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 # integer or float (no exponent)
39 my $integer_or_float_re = qr/ ^ -? \d+ (:? \. \d+ )? $ /x;
41 # Also includes rationals
42 my $numeric_re = qr! $integer_or_float_re | ^ -? \d+ / \d+ $ !x;
43 return $_[0] if $_[0] =~ $numeric_re;
45 (my $loose = $_[0]) =~ s/[-_ \t]//g;
47 return $loose if $loose !~ / ^ (?: is | to )? l $/x;
48 return 'l_' if $_[0] =~ / l .* _ /x; # If original had a trailing '_'
53 ## "SWASH" == "SWATCH HASH". A "swatch" is a swatch of the Unicode landscape.
54 ## It's a data structure that encodes a set of Unicode characters.
58 # If a floating point number is within this distance from the value of a
59 # fraction, it is considered to be that fraction, even if many more digits
60 # are specified that don't exactly match.
61 my $min_floating_slop;
63 # To guard against this program calling something that in turn ends up
64 # calling this program with the same inputs, and hence infinitely
65 # recursing, we keep a stack of the properties that are currently in
66 # progress, pushed upon entry, popped upon return.
70 my ($class, $type, $list, $minbits, $none) = @_;
74 $class = "" unless defined $class;
75 print STDERR __LINE__, ": class=$class, type=$type, list=",
76 (defined $list) ? $list : ':undef:',
77 ", minbits=$minbits, none=$none\n" if DEBUG;
80 ## Get the list of codepoints for the type.
81 ## Called from swash_init (see utf8.c) or SWASHNEW itself.
83 ## Callers of swash_init:
84 ## op.c:pmtrans -- for tr/// and y///
85 ## Unicode::UCD::prop_invlist
86 ## Unicode::UCD::prop_invmap
88 ## Given a $type, our goal is to fill $list with the set of codepoint
89 ## ranges. If $type is false, $list passed is used.
92 ## For binary properties, $minbits must be 1.
93 ## For character mappings (case and transliteration), $minbits must
94 ## be a number except 1.
96 ## $list (or that filled according to $type):
97 ## Refer to perlunicode.pod, "User-Defined Character Properties."
99 ## For binary properties, only characters with the property value
100 ## of True should be listed. The 3rd column, if any, will be ignored
102 ## $none is undocumented, so I'm (khw) trying to do some documentation
103 ## of it now. It appears to be if there is a mapping in an input file
104 ## that maps to 'XXXX', then that is replaced by $none+1, expressed in
105 ## hexadecimal. It is used somehow in tr///.
107 ## To make the parsing of $type clear, this code takes the a rather
108 ## unorthodox approach of last'ing out of the block once we have the
109 ## info we need. Were this to be a subroutine, the 'last' would just
112 # If a problem is found $type is returned;
113 # Upon success, a new (or cached) blessed object is returned with
114 # keys TYPE, BITS, EXTRAS, LIST, and NONE with values having the
115 # same meanings as the input parameters.
116 # SPECIALS contains a reference to any special-treatment hash in the
118 # INVERT_IT is non-zero if the result should be inverted before use
119 # USER_DEFINED is non-zero if the result came from a user-defined
120 my $file; ## file to load data from, and also part of the %Cache key.
122 # Change this to get a different set of Unicode tables
123 my $unicore_dir = 'unicore';
125 my $list_is_from_mktables = 0; # Is $list returned from a mktables
126 # generated file? If so, we know it's
131 # Verify that this isn't a recursive call for this property.
132 # Can't use croak, as it may try to recurse to here itself.
133 my $class_type = $class . "::$type";
134 if (grep { $_ eq $class_type } @recursed) {
135 CORE::die "panic: Infinite recursion in SWASHNEW for '$type'\n";
137 push @recursed, $class_type;
142 # regcomp.c surrounds the property name with '__" and '_i' if this
143 # is to be caseless matching.
144 my $caseless = $type =~ s/^(.*)__(.*)_i$/$1$2/;
146 print STDERR __LINE__, ": type=$type, caseless=$caseless\n" if DEBUG;
151 ## It could be a user-defined property. Look in current
152 ## package if no package given
156 my $caller0 = caller(0);
157 my $caller1 = $type =~ s/(.+):://
163 if (defined $caller1 && $type =~ /^I[ns]\w+$/) {
164 my $prop = "${caller1}::$type";
165 if (exists &{$prop}) {
166 # stolen from Scalar::Util::PP::tainted()
169 local($@, $SIG{__DIE__}, $SIG{__WARN__});
172 eval { kill 0 * $prop };
173 $tainted = 1 if $@ =~ /^Insecure/;
175 die "Insecure user-defined property \\p{$prop}\n"
178 $list = &{$prop}($caseless);
184 # During Perl's compilation, this routine may be called before
185 # the tables are constructed. If so, we have a chicken/egg
186 # problem. If we die, the tables never get constructed, so
187 # keep going, but return an empty table so only what the code
188 # has compiled in internally (currently ASCII/Latin1 range
189 # matching) will work.
191 # Poor man's constant, to avoid a run-time check.
193 = \! defined &DynaLoader::boot_DynaLoader;
196 eval "require '$unicore_dir/Heavy.pl'";
198 print STDERR __LINE__, ": '$@'\n" if DEBUG;
199 pop @recursed if @recursed;
204 require "$unicore_dir/Heavy.pl";
206 BEGIN { delete $utf8::{miniperl} }
208 # All property names are matched caselessly
209 my $property_and_table = CORE::lc $type;
210 print STDERR __LINE__, ": $property_and_table\n" if DEBUG;
212 # See if is of the compound form 'property=value', where the
213 # value indicates the table we should use.
214 my ($property, $table, @remainder) =
215 split /\s*[:=]\s*/, $property_and_table, -1;
217 pop @recursed if @recursed;
222 if (! defined $table) {
224 # Here, is the single form. The property becomes empty, and
225 # the whole value is the table.
227 $prefix = $property = "";
229 print STDERR __LINE__, ": $property\n" if DEBUG;
231 # Here it is the compound property=table form. The property
232 # name is always loosely matched, and always can have an
233 # optional 'is' prefix (which isn't true in the single
235 $property = _loose_name($property) =~ s/^is//r;
237 # And convert to canonical form. Quit if not valid.
238 $property = $utf8::loose_property_name_of{$property};
239 if (! defined $property) {
240 pop @recursed if @recursed;
244 $prefix = "$property=";
246 # If the rhs looks like it is a number...
247 print STDERR __LINE__, ": table=$table\n" if DEBUG;
249 if ($table =~ $number) {
250 print STDERR __LINE__, ": table=$table\n" if DEBUG;
252 # Split on slash, in case it is a rational, like \p{1/5}
253 my @parts = split m{ \s* / \s* }x, $table, -1;
254 print __LINE__, ": $type\n" if @parts > 2 && DEBUG;
256 foreach my $part (@parts) {
257 print __LINE__, ": part=$part\n" if DEBUG;
259 $part =~ s/^\+\s*//; # Remove leading plus
260 $part =~ s/^-\s*/-/; # Remove blanks after unary
263 # Remove underscores between digits.
264 $part =~ s/(?<= [0-9] ) _ (?= [0-9] ) //xg;
266 # No leading zeros (but don't make a single '0'
267 # into a null string)
268 $part =~ s/ ^ ( -? ) 0+ /$1/x;
269 $part .= '0' if $part eq '-' || $part eq "";
271 # No trailing zeros after a decimal point
272 $part =~ s/ ( \. [0-9]*? ) 0+ $ /$1/x;
274 # Begin with a 0 if a leading decimal point
275 $part =~ s/ ^ ( -? ) \. /${1}0./x;
277 # Ensure not a trailing decimal point: turn into an
279 $part =~ s/ \. $ //x;
281 print STDERR __LINE__, ": part=$part\n" if DEBUG;
282 #return $type if $part eq "";
288 # If denominator is negative, get rid of it, and ...
289 if ($parts[1] =~ s/^-//) {
291 # If numerator is also negative, convert the
292 # whole thing to positive, else move the minus
294 if ($parts[0] !~ s/^-//) {
295 $parts[0] = '-' . $parts[0];
298 $table = join '/', @parts;
300 elsif ($property ne 'nv' || $parts[0] !~ /\./) {
302 # Here is not numeric value, or doesn't have a
303 # decimal point. No further manipulation is
304 # necessary. (Note the hard-coded property name.
305 # This could fail if other properties eventually
306 # had fractions as well; perhaps the cjk ones
307 # could evolve to do that. This hard-coding could
308 # be fixed by mktables generating a list of
309 # properties that could have fractions.)
313 # Here is a floating point numeric_value. Convert
314 # to rational. Get a normalized form, like
315 # 5.00E-01, and look that up in the hash
317 my $float = sprintf "%.*e",
321 if (exists $utf8::nv_floating_to_rational{$float}) {
322 $table = $utf8::nv_floating_to_rational{$float};
324 pop @recursed if @recursed;
328 print STDERR __LINE__, ": $property=$table\n" if DEBUG;
332 # Combine lhs (if any) and rhs to get something that matches
333 # the syntax of the lookups.
334 $property_and_table = "$prefix$table";
335 print STDERR __LINE__, ": $property_and_table\n" if DEBUG;
337 # First try stricter matching.
338 $file = $utf8::stricter_to_file_of{$property_and_table};
340 # If didn't find it, try again with looser matching by editing
341 # out the applicable characters on the rhs and looking up
343 my $strict_property_and_table;
344 if (! defined $file) {
346 # This isn't used unless the name begins with 'to'
347 $strict_property_and_table = $property_and_table =~ s/^to//r;
348 $table = _loose_name($table);
349 $property_and_table = "$prefix$table";
350 print STDERR __LINE__, ": $property_and_table\n" if DEBUG;
351 $file = $utf8::loose_to_file_of{$property_and_table};
354 # Add the constant and go fetch it in.
357 # If the file name contains a !, it means to invert. The
358 # 0+ makes sure result is numeric
359 $invert_it = 0 + $file =~ s/!//;
361 if ($utf8::why_deprecated{$file}) {
362 warnings::warnif('deprecated', "Use of '$type' in \\p{} or \\P{} is deprecated because: $utf8::why_deprecated{$file};");
366 && exists $utf8::caseless_equivalent{$property_and_table})
368 $file = $utf8::caseless_equivalent{$property_and_table};
371 # The pseudo-directory '#' means that there really isn't a
372 # file to read, the data is in-line as part of the string;
373 # we extract it below.
374 $file = "$unicore_dir/lib/$file.pl" unless $file =~ m!^#/!;
377 print STDERR __LINE__, ": didn't find $property_and_table\n" if DEBUG;
380 ## Last attempt -- see if it's a standard "To" name
381 ## (e.g. "ToLower") ToTitle is used by ucfirst().
382 ## The user-level way to access ToDigit() and ToFold()
383 ## is to use Unicode::UCD.
385 # Only check if caller wants non-binary
387 if ($property_and_table =~ s/^to//) {
388 # Look input up in list of properties for which we have
389 # mapping files. First do it with the strict approach
390 if (defined ($file = $utf8::strict_property_to_file_of{
391 $strict_property_and_table}))
393 $type = $utf8::file_to_swash_name{$file};
394 print STDERR __LINE__, ": type set to $type\n"
396 $file = "$unicore_dir/$file.pl";
399 elsif (defined ($file =
400 $utf8::loose_property_to_file_of{$property_and_table}))
402 $type = $utf8::file_to_swash_name{$file};
403 print STDERR __LINE__, ": type set to $type\n"
405 $file = "$unicore_dir/$file.pl";
407 } # If that fails see if there is a corresponding binary
409 elsif (defined ($file =
410 $utf8::loose_to_file_of{$property_and_table}))
413 # Here, there is no map file for the property we
414 # are trying to get the map of, but this is a
415 # binary property, and there is a file for it that
416 # can easily be translated to a mapping, so use
417 # that, treating this as a binary property.
418 # Setting 'minbits' here causes it to be stored as
419 # such in the cache, so if someone comes along
420 # later looking for just a binary, they get it.
423 # The 0+ makes sure is numeric
424 $invert_it = 0 + $file =~ s/!//;
425 $file = "$unicore_dir/lib/$file.pl"
426 unless $file =~ m!^#/!;
433 ## If we reach this line, it's because we couldn't figure
434 ## out what to do with $type. Ouch.
437 pop @recursed if @recursed;
439 } # end of GETFILE block
442 print STDERR __LINE__, ": found it (file='$file')\n" if DEBUG;
445 ## If we reach here, it was due to a 'last GETFILE' above
446 ## (exception: user-defined properties and mappings), so we
447 ## have a filename, so now we load it if we haven't already.
449 # The pseudo-directory '#' means the result isn't really a
450 # file, but is in-line, with semi-colons to be turned into
451 # new-lines. Since it is in-line there is no advantage to
453 if ($file =~ s!^#/!!) {
454 $list = $utf8::inline_definitions[$file];
457 # Here, we have an actual file to read in and load, but it
458 # may already have been read-in and cached. The cache key
459 # is the class and file to load, and whether the results
460 # need to be inverted.
461 my $found = $Cache{$class, $file, $invert_it};
462 if ($found and ref($found) eq $class) {
463 print STDERR __LINE__, ": Returning cached swash for '$class,$file,$invert_it' for \\p{$type}\n" if DEBUG;
464 pop @recursed if @recursed;
470 $list = do $file; die $@ if $@;
473 $list_is_from_mktables = 1;
475 } # End of $type is non-null
477 # Here, either $type was null, or we found the requested property and
484 # mktables lists don't have extras, like '&utf8::prop', so don't need
485 # to separate them; also lists are already sorted, so don't need to do
487 if ($list && ! $list_is_from_mktables) {
488 my $taint = substr($list,0,0); # maintain taint
490 # Separate the extras from the code point list, and make sure
491 # user-defined properties and tr/// are well-behaved for
493 if ($user_defined || $none) {
494 my @tmp = split(/^/m, $list);
498 # The extras are anything that doesn't begin with a hex digit.
499 $extras = join '', $taint, grep /^[^0-9a-fA-F]/, @tmp;
501 # Remove the extras, and sort the remaining entries by the
502 # numeric value of their beginning hex digits, removing any
504 $list = join '', $taint,
506 sort { $a->[0] <=> $b->[0] }
507 map { /^([0-9a-fA-F]+)/ && !$seen{$1}++ ? [ CORE::hex($1), $_ ] : () }
508 @tmp; # XXX doesn't do ranges right
511 # mktables has gone to some trouble to make non-user defined
512 # properties well-behaved, so we can skip the effort we do for
513 # user-defined ones. Any extras are at the very beginning of
516 # This regex splits out the first lines of $list into $1 and
517 # strips them off from $list, until we get one that begins
518 # with a hex number, alone on the line, or followed by a tab.
519 # Either portion may be empty.
520 $list =~ s/ \A ( .*? )
521 (?: \z | (?= ^ [0-9a-fA-F]+ (?: \t | $) ) )
524 $extras = "$taint$1";
529 my $hextra = sprintf "%04x", $none + 1;
530 $list =~ s/\tXXXX$/\t$hextra/mg;
533 if ($minbits != 1 && $minbits < 32) { # not binary property
535 while ($list =~ /^([0-9a-fA-F]+)(?:[\t]([0-9a-fA-F]+)?)(?:[ \t]([0-9a-fA-F]+))?/mg) {
536 my $min = CORE::hex $1;
537 my $max = defined $2 ? CORE::hex $2 : $min;
538 my $val = defined $3 ? CORE::hex $3 : 0;
539 $val += $max - $min if defined $3;
540 $top = $val if $val > $top;
544 $top > 0xff ? 16 : 8;
545 $bits = $topbits if $bits < $topbits;
550 for my $x ($extras) {
551 my $taint = substr($x,0,0); # maintain taint
553 while ($x =~ /^([^0-9a-fA-F\n])(.*)/mg) {
554 my $char = "$1$taint";
555 my $name = "$2$taint";
556 print STDERR __LINE__, ": char [$char] => name [$name]\n"
558 if ($char =~ /[-+!&]/) {
559 my ($c,$t) = split(/::/, $name, 2); # bogus use of ::, really
562 $subobj = utf8->SWASHNEW($t, "", $minbits, 0);
564 elsif (exists &$name) {
565 $subobj = utf8->SWASHNEW($name, "", $minbits, 0);
567 elsif ($c =~ /^([0-9a-fA-F]+)/) {
568 $subobj = utf8->SWASHNEW("", $c, $minbits, 0);
570 print STDERR __LINE__, ": returned from getting sub object for $name\n" if DEBUG;
572 pop @recursed if @recursed && $type;
575 push @extras, $name => $subobj;
576 $bits = $subobj->{BITS} if $bits < $subobj->{BITS};
577 $user_defined = $subobj->{USER_DEFINED}
578 if $subobj->{USER_DEFINED};
585 print STDERR __LINE__, ": CLASS = $class, TYPE => $type, BITS => $bits, NONE => $none, INVERT_IT => $invert_it, USER_DEFINED => $user_defined";
586 print STDERR "\nLIST =>\n$list" if defined $list;
587 print STDERR "\nEXTRAS =>\n$extras" if defined $extras;
597 USER_DEFINED => $user_defined,
602 $Cache{$class, $file, $invert_it} = $SWASH;
604 && exists $utf8::SwashInfo{$type}
605 && exists $utf8::SwashInfo{$type}{'specials_name'})
607 my $specials_name = $utf8::SwashInfo{$type}{'specials_name'};
609 print STDERR "\nspecials_name => $specials_name\n" if DEBUG;
610 $SWASH->{'SPECIALS'} = \%$specials_name;
612 $SWASH->{'INVERT_IT'} = $invert_it;
615 pop @recursed if @recursed && $type;
621 # Now SWASHGET is recasted into a C function S_swatch_get (see utf8.c).