12 sub croak { require Carp; Carp::croak(@_) }
15 # Given a lowercase property or property-value name, return its
16 # standardized version that is expected for look-up in the 'loose' hashes
17 # in Heavy.pl (hence, this depends on what mktables does). This squeezes
18 # out blanks, underscores and dashes. The complication stems from the
19 # grandfathered-in 'L_', which retains a single trailing underscore.
21 my $loose = $_[0] =~ s/[-\s_]//rg;
23 return $loose if $loose !~ / ^ (?: is | to )? l $/x;
24 return 'l_' if $_[0] =~ / l .* _ /x; # If original had a trailing '_'
29 ## "SWASH" == "SWATCH HASH". A "swatch" is a swatch of the Unicode landscape.
30 ## It's a data structure that encodes a set of Unicode characters.
34 # If a floating point number is within this distance from the value of a
35 # fraction, it is considered to be that fraction, even if many more digits
36 # are specified that don't exactly match.
37 my $min_floating_slop;
39 # To guard against this program calling something that in turn ends up
40 # calling this program with the same inputs, and hence infinitely
41 # recursing, we keep a stack of the properties that are currently in
42 # progress, pushed upon entry, popped upon return.
46 my ($class, $type, $list, $minbits, $none) = @_;
50 $class = "" unless defined $class;
51 print STDERR __LINE__, ": class=$class, type=$type, list=",
52 (defined $list) ? $list : ':undef:',
53 ", minbits=$minbits, none=$none\n" if DEBUG;
56 ## Get the list of codepoints for the type.
57 ## Called from swash_init (see utf8.c) or SWASHNEW itself.
59 ## Callers of swash_init:
60 ## op.c:pmtrans -- for tr/// and y///
61 ## regexec.c:regclass_swash -- for /[]/, \p, and \P
62 ## utf8.c:is_utf8_common -- for common Unicode properties
63 ## utf8.c:to_utf8_case -- for lc, uc, ucfirst, etc. and //i
64 ## Unicode::UCD::prop_invlist
65 ## Unicode::UCD::prop_invmap
67 ## Given a $type, our goal is to fill $list with the set of codepoint
68 ## ranges. If $type is false, $list passed is used.
71 ## For binary properties, $minbits must be 1.
72 ## For character mappings (case and transliteration), $minbits must
73 ## be a number except 1.
75 ## $list (or that filled according to $type):
76 ## Refer to perlunicode.pod, "User-Defined Character Properties."
78 ## For binary properties, only characters with the property value
79 ## of True should be listed. The 3rd column, if any, will be ignored
81 ## $none is undocumented, so I'm (khw) trying to do some documentation
82 ## of it now. It appears to be if there is a mapping in an input file
83 ## that maps to 'XXXX', then that is replaced by $none+1, expressed in
84 ## hexadecimal. It is used somehow in tr///.
86 ## To make the parsing of $type clear, this code takes the a rather
87 ## unorthodox approach of last'ing out of the block once we have the
88 ## info we need. Were this to be a subroutine, the 'last' would just
91 # If a problem is found $type is returned;
92 # Upon success, a new (or cached) blessed object is returned with
93 # keys TYPE, BITS, EXTRAS, LIST, and NONE with values having the
94 # same meanings as the input parameters.
95 # SPECIALS contains a reference to any special-treatment hash in the
96 # INVERT_IT is non-zero if the result should be inverted before use
97 # USER_DEFINED is non-zero if the result came from a user-defined
99 my $file; ## file to load data from, and also part of the %Cache key.
101 # Change this to get a different set of Unicode tables
102 my $unicore_dir = 'unicore';
104 my $list_is_from_mktables = 0; # Is $list returned from a mktables
105 # generated file? If so, we know it's
110 # Verify that this isn't a recursive call for this property.
111 # Can't use croak, as it may try to recurse to here itself.
112 my $class_type = $class . "::$type";
113 if (grep { $_ eq $class_type } @recursed) {
114 CORE::die "panic: Infinite recursion in SWASHNEW for '$type'\n";
116 push @recursed, $class_type;
121 # regcomp.c surrounds the property name with '__" and '_i' if this
122 # is to be caseless matching.
123 my $caseless = $type =~ s/^(.*)__(.*)_i$/$1$2/;
125 print STDERR __LINE__, ": type=$type, caseless=$caseless\n" if DEBUG;
130 ## It could be a user-defined property. Look in current
131 ## package if no package given
135 my $caller0 = caller(0);
136 my $caller1 = $type =~ s/(.+)::// ? $1 : $caller0 eq 'main' ?
139 if (defined $caller1 && $type =~ /^I[ns]\w+$/) {
140 my $prop = "${caller1}::$type";
141 if (exists &{$prop}) {
142 # stolen from Scalar::Util::PP::tainted()
145 local($@, $SIG{__DIE__}, $SIG{__WARN__});
148 eval { kill 0 * $prop };
149 $tainted = 1 if $@ =~ /^Insecure/;
151 die "Insecure user-defined property \\p{$prop}\n"
154 $list = &{$prop}($caseless);
160 # During Perl's compilation, this routine may be called before
161 # the tables are constructed. If so, we have a chicken/egg
162 # problem. If we die, the tables never get constructed, so
163 # keep going, but return an empty table so only what the code
164 # has compiled in internally (currently ASCII/Latin1 range
165 # matching) will work.
167 # Poor man's constant, to avoid a run-time check.
169 = \! defined &DynaLoader::boot_DynaLoader;
172 eval "require '$unicore_dir/Heavy.pl'";
176 require "$unicore_dir/Heavy.pl";
178 BEGIN { delete $utf8::{miniperl} }
180 # All property names are matched caselessly
181 my $property_and_table = CORE::lc $type;
182 print STDERR __LINE__, ": $property_and_table\n" if DEBUG;
184 # See if is of the compound form 'property=value', where the
185 # value indicates the table we should use.
186 my ($property, $table, @remainder) =
187 split /\s*[:=]\s*/, $property_and_table, -1;
189 pop @recursed if @recursed;
194 if (! defined $table) {
196 # Here, is the single form. The property becomes empty, and
197 # the whole value is the table.
199 $prefix = $property = "";
201 print STDERR __LINE__, ": $property\n" if DEBUG;
203 # Here it is the compound property=table form. The property
204 # name is always loosely matched, and always can have an
205 # optional 'is' prefix (which isn't true in the single
207 $property = _loose_name($property) =~ s/^is//r;
209 # And convert to canonical form. Quit if not valid.
210 $property = $utf8::loose_property_name_of{$property};
211 if (! defined $property) {
212 pop @recursed if @recursed;
216 $prefix = "$property=";
218 # If the rhs looks like it is a number...
219 print STDERR __LINE__, ": table=$table\n" if DEBUG;
220 if ($table =~ qr{ ^ [ \s 0-9 _ + / . -]+ $ }x) {
221 print STDERR __LINE__, ": table=$table\n" if DEBUG;
223 # Don't allow leading nor trailing slashes
224 if ($table =~ / ^ \/ | \/ $ /x) {
225 pop @recursed if @recursed;
229 # Split on slash, in case it is a rational, like \p{1/5}
230 my @parts = split qr{ \s* / \s* }x, $table, -1;
231 print __LINE__, ": $type\n" if @parts > 2 && DEBUG;
233 # Can have maximum of one slash
235 pop @recursed if @recursed;
239 foreach my $part (@parts) {
240 print __LINE__, ": part=$part\n" if DEBUG;
242 $part =~ s/^\+\s*//; # Remove leading plus
243 $part =~ s/^-\s*/-/; # Remove blanks after unary
246 # Remove underscores between digits.
247 $part =~ s/( ?<= [0-9] ) _ (?= [0-9] ) //xg;
249 # No leading zeros (but don't make a single '0'
250 # into a null string)
251 $part =~ s/ ^ ( -? ) 0+ /$1/x;
252 $part .= '0' if $part eq '-' || $part eq "";
254 # No trailing zeros after a decimal point
255 $part =~ s/ ( \. .*? ) 0+ $ /$1/x;
257 # Begin with a 0 if a leading decimal point
258 $part =~ s/ ^ ( -? ) \. /${1}0./x;
260 # Ensure not a trailing decimal point: turn into an
262 $part =~ s/ \. $ //x;
264 print STDERR __LINE__, ": part=$part\n" if DEBUG;
265 #return $type if $part eq "";
267 # Result better look like a number. (This test is
268 # needed because, for example could have a plus in
270 if ($part !~ / ^ -? [0-9]+ ( \. [0-9]+)? $ /x) {
271 pop @recursed if @recursed;
279 # If denominator is negative, get rid of it, and ...
280 if ($parts[1] =~ s/^-//) {
282 # If numerator is also negative, convert the
283 # whole thing to positive, or move the minus to
285 if ($parts[0] !~ s/^-//) {
286 $parts[0] = '-' . $parts[0];
289 $table = join '/', @parts;
291 elsif ($property ne 'nv' || $parts[0] !~ /\./) {
293 # Here is not numeric value, or doesn't have a
294 # decimal point. No further manipulation is
295 # necessary. (Note the hard-coded property name.
296 # This could fail if other properties eventually
297 # had fractions as well; perhaps the cjk ones
298 # could evolve to do that. This hard-coding could
299 # be fixed by mktables generating a list of
300 # properties that could have fractions.)
304 # Here is a floating point numeric_value. Try to
305 # convert to rational. First see if is in the list
307 if (exists $utf8::nv_floating_to_rational{$parts[0]}) {
308 $table = $utf8::nv_floating_to_rational{$parts[0]};
311 # Here not in the list. See if is close
312 # enough to something in the list. First
313 # determine what 'close enough' means. It has
314 # to be as tight as what mktables says is the
315 # maximum slop, and as tight as how many
316 # digits we were passed. That is, if the user
317 # said .667, .6667, .66667, etc. we match as
318 # many digits as they passed until get to
319 # where it doesn't matter any more due to the
320 # machine's precision. If they said .6666668,
322 (my $fraction = $parts[0]) =~ s/^.*\.//;
323 my $epsilon = 10 ** - (length($fraction));
324 if ($epsilon > $utf8::max_floating_slop) {
325 $epsilon = $utf8::max_floating_slop;
328 # But it can't be tighter than the minimum
329 # precision for this machine. If haven't
330 # already calculated that minimum, do so now.
331 if (! defined $min_floating_slop) {
333 # Keep going down an order of magnitude
334 # until find that adding this quantity to
335 # 1 remains 1; but put an upper limit on
336 # this so in case this algorithm doesn't
337 # work properly on some platform, that we
338 # won't loop forever.
340 $min_floating_slop = 1;
341 while (1+ $min_floating_slop != 1
344 my $next = $min_floating_slop / 10;
345 last if $next == 0; # If underflows,
347 $min_floating_slop = $next;
348 print STDERR __LINE__, ": min_float_slop=$min_floating_slop\n" if DEBUG;
351 # Back off a couple orders of magnitude,
353 $min_floating_slop *= 100;
356 if ($epsilon < $min_floating_slop) {
357 $epsilon = $min_floating_slop;
359 print STDERR __LINE__, ": fraction=.$fraction; epsilon=$epsilon\n" if DEBUG;
363 # And for each possible rational in the table,
364 # see if it is within epsilon of the input.
366 (keys %utf8::nv_floating_to_rational)
368 print STDERR __LINE__, ": epsilon=$epsilon, official=$official, diff=", abs($parts[0] - $official), "\n" if DEBUG;
369 if (abs($parts[0] - $official) < $epsilon) {
371 $utf8::nv_floating_to_rational{$official};
376 # Quit if didn't find one.
377 if (! defined $table) {
378 pop @recursed if @recursed;
383 print STDERR __LINE__, ": $property=$table\n" if DEBUG;
387 # Combine lhs (if any) and rhs to get something that matches
388 # the syntax of the lookups.
389 $property_and_table = "$prefix$table";
390 print STDERR __LINE__, ": $property_and_table\n" if DEBUG;
392 # First try stricter matching.
393 $file = $utf8::stricter_to_file_of{$property_and_table};
395 # If didn't find it, try again with looser matching by editing
396 # out the applicable characters on the rhs and looking up
398 if (! defined $file) {
399 $table = _loose_name($table);
400 $property_and_table = "$prefix$table";
401 print STDERR __LINE__, ": $property_and_table\n" if DEBUG;
402 $file = $utf8::loose_to_file_of{$property_and_table};
405 # Add the constant and go fetch it in.
408 # A beginning ! means to invert. The 0+ makes sure is
410 $invert_it = 0 + $file =~ s/^!//;
412 if ($utf8::why_deprecated{$file}) {
413 warnings::warnif('deprecated', "Use of '$type' in \\p{} or \\P{} is deprecated because: $utf8::why_deprecated{$file};");
417 && exists $utf8::caseless_equivalent{$property_and_table})
419 $file = $utf8::caseless_equivalent{$property_and_table};
421 $file= "$unicore_dir/lib/$file.pl";
424 print STDERR __LINE__, ": didn't find $property_and_table\n" if DEBUG;
427 ## Last attempt -- see if it's a standard "To" name
428 ## (e.g. "ToLower") ToTitle is used by ucfirst().
429 ## The user-level way to access ToDigit() and ToFold()
430 ## is to use Unicode::UCD.
432 # Only check if caller wants non-binary
434 if ($minbits != 1 && $property_and_table =~ s/^to//) {{
435 # Look input up in list of properties for which we have
438 $utf8::loose_property_to_file_of{$property_and_table}))
440 $type = $utf8::file_to_swash_name{$file};
441 print STDERR __LINE__, ": type set to $type\n" if DEBUG;
442 $file = "$unicore_dir/$file.pl";
444 } # If that fails see if there is a corresponding binary
446 elsif (defined ($file =
447 $utf8::loose_to_file_of{$property_and_table}))
450 # Here, there is no map file for the property we are
451 # trying to get the map of, but this is a binary
452 # property, and there is a file for it that can easily
453 # be translated to a mapping.
455 # In the case of properties that are forced to binary,
456 # they are a combination. We return the actual
457 # mapping instead of the binary. If the input is
458 # something like 'Tocjkkiicore', it will be found in
459 # %loose_property_to_file_of above as => 'To/kIICore'.
460 # But the form like ToIskiicore won't be. To fix
461 # this, it was easiest to do it here. These
462 # properties are the complements of the default
463 # property, so there is an entry in %loose_to_file_of
464 # that is 'iskiicore' => '!kIICore/N', If we find such
465 # an entry, strip off things and try again, which
466 # should find the entry in %loose_property_to_file_of.
467 # Actual binary properties that are of this form, such
468 # as this entry: 'ishrkt' => '!Perl/Any' will also be
469 # retried, but won't be in %loose_property_to_file_of,
470 # and instead the next time through, it will find
471 # 'hrkt' => '!Perl/Any' and proceed.
474 && $property_and_table =~ s/^is//;
476 # This is a binary property. Setting this here causes
477 # it to be stored as such in the cache, so if someone
478 # comes along later looking for just a binary, they
482 # The 0+ makes sure is numeric
483 $invert_it = 0 + $file =~ s/^!//;
484 $file = "$unicore_dir/lib/$file.pl";
490 ## If we reach this line, it's because we couldn't figure
491 ## out what to do with $type. Ouch.
494 pop @recursed if @recursed;
496 } # end of GETFILE block
499 print STDERR __LINE__, ": found it (file='$file')\n" if DEBUG;
502 ## If we reach here, it was due to a 'last GETFILE' above
503 ## (exception: user-defined properties and mappings), so we
504 ## have a filename, so now we load it if we haven't already.
505 ## If we have, return the cached results. The cache key is the
506 ## class and file to load, and whether the results need to be
509 my $found = $Cache{$class, $file, $invert_it};
510 if ($found and ref($found) eq $class) {
511 print STDERR __LINE__, ": Returning cached swash for '$class,$file,$invert_it' for \\p{$type}\n" if DEBUG;
512 pop @recursed if @recursed;
518 $list = do $file; die $@ if $@;
519 $list_is_from_mktables = 1;
521 } # End of $type is non-null
523 # Here, either $type was null, or we found the requested property and
530 # mktables lists don't have extras, like '&utf8::prop', so don't need
531 # to separate them; also lists are already sorted, so don't need to do
533 if ($list && ! $list_is_from_mktables) {
534 my $taint = substr($list,0,0); # maintain taint
536 # Separate the extras from the code point list, and for
537 # user-defined properties, make sure the latter are well-behaved
538 # for downstream code.
540 my @tmp = split(/^/m, $list);
544 # The extras are anything that doesn't begin with a hex digit.
545 $extras = join '', $taint, grep /^[^0-9a-fA-F]/, @tmp;
547 # Remove the extras, and sort the remaining entries by the
548 # numeric value of their beginning hex digits, removing any
550 $list = join '', $taint,
552 sort { $a->[0] <=> $b->[0] }
553 map { /^([0-9a-fA-F]+)/; [ CORE::hex($1), $_ ] }
554 grep { /^([0-9a-fA-F]+)/ and not $seen{$1}++ } @tmp; # XXX doesn't do ranges right
557 # mktables has gone to some trouble to make non-user defined
558 # properties well-behaved, so we can skip the effort we do for
559 # user-defined ones. Any extras are at the very beginning of
562 # This regex splits out the first lines of $list into $1 and
563 # strips them off from $list, until we get one that begins
564 # with a hex number, alone on the line, or followed by a tab.
565 # Either portion may be empty.
566 $list =~ s/ \A ( .*? )
567 (?: \z | (?= ^ [0-9a-fA-F]+ (?: \t | $) ) )
570 $extras = "$taint$1";
575 my $hextra = sprintf "%04x", $none + 1;
576 $list =~ s/\tXXXX$/\t$hextra/mg;
579 if ($minbits != 1 && $minbits < 32) { # not binary property
581 while ($list =~ /^([0-9a-fA-F]+)(?:[\t]([0-9a-fA-F]+)?)(?:[ \t]([0-9a-fA-F]+))?/mg) {
582 my $min = CORE::hex $1;
583 my $max = defined $2 ? CORE::hex $2 : $min;
584 my $val = defined $3 ? CORE::hex $3 : 0;
585 $val += $max - $min if defined $3;
586 $top = $val if $val > $top;
590 $top > 0xff ? 16 : 8;
591 $bits = $topbits if $bits < $topbits;
596 for my $x ($extras) {
597 my $taint = substr($x,0,0); # maintain taint
599 while ($x =~ /^([^0-9a-fA-F\n])(.*)/mg) {
600 my $char = "$1$taint";
601 my $name = "$2$taint";
602 print STDERR __LINE__, ": char [$char] => name [$name]\n"
604 if ($char =~ /[-+!&]/) {
605 my ($c,$t) = split(/::/, $name, 2); # bogus use of ::, really
608 $subobj = utf8->SWASHNEW($t, "", $minbits, 0);
610 elsif (exists &$name) {
611 $subobj = utf8->SWASHNEW($name, "", $minbits, 0);
613 elsif ($c =~ /^([0-9a-fA-F]+)/) {
614 $subobj = utf8->SWASHNEW("", $c, $minbits, 0);
616 print STDERR __LINE__, ": returned from getting sub object for $name\n" if DEBUG;
618 pop @recursed if @recursed && $type;
621 push @extras, $name => $subobj;
622 $bits = $subobj->{BITS} if $bits < $subobj->{BITS};
623 $user_defined = $subobj->{USER_DEFINED}
624 if $subobj->{USER_DEFINED};
631 print STDERR __LINE__, ": CLASS = $class, TYPE => $type, BITS => $bits, NONE => $none, INVERT_IT => $invert_it, USER_DEFINED => $user_defined";
632 print STDERR "\nLIST =>\n$list" if defined $list;
633 print STDERR "\nEXTRAS =>\n$extras" if defined $extras;
643 USER_DEFINED => $user_defined,
648 $Cache{$class, $file, $invert_it} = $SWASH;
650 && exists $utf8::SwashInfo{$type}
651 && exists $utf8::SwashInfo{$type}{'specials_name'})
653 my $specials_name = $utf8::SwashInfo{$type}{'specials_name'};
655 print STDERR "\nspecials_name => $specials_name\n" if DEBUG;
656 $SWASH->{'SPECIALS'} = \%$specials_name;
658 $SWASH->{'INVERT_IT'} = $invert_it;
661 pop @recursed if @recursed && $type;
667 # Now SWASHGET is recasted into a C function S_swatch_get (see utf8.c).