Commit | Line | Data |
---|---|---|
a0ed51b3 | 1 | package utf8; |
cf25bb62 JH |
2 | use strict; |
3 | use warnings; | |
2783468b KW |
4 | use re "/aa"; # So we won't even try to look at above Latin1, potentially |
5 | # resulting in a recursive call | |
a0ed51b3 | 6 | |
15732964 | 7 | sub DEBUG () { 0 } |
4b17cb47 | 8 | $|=1 if DEBUG; |
a0ed51b3 LW |
9 | |
10 | sub DESTROY {} | |
11 | ||
5beb625e JH |
12 | my %Cache; |
13 | ||
86916d66 JH |
14 | sub croak { require Carp; Carp::croak(@_) } |
15 | ||
c4ab2516 | 16 | sub _loose_name ($) { |
45376db6 KW |
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. | |
22 | ||
23 | my $loose = $_[0] =~ s/[-\s_]//rg; | |
24 | ||
cf2cd801 | 25 | return $loose if $loose !~ / ^ (?: is | to )? l $/x; |
45376db6 KW |
26 | return 'l_' if $_[0] =~ / l .* _ /x; # If original had a trailing '_' |
27 | return $loose; | |
28 | } | |
29 | ||
5beb625e | 30 | ## |
3a2263fe RGS |
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. | |
5beb625e JH |
33 | ## |
34 | ||
99870f4d KW |
35 | { |
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 $min_floating_slop; | |
40 | ||
3dd10fe8 KW |
41 | # To guard against this program calling something that in turn ends up |
42 | # calling this program with the same inputs, and hence infinitely | |
43 | # recursing, we keep a stack of the properties that are currently in | |
44 | # progress, pushed upon entry, popped upon return. | |
45 | my @recursed; | |
46 | ||
99870f4d KW |
47 | sub SWASHNEW { |
48 | my ($class, $type, $list, $minbits, $none) = @_; | |
0fada5e1 | 49 | my $user_defined = 0; |
99870f4d KW |
50 | local $^D = 0 if $^D; |
51 | ||
3dd10fe8 | 52 | $class = "" unless defined $class; |
5bcf08c4 KW |
53 | print STDERR __LINE__, ": class=$class, type=$type, list=", |
54 | (defined $list) ? $list : ':undef:', | |
55 | ", minbits=$minbits, none=$none\n" if DEBUG; | |
99870f4d KW |
56 | |
57 | ## | |
58 | ## Get the list of codepoints for the type. | |
59 | ## Called from swash_init (see utf8.c) or SWASHNEW itself. | |
60 | ## | |
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 | |
681d705c | 66 | ## Unicode::UCD::prop_invlist |
62b3b855 | 67 | ## Unicode::UCD::prop_invmap |
99870f4d KW |
68 | ## |
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. | |
71 | ## | |
72 | ## $minbits: | |
73 | ## For binary properties, $minbits must be 1. | |
74 | ## For character mappings (case and transliteration), $minbits must | |
75 | ## be a number except 1. | |
76 | ## | |
77 | ## $list (or that filled according to $type): | |
78 | ## Refer to perlunicode.pod, "User-Defined Character Properties." | |
79 | ## | |
80 | ## For binary properties, only characters with the property value | |
81 | ## of True should be listed. The 3rd column, if any, will be ignored | |
82 | ## | |
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 | |
83fe8199 | 86 | ## hexadecimal. It is used somehow in tr///. |
99870f4d KW |
87 | ## |
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 | |
91 | ## be a 'return'. | |
92 | ## | |
fbe6f3cf KW |
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. | |
a6517ebf KW |
97 | # SPECIALS contains a reference to any special-treatment hash in the |
98 | # INVERT_IT is non-zero if the result should be inverted before use | |
0fada5e1 KW |
99 | # USER_DEFINED is non-zero if the result came from a user-defined |
100 | # property. | |
99870f4d | 101 | my $file; ## file to load data from, and also part of the %Cache key. |
99870f4d KW |
102 | |
103 | # Change this to get a different set of Unicode tables | |
104 | my $unicore_dir = 'unicore'; | |
a6517ebf | 105 | my $invert_it = 0; |
7b953548 KW |
106 | my $list_is_from_mktables = 0; # Is $list returned from a mktables |
107 | # generated file? If so, we know it's | |
108 | # well behaved. | |
99870f4d KW |
109 | |
110 | if ($type) | |
cf25bb62 | 111 | { |
3dd10fe8 | 112 | # Verify that this isn't a recursive call for this property. |
dbe7a391 | 113 | # Can't use croak, as it may try to recurse to here itself. |
3dd10fe8 KW |
114 | my $class_type = $class . "::$type"; |
115 | if (grep { $_ eq $class_type } @recursed) { | |
116 | CORE::die "panic: Infinite recursion in SWASHNEW for '$type'\n"; | |
117 | } | |
118 | push @recursed, $class_type; | |
119 | ||
99870f4d KW |
120 | $type =~ s/^\s+//; |
121 | $type =~ s/\s+$//; | |
122 | ||
4ff124d1 KW |
123 | # regcomp.c surrounds the property name with '__" and '_i' if this |
124 | # is to be caseless matching. | |
20b1a891 | 125 | my $caseless = $type =~ s/^(.*)__(.*)_i$/$1$2/; |
4ff124d1 KW |
126 | |
127 | print STDERR __LINE__, ": type=$type, caseless=$caseless\n" if DEBUG; | |
99870f4d KW |
128 | |
129 | GETFILE: | |
130 | { | |
131 | ## | |
a44f4b74 KW |
132 | ## It could be a user-defined property. Look in current |
133 | ## package if no package given | |
99870f4d KW |
134 | ## |
135 | ||
20b1a891 KW |
136 | |
137 | my $caller0 = caller(0); | |
138 | my $caller1 = $type =~ s/(.+)::// ? $1 : $caller0 eq 'main' ? | |
139 | 'main' : caller(1); | |
99870f4d | 140 | |
d658a8a8 | 141 | if (defined $caller1 && $type =~ /^I[ns]\w+$/) { |
99870f4d KW |
142 | my $prop = "${caller1}::$type"; |
143 | if (exists &{$prop}) { | |
0e9be77f DM |
144 | # stolen from Scalar::Util::PP::tainted() |
145 | my $tainted; | |
146 | { | |
147 | local($@, $SIG{__DIE__}, $SIG{__WARN__}); | |
148 | local $^W = 0; | |
149 | no warnings; | |
150 | eval { kill 0 * $prop }; | |
151 | $tainted = 1 if $@ =~ /^Insecure/; | |
152 | } | |
153 | die "Insecure user-defined property \\p{$prop}\n" | |
154 | if $tainted; | |
99870f4d | 155 | no strict 'refs'; |
4ff124d1 | 156 | $list = &{$prop}($caseless); |
0fada5e1 | 157 | $user_defined = 1; |
99870f4d KW |
158 | last GETFILE; |
159 | } | |
160 | } | |
161 | ||
a44f4b74 KW |
162 | # During Perl's compilation, this routine may be called before |
163 | # the tables are constructed. If so, we have a chicken/egg | |
164 | # problem. If we die, the tables never get constructed, so | |
165 | # keep going, but return an empty table so only what the code | |
166 | # has compiled in internally (currently ASCII/Latin1 range | |
167 | # matching) will work. | |
1363cbd0 FC |
168 | BEGIN { |
169 | # Poor man's constant, to avoid a run-time check. | |
170 | $utf8::{miniperl} | |
171 | = \! defined &DynaLoader::boot_DynaLoader; | |
172 | } | |
173 | if (miniperl) { | |
a44f4b74 KW |
174 | eval "require '$unicore_dir/Heavy.pl'"; |
175 | last GETFILE if $@; | |
176 | } | |
177 | else { | |
178 | require "$unicore_dir/Heavy.pl"; | |
179 | } | |
1363cbd0 | 180 | BEGIN { delete $utf8::{miniperl} } |
99870f4d | 181 | |
4ff124d1 | 182 | # All property names are matched caselessly |
49801ec2 | 183 | my $property_and_table = CORE::lc $type; |
99870f4d KW |
184 | print STDERR __LINE__, ": $property_and_table\n" if DEBUG; |
185 | ||
186 | # See if is of the compound form 'property=value', where the | |
187 | # value indicates the table we should use. | |
188 | my ($property, $table, @remainder) = | |
189 | split /\s*[:=]\s*/, $property_and_table, -1; | |
3dd10fe8 KW |
190 | if (@remainder) { |
191 | pop @recursed if @recursed; | |
192 | return $type; | |
193 | } | |
99870f4d KW |
194 | |
195 | my $prefix; | |
196 | if (! defined $table) { | |
197 | ||
198 | # Here, is the single form. The property becomes empty, and | |
199 | # the whole value is the table. | |
200 | $table = $property; | |
201 | $prefix = $property = ""; | |
202 | } else { | |
203 | print STDERR __LINE__, ": $property\n" if DEBUG; | |
204 | ||
205 | # Here it is the compound property=table form. The property | |
807807b7 KW |
206 | # name is always loosely matched, and always can have an |
207 | # optional 'is' prefix (which isn't true in the single | |
208 | # form). | |
c4ab2516 | 209 | $property = _loose_name($property) =~ s/^is//r; |
99870f4d KW |
210 | |
211 | # And convert to canonical form. Quit if not valid. | |
212 | $property = $utf8::loose_property_name_of{$property}; | |
3dd10fe8 KW |
213 | if (! defined $property) { |
214 | pop @recursed if @recursed; | |
215 | return $type; | |
216 | } | |
99870f4d KW |
217 | |
218 | $prefix = "$property="; | |
219 | ||
220 | # If the rhs looks like it is a number... | |
221 | print STDERR __LINE__, ": table=$table\n" if DEBUG; | |
222 | if ($table =~ qr{ ^ [ \s 0-9 _ + / . -]+ $ }x) { | |
223 | print STDERR __LINE__, ": table=$table\n" if DEBUG; | |
224 | ||
225 | # Don't allow leading nor trailing slashes | |
3dd10fe8 KW |
226 | if ($table =~ / ^ \/ | \/ $ /x) { |
227 | pop @recursed if @recursed; | |
228 | return $type; | |
229 | } | |
99870f4d KW |
230 | |
231 | # Split on slash, in case it is a rational, like \p{1/5} | |
232 | my @parts = split qr{ \s* / \s* }x, $table, -1; | |
233 | print __LINE__, ": $type\n" if @parts > 2 && DEBUG; | |
234 | ||
235 | # Can have maximum of one slash | |
3dd10fe8 KW |
236 | if (@parts > 2) { |
237 | pop @recursed if @recursed; | |
238 | return $type; | |
239 | } | |
99870f4d KW |
240 | |
241 | foreach my $part (@parts) { | |
242 | print __LINE__, ": part=$part\n" if DEBUG; | |
243 | ||
244 | $part =~ s/^\+\s*//; # Remove leading plus | |
245 | $part =~ s/^-\s*/-/; # Remove blanks after unary | |
246 | # minus | |
247 | ||
248 | # Remove underscores between digits. | |
eb568297 | 249 | $part =~ s/(?<= [0-9] ) _ (?= [0-9] ) //xg; |
99870f4d KW |
250 | |
251 | # No leading zeros (but don't make a single '0' | |
252 | # into a null string) | |
253 | $part =~ s/ ^ ( -? ) 0+ /$1/x; | |
254 | $part .= '0' if $part eq '-' || $part eq ""; | |
255 | ||
256 | # No trailing zeros after a decimal point | |
257 | $part =~ s/ ( \. .*? ) 0+ $ /$1/x; | |
258 | ||
259 | # Begin with a 0 if a leading decimal point | |
260 | $part =~ s/ ^ ( -? ) \. /${1}0./x; | |
261 | ||
262 | # Ensure not a trailing decimal point: turn into an | |
263 | # integer | |
264 | $part =~ s/ \. $ //x; | |
265 | ||
266 | print STDERR __LINE__, ": part=$part\n" if DEBUG; | |
267 | #return $type if $part eq ""; | |
268 | ||
269 | # Result better look like a number. (This test is | |
270 | # needed because, for example could have a plus in | |
271 | # the middle.) | |
3dd10fe8 KW |
272 | if ($part !~ / ^ -? [0-9]+ ( \. [0-9]+)? $ /x) { |
273 | pop @recursed if @recursed; | |
274 | return $type; | |
275 | } | |
99870f4d KW |
276 | } |
277 | ||
278 | # If a rational... | |
279 | if (@parts == 2) { | |
280 | ||
281 | # If denominator is negative, get rid of it, and ... | |
282 | if ($parts[1] =~ s/^-//) { | |
283 | ||
284 | # If numerator is also negative, convert the | |
285 | # whole thing to positive, or move the minus to | |
286 | # the numerator | |
287 | if ($parts[0] !~ s/^-//) { | |
288 | $parts[0] = '-' . $parts[0]; | |
289 | } | |
290 | } | |
291 | $table = join '/', @parts; | |
292 | } | |
293 | elsif ($property ne 'nv' || $parts[0] !~ /\./) { | |
294 | ||
295 | # Here is not numeric value, or doesn't have a | |
296 | # decimal point. No further manipulation is | |
297 | # necessary. (Note the hard-coded property name. | |
298 | # This could fail if other properties eventually | |
299 | # had fractions as well; perhaps the cjk ones | |
300 | # could evolve to do that. This hard-coding could | |
301 | # be fixed by mktables generating a list of | |
302 | # properties that could have fractions.) | |
303 | $table = $parts[0]; | |
304 | } else { | |
305 | ||
306 | # Here is a floating point numeric_value. Try to | |
307 | # convert to rational. First see if is in the list | |
308 | # of known ones. | |
309 | if (exists $utf8::nv_floating_to_rational{$parts[0]}) { | |
310 | $table = $utf8::nv_floating_to_rational{$parts[0]}; | |
311 | } else { | |
312 | ||
313 | # Here not in the list. See if is close | |
314 | # enough to something in the list. First | |
315 | # determine what 'close enough' means. It has | |
316 | # to be as tight as what mktables says is the | |
317 | # maximum slop, and as tight as how many | |
318 | # digits we were passed. That is, if the user | |
319 | # said .667, .6667, .66667, etc. we match as | |
320 | # many digits as they passed until get to | |
321 | # where it doesn't matter any more due to the | |
322 | # machine's precision. If they said .6666668, | |
323 | # we fail. | |
324 | (my $fraction = $parts[0]) =~ s/^.*\.//; | |
325 | my $epsilon = 10 ** - (length($fraction)); | |
326 | if ($epsilon > $utf8::max_floating_slop) { | |
327 | $epsilon = $utf8::max_floating_slop; | |
328 | } | |
329 | ||
330 | # But it can't be tighter than the minimum | |
331 | # precision for this machine. If haven't | |
332 | # already calculated that minimum, do so now. | |
333 | if (! defined $min_floating_slop) { | |
334 | ||
335 | # Keep going down an order of magnitude | |
336 | # until find that adding this quantity to | |
337 | # 1 remains 1; but put an upper limit on | |
338 | # this so in case this algorithm doesn't | |
339 | # work properly on some platform, that we | |
340 | # won't loop forever. | |
341 | my $count = 0; | |
342 | $min_floating_slop = 1; | |
343 | while (1+ $min_floating_slop != 1 | |
344 | && $count++ < 50) | |
345 | { | |
346 | my $next = $min_floating_slop / 10; | |
347 | last if $next == 0; # If underflows, | |
348 | # use previous one | |
349 | $min_floating_slop = $next; | |
350 | print STDERR __LINE__, ": min_float_slop=$min_floating_slop\n" if DEBUG; | |
351 | } | |
352 | ||
353 | # Back off a couple orders of magnitude, | |
354 | # just to be safe. | |
355 | $min_floating_slop *= 100; | |
356 | } | |
357 | ||
358 | if ($epsilon < $min_floating_slop) { | |
359 | $epsilon = $min_floating_slop; | |
360 | } | |
361 | print STDERR __LINE__, ": fraction=.$fraction; epsilon=$epsilon\n" if DEBUG; | |
362 | ||
363 | undef $table; | |
364 | ||
365 | # And for each possible rational in the table, | |
366 | # see if it is within epsilon of the input. | |
367 | foreach my $official | |
368 | (keys %utf8::nv_floating_to_rational) | |
369 | { | |
370 | print STDERR __LINE__, ": epsilon=$epsilon, official=$official, diff=", abs($parts[0] - $official), "\n" if DEBUG; | |
371 | if (abs($parts[0] - $official) < $epsilon) { | |
372 | $table = | |
373 | $utf8::nv_floating_to_rational{$official}; | |
374 | last; | |
375 | } | |
376 | } | |
377 | ||
378 | # Quit if didn't find one. | |
3dd10fe8 KW |
379 | if (! defined $table) { |
380 | pop @recursed if @recursed; | |
381 | return $type; | |
382 | } | |
99870f4d KW |
383 | } |
384 | } | |
385 | print STDERR __LINE__, ": $property=$table\n" if DEBUG; | |
386 | } | |
387 | } | |
388 | ||
389 | # Combine lhs (if any) and rhs to get something that matches | |
390 | # the syntax of the lookups. | |
391 | $property_and_table = "$prefix$table"; | |
392 | print STDERR __LINE__, ": $property_and_table\n" if DEBUG; | |
393 | ||
394 | # First try stricter matching. | |
395 | $file = $utf8::stricter_to_file_of{$property_and_table}; | |
396 | ||
397 | # If didn't find it, try again with looser matching by editing | |
398 | # out the applicable characters on the rhs and looking up | |
399 | # again. | |
400 | if (! defined $file) { | |
c4ab2516 | 401 | $table = _loose_name($table); |
99870f4d KW |
402 | $property_and_table = "$prefix$table"; |
403 | print STDERR __LINE__, ": $property_and_table\n" if DEBUG; | |
404 | $file = $utf8::loose_to_file_of{$property_and_table}; | |
405 | } | |
406 | ||
407 | # Add the constant and go fetch it in. | |
408 | if (defined $file) { | |
a6517ebf | 409 | |
eb092534 KW |
410 | # If the file name contains a !, it means to invert. The |
411 | # 0+ makes sure result is numeric | |
412 | $invert_it = 0 + $file =~ s/!//; | |
a6517ebf | 413 | |
99870f4d KW |
414 | if ($utf8::why_deprecated{$file}) { |
415 | warnings::warnif('deprecated', "Use of '$type' in \\p{} or \\P{} is deprecated because: $utf8::why_deprecated{$file};"); | |
416 | } | |
4ff124d1 KW |
417 | |
418 | if ($caseless | |
419 | && exists $utf8::caseless_equivalent{$property_and_table}) | |
420 | { | |
421 | $file = $utf8::caseless_equivalent{$property_and_table}; | |
422 | } | |
eb092534 KW |
423 | |
424 | # The pseudo-directory '#' means that there really isn't a | |
425 | # file to read, the data is in-line as part of the string; | |
426 | # we extract it below. | |
427 | $file = "$unicore_dir/lib/$file.pl" unless $file =~ m!^#/!; | |
99870f4d KW |
428 | last GETFILE; |
429 | } | |
430 | print STDERR __LINE__, ": didn't find $property_and_table\n" if DEBUG; | |
431 | ||
432 | ## | |
99870f4d KW |
433 | ## Last attempt -- see if it's a standard "To" name |
434 | ## (e.g. "ToLower") ToTitle is used by ucfirst(). | |
435 | ## The user-level way to access ToDigit() and ToFold() | |
436 | ## is to use Unicode::UCD. | |
437 | ## | |
cf2cd801 | 438 | # Only check if caller wants non-binary |
6aec79bd KW |
439 | my $retried = 0; |
440 | if ($minbits != 1 && $property_and_table =~ s/^to//) {{ | |
cf2cd801 KW |
441 | # Look input up in list of properties for which we have |
442 | # mapping files. | |
443 | if (defined ($file = | |
444 | $utf8::loose_property_to_file_of{$property_and_table})) | |
445 | { | |
446 | $type = $utf8::file_to_swash_name{$file}; | |
447 | print STDERR __LINE__, ": type set to $type\n" if DEBUG; | |
448 | $file = "$unicore_dir/$file.pl"; | |
449 | last GETFILE; | |
6aec79bd KW |
450 | } # If that fails see if there is a corresponding binary |
451 | # property file | |
452 | elsif (defined ($file = | |
453 | $utf8::loose_to_file_of{$property_and_table})) | |
454 | { | |
455 | ||
456 | # Here, there is no map file for the property we are | |
457 | # trying to get the map of, but this is a binary | |
458 | # property, and there is a file for it that can easily | |
459 | # be translated to a mapping. | |
460 | ||
461 | # In the case of properties that are forced to binary, | |
462 | # they are a combination. We return the actual | |
463 | # mapping instead of the binary. If the input is | |
464 | # something like 'Tocjkkiicore', it will be found in | |
465 | # %loose_property_to_file_of above as => 'To/kIICore'. | |
466 | # But the form like ToIskiicore won't be. To fix | |
467 | # this, it was easiest to do it here. These | |
468 | # properties are the complements of the default | |
469 | # property, so there is an entry in %loose_to_file_of | |
470 | # that is 'iskiicore' => '!kIICore/N', If we find such | |
471 | # an entry, strip off things and try again, which | |
472 | # should find the entry in %loose_property_to_file_of. | |
473 | # Actual binary properties that are of this form, such | |
474 | # as this entry: 'ishrkt' => '!Perl/Any' will also be | |
475 | # retried, but won't be in %loose_property_to_file_of, | |
476 | # and instead the next time through, it will find | |
477 | # 'hrkt' => '!Perl/Any' and proceed. | |
478 | redo if ! $retried | |
479 | && $file =~ /^!/ | |
480 | && $property_and_table =~ s/^is//; | |
481 | ||
482 | # This is a binary property. Setting this here causes | |
483 | # it to be stored as such in the cache, so if someone | |
484 | # comes along later looking for just a binary, they | |
485 | # get it. | |
486 | $minbits = 1; | |
487 | ||
4a7e937e | 488 | # The 0+ makes sure is numeric |
eb092534 KW |
489 | $invert_it = 0 + $file =~ s/!//; |
490 | $file = "$unicore_dir/lib/$file.pl" unless $file =~ m!^#/!; | |
6aec79bd | 491 | last GETFILE; |
cf2cd801 | 492 | } |
6aec79bd | 493 | } } |
99870f4d KW |
494 | |
495 | ## | |
496 | ## If we reach this line, it's because we couldn't figure | |
497 | ## out what to do with $type. Ouch. | |
498 | ## | |
499 | ||
3dd10fe8 | 500 | pop @recursed if @recursed; |
99870f4d | 501 | return $type; |
c9db9632 | 502 | } # end of GETFILE block |
cf25bb62 | 503 | |
99870f4d KW |
504 | if (defined $file) { |
505 | print STDERR __LINE__, ": found it (file='$file')\n" if DEBUG; | |
506 | ||
507 | ## | |
508 | ## If we reach here, it was due to a 'last GETFILE' above | |
509 | ## (exception: user-defined properties and mappings), so we | |
510 | ## have a filename, so now we load it if we haven't already. | |
eb092534 KW |
511 | |
512 | # The pseudo-directory '#' means the result isn't really a | |
513 | # file, but is in-line, with semi-colons to be turned into | |
514 | # new-lines. Since it is in-line there is no advantage to | |
515 | # caching the result | |
516 | if ($file =~ s!^#/!!) { | |
3854b4b8 | 517 | $list = $utf8::inline_definitions[$file]; |
eb092534 KW |
518 | } |
519 | else { | |
520 | # Here, we have an actual file to read in and load, but it | |
521 | # may already have been read-in and cached. The cache key | |
522 | # is the class and file to load, and whether the results | |
523 | # need to be inverted. | |
1f9f7d4c KW |
524 | my $found = $Cache{$class, $file, $invert_it}; |
525 | if ($found and ref($found) eq $class) { | |
526 | print STDERR __LINE__, ": Returning cached swash for '$class,$file,$invert_it' for \\p{$type}\n" if DEBUG; | |
527 | pop @recursed if @recursed; | |
528 | return $found; | |
529 | } | |
99870f4d | 530 | |
1f9f7d4c KW |
531 | local $@; |
532 | local $!; | |
533 | $list = do $file; die $@ if $@; | |
eb092534 KW |
534 | } |
535 | ||
7b953548 | 536 | $list_is_from_mktables = 1; |
0a0ffbce | 537 | } |
c9db9632 KW |
538 | } # End of $type is non-null |
539 | ||
540 | # Here, either $type was null, or we found the requested property and | |
541 | # read it into $list | |
cf25bb62 | 542 | |
7b953548 | 543 | my $extras = ""; |
c9db9632 | 544 | |
99870f4d KW |
545 | my $bits = $minbits; |
546 | ||
7b953548 KW |
547 | # mktables lists don't have extras, like '&utf8::prop', so don't need |
548 | # to separate them; also lists are already sorted, so don't need to do | |
549 | # that. | |
550 | if ($list && ! $list_is_from_mktables) { | |
0e9be77f | 551 | my $taint = substr($list,0,0); # maintain taint |
9b225781 | 552 | |
cb6d3474 KW |
553 | # Separate the extras from the code point list, and make sure |
554 | # user-defined properties and tr/// are well-behaved for | |
555 | # downstream code. | |
556 | if ($user_defined || $none) { | |
9b225781 KW |
557 | my @tmp = split(/^/m, $list); |
558 | my %seen; | |
559 | no warnings; | |
560 | ||
561 | # The extras are anything that doesn't begin with a hex digit. | |
562 | $extras = join '', $taint, grep /^[^0-9a-fA-F]/, @tmp; | |
563 | ||
564 | # Remove the extras, and sort the remaining entries by the | |
565 | # numeric value of their beginning hex digits, removing any | |
566 | # duplicates. | |
567 | $list = join '', $taint, | |
568 | map { $_->[1] } | |
569 | sort { $a->[0] <=> $b->[0] } | |
570 | map { /^([0-9a-fA-F]+)/; [ CORE::hex($1), $_ ] } | |
571 | grep { /^([0-9a-fA-F]+)/ and not $seen{$1}++ } @tmp; # XXX doesn't do ranges right | |
4de6d205 KW |
572 | } |
573 | else { | |
574 | # mktables has gone to some trouble to make non-user defined | |
575 | # properties well-behaved, so we can skip the effort we do for | |
576 | # user-defined ones. Any extras are at the very beginning of | |
577 | # the string. | |
578 | ||
579 | # This regex splits out the first lines of $list into $1 and | |
580 | # strips them off from $list, until we get one that begins | |
581 | # with a hex number, alone on the line, or followed by a tab. | |
582 | # Either portion may be empty. | |
583 | $list =~ s/ \A ( .*? ) | |
584 | (?: \z | (?= ^ [0-9a-fA-F]+ (?: \t | $) ) ) | |
585 | //msx; | |
586 | ||
587 | $extras = "$taint$1"; | |
588 | } | |
99870f4d | 589 | } |
a0ed51b3 | 590 | |
99870f4d KW |
591 | if ($none) { |
592 | my $hextra = sprintf "%04x", $none + 1; | |
593 | $list =~ s/\tXXXX$/\t$hextra/mg; | |
594 | } | |
a0ed51b3 | 595 | |
99870f4d KW |
596 | if ($minbits != 1 && $minbits < 32) { # not binary property |
597 | my $top = 0; | |
598 | while ($list =~ /^([0-9a-fA-F]+)(?:[\t]([0-9a-fA-F]+)?)(?:[ \t]([0-9a-fA-F]+))?/mg) { | |
599 | my $min = CORE::hex $1; | |
600 | my $max = defined $2 ? CORE::hex $2 : $min; | |
601 | my $val = defined $3 ? CORE::hex $3 : 0; | |
602 | $val += $max - $min if defined $3; | |
603 | $top = $val if $val > $top; | |
604 | } | |
605 | my $topbits = | |
606 | $top > 0xffff ? 32 : | |
607 | $top > 0xff ? 16 : 8; | |
608 | $bits = $topbits if $bits < $topbits; | |
609 | } | |
a0ed51b3 | 610 | |
99870f4d KW |
611 | my @extras; |
612 | if ($extras) { | |
613 | for my $x ($extras) { | |
0e9be77f | 614 | my $taint = substr($x,0,0); # maintain taint |
99870f4d KW |
615 | pos $x = 0; |
616 | while ($x =~ /^([^0-9a-fA-F\n])(.*)/mg) { | |
0e9be77f DM |
617 | my $char = "$1$taint"; |
618 | my $name = "$2$taint"; | |
619 | print STDERR __LINE__, ": char [$char] => name [$name]\n" | |
620 | if DEBUG; | |
99870f4d KW |
621 | if ($char =~ /[-+!&]/) { |
622 | my ($c,$t) = split(/::/, $name, 2); # bogus use of ::, really | |
623 | my $subobj; | |
624 | if ($c eq 'utf8') { | |
625 | $subobj = utf8->SWASHNEW($t, "", $minbits, 0); | |
626 | } | |
627 | elsif (exists &$name) { | |
628 | $subobj = utf8->SWASHNEW($name, "", $minbits, 0); | |
629 | } | |
630 | elsif ($c =~ /^([0-9a-fA-F]+)/) { | |
631 | $subobj = utf8->SWASHNEW("", $c, $minbits, 0); | |
632 | } | |
25df5e40 | 633 | print STDERR __LINE__, ": returned from getting sub object for $name\n" if DEBUG; |
3dd10fe8 KW |
634 | if (! ref $subobj) { |
635 | pop @recursed if @recursed && $type; | |
636 | return $subobj; | |
637 | } | |
99870f4d KW |
638 | push @extras, $name => $subobj; |
639 | $bits = $subobj->{BITS} if $bits < $subobj->{BITS}; | |
04e9cbbb KW |
640 | $user_defined = $subobj->{USER_DEFINED} |
641 | if $subobj->{USER_DEFINED}; | |
99870f4d KW |
642 | } |
643 | } | |
644 | } | |
645 | } | |
a0ed51b3 | 646 | |
99870f4d | 647 | if (DEBUG) { |
0fada5e1 | 648 | print STDERR __LINE__, ": CLASS = $class, TYPE => $type, BITS => $bits, NONE => $none, INVERT_IT => $invert_it, USER_DEFINED => $user_defined"; |
99870f4d KW |
649 | print STDERR "\nLIST =>\n$list" if defined $list; |
650 | print STDERR "\nEXTRAS =>\n$extras" if defined $extras; | |
651 | print STDERR "\n"; | |
652 | } | |
a0ed51b3 | 653 | |
99870f4d KW |
654 | my $SWASH = bless { |
655 | TYPE => $type, | |
656 | BITS => $bits, | |
657 | EXTRAS => $extras, | |
658 | LIST => $list, | |
659 | NONE => $none, | |
0fada5e1 | 660 | USER_DEFINED => $user_defined, |
99870f4d KW |
661 | @extras, |
662 | } => $class; | |
663 | ||
664 | if ($file) { | |
4a7e937e | 665 | $Cache{$class, $file, $invert_it} = $SWASH; |
64870e05 KW |
666 | if ($type |
667 | && exists $utf8::SwashInfo{$type} | |
668 | && exists $utf8::SwashInfo{$type}{'specials_name'}) | |
669 | { | |
670 | my $specials_name = $utf8::SwashInfo{$type}{'specials_name'}; | |
671 | no strict "refs"; | |
69b36ee0 | 672 | print STDERR "\nspecials_name => $specials_name\n" if DEBUG; |
64870e05 KW |
673 | $SWASH->{'SPECIALS'} = \%$specials_name; |
674 | } | |
a6517ebf | 675 | $SWASH->{'INVERT_IT'} = $invert_it; |
99870f4d | 676 | } |
5beb625e | 677 | |
3dd10fe8 KW |
678 | pop @recursed if @recursed && $type; |
679 | ||
99870f4d | 680 | return $SWASH; |
5beb625e | 681 | } |
a0ed51b3 LW |
682 | } |
683 | ||
b0e3252e | 684 | # Now SWASHGET is recasted into a C function S_swatch_get (see utf8.c). |
a0ed51b3 LW |
685 | |
686 | 1; |