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(@_) }
17 # Given a lowercase property or property-value name, return its
18 # standardized version that is expected for look-up in the 'loose' hashes
19 # in Heavy.pl (hence, this depends on what mktables does). This squeezes
20 # out blanks, underscores and dashes. The complication stems from the
21 # grandfathered-in 'L_', which retains a single trailing underscore.
23 my $loose = $_[0] =~ s/[-\s_]//rg;
25 return $loose if $loose !~ / ^ (?: is | to )? l $/x;
26 return 'l_' if $_[0] =~ / l .* _ /x; # If original had a trailing '_'
31 ## "SWASH" == "SWATCH HASH". A "swatch" is a swatch of the Unicode landscape.
32 ## It's a data structure that encodes a set of Unicode characters.
36 # If a floating point number is within this distance from the value of a
37 # fraction, it is considered to be that fraction, even if many more digits
38 # are specified that don't exactly match.
39 my $min_floating_slop;
41 # To guard against this program calling something that in turn ends up
42 # calling this program with the same inputs, and hence infinitely
43 # recursing, we keep a stack of the properties that are currently in
44 # progress, pushed upon entry, popped upon return.
48 my ($class, $type, $list, $minbits, $none) = @_;
52 $class = "" unless defined $class;
53 print STDERR __LINE__, ": class=$class, type=$type, list=",
54 (defined $list) ? $list : ':undef:',
55 ", minbits=$minbits, none=$none\n" if DEBUG;
58 ## Get the list of codepoints for the type.
59 ## Called from swash_init (see utf8.c) or SWASHNEW itself.
61 ## Callers of swash_init:
62 ## op.c:pmtrans -- for tr/// and y///
63 ## regexec.c:regclass_swash -- for /[]/, \p, and \P
64 ## utf8.c:is_utf8_common -- for common Unicode properties
65 ## utf8.c:to_utf8_case -- for lc, uc, ucfirst, etc. and //i
66 ## Unicode::UCD::prop_invlist
67 ## Unicode::UCD::prop_invmap
69 ## Given a $type, our goal is to fill $list with the set of codepoint
70 ## ranges. If $type is false, $list passed is used.
73 ## For binary properties, $minbits must be 1.
74 ## For character mappings (case and transliteration), $minbits must
75 ## be a number except 1.
77 ## $list (or that filled according to $type):
78 ## Refer to perlunicode.pod, "User-Defined Character Properties."
80 ## For binary properties, only characters with the property value
81 ## of True should be listed. The 3rd column, if any, will be ignored
83 ## $none is undocumented, so I'm (khw) trying to do some documentation
84 ## of it now. It appears to be if there is a mapping in an input file
85 ## that maps to 'XXXX', then that is replaced by $none+1, expressed in
86 ## hexadecimal. It is used somehow in tr///.
88 ## To make the parsing of $type clear, this code takes the a rather
89 ## unorthodox approach of last'ing out of the block once we have the
90 ## info we need. Were this to be a subroutine, the 'last' would just
93 # If a problem is found $type is returned;
94 # Upon success, a new (or cached) blessed object is returned with
95 # keys TYPE, BITS, EXTRAS, LIST, and NONE with values having the
96 # same meanings as the input parameters.
97 # SPECIALS contains a reference to any special-treatment hash in the
99 # INVERT_IT is non-zero if the result should be inverted before use
100 # USER_DEFINED is non-zero if the result came from a user-defined
101 my $file; ## file to load data from, and also part of the %Cache key.
103 # Change this to get a different set of Unicode tables
104 my $unicore_dir = 'unicore';
106 my $list_is_from_mktables = 0; # Is $list returned from a mktables
107 # generated file? If so, we know it's
112 # Verify that this isn't a recursive call for this property.
113 # Can't use croak, as it may try to recurse to here itself.
114 my $class_type = $class . "::$type";
115 if (grep { $_ eq $class_type } @recursed) {
116 CORE::die "panic: Infinite recursion in SWASHNEW for '$type'\n";
118 push @recursed, $class_type;
123 # regcomp.c surrounds the property name with '__" and '_i' if this
124 # is to be caseless matching.
125 my $caseless = $type =~ s/^(.*)__(.*)_i$/$1$2/;
127 print STDERR __LINE__, ": type=$type, caseless=$caseless\n" if DEBUG;
132 ## It could be a user-defined property. Look in current
133 ## package if no package given
137 my $caller0 = caller(0);
138 my $caller1 = $type =~ s/(.+):://
144 if (defined $caller1 && $type =~ /^I[ns]\w+$/) {
145 my $prop = "${caller1}::$type";
146 if (exists &{$prop}) {
147 # stolen from Scalar::Util::PP::tainted()
150 local($@, $SIG{__DIE__}, $SIG{__WARN__});
153 eval { kill 0 * $prop };
154 $tainted = 1 if $@ =~ /^Insecure/;
156 die "Insecure user-defined property \\p{$prop}\n"
159 $list = &{$prop}($caseless);
165 # During Perl's compilation, this routine may be called before
166 # the tables are constructed. If so, we have a chicken/egg
167 # problem. If we die, the tables never get constructed, so
168 # keep going, but return an empty table so only what the code
169 # has compiled in internally (currently ASCII/Latin1 range
170 # matching) will work.
172 # Poor man's constant, to avoid a run-time check.
174 = \! defined &DynaLoader::boot_DynaLoader;
177 eval "require '$unicore_dir/Heavy.pl'";
179 print STDERR __LINE__, ": '$@'\n" if DEBUG;
180 pop @recursed if @recursed;
185 require "$unicore_dir/Heavy.pl";
187 BEGIN { delete $utf8::{miniperl} }
189 # All property names are matched caselessly
190 my $property_and_table = CORE::lc $type;
191 print STDERR __LINE__, ": $property_and_table\n" if DEBUG;
193 # See if is of the compound form 'property=value', where the
194 # value indicates the table we should use.
195 my ($property, $table, @remainder) =
196 split /\s*[:=]\s*/, $property_and_table, -1;
198 pop @recursed if @recursed;
203 if (! defined $table) {
205 # Here, is the single form. The property becomes empty, and
206 # the whole value is the table.
208 $prefix = $property = "";
210 print STDERR __LINE__, ": $property\n" if DEBUG;
212 # Here it is the compound property=table form. The property
213 # name is always loosely matched, and always can have an
214 # optional 'is' prefix (which isn't true in the single
216 $property = _loose_name($property) =~ s/^is//r;
218 # And convert to canonical form. Quit if not valid.
219 $property = $utf8::loose_property_name_of{$property};
220 if (! defined $property) {
221 pop @recursed if @recursed;
225 $prefix = "$property=";
227 # If the rhs looks like it is a number...
228 print STDERR __LINE__, ": table=$table\n" if DEBUG;
229 if ($table =~ qr{ ^ [ \s 0-9 _ + / . -]+ $ }x) {
230 print STDERR __LINE__, ": table=$table\n" if DEBUG;
232 # Don't allow leading nor trailing slashes
233 if ($table =~ / ^ \/ | \/ $ /x) {
234 pop @recursed if @recursed;
238 # Split on slash, in case it is a rational, like \p{1/5}
239 my @parts = split qr{ \s* / \s* }x, $table, -1;
240 print __LINE__, ": $type\n" if @parts > 2 && DEBUG;
242 # Can have maximum of one slash
244 pop @recursed if @recursed;
248 foreach my $part (@parts) {
249 print __LINE__, ": part=$part\n" if DEBUG;
251 $part =~ s/^\+\s*//; # Remove leading plus
252 $part =~ s/^-\s*/-/; # Remove blanks after unary
255 # Remove underscores between digits.
256 $part =~ s/(?<= [0-9] ) _ (?= [0-9] ) //xg;
258 # No leading zeros (but don't make a single '0'
259 # into a null string)
260 $part =~ s/ ^ ( -? ) 0+ /$1/x;
261 $part .= '0' if $part eq '-' || $part eq "";
263 # No trailing zeros after a decimal point
264 $part =~ s/ ( \. .*? ) 0+ $ /$1/x;
266 # Begin with a 0 if a leading decimal point
267 $part =~ s/ ^ ( -? ) \. /${1}0./x;
269 # Ensure not a trailing decimal point: turn into an
271 $part =~ s/ \. $ //x;
273 print STDERR __LINE__, ": part=$part\n" if DEBUG;
274 #return $type if $part eq "";
276 # Result better look like a number. (This test is
277 # needed because, for example could have a plus in
279 if ($part !~ / ^ -? [0-9]+ ( \. [0-9]+)? $ /x) {
280 pop @recursed if @recursed;
288 # If denominator is negative, get rid of it, and ...
289 if ($parts[1] =~ s/^-//) {
291 # If numerator is also negative, convert the
292 # whole thing to positive, or move the minus to
294 if ($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. Try to
314 # convert to rational. First see if is in the list
316 if (exists $utf8::nv_floating_to_rational{$parts[0]}) {
317 $table = $utf8::nv_floating_to_rational{$parts[0]};
320 # Here not in the list. See if is close
321 # enough to something in the list. First
322 # determine what 'close enough' means. It has
323 # to be as tight as what mktables says is the
324 # maximum slop, and as tight as how many
325 # digits we were passed. That is, if the user
326 # said .667, .6667, .66667, etc. we match as
327 # many digits as they passed until get to
328 # where it doesn't matter any more due to the
329 # machine's precision. If they said .6666668,
331 (my $fraction = $parts[0]) =~ s/^.*\.//;
332 my $epsilon = 10 ** - (length($fraction));
333 if ($epsilon > $utf8::max_floating_slop) {
334 $epsilon = $utf8::max_floating_slop;
337 # But it can't be tighter than the minimum
338 # precision for this machine. If haven't
339 # already calculated that minimum, do so now.
340 if (! defined $min_floating_slop) {
342 # Keep going down an order of magnitude
343 # until find that adding this quantity to
344 # 1 remains 1; but put an upper limit on
345 # this so in case this algorithm doesn't
346 # work properly on some platform, that we
347 # won't loop forever.
349 $min_floating_slop = 1;
350 while (1+ $min_floating_slop != 1
353 my $next = $min_floating_slop / 10;
354 last if $next == 0; # If underflows,
356 $min_floating_slop = $next;
357 print STDERR __LINE__, ": min_float_slop=$min_floating_slop\n" if DEBUG;
360 # Back off a couple orders of magnitude,
362 $min_floating_slop *= 100;
365 if ($epsilon < $min_floating_slop) {
366 $epsilon = $min_floating_slop;
368 print STDERR __LINE__, ": fraction=.$fraction; epsilon=$epsilon\n" if DEBUG;
372 # And for each possible rational in the table,
373 # see if it is within epsilon of the input.
375 (keys %utf8::nv_floating_to_rational)
377 print STDERR __LINE__, ": epsilon=$epsilon, official=$official, diff=", abs($parts[0] - $official), "\n" if DEBUG;
378 if (abs($parts[0] - $official) < $epsilon) {
380 $utf8::nv_floating_to_rational{$official};
385 # Quit if didn't find one.
386 if (! defined $table) {
387 pop @recursed if @recursed;
392 print STDERR __LINE__, ": $property=$table\n" if DEBUG;
396 # Combine lhs (if any) and rhs to get something that matches
397 # the syntax of the lookups.
398 $property_and_table = "$prefix$table";
399 print STDERR __LINE__, ": $property_and_table\n" if DEBUG;
401 # First try stricter matching.
402 $file = $utf8::stricter_to_file_of{$property_and_table};
404 # If didn't find it, try again with looser matching by editing
405 # out the applicable characters on the rhs and looking up
407 if (! defined $file) {
408 $table = _loose_name($table);
409 $property_and_table = "$prefix$table";
410 print STDERR __LINE__, ": $property_and_table\n" if DEBUG;
411 $file = $utf8::loose_to_file_of{$property_and_table};
414 # Add the constant and go fetch it in.
417 # If the file name contains a !, it means to invert. The
418 # 0+ makes sure result is numeric
419 $invert_it = 0 + $file =~ s/!//;
421 if ($utf8::why_deprecated{$file}) {
422 warnings::warnif('deprecated', "Use of '$type' in \\p{} or \\P{} is deprecated because: $utf8::why_deprecated{$file};");
426 && exists $utf8::caseless_equivalent{$property_and_table})
428 $file = $utf8::caseless_equivalent{$property_and_table};
431 # The pseudo-directory '#' means that there really isn't a
432 # file to read, the data is in-line as part of the string;
433 # we extract it below.
434 $file = "$unicore_dir/lib/$file.pl" unless $file =~ m!^#/!;
437 print STDERR __LINE__, ": didn't find $property_and_table\n" if DEBUG;
440 ## Last attempt -- see if it's a standard "To" name
441 ## (e.g. "ToLower") ToTitle is used by ucfirst().
442 ## The user-level way to access ToDigit() and ToFold()
443 ## is to use Unicode::UCD.
445 # Only check if caller wants non-binary
447 if ($minbits != 1 && $property_and_table =~ s/^to//) {{
448 # Look input up in list of properties for which we have
451 $utf8::loose_property_to_file_of{$property_and_table}))
453 $type = $utf8::file_to_swash_name{$file};
454 print STDERR __LINE__, ": type set to $type\n" if DEBUG;
455 $file = "$unicore_dir/$file.pl";
457 } # If that fails see if there is a corresponding binary
459 elsif (defined ($file =
460 $utf8::loose_to_file_of{$property_and_table}))
463 # Here, there is no map file for the property we are
464 # trying to get the map of, but this is a binary
465 # property, and there is a file for it that can easily
466 # be translated to a mapping.
468 # In the case of properties that are forced to binary,
469 # they are a combination. We return the actual
470 # mapping instead of the binary. If the input is
471 # something like 'Tocjkkiicore', it will be found in
472 # %loose_property_to_file_of above as => 'To/kIICore'.
473 # But the form like ToIskiicore won't be. To fix
474 # this, it was easiest to do it here. These
475 # properties are the complements of the default
476 # property, so there is an entry in %loose_to_file_of
477 # that is 'iskiicore' => '!kIICore/N', If we find such
478 # an entry, strip off things and try again, which
479 # should find the entry in %loose_property_to_file_of.
480 # Actual binary properties that are of this form, such
481 # as this entry: 'ishrkt' => '!Perl/Any' will also be
482 # retried, but won't be in %loose_property_to_file_of,
483 # and instead the next time through, it will find
484 # 'hrkt' => '!Perl/Any' and proceed.
487 && $property_and_table =~ s/^is//;
489 # This is a binary property. Setting this here causes
490 # it to be stored as such in the cache, so if someone
491 # comes along later looking for just a binary, they
495 # The 0+ makes sure is numeric
496 $invert_it = 0 + $file =~ s/!//;
497 $file = "$unicore_dir/lib/$file.pl" unless $file =~ m!^#/!;
503 ## If we reach this line, it's because we couldn't figure
504 ## out what to do with $type. Ouch.
507 pop @recursed if @recursed;
509 } # end of GETFILE block
512 print STDERR __LINE__, ": found it (file='$file')\n" if DEBUG;
515 ## If we reach here, it was due to a 'last GETFILE' above
516 ## (exception: user-defined properties and mappings), so we
517 ## have a filename, so now we load it if we haven't already.
519 # The pseudo-directory '#' means the result isn't really a
520 # file, but is in-line, with semi-colons to be turned into
521 # new-lines. Since it is in-line there is no advantage to
523 if ($file =~ s!^#/!!) {
524 $list = $utf8::inline_definitions[$file];
527 # Here, we have an actual file to read in and load, but it
528 # may already have been read-in and cached. The cache key
529 # is the class and file to load, and whether the results
530 # need to be inverted.
531 my $found = $Cache{$class, $file, $invert_it};
532 if ($found and ref($found) eq $class) {
533 print STDERR __LINE__, ": Returning cached swash for '$class,$file,$invert_it' for \\p{$type}\n" if DEBUG;
534 pop @recursed if @recursed;
540 $list = do $file; die $@ if $@;
543 $list_is_from_mktables = 1;
545 } # End of $type is non-null
547 # Here, either $type was null, or we found the requested property and
554 # mktables lists don't have extras, like '&utf8::prop', so don't need
555 # to separate them; also lists are already sorted, so don't need to do
557 if ($list && ! $list_is_from_mktables) {
558 my $taint = substr($list,0,0); # maintain taint
560 # Separate the extras from the code point list, and make sure
561 # user-defined properties and tr/// are well-behaved for
563 if ($user_defined || $none) {
564 my @tmp = split(/^/m, $list);
568 # The extras are anything that doesn't begin with a hex digit.
569 $extras = join '', $taint, grep /^[^0-9a-fA-F]/, @tmp;
571 # Remove the extras, and sort the remaining entries by the
572 # numeric value of their beginning hex digits, removing any
574 $list = join '', $taint,
576 sort { $a->[0] <=> $b->[0] }
577 map { /^([0-9a-fA-F]+)/ && !$seen{$1}++ ? [ CORE::hex($1), $_ ] : () }
578 @tmp; # XXX doesn't do ranges right
581 # mktables has gone to some trouble to make non-user defined
582 # properties well-behaved, so we can skip the effort we do for
583 # user-defined ones. Any extras are at the very beginning of
586 # This regex splits out the first lines of $list into $1 and
587 # strips them off from $list, until we get one that begins
588 # with a hex number, alone on the line, or followed by a tab.
589 # Either portion may be empty.
590 $list =~ s/ \A ( .*? )
591 (?: \z | (?= ^ [0-9a-fA-F]+ (?: \t | $) ) )
594 $extras = "$taint$1";
599 my $hextra = sprintf "%04x", $none + 1;
600 $list =~ s/\tXXXX$/\t$hextra/mg;
603 if ($minbits != 1 && $minbits < 32) { # not binary property
605 while ($list =~ /^([0-9a-fA-F]+)(?:[\t]([0-9a-fA-F]+)?)(?:[ \t]([0-9a-fA-F]+))?/mg) {
606 my $min = CORE::hex $1;
607 my $max = defined $2 ? CORE::hex $2 : $min;
608 my $val = defined $3 ? CORE::hex $3 : 0;
609 $val += $max - $min if defined $3;
610 $top = $val if $val > $top;
614 $top > 0xff ? 16 : 8;
615 $bits = $topbits if $bits < $topbits;
620 for my $x ($extras) {
621 my $taint = substr($x,0,0); # maintain taint
623 while ($x =~ /^([^0-9a-fA-F\n])(.*)/mg) {
624 my $char = "$1$taint";
625 my $name = "$2$taint";
626 print STDERR __LINE__, ": char [$char] => name [$name]\n"
628 if ($char =~ /[-+!&]/) {
629 my ($c,$t) = split(/::/, $name, 2); # bogus use of ::, really
632 $subobj = utf8->SWASHNEW($t, "", $minbits, 0);
634 elsif (exists &$name) {
635 $subobj = utf8->SWASHNEW($name, "", $minbits, 0);
637 elsif ($c =~ /^([0-9a-fA-F]+)/) {
638 $subobj = utf8->SWASHNEW("", $c, $minbits, 0);
640 print STDERR __LINE__, ": returned from getting sub object for $name\n" if DEBUG;
642 pop @recursed if @recursed && $type;
645 push @extras, $name => $subobj;
646 $bits = $subobj->{BITS} if $bits < $subobj->{BITS};
647 $user_defined = $subobj->{USER_DEFINED}
648 if $subobj->{USER_DEFINED};
655 print STDERR __LINE__, ": CLASS = $class, TYPE => $type, BITS => $bits, NONE => $none, INVERT_IT => $invert_it, USER_DEFINED => $user_defined";
656 print STDERR "\nLIST =>\n$list" if defined $list;
657 print STDERR "\nEXTRAS =>\n$extras" if defined $extras;
667 USER_DEFINED => $user_defined,
672 $Cache{$class, $file, $invert_it} = $SWASH;
674 && exists $utf8::SwashInfo{$type}
675 && exists $utf8::SwashInfo{$type}{'specials_name'})
677 my $specials_name = $utf8::SwashInfo{$type}{'specials_name'};
679 print STDERR "\nspecials_name => $specials_name\n" if DEBUG;
680 $SWASH->{'SPECIALS'} = \%$specials_name;
682 $SWASH->{'INVERT_IT'} = $invert_it;
685 pop @recursed if @recursed && $type;
691 # Now SWASHGET is recasted into a C function S_swatch_get (see utf8.c).