it as is for now. if ($txt =~ m/^$hex\n/m) { # The name starts with the next character and goes up to the # next new-line. Using capturing parentheses above instead of # @+ more than doubles the execution time in Perl 5.13 $return = substr($txt, $+[0], index($txt, "\n", $+[0]) - $+[0]); # If not one of these 4 code points, return what we've found. if ($hex !~ / ^ $no_name_code_points_re $ /x) { $viacode{$hex} = $return; return $return; } # For backwards compatibility, we don't return the official name of # the 4 code points if there are user-defined aliases for them -- so # continue looking. } } # See if there is a user name for it, before giving up completely. # First get the scoped aliases, give up if have none. my $H_ref = (caller(1))[10]; return if ! defined $return && (! defined $H_ref || ! exists $H_ref->{charnames_stringified_inverse_ords}); my %code_point_aliases; if (defined $H_ref->{charnames_stringified_inverse_ords}) { %code_point_aliases = split ',', $H_ref->{charnames_stringified_inverse_ords}; return $code_point_aliases{$hex} if exists $code_point_aliases{$hex}; } # Here there is no user-defined alias, return any official one. return $return if defined $return; if (CORE::hex($hex) > 0x10FFFF && warnings::enabled('non_unicode')) { carp "Unicode characters only allocated up to U+10FFFF (you asked for U+$hex)"; } return; } # viacode 1; # ex: set ts=8 sts=2 sw=2 et: