4 use re "/aa"; # So we won't even try to look at above Latin1, potentially
5 # resulting in a recursive call
14 sub croak { require Carp; Carp::croak(@_) }
16 # Digits may be separated by a single underscore
17 my $digits = qr/ ( [0-9] _? )+ (?!:_) /x;
19 # A sign can be surrounded by white space
20 my $sign = qr/ \s* [+-]? \s* /x;
22 my $f_float = qr/ $sign $digits+ \. $digits* # e.g., 5.0, 5.
23 | $sign $digits* \. $digits+/x; # 0.7, .7
25 # A number may be an integer, a rational, or a float with an optional exponent
26 # We (shudder) accept a signed denominator
27 my $number = qr{ ^ $sign $digits+ $
28 | ^ $sign $digits+ \/ $sign $digits+ $
29 | ^ $f_float (?: [Ee] [+-]? $digits )? $}x;
32 # Given a lowercase property or property-value name, return its
33 # standardized version that is expected for look-up in the 'loose' hashes
34 # in Heavy.pl (hence, this depends on what mktables does). This squeezes
35 # out blanks, underscores and dashes. The complication stems from the
36 # grandfathered-in 'L_', which retains a single trailing underscore.
38 (my $loose = $_[0]) =~ s/[-_ \t]//g;
40 return $loose if $loose !~ / ^ (?: is | to )? l $/x;
41 return 'l_' if $_[0] =~ / l .* _ /x; # If original had a trailing '_'
46 ## "SWASH" == "SWATCH HASH". A "swatch" is a swatch of the Unicode landscape.
47 ## It's a data structure that encodes a set of Unicode characters.
51 # If a floating point number is within this distance from the value of a
52 # fraction, it is considered to be that fraction, even if many more digits
53 # are specified that don't exactly match.
54 my $min_floating_slop;
56 # To guard against this program calling something that in turn ends up
57 # calling this program with the same inputs, and hence infinitely
58 # recursing, we keep a stack of the properties that are currently in
59 # progress, pushed upon entry, popped upon return.
63 my ($class, $type, $list, $minbits, $none) = @_;
67 $class = "" unless defined $class;
68 print STDERR __LINE__, ": class=$class, type=$type, list=",
69 (defined $list) ? $list : ':undef:',
70 ", minbits=$minbits, none=$none\n" if DEBUG;
73 ## Get the list of codepoints for the type.
74 ## Called from swash_init (see utf8.c) or SWASHNEW itself.
76 ## Callers of swash_init:
77 ## op.c:pmtrans -- for tr/// and y///
78 ## Unicode::UCD::prop_invlist
79 ## Unicode::UCD::prop_invmap
81 ## Given a $type, our goal is to fill $list with the set of codepoint
82 ## ranges. If $type is false, $list passed is used.
85 ## For binary properties, $minbits must be 1.
86 ## For character mappings (case and transliteration), $minbits must
87 ## be a number except 1.
89 ## $list (or that filled according to $type):
90 ## Refer to perlunicode.pod, "User-Defined Character Properties."
92 ## For binary properties, only characters with the property value
93 ## of True should be listed. The 3rd column, if any, will be ignored
95 ## $none is undocumented, so I'm (khw) trying to do some documentation
96 ## of it now. It appears to be if there is a mapping in an input file
97 ## that maps to 'XXXX', then that is replaced by $none+1, expressed in
98 ## hexadecimal. It is used somehow in tr///.
100 ## To make the parsing of $type clear, this code takes the a rather
101 ## unorthodox approach of last'ing out of the block once we have the
102 ## info we need. Were this to be a subroutine, the 'last' would just
105 # If a problem is found $type is returned;
106 # Upon success, a new (or cached) blessed object is returned with
107 # keys TYPE, BITS, EXTRAS, LIST, and NONE with values having the
108 # same meanings as the input parameters.
109 # SPECIALS contains a reference to any special-treatment hash in the
111 # INVERT_IT is non-zero if the result should be inverted before use
112 # USER_DEFINED is non-zero if the result came from a user-defined
113 my $file; ## file to load data from, and also part of the %Cache key.
115 # Change this to get a different set of Unicode tables
116 my $unicore_dir = 'unicore';
118 my $list_is_from_mktables = 0; # Is $list returned from a mktables
119 # generated file? If so, we know it's
124 # Verify that this isn't a recursive call for this property.
125 # Can't use croak, as it may try to recurse to here itself.
126 my $class_type = $class . "::$type";
127 if (grep { $_ eq $class_type } @recursed) {
128 CORE::die "panic: Infinite recursion in SWASHNEW for '$type'\n";
130 push @recursed, $class_type;
135 # regcomp.c surrounds the property name with '__" and '_i' if this
136 # is to be caseless matching.
137 my $caseless = $type =~ s/^(.*)__(.*)_i$/$1$2/;
139 print STDERR __LINE__, ": type=$type, caseless=$caseless\n" if DEBUG;
144 ## It could be a user-defined property. Look in current
145 ## package if no package given
149 my $caller0 = caller(0);
150 my $caller1 = $type =~ s/(.+):://
156 if (defined $caller1 && $type =~ /^I[ns]\w+$/) {
157 my $prop = "${caller1}::$type";
158 if (exists &{$prop}) {
159 # stolen from Scalar::Util::PP::tainted()
162 local($@, $SIG{__DIE__}, $SIG{__WARN__});
165 eval { kill 0 * $prop };
166 $tainted = 1 if $@ =~ /^Insecure/;
168 die "Insecure user-defined property \\p{$prop}\n"
171 $list = &{$prop}($caseless);
177 # During Perl's compilation, this routine may be called before
178 # the tables are constructed. If so, we have a chicken/egg
179 # problem. If we die, the tables never get constructed, so
180 # keep going, but return an empty table so only what the code
181 # has compiled in internally (currently ASCII/Latin1 range
182 # matching) will work.
184 # Poor man's constant, to avoid a run-time check.
186 = \! defined &DynaLoader::boot_DynaLoader;
189 eval "require '$unicore_dir/Heavy.pl'";
191 print STDERR __LINE__, ": '$@'\n" if DEBUG;
192 pop @recursed if @recursed;
197 require "$unicore_dir/Heavy.pl";
199 BEGIN { delete $utf8::{miniperl} }
201 # All property names are matched caselessly
202 my $property_and_table = CORE::lc $type;
203 print STDERR __LINE__, ": $property_and_table\n" if DEBUG;
205 # See if is of the compound form 'property=value', where the
206 # value indicates the table we should use.
207 my ($property, $table, @remainder) =
208 split /\s*[:=]\s*/, $property_and_table, -1;
210 pop @recursed if @recursed;
215 if (! defined $table) {
217 # Here, is the single form. The property becomes empty, and
218 # the whole value is the table.
220 $prefix = $property = "";
222 print STDERR __LINE__, ": $property\n" if DEBUG;
224 # Here it is the compound property=table form. The property
225 # name is always loosely matched, and always can have an
226 # optional 'is' prefix (which isn't true in the single
228 $property = _loose_name($property) =~ s/^is//r;
230 # And convert to canonical form. Quit if not valid.
231 $property = $utf8::loose_property_name_of{$property};
232 if (! defined $property) {
233 pop @recursed if @recursed;
237 $prefix = "$property=";
239 # If the rhs looks like it is a number...
240 print STDERR __LINE__, ": table=$table\n" if DEBUG;
242 if ($table =~ $number) {
243 print STDERR __LINE__, ": table=$table\n" if DEBUG;
245 # Split on slash, in case it is a rational, like \p{1/5}
246 my @parts = split m{ \s* / \s* }x, $table, -1;
247 print __LINE__, ": $type\n" if @parts > 2 && DEBUG;
249 foreach my $part (@parts) {
250 print __LINE__, ": part=$part\n" if DEBUG;
252 $part =~ s/^\+\s*//; # Remove leading plus
253 $part =~ s/^-\s*/-/; # Remove blanks after unary
256 # Remove underscores between digits.
257 $part =~ s/(?<= [0-9] ) _ (?= [0-9] ) //xg;
259 # No leading zeros (but don't make a single '0'
260 # into a null string)
261 $part =~ s/ ^ ( -? ) 0+ /$1/x;
262 $part .= '0' if $part eq '-' || $part eq "";
264 # No trailing zeros after a decimal point
265 $part =~ s/ ( \. [0-9]*? ) 0+ $ /$1/x;
267 # Begin with a 0 if a leading decimal point
268 $part =~ s/ ^ ( -? ) \. /${1}0./x;
270 # Ensure not a trailing decimal point: turn into an
272 $part =~ s/ \. $ //x;
274 print STDERR __LINE__, ": part=$part\n" if DEBUG;
275 #return $type if $part eq "";
281 # If denominator is negative, get rid of it, and ...
282 if ($parts[1] =~ s/^-//) {
284 # If numerator is also negative, convert the
285 # whole thing to positive, else move the minus
287 if ($parts[0] !~ s/^-//) {
288 $parts[0] = '-' . $parts[0];
291 $table = join '/', @parts;
293 elsif ($property ne 'nv' || $parts[0] !~ /\./) {
295 # Here is not numeric value, or doesn't have a
296 # decimal point. No further manipulation is
297 # necessary. (Note the hard-coded property name.
298 # This could fail if other properties eventually
299 # had fractions as well; perhaps the cjk ones
300 # could evolve to do that. This hard-coding could
301 # be fixed by mktables generating a list of
302 # properties that could have fractions.)
306 # Here is a floating point numeric_value. Convert
307 # to rational. Get a normalized form, like
308 # 5.00E-01, and look that up in the hash
310 my $float = sprintf "%.*e",
314 if (exists $utf8::nv_floating_to_rational{$float}) {
315 $table = $utf8::nv_floating_to_rational{$float};
317 pop @recursed if @recursed;
321 print STDERR __LINE__, ": $property=$table\n" if DEBUG;
325 # Combine lhs (if any) and rhs to get something that matches
326 # the syntax of the lookups.
327 $property_and_table = "$prefix$table";
328 print STDERR __LINE__, ": $property_and_table\n" if DEBUG;
330 # First try stricter matching.
331 $file = $utf8::stricter_to_file_of{$property_and_table};
333 # If didn't find it, try again with looser matching by editing
334 # out the applicable characters on the rhs and looking up
336 my $strict_property_and_table;
337 if (! defined $file) {
339 # This isn't used unless the name begins with 'to'
340 $strict_property_and_table = $property_and_table =~ s/^to//r;
341 $table = _loose_name($table);
342 $property_and_table = "$prefix$table";
343 print STDERR __LINE__, ": $property_and_table\n" if DEBUG;
344 $file = $utf8::loose_to_file_of{$property_and_table};
347 # Add the constant and go fetch it in.
350 # If the file name contains a !, it means to invert. The
351 # 0+ makes sure result is numeric
352 $invert_it = 0 + $file =~ s/!//;
354 if ($utf8::why_deprecated{$file}) {
355 warnings::warnif('deprecated', "Use of '$type' in \\p{} or \\P{} is deprecated because: $utf8::why_deprecated{$file};");
359 && exists $utf8::caseless_equivalent{$property_and_table})
361 $file = $utf8::caseless_equivalent{$property_and_table};
364 # The pseudo-directory '#' means that there really isn't a
365 # file to read, the data is in-line as part of the string;
366 # we extract it below.
367 $file = "$unicore_dir/lib/$file.pl" unless $file =~ m!^#/!;
370 print STDERR __LINE__, ": didn't find $property_and_table\n" if DEBUG;
373 ## Last attempt -- see if it's a standard "To" name
374 ## (e.g. "ToLower") ToTitle is used by ucfirst().
375 ## The user-level way to access ToDigit() and ToFold()
376 ## is to use Unicode::UCD.
378 # Only check if caller wants non-binary
380 if ($property_and_table =~ s/^to//) {
381 # Look input up in list of properties for which we have
382 # mapping files. First do it with the strict approach
383 if (defined ($file = $utf8::strict_property_to_file_of{
384 $strict_property_and_table}))
386 $type = $utf8::file_to_swash_name{$file};
387 print STDERR __LINE__, ": type set to $type\n"
389 $file = "$unicore_dir/$file.pl";
392 elsif (defined ($file =
393 $utf8::loose_property_to_file_of{$property_and_table}))
395 $type = $utf8::file_to_swash_name{$file};
396 print STDERR __LINE__, ": type set to $type\n"
398 $file = "$unicore_dir/$file.pl";
400 } # If that fails see if there is a corresponding binary
402 elsif (defined ($file =
403 $utf8::loose_to_file_of{$property_and_table}))
406 # Here, there is no map file for the property we
407 # are trying to get the map of, but this is a
408 # binary property, and there is a file for it that
409 # can easily be translated to a mapping, so use
410 # that, treating this as a binary property.
411 # Setting 'minbits' here causes it to be stored as
412 # such in the cache, so if someone comes along
413 # later looking for just a binary, they get it.
416 # The 0+ makes sure is numeric
417 $invert_it = 0 + $file =~ s/!//;
418 $file = "$unicore_dir/lib/$file.pl"
419 unless $file =~ m!^#/!;
426 ## If we reach this line, it's because we couldn't figure
427 ## out what to do with $type. Ouch.
430 pop @recursed if @recursed;
432 } # end of GETFILE block
435 print STDERR __LINE__, ": found it (file='$file')\n" if DEBUG;
438 ## If we reach here, it was due to a 'last GETFILE' above
439 ## (exception: user-defined properties and mappings), so we
440 ## have a filename, so now we load it if we haven't already.
442 # The pseudo-directory '#' means the result isn't really a
443 # file, but is in-line, with semi-colons to be turned into
444 # new-lines. Since it is in-line there is no advantage to
446 if ($file =~ s!^#/!!) {
447 $list = $utf8::inline_definitions[$file];
450 # Here, we have an actual file to read in and load, but it
451 # may already have been read-in and cached. The cache key
452 # is the class and file to load, and whether the results
453 # need to be inverted.
454 my $found = $Cache{$class, $file, $invert_it};
455 if ($found and ref($found) eq $class) {
456 print STDERR __LINE__, ": Returning cached swash for '$class,$file,$invert_it' for \\p{$type}\n" if DEBUG;
457 pop @recursed if @recursed;
463 $list = do $file; die $@ if $@;
466 $list_is_from_mktables = 1;
468 } # End of $type is non-null
470 # Here, either $type was null, or we found the requested property and
477 # mktables lists don't have extras, like '&utf8::prop', so don't need
478 # to separate them; also lists are already sorted, so don't need to do
480 if ($list && ! $list_is_from_mktables) {
481 my $taint = substr($list,0,0); # maintain taint
483 # Separate the extras from the code point list, and make sure
484 # user-defined properties and tr/// are well-behaved for
486 if ($user_defined || $none) {
487 my @tmp = split(/^/m, $list);
491 # The extras are anything that doesn't begin with a hex digit.
492 $extras = join '', $taint, grep /^[^0-9a-fA-F]/, @tmp;
494 # Remove the extras, and sort the remaining entries by the
495 # numeric value of their beginning hex digits, removing any
497 $list = join '', $taint,
499 sort { $a->[0] <=> $b->[0] }
500 map { /^([0-9a-fA-F]+)/ && !$seen{$1}++ ? [ CORE::hex($1), $_ ] : () }
501 @tmp; # XXX doesn't do ranges right
504 # mktables has gone to some trouble to make non-user defined
505 # properties well-behaved, so we can skip the effort we do for
506 # user-defined ones. Any extras are at the very beginning of
509 # This regex splits out the first lines of $list into $1 and
510 # strips them off from $list, until we get one that begins
511 # with a hex number, alone on the line, or followed by a tab.
512 # Either portion may be empty.
513 $list =~ s/ \A ( .*? )
514 (?: \z | (?= ^ [0-9a-fA-F]+ (?: \t | $) ) )
517 $extras = "$taint$1";
522 my $hextra = sprintf "%04x", $none + 1;
523 $list =~ s/\tXXXX$/\t$hextra/mg;
526 if ($minbits != 1 && $minbits < 32) { # not binary property
528 while ($list =~ /^([0-9a-fA-F]+)(?:[\t]([0-9a-fA-F]+)?)(?:[ \t]([0-9a-fA-F]+))?/mg) {
529 my $min = CORE::hex $1;
530 my $max = defined $2 ? CORE::hex $2 : $min;
531 my $val = defined $3 ? CORE::hex $3 : 0;
532 $val += $max - $min if defined $3;
533 $top = $val if $val > $top;
537 $top > 0xff ? 16 : 8;
538 $bits = $topbits if $bits < $topbits;
543 for my $x ($extras) {
544 my $taint = substr($x,0,0); # maintain taint
546 while ($x =~ /^([^0-9a-fA-F\n])(.*)/mg) {
547 my $char = "$1$taint";
548 my $name = "$2$taint";
549 print STDERR __LINE__, ": char [$char] => name [$name]\n"
551 if ($char =~ /[-+!&]/) {
552 my ($c,$t) = split(/::/, $name, 2); # bogus use of ::, really
555 $subobj = utf8->SWASHNEW($t, "", $minbits, 0);
557 elsif (exists &$name) {
558 $subobj = utf8->SWASHNEW($name, "", $minbits, 0);
560 elsif ($c =~ /^([0-9a-fA-F]+)/) {
561 $subobj = utf8->SWASHNEW("", $c, $minbits, 0);
563 print STDERR __LINE__, ": returned from getting sub object for $name\n" if DEBUG;
565 pop @recursed if @recursed && $type;
568 push @extras, $name => $subobj;
569 $bits = $subobj->{BITS} if $bits < $subobj->{BITS};
570 $user_defined = $subobj->{USER_DEFINED}
571 if $subobj->{USER_DEFINED};
578 print STDERR __LINE__, ": CLASS = $class, TYPE => $type, BITS => $bits, NONE => $none, INVERT_IT => $invert_it, USER_DEFINED => $user_defined";
579 print STDERR "\nLIST =>\n$list" if defined $list;
580 print STDERR "\nEXTRAS =>\n$extras" if defined $extras;
590 USER_DEFINED => $user_defined,
595 $Cache{$class, $file, $invert_it} = $SWASH;
597 && exists $utf8::SwashInfo{$type}
598 && exists $utf8::SwashInfo{$type}{'specials_name'})
600 my $specials_name = $utf8::SwashInfo{$type}{'specials_name'};
602 print STDERR "\nspecials_name => $specials_name\n" if DEBUG;
603 $SWASH->{'SPECIALS'} = \%$specials_name;
605 $SWASH->{'INVERT_IT'} = $invert_it;
608 pop @recursed if @recursed && $type;
614 # Now SWASHGET is recasted into a C function S_swatch_get (see utf8.c).