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 ## regexec.c:regclass_swash -- for /[]/, \p, and \P
79 ## utf8.c:is_utf8_common -- for common Unicode properties
80 ## utf8.c:S__to_utf8_case -- for lc, uc, ucfirst, etc. and //i
81 ## Unicode::UCD::prop_invlist
82 ## Unicode::UCD::prop_invmap
84 ## Given a $type, our goal is to fill $list with the set of codepoint
85 ## ranges. If $type is false, $list passed is used.
88 ## For binary properties, $minbits must be 1.
89 ## For character mappings (case and transliteration), $minbits must
90 ## be a number except 1.
92 ## $list (or that filled according to $type):
93 ## Refer to perlunicode.pod, "User-Defined Character Properties."
95 ## For binary properties, only characters with the property value
96 ## of True should be listed. The 3rd column, if any, will be ignored
98 ## $none is undocumented, so I'm (khw) trying to do some documentation
99 ## of it now. It appears to be if there is a mapping in an input file
100 ## that maps to 'XXXX', then that is replaced by $none+1, expressed in
101 ## hexadecimal. It is used somehow in tr///.
103 ## To make the parsing of $type clear, this code takes the a rather
104 ## unorthodox approach of last'ing out of the block once we have the
105 ## info we need. Were this to be a subroutine, the 'last' would just
108 # If a problem is found $type is returned;
109 # Upon success, a new (or cached) blessed object is returned with
110 # keys TYPE, BITS, EXTRAS, LIST, and NONE with values having the
111 # same meanings as the input parameters.
112 # SPECIALS contains a reference to any special-treatment hash in the
114 # INVERT_IT is non-zero if the result should be inverted before use
115 # USER_DEFINED is non-zero if the result came from a user-defined
116 my $file; ## file to load data from, and also part of the %Cache key.
118 # Change this to get a different set of Unicode tables
119 my $unicore_dir = 'unicore';
121 my $list_is_from_mktables = 0; # Is $list returned from a mktables
122 # generated file? If so, we know it's
127 # Verify that this isn't a recursive call for this property.
128 # Can't use croak, as it may try to recurse to here itself.
129 my $class_type = $class . "::$type";
130 if (grep { $_ eq $class_type } @recursed) {
131 CORE::die "panic: Infinite recursion in SWASHNEW for '$type'\n";
133 push @recursed, $class_type;
138 # regcomp.c surrounds the property name with '__" and '_i' if this
139 # is to be caseless matching.
140 my $caseless = $type =~ s/^(.*)__(.*)_i$/$1$2/;
142 print STDERR __LINE__, ": type=$type, caseless=$caseless\n" if DEBUG;
147 ## It could be a user-defined property. Look in current
148 ## package if no package given
152 my $caller0 = caller(0);
153 my $caller1 = $type =~ s/(.+):://
159 if (defined $caller1 && $type =~ /^I[ns]\w+$/) {
160 my $prop = "${caller1}::$type";
161 if (exists &{$prop}) {
162 # stolen from Scalar::Util::PP::tainted()
165 local($@, $SIG{__DIE__}, $SIG{__WARN__});
168 eval { kill 0 * $prop };
169 $tainted = 1 if $@ =~ /^Insecure/;
171 die "Insecure user-defined property \\p{$prop}\n"
174 $list = &{$prop}($caseless);
180 # During Perl's compilation, this routine may be called before
181 # the tables are constructed. If so, we have a chicken/egg
182 # problem. If we die, the tables never get constructed, so
183 # keep going, but return an empty table so only what the code
184 # has compiled in internally (currently ASCII/Latin1 range
185 # matching) will work.
187 # Poor man's constant, to avoid a run-time check.
189 = \! defined &DynaLoader::boot_DynaLoader;
192 eval "require '$unicore_dir/Heavy.pl'";
194 print STDERR __LINE__, ": '$@'\n" if DEBUG;
195 pop @recursed if @recursed;
200 require "$unicore_dir/Heavy.pl";
202 BEGIN { delete $utf8::{miniperl} }
204 # All property names are matched caselessly
205 my $property_and_table = CORE::lc $type;
206 print STDERR __LINE__, ": $property_and_table\n" if DEBUG;
208 # See if is of the compound form 'property=value', where the
209 # value indicates the table we should use.
210 my ($property, $table, @remainder) =
211 split /\s*[:=]\s*/, $property_and_table, -1;
213 pop @recursed if @recursed;
218 if (! defined $table) {
220 # Here, is the single form. The property becomes empty, and
221 # the whole value is the table.
223 $prefix = $property = "";
225 print STDERR __LINE__, ": $property\n" if DEBUG;
227 # Here it is the compound property=table form. The property
228 # name is always loosely matched, and always can have an
229 # optional 'is' prefix (which isn't true in the single
231 $property = _loose_name($property) =~ s/^is//r;
233 # And convert to canonical form. Quit if not valid.
234 $property = $utf8::loose_property_name_of{$property};
235 if (! defined $property) {
236 pop @recursed if @recursed;
240 $prefix = "$property=";
242 # If the rhs looks like it is a number...
243 print STDERR __LINE__, ": table=$table\n" if DEBUG;
245 if ($table =~ $number) {
246 print STDERR __LINE__, ": table=$table\n" if DEBUG;
248 # Split on slash, in case it is a rational, like \p{1/5}
249 my @parts = split m{ \s* / \s* }x, $table, -1;
250 print __LINE__, ": $type\n" if @parts > 2 && DEBUG;
252 foreach my $part (@parts) {
253 print __LINE__, ": part=$part\n" if DEBUG;
255 $part =~ s/^\+\s*//; # Remove leading plus
256 $part =~ s/^-\s*/-/; # Remove blanks after unary
259 # Remove underscores between digits.
260 $part =~ s/(?<= [0-9] ) _ (?= [0-9] ) //xg;
262 # No leading zeros (but don't make a single '0'
263 # into a null string)
264 $part =~ s/ ^ ( -? ) 0+ /$1/x;
265 $part .= '0' if $part eq '-' || $part eq "";
267 # No trailing zeros after a decimal point
268 $part =~ s/ ( \. [0-9]*? ) 0+ $ /$1/x;
270 # Begin with a 0 if a leading decimal point
271 $part =~ s/ ^ ( -? ) \. /${1}0./x;
273 # Ensure not a trailing decimal point: turn into an
275 $part =~ s/ \. $ //x;
277 print STDERR __LINE__, ": part=$part\n" if DEBUG;
278 #return $type if $part eq "";
284 # If denominator is negative, get rid of it, and ...
285 if ($parts[1] =~ s/^-//) {
287 # If numerator is also negative, convert the
288 # whole thing to positive, else move the minus
290 if ($parts[0] !~ s/^-//) {
291 $parts[0] = '-' . $parts[0];
294 $table = join '/', @parts;
296 elsif ($property ne 'nv' || $parts[0] !~ /\./) {
298 # Here is not numeric value, or doesn't have a
299 # decimal point. No further manipulation is
300 # necessary. (Note the hard-coded property name.
301 # This could fail if other properties eventually
302 # had fractions as well; perhaps the cjk ones
303 # could evolve to do that. This hard-coding could
304 # be fixed by mktables generating a list of
305 # properties that could have fractions.)
309 # Here is a floating point numeric_value. Convert
310 # to rational. Get a normalized form, like
311 # 5.00E-01, and look that up in the hash
313 my $float = sprintf "%.*e",
317 if (exists $utf8::nv_floating_to_rational{$float}) {
318 $table = $utf8::nv_floating_to_rational{$float};
320 pop @recursed if @recursed;
324 print STDERR __LINE__, ": $property=$table\n" if DEBUG;
328 # Combine lhs (if any) and rhs to get something that matches
329 # the syntax of the lookups.
330 $property_and_table = "$prefix$table";
331 print STDERR __LINE__, ": $property_and_table\n" if DEBUG;
333 # First try stricter matching.
334 $file = $utf8::stricter_to_file_of{$property_and_table};
336 # If didn't find it, try again with looser matching by editing
337 # out the applicable characters on the rhs and looking up
339 my $strict_property_and_table;
340 if (! defined $file) {
342 # This isn't used unless the name begins with 'to'
343 $strict_property_and_table = $property_and_table =~ s/^to//r;
344 $table = _loose_name($table);
345 $property_and_table = "$prefix$table";
346 print STDERR __LINE__, ": $property_and_table\n" if DEBUG;
347 $file = $utf8::loose_to_file_of{$property_and_table};
350 # Add the constant and go fetch it in.
353 # If the file name contains a !, it means to invert. The
354 # 0+ makes sure result is numeric
355 $invert_it = 0 + $file =~ s/!//;
357 if ($utf8::why_deprecated{$file}) {
358 warnings::warnif('deprecated', "Use of '$type' in \\p{} or \\P{} is deprecated because: $utf8::why_deprecated{$file};");
362 && exists $utf8::caseless_equivalent{$property_and_table})
364 $file = $utf8::caseless_equivalent{$property_and_table};
367 # The pseudo-directory '#' means that there really isn't a
368 # file to read, the data is in-line as part of the string;
369 # we extract it below.
370 $file = "$unicore_dir/lib/$file.pl" unless $file =~ m!^#/!;
373 print STDERR __LINE__, ": didn't find $property_and_table\n" if DEBUG;
376 ## Last attempt -- see if it's a standard "To" name
377 ## (e.g. "ToLower") ToTitle is used by ucfirst().
378 ## The user-level way to access ToDigit() and ToFold()
379 ## is to use Unicode::UCD.
381 # Only check if caller wants non-binary
383 if ($property_and_table =~ s/^to//) {
384 # Look input up in list of properties for which we have
385 # mapping files. First do it with the strict approach
386 if (defined ($file = $utf8::strict_property_to_file_of{
387 $strict_property_and_table}))
389 $type = $utf8::file_to_swash_name{$file};
390 print STDERR __LINE__, ": type set to $type\n"
392 $file = "$unicore_dir/$file.pl";
395 elsif (defined ($file =
396 $utf8::loose_property_to_file_of{$property_and_table}))
398 $type = $utf8::file_to_swash_name{$file};
399 print STDERR __LINE__, ": type set to $type\n"
401 $file = "$unicore_dir/$file.pl";
403 } # If that fails see if there is a corresponding binary
405 elsif (defined ($file =
406 $utf8::loose_to_file_of{$property_and_table}))
409 # Here, there is no map file for the property we
410 # are trying to get the map of, but this is a
411 # binary property, and there is a file for it that
412 # can easily be translated to a mapping, so use
413 # that, treating this as a binary property.
414 # Setting 'minbits' here causes it to be stored as
415 # such in the cache, so if someone comes along
416 # later looking for just a binary, they get it.
419 # The 0+ makes sure is numeric
420 $invert_it = 0 + $file =~ s/!//;
421 $file = "$unicore_dir/lib/$file.pl"
422 unless $file =~ m!^#/!;
429 ## If we reach this line, it's because we couldn't figure
430 ## out what to do with $type. Ouch.
433 pop @recursed if @recursed;
435 } # end of GETFILE block
438 print STDERR __LINE__, ": found it (file='$file')\n" if DEBUG;
441 ## If we reach here, it was due to a 'last GETFILE' above
442 ## (exception: user-defined properties and mappings), so we
443 ## have a filename, so now we load it if we haven't already.
445 # The pseudo-directory '#' means the result isn't really a
446 # file, but is in-line, with semi-colons to be turned into
447 # new-lines. Since it is in-line there is no advantage to
449 if ($file =~ s!^#/!!) {
450 $list = $utf8::inline_definitions[$file];
453 # Here, we have an actual file to read in and load, but it
454 # may already have been read-in and cached. The cache key
455 # is the class and file to load, and whether the results
456 # need to be inverted.
457 my $found = $Cache{$class, $file, $invert_it};
458 if ($found and ref($found) eq $class) {
459 print STDERR __LINE__, ": Returning cached swash for '$class,$file,$invert_it' for \\p{$type}\n" if DEBUG;
460 pop @recursed if @recursed;
466 $list = do $file; die $@ if $@;
469 $list_is_from_mktables = 1;
471 } # End of $type is non-null
473 # Here, either $type was null, or we found the requested property and
480 # mktables lists don't have extras, like '&utf8::prop', so don't need
481 # to separate them; also lists are already sorted, so don't need to do
483 if ($list && ! $list_is_from_mktables) {
484 my $taint = substr($list,0,0); # maintain taint
486 # Separate the extras from the code point list, and make sure
487 # user-defined properties and tr/// are well-behaved for
489 if ($user_defined || $none) {
490 my @tmp = split(/^/m, $list);
494 # The extras are anything that doesn't begin with a hex digit.
495 $extras = join '', $taint, grep /^[^0-9a-fA-F]/, @tmp;
497 # Remove the extras, and sort the remaining entries by the
498 # numeric value of their beginning hex digits, removing any
500 $list = join '', $taint,
502 sort { $a->[0] <=> $b->[0] }
503 map { /^([0-9a-fA-F]+)/ && !$seen{$1}++ ? [ CORE::hex($1), $_ ] : () }
504 @tmp; # XXX doesn't do ranges right
507 # mktables has gone to some trouble to make non-user defined
508 # properties well-behaved, so we can skip the effort we do for
509 # user-defined ones. Any extras are at the very beginning of
512 # This regex splits out the first lines of $list into $1 and
513 # strips them off from $list, until we get one that begins
514 # with a hex number, alone on the line, or followed by a tab.
515 # Either portion may be empty.
516 $list =~ s/ \A ( .*? )
517 (?: \z | (?= ^ [0-9a-fA-F]+ (?: \t | $) ) )
520 $extras = "$taint$1";
525 my $hextra = sprintf "%04x", $none + 1;
526 $list =~ s/\tXXXX$/\t$hextra/mg;
529 if ($minbits != 1 && $minbits < 32) { # not binary property
531 while ($list =~ /^([0-9a-fA-F]+)(?:[\t]([0-9a-fA-F]+)?)(?:[ \t]([0-9a-fA-F]+))?/mg) {
532 my $min = CORE::hex $1;
533 my $max = defined $2 ? CORE::hex $2 : $min;
534 my $val = defined $3 ? CORE::hex $3 : 0;
535 $val += $max - $min if defined $3;
536 $top = $val if $val > $top;
540 $top > 0xff ? 16 : 8;
541 $bits = $topbits if $bits < $topbits;
546 for my $x ($extras) {
547 my $taint = substr($x,0,0); # maintain taint
549 while ($x =~ /^([^0-9a-fA-F\n])(.*)/mg) {
550 my $char = "$1$taint";
551 my $name = "$2$taint";
552 print STDERR __LINE__, ": char [$char] => name [$name]\n"
554 if ($char =~ /[-+!&]/) {
555 my ($c,$t) = split(/::/, $name, 2); # bogus use of ::, really
558 $subobj = utf8->SWASHNEW($t, "", $minbits, 0);
560 elsif (exists &$name) {
561 $subobj = utf8->SWASHNEW($name, "", $minbits, 0);
563 elsif ($c =~ /^([0-9a-fA-F]+)/) {
564 $subobj = utf8->SWASHNEW("", $c, $minbits, 0);
566 print STDERR __LINE__, ": returned from getting sub object for $name\n" if DEBUG;
568 pop @recursed if @recursed && $type;
571 push @extras, $name => $subobj;
572 $bits = $subobj->{BITS} if $bits < $subobj->{BITS};
573 $user_defined = $subobj->{USER_DEFINED}
574 if $subobj->{USER_DEFINED};
581 print STDERR __LINE__, ": CLASS = $class, TYPE => $type, BITS => $bits, NONE => $none, INVERT_IT => $invert_it, USER_DEFINED => $user_defined";
582 print STDERR "\nLIST =>\n$list" if defined $list;
583 print STDERR "\nEXTRAS =>\n$extras" if defined $extras;
593 USER_DEFINED => $user_defined,
598 $Cache{$class, $file, $invert_it} = $SWASH;
600 && exists $utf8::SwashInfo{$type}
601 && exists $utf8::SwashInfo{$type}{'specials_name'})
603 my $specials_name = $utf8::SwashInfo{$type}{'specials_name'};
605 print STDERR "\nspecials_name => $specials_name\n" if DEBUG;
606 $SWASH->{'SPECIALS'} = \%$specials_name;
608 $SWASH->{'INVERT_IT'} = $invert_it;
611 pop @recursed if @recursed && $type;
617 # Now SWASHGET is recasted into a C function S_swatch_get (see utf8.c).