11 sub croak { require Carp; Carp::croak(@_) }
14 ## "SWASH" == "SWATCH HASH". A "swatch" is a swatch of the Unicode landscape.
15 ## It's a data structure that encodes a set of Unicode characters.
19 # If a floating point number is within this distance from the value of a
20 # fraction, it is considered to be that fraction, even if many more digits
21 # are specified that don't exactly match.
22 my $min_floating_slop;
24 # To guard against this program calling something that in turn ends up
25 # calling this program with the same inputs, and hence infinitely
26 # recursing, we keep a stack of the properties that are currently in
27 # progress, pushed upon entry, popped upon return.
31 my ($class, $type, $list, $minbits, $none) = @_;
34 $class = "" unless defined $class;
35 print STDERR __LINE__, ": class=$class, type=$type, list=",
36 (defined $list) ? $list : ':undef:',
37 ", minbits=$minbits, none=$none\n" if DEBUG;
40 ## Get the list of codepoints for the type.
41 ## Called from swash_init (see utf8.c) or SWASHNEW itself.
43 ## Callers of swash_init:
44 ## op.c:pmtrans -- for tr/// and y///
45 ## regexec.c:regclass_swash -- for /[]/, \p, and \P
46 ## utf8.c:is_utf8_common -- for common Unicode properties
47 ## utf8.c:to_utf8_case -- for lc, uc, ucfirst, etc. and //i
49 ## Given a $type, our goal is to fill $list with the set of codepoint
50 ## ranges. If $type is false, $list passed is used.
53 ## For binary properties, $minbits must be 1.
54 ## For character mappings (case and transliteration), $minbits must
55 ## be a number except 1.
57 ## $list (or that filled according to $type):
58 ## Refer to perlunicode.pod, "User-Defined Character Properties."
60 ## For binary properties, only characters with the property value
61 ## of True should be listed. The 3rd column, if any, will be ignored
63 ## $none is undocumented, so I'm (khw) trying to do some documentation
64 ## of it now. It appears to be if there is a mapping in an input file
65 ## that maps to 'XXXX', then that is replaced by $none+1, expressed in
66 ## hexadecimal. It is used somehow in tr///.
68 ## To make the parsing of $type clear, this code takes the a rather
69 ## unorthodox approach of last'ing out of the block once we have the
70 ## info we need. Were this to be a subroutine, the 'last' would just
73 # If a problem is found $type is returned;
74 # Upon success, a new (or cached) blessed object is returned with
75 # keys TYPE, BITS, EXTRAS, LIST, and NONE with values having the
76 # same meanings as the input parameters.
77 # SPECIALS contains a reference to any special-treatment hash in the
78 # INVERT_IT is non-zero if the result should be inverted before use
79 my $file; ## file to load data from, and also part of the %Cache key.
82 # Change this to get a different set of Unicode tables
83 my $unicore_dir = 'unicore';
89 # Verify that this isn't a recursive call for this property.
90 # Can't use croak, as it may try to recurse here itself.
91 my $class_type = $class . "::$type";
92 if (grep { $_ eq $class_type } @recursed) {
93 CORE::die "panic: Infinite recursion in SWASHNEW for '$type'\n";
95 push @recursed, $class_type;
100 # regcomp.c surrounds the property name with '__" and '_i' if this
101 # is to be caseless matching.
102 my $caseless = $type =~ s/^__(.*)_i$/$1/;
104 print STDERR __LINE__, ": type=$type, caseless=$caseless\n" if DEBUG;
109 ## It could be a user-defined property. Look in current
110 ## package if no package given
113 my $caller1 = $type =~ s/(.+)::// ? $1 : caller(1);
115 if (defined $caller1 && $type =~ /^I[ns]\w+$/) {
116 my $prop = "${caller1}::$type";
117 if (exists &{$prop}) {
118 # stolen from Scalar::Util::PP::tainted()
121 local($@, $SIG{__DIE__}, $SIG{__WARN__});
124 eval { kill 0 * $prop };
125 $tainted = 1 if $@ =~ /^Insecure/;
127 die "Insecure user-defined property \\p{$prop}\n"
130 $list = &{$prop}($caseless);
135 # During Perl's compilation, this routine may be called before
136 # the tables are constructed. If so, we have a chicken/egg
137 # problem. If we die, the tables never get constructed, so
138 # keep going, but return an empty table so only what the code
139 # has compiled in internally (currently ASCII/Latin1 range
140 # matching) will work.
142 # Poor man's constant, to avoid a run-time check.
144 = \! defined &DynaLoader::boot_DynaLoader;
147 eval "require '$unicore_dir/Heavy.pl'";
151 require "$unicore_dir/Heavy.pl";
153 BEGIN { delete $utf8::{miniperl} }
155 # All property names are matched caselessly
156 my $property_and_table = CORE::lc $type;
157 print STDERR __LINE__, ": $property_and_table\n" if DEBUG;
159 # See if is of the compound form 'property=value', where the
160 # value indicates the table we should use.
161 my ($property, $table, @remainder) =
162 split /\s*[:=]\s*/, $property_and_table, -1;
164 pop @recursed if @recursed;
169 if (! defined $table) {
171 # Here, is the single form. The property becomes empty, and
172 # the whole value is the table.
174 $prefix = $property = "";
176 print STDERR __LINE__, ": $property\n" if DEBUG;
178 # Here it is the compound property=table form. The property
179 # name is always loosely matched, which means remove any of
181 $property =~ s/[_\s-]//g;
183 # And convert to canonical form. Quit if not valid.
184 $property = $utf8::loose_property_name_of{$property};
185 if (! defined $property) {
186 pop @recursed if @recursed;
190 $prefix = "$property=";
192 # If the rhs looks like it is a number...
193 print STDERR __LINE__, ": table=$table\n" if DEBUG;
194 if ($table =~ qr{ ^ [ \s 0-9 _ + / . -]+ $ }x) {
195 print STDERR __LINE__, ": table=$table\n" if DEBUG;
197 # Don't allow leading nor trailing slashes
198 if ($table =~ / ^ \/ | \/ $ /x) {
199 pop @recursed if @recursed;
203 # Split on slash, in case it is a rational, like \p{1/5}
204 my @parts = split qr{ \s* / \s* }x, $table, -1;
205 print __LINE__, ": $type\n" if @parts > 2 && DEBUG;
207 # Can have maximum of one slash
209 pop @recursed if @recursed;
213 foreach my $part (@parts) {
214 print __LINE__, ": part=$part\n" if DEBUG;
216 $part =~ s/^\+\s*//; # Remove leading plus
217 $part =~ s/^-\s*/-/; # Remove blanks after unary
220 # Remove underscores between digits.
221 $part =~ s/( ?<= [0-9] ) _ (?= [0-9] ) //xg;
223 # No leading zeros (but don't make a single '0'
224 # into a null string)
225 $part =~ s/ ^ ( -? ) 0+ /$1/x;
226 $part .= '0' if $part eq '-' || $part eq "";
228 # No trailing zeros after a decimal point
229 $part =~ s/ ( \. .*? ) 0+ $ /$1/x;
231 # Begin with a 0 if a leading decimal point
232 $part =~ s/ ^ ( -? ) \. /${1}0./x;
234 # Ensure not a trailing decimal point: turn into an
236 $part =~ s/ \. $ //x;
238 print STDERR __LINE__, ": part=$part\n" if DEBUG;
239 #return $type if $part eq "";
241 # Result better look like a number. (This test is
242 # needed because, for example could have a plus in
244 if ($part !~ / ^ -? [0-9]+ ( \. [0-9]+)? $ /x) {
245 pop @recursed if @recursed;
253 # If denominator is negative, get rid of it, and ...
254 if ($parts[1] =~ s/^-//) {
256 # If numerator is also negative, convert the
257 # whole thing to positive, or move the minus to
259 if ($parts[0] !~ s/^-//) {
260 $parts[0] = '-' . $parts[0];
263 $table = join '/', @parts;
265 elsif ($property ne 'nv' || $parts[0] !~ /\./) {
267 # Here is not numeric value, or doesn't have a
268 # decimal point. No further manipulation is
269 # necessary. (Note the hard-coded property name.
270 # This could fail if other properties eventually
271 # had fractions as well; perhaps the cjk ones
272 # could evolve to do that. This hard-coding could
273 # be fixed by mktables generating a list of
274 # properties that could have fractions.)
278 # Here is a floating point numeric_value. Try to
279 # convert to rational. First see if is in the list
281 if (exists $utf8::nv_floating_to_rational{$parts[0]}) {
282 $table = $utf8::nv_floating_to_rational{$parts[0]};
285 # Here not in the list. See if is close
286 # enough to something in the list. First
287 # determine what 'close enough' means. It has
288 # to be as tight as what mktables says is the
289 # maximum slop, and as tight as how many
290 # digits we were passed. That is, if the user
291 # said .667, .6667, .66667, etc. we match as
292 # many digits as they passed until get to
293 # where it doesn't matter any more due to the
294 # machine's precision. If they said .6666668,
296 (my $fraction = $parts[0]) =~ s/^.*\.//;
297 my $epsilon = 10 ** - (length($fraction));
298 if ($epsilon > $utf8::max_floating_slop) {
299 $epsilon = $utf8::max_floating_slop;
302 # But it can't be tighter than the minimum
303 # precision for this machine. If haven't
304 # already calculated that minimum, do so now.
305 if (! defined $min_floating_slop) {
307 # Keep going down an order of magnitude
308 # until find that adding this quantity to
309 # 1 remains 1; but put an upper limit on
310 # this so in case this algorithm doesn't
311 # work properly on some platform, that we
312 # won't loop forever.
314 $min_floating_slop = 1;
315 while (1+ $min_floating_slop != 1
318 my $next = $min_floating_slop / 10;
319 last if $next == 0; # If underflows,
321 $min_floating_slop = $next;
322 print STDERR __LINE__, ": min_float_slop=$min_floating_slop\n" if DEBUG;
325 # Back off a couple orders of magnitude,
327 $min_floating_slop *= 100;
330 if ($epsilon < $min_floating_slop) {
331 $epsilon = $min_floating_slop;
333 print STDERR __LINE__, ": fraction=.$fraction; epsilon=$epsilon\n" if DEBUG;
337 # And for each possible rational in the table,
338 # see if it is within epsilon of the input.
340 (keys %utf8::nv_floating_to_rational)
342 print STDERR __LINE__, ": epsilon=$epsilon, official=$official, diff=", abs($parts[0] - $official), "\n" if DEBUG;
343 if (abs($parts[0] - $official) < $epsilon) {
345 $utf8::nv_floating_to_rational{$official};
350 # Quit if didn't find one.
351 if (! defined $table) {
352 pop @recursed if @recursed;
357 print STDERR __LINE__, ": $property=$table\n" if DEBUG;
361 # Combine lhs (if any) and rhs to get something that matches
362 # the syntax of the lookups.
363 $property_and_table = "$prefix$table";
364 print STDERR __LINE__, ": $property_and_table\n" if DEBUG;
366 # First try stricter matching.
367 $file = $utf8::stricter_to_file_of{$property_and_table};
369 # If didn't find it, try again with looser matching by editing
370 # out the applicable characters on the rhs and looking up
372 if (! defined $file) {
373 $table =~ s/ [_\s-] //xg;
374 $property_and_table = "$prefix$table";
375 print STDERR __LINE__, ": $property_and_table\n" if DEBUG;
376 $file = $utf8::loose_to_file_of{$property_and_table};
379 # Add the constant and go fetch it in.
382 # A beginning ! means to invert
383 $invert_it = $file =~ s/^!//;
385 if ($utf8::why_deprecated{$file}) {
386 warnings::warnif('deprecated', "Use of '$type' in \\p{} or \\P{} is deprecated because: $utf8::why_deprecated{$file};");
390 && exists $utf8::caseless_equivalent{$property_and_table})
392 $file = $utf8::caseless_equivalent{$property_and_table};
394 $file= "$unicore_dir/lib/$file.pl";
397 print STDERR __LINE__, ": didn't find $property_and_table\n" if DEBUG;
400 ## See if it's a user-level "To".
403 my $caller0 = caller(0);
405 if (defined $caller0 && $type =~ /^To(?:\w+)$/) {
406 my $map = $caller0 . "::" . $type;
408 if (exists &{$map}) {
412 warnings::warnif('deprecated', "User-defined case-mapping '$type' is deprecated");
418 ## Last attempt -- see if it's a standard "To" name
419 ## (e.g. "ToLower") ToTitle is used by ucfirst().
420 ## The user-level way to access ToDigit() and ToFold()
421 ## is to use Unicode::UCD.
423 if ($type =~ /^To(Digit|Fold|Lower|Title|Upper)$/) {
425 # Fail if wanting a binary property, as these aren't.
427 pop @recursed if @recursed;
430 $file = "$unicore_dir/To/$1.pl";
431 ## would like to test to see if $file actually exists....
436 ## If we reach this line, it's because we couldn't figure
437 ## out what to do with $type. Ouch.
440 pop @recursed if @recursed;
445 print STDERR __LINE__, ": found it (file='$file')\n" if DEBUG;
448 ## If we reach here, it was due to a 'last GETFILE' above
449 ## (exception: user-defined properties and mappings), so we
450 ## have a filename, so now we load it if we haven't already.
451 ## If we have, return the cached results. The cache key is the
452 ## class and file to load.
454 my $found = $Cache{$class, $file};
455 if ($found and ref($found) eq $class) {
456 print STDERR __LINE__, ": Returning cached '$file' for \\p{$type}; invert_it=$invert_it\n" if DEBUG;
457 pop @recursed if @recursed;
458 $found->{'INVERT_IT'} = $invert_it;
464 $list = do $file; die $@ if $@;
467 $ListSorted = 1; ## we know that these lists are sorted
474 my $taint = substr($list,0,0); # maintain taint
475 my @tmp = split(/^/m, $list);
478 $extras = join '', $taint, grep /^[^0-9a-fA-F]/, @tmp;
479 $list = join '', $taint,
481 sort { $a->[0] <=> $b->[0] }
482 map { /^([0-9a-fA-F]+)/; [ CORE::hex($1), $_ ] }
483 grep { /^([0-9a-fA-F]+)/ and not $seen{$1}++ } @tmp; # XXX doesn't do ranges right
487 my $hextra = sprintf "%04x", $none + 1;
488 $list =~ s/\tXXXX$/\t$hextra/mg;
491 if ($minbits != 1 && $minbits < 32) { # not binary property
493 while ($list =~ /^([0-9a-fA-F]+)(?:[\t]([0-9a-fA-F]+)?)(?:[ \t]([0-9a-fA-F]+))?/mg) {
494 my $min = CORE::hex $1;
495 my $max = defined $2 ? CORE::hex $2 : $min;
496 my $val = defined $3 ? CORE::hex $3 : 0;
497 $val += $max - $min if defined $3;
498 $top = $val if $val > $top;
502 $top > 0xff ? 16 : 8;
503 $bits = $topbits if $bits < $topbits;
508 for my $x ($extras) {
509 my $taint = substr($x,0,0); # maintain taint
511 while ($x =~ /^([^0-9a-fA-F\n])(.*)/mg) {
512 my $char = "$1$taint";
513 my $name = "$2$taint";
514 print STDERR __LINE__, ": char [$char] => name [$name]\n"
516 if ($char =~ /[-+!&]/) {
517 my ($c,$t) = split(/::/, $name, 2); # bogus use of ::, really
520 $subobj = utf8->SWASHNEW($t, "", $minbits, 0);
522 elsif (exists &$name) {
523 $subobj = utf8->SWASHNEW($name, "", $minbits, 0);
525 elsif ($c =~ /^([0-9a-fA-F]+)/) {
526 $subobj = utf8->SWASHNEW("", $c, $minbits, 0);
529 pop @recursed if @recursed && $type;
532 push @extras, $name => $subobj;
533 $bits = $subobj->{BITS} if $bits < $subobj->{BITS};
540 print STDERR __LINE__, ": CLASS = $class, TYPE => $type, BITS => $bits, NONE => $none, INVERT_IT => $invert_it";
541 print STDERR "\nLIST =>\n$list" if defined $list;
542 print STDERR "\nEXTRAS =>\n$extras" if defined $extras;
556 $Cache{$class, $file} = $SWASH;
558 && exists $utf8::SwashInfo{$type}
559 && exists $utf8::SwashInfo{$type}{'specials_name'})
561 my $specials_name = $utf8::SwashInfo{$type}{'specials_name'};
563 print STDERR "\nspecials_name => $SWASH->{'SPECIALS'}\n" if DEBUG;
564 $SWASH->{'SPECIALS'} = \%$specials_name;
566 $SWASH->{'INVERT_IT'} = $invert_it;
569 pop @recursed if @recursed && $type;
575 # Now SWASHGET is recasted into a C function S_swash_get (see utf8.c).