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.
102 # Change this to get a different set of Unicode tables
103 my $unicore_dir = 'unicore';
109 # Verify that this isn't a recursive call for this property.
110 # Can't use croak, as it may try to recurse here itself.
111 my $class_type = $class . "::$type";
112 if (grep { $_ eq $class_type } @recursed) {
113 CORE::die "panic: Infinite recursion in SWASHNEW for '$type'\n";
115 push @recursed, $class_type;
120 # regcomp.c surrounds the property name with '__" and '_i' if this
121 # is to be caseless matching.
122 my $caseless = $type =~ s/^__(.*)_i$/$1/;
124 print STDERR __LINE__, ": type=$type, caseless=$caseless\n" if DEBUG;
129 ## It could be a user-defined property. Look in current
130 ## package if no package given
133 my $caller1 = $type =~ s/(.+)::// ? $1 : caller(1);
135 if (defined $caller1 && $type =~ /^I[ns]\w+$/) {
136 my $prop = "${caller1}::$type";
137 if (exists &{$prop}) {
138 # stolen from Scalar::Util::PP::tainted()
141 local($@, $SIG{__DIE__}, $SIG{__WARN__});
144 eval { kill 0 * $prop };
145 $tainted = 1 if $@ =~ /^Insecure/;
147 die "Insecure user-defined property \\p{$prop}\n"
150 $list = &{$prop}($caseless);
156 # During Perl's compilation, this routine may be called before
157 # the tables are constructed. If so, we have a chicken/egg
158 # problem. If we die, the tables never get constructed, so
159 # keep going, but return an empty table so only what the code
160 # has compiled in internally (currently ASCII/Latin1 range
161 # matching) will work.
163 # Poor man's constant, to avoid a run-time check.
165 = \! defined &DynaLoader::boot_DynaLoader;
168 eval "require '$unicore_dir/Heavy.pl'";
172 require "$unicore_dir/Heavy.pl";
174 BEGIN { delete $utf8::{miniperl} }
176 # All property names are matched caselessly
177 my $property_and_table = CORE::lc $type;
178 print STDERR __LINE__, ": $property_and_table\n" if DEBUG;
180 # See if is of the compound form 'property=value', where the
181 # value indicates the table we should use.
182 my ($property, $table, @remainder) =
183 split /\s*[:=]\s*/, $property_and_table, -1;
185 pop @recursed if @recursed;
190 if (! defined $table) {
192 # Here, is the single form. The property becomes empty, and
193 # the whole value is the table.
195 $prefix = $property = "";
197 print STDERR __LINE__, ": $property\n" if DEBUG;
199 # Here it is the compound property=table form. The property
200 # name is always loosely matched, and always can have an
201 # optional 'is' prefix (which isn't true in the single
203 $property = _loose_name($property) =~ s/^is//r;
205 # And convert to canonical form. Quit if not valid.
206 $property = $utf8::loose_property_name_of{$property};
207 if (! defined $property) {
208 pop @recursed if @recursed;
212 $prefix = "$property=";
214 # If the rhs looks like it is a number...
215 print STDERR __LINE__, ": table=$table\n" if DEBUG;
216 if ($table =~ qr{ ^ [ \s 0-9 _ + / . -]+ $ }x) {
217 print STDERR __LINE__, ": table=$table\n" if DEBUG;
219 # Don't allow leading nor trailing slashes
220 if ($table =~ / ^ \/ | \/ $ /x) {
221 pop @recursed if @recursed;
225 # Split on slash, in case it is a rational, like \p{1/5}
226 my @parts = split qr{ \s* / \s* }x, $table, -1;
227 print __LINE__, ": $type\n" if @parts > 2 && DEBUG;
229 # Can have maximum of one slash
231 pop @recursed if @recursed;
235 foreach my $part (@parts) {
236 print __LINE__, ": part=$part\n" if DEBUG;
238 $part =~ s/^\+\s*//; # Remove leading plus
239 $part =~ s/^-\s*/-/; # Remove blanks after unary
242 # Remove underscores between digits.
243 $part =~ s/( ?<= [0-9] ) _ (?= [0-9] ) //xg;
245 # No leading zeros (but don't make a single '0'
246 # into a null string)
247 $part =~ s/ ^ ( -? ) 0+ /$1/x;
248 $part .= '0' if $part eq '-' || $part eq "";
250 # No trailing zeros after a decimal point
251 $part =~ s/ ( \. .*? ) 0+ $ /$1/x;
253 # Begin with a 0 if a leading decimal point
254 $part =~ s/ ^ ( -? ) \. /${1}0./x;
256 # Ensure not a trailing decimal point: turn into an
258 $part =~ s/ \. $ //x;
260 print STDERR __LINE__, ": part=$part\n" if DEBUG;
261 #return $type if $part eq "";
263 # Result better look like a number. (This test is
264 # needed because, for example could have a plus in
266 if ($part !~ / ^ -? [0-9]+ ( \. [0-9]+)? $ /x) {
267 pop @recursed if @recursed;
275 # If denominator is negative, get rid of it, and ...
276 if ($parts[1] =~ s/^-//) {
278 # If numerator is also negative, convert the
279 # whole thing to positive, or move the minus to
281 if ($parts[0] !~ s/^-//) {
282 $parts[0] = '-' . $parts[0];
285 $table = join '/', @parts;
287 elsif ($property ne 'nv' || $parts[0] !~ /\./) {
289 # Here is not numeric value, or doesn't have a
290 # decimal point. No further manipulation is
291 # necessary. (Note the hard-coded property name.
292 # This could fail if other properties eventually
293 # had fractions as well; perhaps the cjk ones
294 # could evolve to do that. This hard-coding could
295 # be fixed by mktables generating a list of
296 # properties that could have fractions.)
300 # Here is a floating point numeric_value. Try to
301 # convert to rational. First see if is in the list
303 if (exists $utf8::nv_floating_to_rational{$parts[0]}) {
304 $table = $utf8::nv_floating_to_rational{$parts[0]};
307 # Here not in the list. See if is close
308 # enough to something in the list. First
309 # determine what 'close enough' means. It has
310 # to be as tight as what mktables says is the
311 # maximum slop, and as tight as how many
312 # digits we were passed. That is, if the user
313 # said .667, .6667, .66667, etc. we match as
314 # many digits as they passed until get to
315 # where it doesn't matter any more due to the
316 # machine's precision. If they said .6666668,
318 (my $fraction = $parts[0]) =~ s/^.*\.//;
319 my $epsilon = 10 ** - (length($fraction));
320 if ($epsilon > $utf8::max_floating_slop) {
321 $epsilon = $utf8::max_floating_slop;
324 # But it can't be tighter than the minimum
325 # precision for this machine. If haven't
326 # already calculated that minimum, do so now.
327 if (! defined $min_floating_slop) {
329 # Keep going down an order of magnitude
330 # until find that adding this quantity to
331 # 1 remains 1; but put an upper limit on
332 # this so in case this algorithm doesn't
333 # work properly on some platform, that we
334 # won't loop forever.
336 $min_floating_slop = 1;
337 while (1+ $min_floating_slop != 1
340 my $next = $min_floating_slop / 10;
341 last if $next == 0; # If underflows,
343 $min_floating_slop = $next;
344 print STDERR __LINE__, ": min_float_slop=$min_floating_slop\n" if DEBUG;
347 # Back off a couple orders of magnitude,
349 $min_floating_slop *= 100;
352 if ($epsilon < $min_floating_slop) {
353 $epsilon = $min_floating_slop;
355 print STDERR __LINE__, ": fraction=.$fraction; epsilon=$epsilon\n" if DEBUG;
359 # And for each possible rational in the table,
360 # see if it is within epsilon of the input.
362 (keys %utf8::nv_floating_to_rational)
364 print STDERR __LINE__, ": epsilon=$epsilon, official=$official, diff=", abs($parts[0] - $official), "\n" if DEBUG;
365 if (abs($parts[0] - $official) < $epsilon) {
367 $utf8::nv_floating_to_rational{$official};
372 # Quit if didn't find one.
373 if (! defined $table) {
374 pop @recursed if @recursed;
379 print STDERR __LINE__, ": $property=$table\n" if DEBUG;
383 # Combine lhs (if any) and rhs to get something that matches
384 # the syntax of the lookups.
385 $property_and_table = "$prefix$table";
386 print STDERR __LINE__, ": $property_and_table\n" if DEBUG;
388 # First try stricter matching.
389 $file = $utf8::stricter_to_file_of{$property_and_table};
391 # If didn't find it, try again with looser matching by editing
392 # out the applicable characters on the rhs and looking up
394 if (! defined $file) {
395 $table = _loose_name($table);
396 $property_and_table = "$prefix$table";
397 print STDERR __LINE__, ": $property_and_table\n" if DEBUG;
398 $file = $utf8::loose_to_file_of{$property_and_table};
401 # Add the constant and go fetch it in.
404 # A beginning ! means to invert
405 $invert_it = $file =~ s/^!//;
407 if ($utf8::why_deprecated{$file}) {
408 warnings::warnif('deprecated', "Use of '$type' in \\p{} or \\P{} is deprecated because: $utf8::why_deprecated{$file};");
412 && exists $utf8::caseless_equivalent{$property_and_table})
414 $file = $utf8::caseless_equivalent{$property_and_table};
416 $file= "$unicore_dir/lib/$file.pl";
419 print STDERR __LINE__, ": didn't find $property_and_table\n" if DEBUG;
422 ## Last attempt -- see if it's a standard "To" name
423 ## (e.g. "ToLower") ToTitle is used by ucfirst().
424 ## The user-level way to access ToDigit() and ToFold()
425 ## is to use Unicode::UCD.
427 # Only check if caller wants non-binary
429 if ($minbits != 1 && $property_and_table =~ s/^to//) {{
430 # Look input up in list of properties for which we have
433 $utf8::loose_property_to_file_of{$property_and_table}))
435 $type = $utf8::file_to_swash_name{$file};
436 print STDERR __LINE__, ": type set to $type\n" if DEBUG;
437 $file = "$unicore_dir/$file.pl";
439 } # If that fails see if there is a corresponding binary
441 elsif (defined ($file =
442 $utf8::loose_to_file_of{$property_and_table}))
445 # Here, there is no map file for the property we are
446 # trying to get the map of, but this is a binary
447 # property, and there is a file for it that can easily
448 # be translated to a mapping.
450 # In the case of properties that are forced to binary,
451 # they are a combination. We return the actual
452 # mapping instead of the binary. If the input is
453 # something like 'Tocjkkiicore', it will be found in
454 # %loose_property_to_file_of above as => 'To/kIICore'.
455 # But the form like ToIskiicore won't be. To fix
456 # this, it was easiest to do it here. These
457 # properties are the complements of the default
458 # property, so there is an entry in %loose_to_file_of
459 # that is 'iskiicore' => '!kIICore/N', If we find such
460 # an entry, strip off things and try again, which
461 # should find the entry in %loose_property_to_file_of.
462 # Actual binary properties that are of this form, such
463 # as this entry: 'ishrkt' => '!Perl/Any' will also be
464 # retried, but won't be in %loose_property_to_file_of,
465 # and instead the next time through, it will find
466 # 'hrkt' => '!Perl/Any' and proceed.
469 && $property_and_table =~ s/^is//;
471 # This is a binary property. Setting this here causes
472 # it to be stored as such in the cache, so if someone
473 # comes along later looking for just a binary, they
477 $invert_it = $file =~ s/^!//;
478 $file = "$unicore_dir/lib/$file.pl";
484 ## If we reach this line, it's because we couldn't figure
485 ## out what to do with $type. Ouch.
488 pop @recursed if @recursed;
490 } # end of GETFILE block
493 print STDERR __LINE__, ": found it (file='$file')\n" if DEBUG;
496 ## If we reach here, it was due to a 'last GETFILE' above
497 ## (exception: user-defined properties and mappings), so we
498 ## have a filename, so now we load it if we haven't already.
499 ## If we have, return the cached results. The cache key is the
500 ## class and file to load.
502 my $found = $Cache{$class, $file};
503 if ($found and ref($found) eq $class) {
504 print STDERR __LINE__, ": Returning cached '$file' for \\p{$type}; invert_it=$invert_it\n" if DEBUG;
505 pop @recursed if @recursed;
506 $found->{'INVERT_IT'} = $invert_it;
512 $list = do $file; die $@ if $@;
515 $ListSorted = 1; ## we know that these lists are sorted
516 } # End of $type is non-null
518 # Here, either $type was null, or we found the requested property and
526 my $taint = substr($list,0,0); # maintain taint
527 my @tmp = split(/^/m, $list);
530 $extras = join '', $taint, grep /^[^0-9a-fA-F]/, @tmp;
531 $list = join '', $taint,
533 sort { $a->[0] <=> $b->[0] }
534 map { /^([0-9a-fA-F]+)/; [ CORE::hex($1), $_ ] }
535 grep { /^([0-9a-fA-F]+)/ and not $seen{$1}++ } @tmp; # XXX doesn't do ranges right
539 my $hextra = sprintf "%04x", $none + 1;
540 $list =~ s/\tXXXX$/\t$hextra/mg;
543 if ($minbits != 1 && $minbits < 32) { # not binary property
545 while ($list =~ /^([0-9a-fA-F]+)(?:[\t]([0-9a-fA-F]+)?)(?:[ \t]([0-9a-fA-F]+))?/mg) {
546 my $min = CORE::hex $1;
547 my $max = defined $2 ? CORE::hex $2 : $min;
548 my $val = defined $3 ? CORE::hex $3 : 0;
549 $val += $max - $min if defined $3;
550 $top = $val if $val > $top;
554 $top > 0xff ? 16 : 8;
555 $bits = $topbits if $bits < $topbits;
560 for my $x ($extras) {
561 my $taint = substr($x,0,0); # maintain taint
563 while ($x =~ /^([^0-9a-fA-F\n])(.*)/mg) {
564 my $char = "$1$taint";
565 my $name = "$2$taint";
566 print STDERR __LINE__, ": char [$char] => name [$name]\n"
568 if ($char =~ /[-+!&]/) {
569 my ($c,$t) = split(/::/, $name, 2); # bogus use of ::, really
572 $subobj = utf8->SWASHNEW($t, "", $minbits, 0);
574 elsif (exists &$name) {
575 $subobj = utf8->SWASHNEW($name, "", $minbits, 0);
577 elsif ($c =~ /^([0-9a-fA-F]+)/) {
578 $subobj = utf8->SWASHNEW("", $c, $minbits, 0);
581 pop @recursed if @recursed && $type;
584 push @extras, $name => $subobj;
585 $bits = $subobj->{BITS} if $bits < $subobj->{BITS};
586 $user_defined = $subobj->{USER_DEFINED}
587 if $subobj->{USER_DEFINED};
594 print STDERR __LINE__, ": CLASS = $class, TYPE => $type, BITS => $bits, NONE => $none, INVERT_IT => $invert_it, USER_DEFINED => $user_defined";
595 print STDERR "\nLIST =>\n$list" if defined $list;
596 print STDERR "\nEXTRAS =>\n$extras" if defined $extras;
606 USER_DEFINED => $user_defined,
611 $Cache{$class, $file} = $SWASH;
613 && exists $utf8::SwashInfo{$type}
614 && exists $utf8::SwashInfo{$type}{'specials_name'})
616 my $specials_name = $utf8::SwashInfo{$type}{'specials_name'};
618 print STDERR "\nspecials_name => $specials_name\n" if DEBUG;
619 $SWASH->{'SPECIALS'} = \%$specials_name;
621 $SWASH->{'INVERT_IT'} = $invert_it;
624 pop @recursed if @recursed && $type;
630 # Now SWASHGET is recasted into a C function S_swash_get (see utf8.c).