2 package CharClass::Matcher;
6 use warnings FATAL => 'all';
7 use Text::Wrap qw(wrap);
9 $Data::Dumper::Useqq= 1;
10 our $hex_fmt= "0x%02X";
15 sub ASCII_PLATFORM { (ord('A') == 65) }
17 require 'regen/regen_lib.pl';
21 CharClass::Matcher -- Generate C macros that match character classes efficiently
25 perl Porting/regcharclass.pl
29 Dynamically generates macros for detecting special charclasses
30 in latin-1, utf8, and codepoint forms. Macros can be set to return
31 the length (in bytes) of the matched codepoint, and/or the codepoint itself.
33 To regenerate F<regcharclass.h>, run this script from perl-root. No arguments
36 Using WHATEVER as an example the following macros can be produced, depending
37 on the input parameters (how to get each is described by internal comments at
38 the C<__DATA__> line):
42 =item C<is_WHATEVER(s,is_utf8)>
44 =item C<is_WHATEVER_safe(s,e,is_utf8)>
46 Do a lookup as appropriate based on the C<is_utf8> flag. When possible
47 comparisons involving octect<128 are done before checking the C<is_utf8>
48 flag, hopefully saving time.
50 The version without the C<_safe> suffix should be used only when the input is
51 known to be well-formed.
53 =item C<is_WHATEVER_utf8(s)>
55 =item C<is_WHATEVER_utf8_safe(s,e)>
57 Do a lookup assuming the string is encoded in (normalized) UTF8.
59 The version without the C<_safe> suffix should be used only when the input is
60 known to be well-formed.
62 =item C<is_WHATEVER_latin1(s)>
64 =item C<is_WHATEVER_latin1_safe(s,e)>
66 Do a lookup assuming the string is encoded in latin-1 (aka plan octets).
68 The version without the C<_safe> suffix should be used only when it is known
69 that C<s> contains at least one character.
71 =item C<is_WHATEVER_cp(cp)>
73 Check to see if the string matches a given codepoint (hypothetically a
74 U32). The condition is constructed as as to "break out" as early as
75 possible if the codepoint is out of range of the condition.
79 (cp==X || (cp>X && (cp==Y || (cp>Y && ...))))
81 Thus if the character is X+1 only two comparisons will be done. Making
82 matching lookups slower, but non-matching faster.
84 =item C<what_len_WHATEVER_FOO(arg1, ..., len)>
86 A variant form of each of the macro types described above can be generated, in
87 which the code point is returned by the macro, and an extra parameter (in the
88 final position) is added, which is a pointer for the macro to set the byte
89 length of the returned code point.
91 These forms all have a C<what_len> prefix instead of the C<is_>, for example
92 C<what_len_WHATEVER_safe(s,e,is_utf8,len)> and
93 C<what_len_WHATEVER_utf8(s,len)>.
95 These forms should not be used I<except> on small sets of mostly widely
96 separated code points; otherwise the code generated is inefficient. For these
97 cases, it is best to use the C<is_> forms, and then find the code point with
98 C<utf8_to_uvchr_buf>(). This program can fail with a "deep recursion"
99 message on the worst of the inappropriate sets. Examine the generated macro
100 to see if it is acceptable.
102 =item C<what_WHATEVER_FOO(arg1, ...)>
104 A variant form of each of the C<is_> macro types described above can be generated, in
105 which the code point and not the length is returned by the macro. These have
106 the same caveat as L</what_len_WHATEVER_FOO(arg1, ..., len)>, plus they should
107 not be used where the set contains a NULL, as 0 is returned for two different
108 cases: a) the set doesn't include the input code point; b) the set does
109 include it, and it is a NULL.
115 perltidy -st -bt=1 -bbt=0 -pt=0 -sbt=1 -ce -nwls== "%f"
120 Author: Yves Orton (demerphq) 2007. Maintained by Perl5 Porters.
124 No tests directly here (although the regex engine will fail tests
125 if this code is broken). Insufficient documentation and no Getopts
126 handler for using the module as a script.
130 You may distribute under the terms of either the GNU General Public
131 License or the Artistic License, as specified in the README file.
135 # Sub naming convention:
136 # __func : private subroutine, can not be called as a method
137 # _func : private method, not meant for external use
138 # func : public method.
141 #-------------------------------------------------------------------------------
143 # ($cp,$n,$l,$u)=__uni_latin($str);
145 # Return a list of arrays, each of which when interpreted correctly
146 # represent the string in some given encoding with specific conditions.
148 # $cp - list of codepoints that make up the string.
149 # $n - list of octets that make up the string if all codepoints are invariant
150 # regardless of if the string is in UTF-8 or not.
151 # $l - list of octets that make up the string in latin1 encoding if all
152 # codepoints < 256, and at least one codepoint is UTF-8 variant.
153 # $u - list of octets that make up the string in utf8 if any codepoint is
157 #-----------+----------
158 # 0 - 127 : $n (127/128 are the values for ASCII platforms)
168 my $only_has_invariants = 1;
169 for my $ch ( split //, $str ) {
172 push @cp_high, $cp if $cp > 255;
173 $max= $cp if $max < $cp;
174 if (! ASCII_PLATFORM && $only_has_invariants) {
176 $only_has_invariants = 0;
180 utf8::upgrade($temp);
181 my @utf8 = unpack "U0C*", $temp;
182 $only_has_invariants = (@utf8 == 1 && $utf8[0] == $cp);
187 $only_has_invariants = $max < 128 if ASCII_PLATFORM;
188 if ($only_has_invariants) {
191 $l= [@cp] if $max && $max < 256;
195 $u= [ unpack "U0C*", $u ] if defined $u;
197 return ( \@cp, \@cp_high, $n, $l, $u );
201 # $clean= __clean($expr);
203 # Cleanup a ternary expression, removing unnecessary parens and apply some
204 # simplifications using regexes.
213 $parens= qr/ (?> \( (?> (?: (?> [^()]+ ) | (??{ $parens }) )* ) \) ) /x;
215 ## remove redundant parens
216 1 while $expr =~ s/ \( \s* ( $parens ) \s* \) /$1/gx;
219 # repeatedly simplify conditions like
220 # ( (cond1) ? ( (cond2) ? X : Y ) : Y )
222 # ( ( (cond1) && (cond2) ) ? X : Y )
223 # Also similarly handles expressions like:
224 # : (cond1) ? ( (cond2) ? X : Y ) : Y )
225 # Note the inclusion of the close paren in ([:()]) and the open paren in ([()]) is
226 # purely to ensure we have a balanced set of parens in the expression which makes
227 # it easier to understand the pattern in an editor that understands paren's, we do
228 # not expect either of these cases to actually fire. - Yves
234 \? \s* ($parens|[^()?:\s]+?) \s*
235 : \s* ($parens|[^()?:\s]+?) \s*
239 /$1 ( $2 && $3 ) ? $4 : $5 $6/gx;
245 # $text= __macro(@args);
246 # Join args together by newlines, and then neatly add backslashes to the end
247 # of every line as expected by the C pre-processor for #define's.
251 my $str= join "\n", @_;
253 my @lines= map { s/\s+$//; s/\t/ /g; $_ } split /\n/, $str;
254 my $last= pop @lines;
255 $str= join "\n", ( map { sprintf "%-76s\\", $_ } @lines ), $last;
256 1 while $str =~ s/^(\t*) {8}/$1\t/gm;
261 # my $op=__incrdepth($op);
263 # take an 'op' hashref and add one to it and all its childrens depths.
268 return unless ref $op;
270 __incrdepth( $op->{yes} );
271 __incrdepth( $op->{no} );
275 # join two branches of an opcode together with a condition, incrementing
276 # the depth on the yes branch when we do so.
277 # returns the new root opcode of the tree.
279 my ( $cond, $yes, $no )= @_;
282 yes => __incrdepth( $yes ),
292 # my $obj=CLASS->new(op=>'SOMENAME',title=>'blah',txt=>[..]);
294 # Create a new CharClass::Matcher object by parsing the text in
295 # the txt array. Currently applies the following rules:
297 # Element starts with C<0x>, line is evaled the result treated as
298 # a number which is passed to chr().
300 # Element starts with C<">, line is evaled and the result treated
303 # Each string is then stored in the 'strs' subhash as a hash record
304 # made up of the results of __uni_latin1, using the keynames
305 # 'low','latin1','utf8', as well as the synthesized 'LATIN1', 'high', and
306 # 'UTF8' which hold a merge of 'low' and their lowercase equivelents.
308 # Size data is tracked per type in the 'size' subhash.
316 die "in " . __PACKAGE__ . " constructor '$_;' is a mandatory field"
322 title => $opt{title} || '',
324 foreach my $txt ( @{ $opt{txt} } ) {
326 if ( $str =~ /^[""]/ ) {
328 } elsif ($str =~ / - /x ) { # A range: Replace this element on the
329 # list with its expansion
330 my ($lower, $upper) = $str =~ / 0x (.+?) \s* - \s* 0x (.+) /x;
331 die "Format must be like '0xDEAD - 0xBEAF'; instead was '$str'" if ! defined $lower || ! defined $upper;
332 foreach my $cp (hex $lower .. hex $upper) {
333 push @{$opt{txt}}, sprintf "0x%X", $cp;
336 } elsif ($str =~ s/ ^ N (?= 0x ) //x ) {
337 # Otherwise undocumented, a leading N means is already in the
338 # native character set; don't convert.
340 } elsif ( $str =~ /^0x/ ) {
343 # Convert from Unicode/ASCII to native, if necessary
344 $str = utf8::unicode_to_native($str) if ! ASCII_PLATFORM
347 } elsif ( $str =~ / \s* \\p \{ ( .*? ) \} /x) {
349 use Unicode::UCD qw(prop_invlist);
351 my @invlist = prop_invlist($property, '_perl_core_internal_ok');
354 # An empty return could mean an unknown property, or merely
355 # that it is empty. Call in scalar context to differentiate
356 my $count = prop_invlist($property, '_perl_core_internal_ok');
357 die "$property not found" unless defined $count;
360 # Replace this element on the list with the property's expansion
361 for (my $i = 0; $i < @invlist; $i += 2) {
362 foreach my $cp ($invlist[$i] .. $invlist[$i+1] - 1) {
364 # prop_invlist() returns native values; add leading 'N'
366 push @{$opt{txt}}, sprintf "N0x%X", $cp;
370 } elsif ($str =~ / ^ do \s+ ( .* ) /x) {
371 die "do '$1' failed: $!$@" if ! do $1 or $@;
373 } elsif ($str =~ / ^ & \s* ( .* ) /x) { # user-furnished sub() call
374 my @results = eval "$1";
375 die "eval '$1' failed: $@" if $@;
376 push @{$opt{txt}}, @results;
379 die "Unparsable line: $txt\n";
381 my ( $cp, $cp_high, $low, $latin1, $utf8 )= __uni_latin1( $str );
382 my $UTF8= $low || $utf8;
383 my $LATIN1= $low || $latin1;
384 my $high = (scalar grep { $_ < 256 } @$cp) ? 0 : $utf8;
385 #die Dumper($txt,$cp,$low,$latin1,$utf8)
386 # if $txt=~/NEL/ or $utf8 and @$utf8>3;
388 @{ $self->{strs}{$str} }{qw( str txt low utf8 latin1 high cp cp_high UTF8 LATIN1 )}=
389 ( $str, $txt, $low, $utf8, $latin1, $high, $cp, $cp_high, $UTF8, $LATIN1 );
390 my $rec= $self->{strs}{$str};
391 foreach my $key ( qw(low utf8 latin1 high cp cp_high UTF8 LATIN1) ) {
392 $self->{size}{$key}{ 0 + @{ $self->{strs}{$str}{$key} } }++
393 if $self->{strs}{$str}{$key};
395 $self->{has_multi} ||= @$cp > 1;
396 $self->{has_ascii} ||= $latin1 && @$latin1;
397 $self->{has_low} ||= $low && @$low;
398 $self->{has_high} ||= !$low && !$latin1;
400 $self->{val_fmt}= $hex_fmt;
401 $self->{count}= 0 + keys %{ $self->{strs} };
405 # my $trie = make_trie($type,$maxlen);
407 # using the data stored in the object build a trie of a specific type,
408 # and with specific maximum depth. The trie is made up the elements of
409 # the given types array for each string in the object (assuming it is
412 # returns the trie, or undef if there was no relevant data in the object.
416 my ( $self, $type, $maxlen )= @_;
418 my $strs= $self->{strs};
420 foreach my $rec ( values %$strs ) {
421 die "panic: unknown type '$type'"
422 if !exists $rec->{$type};
423 my $dat= $rec->{$type};
425 next if $maxlen && @$dat > $maxlen;
427 foreach my $elem ( @$dat ) {
428 $node->{$elem} ||= {};
429 $node= $node->{$elem};
431 $node->{''}= $rec->{str};
433 return 0 + keys( %trie ) ? \%trie : undef;
439 # This returns a list of the positions of the bits in the input word that
445 push @positions, $position if $word & 1;
452 # my $optree= _optree()
454 # recursively convert a trie to an optree where every node represents
460 my ( $self, $trie, $test_type, $ret_type, $else, $depth )= @_;
461 return unless defined $trie;
462 if ( $self->{has_multi} and $ret_type =~ /cp|both/ ) {
463 die "Can't do 'cp' optree from multi-codepoint strings";
466 $else= 0 unless defined $else;
467 $depth= 0 unless defined $depth;
469 # if we have an emptry string as a key it means we are in an
470 # accepting state and unless we can match further on should
471 # return the value of the '' key.
472 if (exists $trie->{''} ) {
473 # we can now update the "else" value, anything failing to match
474 # after this point should return the value from this.
475 if ( $ret_type eq 'cp' ) {
476 $else= $self->{strs}{ $trie->{''} }{cp}[0];
477 $else= sprintf "$self->{val_fmt}", $else if $else > 9;
478 } elsif ( $ret_type eq 'len' ) {
480 } elsif ( $ret_type eq 'both') {
481 $else= $self->{strs}{ $trie->{''} }{cp}[0];
482 $else= sprintf "$self->{val_fmt}", $else if $else > 9;
483 $else= "len=$depth, $else";
486 # extract the meaningful keys from the trie, filter out '' as
487 # it means we are an accepting state (end of sequence).
488 my @conds= sort { $a <=> $b } grep { length $_ } keys %$trie;
490 # if we havent any keys there is no further we can match and we
491 # can return the "else" value.
492 return $else if !@conds;
495 my $test= $test_type =~ /^cp/ ? "cp" : "((U8*)s)[$depth]";
496 # first we loop over the possible keys/conditions and find out what they look like
497 # we group conditions with the same optree together.
500 local $Data::Dumper::Sortkeys=1;
501 foreach my $cond ( @conds ) {
503 # get the optree for this child/condition
504 my $res= $self->_optree( $trie->{$cond}, $test_type, $ret_type, $else, $depth + 1 );
505 # convert it to a string with Dumper
506 my $res_code= Dumper( $res );
508 push @{$dmp_res{$res_code}{vals}}, $cond;
509 if (!$dmp_res{$res_code}{optree}) {
510 $dmp_res{$res_code}{optree}= $res;
511 push @res_order, $res_code;
515 # now that we have deduped the optrees we construct a new optree containing the merged
519 foreach my $res_code_idx (0 .. $#res_order) {
520 my $res_code= $res_order[$res_code_idx];
521 $node->{vals}= $dmp_res{$res_code}{vals};
522 $node->{test}= $test;
523 $node->{yes}= $dmp_res{$res_code}{optree};
524 $node->{depth}= $depth;
525 if ($res_code_idx < $#res_order) {
526 $node= $node->{no}= {};
536 # my $optree= optree(%opts);
538 # Convert a trie to an optree, wrapper for _optree
543 my $trie= $self->make_trie( $opt{type}, $opt{max_depth} );
544 $opt{ret_type} ||= 'len';
545 my $test_type= $opt{type} =~ /^cp/ ? 'cp' : 'depth';
546 return $self->_optree( $trie, $test_type, $opt{ret_type}, $opt{else}, 0 );
549 # my $optree= generic_optree(%opts);
551 # build a "generic" optree out of the three 'low', 'latin1', 'utf8'
552 # sets of strings, including a branch for handling the string type check.
559 $opt{ret_type} ||= 'len';
560 my $test_type= 'depth';
561 my $else= $opt{else} || 0;
563 my $latin1= $self->make_trie( 'latin1', $opt{max_depth} );
564 my $utf8= $self->make_trie( 'utf8', $opt{max_depth} );
566 $_= $self->_optree( $_, $test_type, $opt{ret_type}, $else, 0 )
570 $else= __cond_join( "( is_utf8 )", $utf8, $latin1 || $else );
571 } elsif ( $latin1 ) {
572 $else= __cond_join( "!( is_utf8 )", $latin1, $else );
574 my $low= $self->make_trie( 'low', $opt{max_depth} );
576 $else= $self->_optree( $low, $test_type, $opt{ret_type}, $else, 0 );
584 # create a string length guarded optree.
590 my $type= $opt{type};
592 die "Can't do a length_optree on type 'cp', makes no sense."
595 my ( @size, $method );
597 if ( $type eq 'generic' ) {
598 $method= 'generic_optree';
600 %{ $self->{size}{low} || {} },
601 %{ $self->{size}{latin1} || {} },
602 %{ $self->{size}{utf8} || {} }
604 @size= sort { $a <=> $b } keys %sizes;
607 @size= sort { $a <=> $b } keys %{ $self->{size}{$type} };
610 my $else= ( $opt{else} ||= 0 );
611 for my $size ( @size ) {
612 my $optree= $self->$method( %opt, type => $type, max_depth => $size );
613 my $cond= "((e)-(s) > " . ( $size - 1 ).")";
614 $else= __cond_join( $cond, $optree, $else );
619 sub calculate_mask(@) {
620 # Look at the input list of byte values. This routine returns an array of
621 # mask/base pairs to generate that list.
624 my $list_count = @list;
626 # Consider a set of byte values, A, B, C .... If we want to determine if
627 # <c> is one of them, we can write c==A || c==B || c==C .... If the
628 # values are consecutive, we can shorten that to A<=c && c<=Z, which uses
629 # far fewer branches. If only some of them are consecutive we can still
630 # save some branches by creating range tests for just those that are
631 # consecutive. _cond_as_str() does this work for looking for ranges.
633 # Another approach is to look at the bit patterns for A, B, C .... and see
634 # if they have some commonalities. That's what this function does. For
635 # example, consider a set consisting of the bytes
636 # 0xF0, 0xF1, 0xF2, and 0xF3. We could write:
637 # 0xF0 <= c && c <= 0xF4
638 # But the following mask/compare also works, and has just one test:
640 # The reason it works is that the set consists of exactly those bytes
641 # whose first 4 bits are 1, and the next two are 0. (The value of the
642 # other 2 bits is immaterial in determining if a byte is in the set or
643 # not.) The mask masks out those 2 irrelevant bits, and the comparison
644 # makes sure that the result matches all bytes which match those 6
645 # material bits exactly. In other words, the set of bytes contains
646 # exactly those whose bottom two bit positions are either 0 or 1. The
647 # same principle applies to bit positions that are not necessarily
648 # adjacent. And it can be applied to bytes that differ in 1 through all 8
649 # bit positions. In order to be a candidate for this optimization, the
650 # number of bytes in the set must be a power of 2.
652 # Consider a different example, the set 0x53, 0x54, 0x73, and 0x74. That
653 # requires 4 tests using either ranges or individual values, and even
654 # though the number in the set is a power of 2, it doesn't qualify for the
655 # mask optimization described above because the number of bits that are
656 # different is too large for that. However, the set can be expressed as
657 # two branches with masks thusly:
658 # (c & 0xDF) == 0x53 || (c & 0xDF) == 0x54
659 # a branch savings of 50%. This is done by splitting the set into two
660 # subsets each of which has 2 elements, and within each set the values
663 # This function attempts to find some way to save some branches using the
664 # mask technique. If not, it returns an empty list; if so, it
665 # returns a list consisting of
666 # [ [compare1, mask1], [compare2, mask2], ...
667 # [compare_n, undef], [compare_m, undef], ...
669 # The <mask> is undef in the above for those bytes that must be tested
672 # This function does not attempt to find the optimal set. To do so would
673 # probably require testing all possible combinations, and keeping track of
674 # the current best one.
676 # There are probably much better algorithms, but this is the one I (khw)
677 # came up with. We start with doing a bit-wise compare of every byte in
678 # the set with every other byte. The results are sorted into arrays of
679 # all those that differ by the same bit positions. These are stored in a
680 # hash with the each key being the bits they differ in. Here is the hash
681 # for the 0x53, 0x54, 0x73, 0x74 set:
709 # The set consisting of values which differ in the 4 bit positions 0, 1,
710 # 2, and 5 from some other value in the set consists of all 4 values.
711 # Likewise all 4 values differ from some other value in the 3 bit
712 # positions 0, 1, and 2; and all 4 values differ from some other value in
713 # the single bit position 5. The keys at the uppermost level in the above
714 # hash, 1, 3, and 4, give the number of bit positions that each sub-key
715 # below it has. For example, the 4 key could have as its value an array
716 # consisting of "0,1,2,5", "0,1,2,6", and "3,4,6,7", if the inputs were
717 # such. The best optimization will group the most values into a single
718 # mask. The most values will be the ones that differ in the most
719 # positions, the ones with the largest value for the topmost key. These
720 # keys, are thus just for convenience of sorting by that number, and do
721 # not have any bearing on the core of the algorithm.
723 # We start with an element from largest number of differing bits. The
724 # largest in this case is 4 bits, and there is only one situation in this
725 # set which has 4 differing bits, "0,1,2,5". We look for any subset of
726 # this set which has 16 values that differ in these 4 bits. There aren't
727 # any, because there are only 4 values in the entire set. We then look at
728 # the next possible thing, which is 3 bits differing in positions "0,1,2".
729 # We look for a subset that has 8 values that differ in these 3 bits.
730 # Again there are none. So we go to look for the next possible thing,
731 # which is a subset of 2**1 values that differ only in bit position 5. 83
732 # and 115 do, so we calculate a mask and base for those and remove them
733 # from every set. Since there is only the one set remaining, we remove
734 # them from just this one. We then look to see if there is another set of
735 # 2 values that differ in bit position 5. 84 and 116 do, so we calculate
736 # a mask and base for those and remove them from every set (again only
737 # this set remains in this example). The set is now empty, and there are
738 # no more sets to look at, so we are done.
740 if ($list_count == 256) { # All 256 is trivially masked
746 # Generate bits-differing lists for each element compared against each
748 for my $i (0 .. $list_count - 2) {
749 for my $j ($i + 1 .. $list_count - 1) {
750 my @bits_that_differ = pop_count($list[$i] ^ $list[$j]);
751 my $differ_count = @bits_that_differ;
752 my $key = join ",", @bits_that_differ;
753 push @{$hash{$differ_count}{$key}}, $list[$i] unless grep { $_ == $list[$i] } @{$hash{$differ_count}{$key}};
754 push @{$hash{$differ_count}{$key}}, $list[$j];
758 print STDERR __LINE__, ": calculate_mask() called: List of values grouped by differing bits: ", Dumper \%hash if DEBUG;
761 foreach my $count (reverse sort { $a <=> $b } keys %hash) {
762 my $need = 2 ** $count; # Need 8 values for 3 differing bits, etc
763 foreach my $bits (sort keys $hash{$count}) {
765 print STDERR __LINE__, ": For $count bit(s) difference ($bits), need $need; have ", scalar @{$hash{$count}{$bits}}, "\n" if DEBUG;
767 # Look only as long as there are at least as many elements in the
768 # subset as are needed
769 while ((my $cur_count = @{$hash{$count}{$bits}}) >= $need) {
771 print STDERR __LINE__, ": Looking at bit positions ($bits): ", Dumper $hash{$count}{$bits} if DEBUG;
773 # Start with the first element in it
774 my $try_base = $hash{$count}{$bits}[0];
775 my @subset = $try_base;
777 # If it succeeds, we return a mask and a base to compare
778 # against the masked value. That base will be the AND of
779 # every element in the subset. Initialize to the one element
781 my $compare = $try_base;
783 # We are trying to find a subset of this that has <need>
784 # elements that differ in the bit positions given by the
785 # string $bits, which is comma separated.
786 my @bits = split ",", $bits;
788 TRY: # Look through the remainder of the list for other
789 # elements that differ only by these bit positions.
791 for (my $i = 1; $i < $cur_count; $i++) {
792 my $try_this = $hash{$count}{$bits}[$i];
793 my @positions = pop_count($try_base ^ $try_this);
795 print STDERR __LINE__, ": $try_base vs $try_this: is (", join(',', @positions), ") a subset of ($bits)?" if DEBUG;;
797 foreach my $pos (@positions) {
798 unless (grep { $pos == $_ } @bits) {
799 print STDERR " No\n" if DEBUG;
800 my $remaining = $cur_count - $i - 1;
801 if ($remaining && @subset + $remaining < $need) {
802 print STDERR __LINE__, ": Can stop trying $try_base, because even if all the remaining $remaining values work, they wouldn't add up to the needed $need when combined with the existing ", scalar @subset, " ones\n" if DEBUG;
809 print STDERR " Yes\n" if DEBUG;
810 push @subset, $try_this;
812 # Add this to the mask base, in case it ultimately
814 $compare &= $try_this;
817 print STDERR __LINE__, ": subset (", join(", ", @subset), ") has ", scalar @subset, " elements; needs $need\n" if DEBUG;
819 if (@subset < $need) {
820 shift @{$hash{$count}{$bits}};
821 next; # Try with next value
826 foreach my $position (@bits) {
827 $mask |= 1 << $position;
829 $mask = ~$mask & 0xFF;
830 push @final_results, [$compare, $mask];
832 printf STDERR "%d: Got it: compare=%d=0x%X; mask=%X\n", __LINE__, $compare, $compare, $mask if DEBUG;
834 # These values are now spoken for. Remove them from future
836 foreach my $remove_count (sort keys %hash) {
837 foreach my $bits (sort keys %{$hash{$remove_count}}) {
838 foreach my $to_remove (@subset) {
839 @{$hash{$remove_count}{$bits}} = grep { $_ != $to_remove } @{$hash{$remove_count}{$bits}};
847 # Any values that remain in the list are ones that have to be tested for
850 foreach my $count (reverse sort { $a <=> $b } keys %hash) {
851 foreach my $bits (sort keys $hash{$count}) {
852 foreach my $remaining (@{$hash{$count}{$bits}}) {
854 # If we already know about this value, just ignore it.
855 next if grep { $remaining == $_ } @individuals;
857 # Otherwise it needs to be returned as something to match
859 push @final_results, [$remaining, undef];
860 push @individuals, $remaining;
865 # Sort by increasing numeric value
866 @final_results = sort { $a->[0] <=> $b->[0] } @final_results;
868 print STDERR __LINE__, ": Final return: ", Dumper \@final_results if DEBUG;
870 return @final_results;
874 # turn a list of conditions into a text expression
875 # - merges ranges of conditions, and joins the result with ||
877 my ( $self, $op, $combine, $opts_ref )= @_;
878 my $cond= $op->{vals};
879 my $test= $op->{test};
880 my $is_cp_ret = $opts_ref->{ret_type} eq "cp";
881 return "( $test )" if !defined $cond;
886 # We skip this if there are optimizations that
887 # we can apply (below) to the individual ranges
888 if ( ($is_cp_ret || $combine) && @ranges && ref $ranges[-1]) {
889 if ( $ranges[-1][0] == $ranges[-1][1] ) {
890 $ranges[-1]= $ranges[-1][0];
891 } elsif ( $ranges[-1][0] + 1 == $ranges[-1][1] ) {
892 $ranges[-1]= $ranges[-1][0];
893 push @ranges, $ranges[-1] + 1;
897 for my $condition ( @$cond ) {
898 if ( !@ranges || $condition != $ranges[-1][1] + 1 ) {
900 push @ranges, [ $condition, $condition ];
907 return $self->_combine( $test, @ranges )
914 "( $self->{val_fmt} <= $test && $test <= $self->{val_fmt} )",
916 : sprintf( "$self->{val_fmt} == $test", $_ );
919 return "( " . join( " || ", @ranges ) . " )";
922 # If the input set has certain characteristics, we can optimize tests
923 # for it. This doesn't apply if returning the code point, as we want
924 # each element of the set individually. The code above is for this
927 return 1 if @$cond == 256; # If all bytes match, is trivially true
932 # See if the entire set shares optimizable characterstics, and if so,
933 # return the optimization. We delay checking for this on sets with
934 # just a single range, as there may be better optimizations available
936 @masks = calculate_mask(@$cond);
938 # Stringify the output of calculate_mask()
941 foreach my $mask_ref (@masks) {
942 if (defined $mask_ref->[1]) {
943 push @return, sprintf "( ( $test & $self->{val_fmt} ) == $self->{val_fmt} )", $mask_ref->[1], $mask_ref->[0];
945 else { # An undefined mask means to use the value as-is
946 push @return, sprintf "$test == $self->{val_fmt}", $mask_ref->[0];
950 # The best possible case below for specifying this set of values via
951 # ranges is 1 branch per range. If our mask method yielded better
952 # results, there is no sense trying something that is bound to be
954 if (@return < @ranges) {
955 return "( " . join( " || ", @return ) . " )";
962 # Here, there was no entire-class optimization that was clearly better
963 # than doing things by ranges. Look at each range.
964 my $range_count_extra = 0;
965 for (my $i = 0; $i < @ranges; $i++) {
966 if (! ref $ranges[$i]) { # Trivial case: no range
967 $ranges[$i] = sprintf "$self->{val_fmt} == $test", $ranges[$i];
969 elsif ($ranges[$i]->[0] == $ranges[$i]->[1]) {
970 $ranges[$i] = # Trivial case: single element range
971 sprintf "$self->{val_fmt} == $test", $ranges[$i]->[0];
976 # Well-formed UTF-8 continuation bytes on ascii platforms must be
977 # in the range 0x80 .. 0xBF. If we know that the input is
978 # well-formed (indicated by not trying to be 'safe'), we can omit
979 # tests that verify that the input is within either of these
980 # bounds. (No legal UTF-8 character can begin with anything in
981 # this range, so we don't have to worry about this being a
982 # continuation byte or not.)
984 && ! $opts_ref->{safe}
985 && $opts_ref->{type} =~ / ^ (?: utf8 | high ) $ /xi)
987 my $lower_limit_is_80 = ($ranges[$i]->[0] == 0x80);
988 my $upper_limit_is_BF = ($ranges[$i]->[1] == 0xBF);
990 # If the range is the entire legal range, it matches any legal
991 # byte, so we can omit both tests. (This should happen only
992 # if the number of ranges is 1.)
993 if ($lower_limit_is_80 && $upper_limit_is_BF) {
996 elsif ($lower_limit_is_80) { # Just use the upper limit test
997 $output = sprintf("( $test <= $self->{val_fmt} )",
1000 elsif ($upper_limit_is_BF) { # Just use the lower limit test
1001 $output = sprintf("( $test >= $self->{val_fmt} )",
1006 # If we didn't change to omit a test above, see if the number of
1007 # elements is a power of 2 (only a single bit in the
1008 # representation of its count will be set) and if so, it may be
1009 # that a mask/compare optimization is possible.
1011 && pop_count($ranges[$i]->[1] - $ranges[$i]->[0] + 1) == 1)
1014 push @list, $_ for ($ranges[$i]->[0] .. $ranges[$i]->[1]);
1015 my @this_masks = calculate_mask(@list);
1017 # Use the mask if there is just one for the whole range.
1018 # Otherwise there is no savings over the two branches that can
1020 if (@this_masks == 1 && defined $this_masks[0][1]) {
1021 $output = sprintf "( $test & $self->{val_fmt} ) == $self->{val_fmt}", $this_masks[0][1], $this_masks[0][0];
1025 if ($output ne "") { # Prefer any optimization
1026 $ranges[$i] = $output;
1029 # No optimization happened. We need a test that the code
1030 # point is within both bounds. But, if the bounds are
1031 # adjacent code points, it is cleaner to say
1032 # 'first == test || second == test'
1034 # 'first <= test && test <= second'
1036 $range_count_extra++; # This range requires 2 branches to
1038 if ($ranges[$i]->[0] + 1 == $ranges[$i]->[1]) {
1040 . join( " || ", ( map
1041 { sprintf "$self->{val_fmt} == $test", $_ }
1045 else { # Full bounds checking
1046 $ranges[$i] = sprintf("( $self->{val_fmt} <= $test && $test <= $self->{val_fmt} )", $ranges[$i]->[0], $ranges[$i]->[1]);
1052 # We have generated the list of bytes in two ways; one trying to use masks
1053 # to cut the number of branches down, and the other to look at individual
1054 # ranges (some of which could be cut down by using a mask for just it).
1055 # We return whichever method uses the fewest branches.
1057 . join( " || ", (@masks && @masks < @ranges + $range_count_extra)
1064 # recursively turn a list of conditions into a fast break-out condition
1065 # used by _cond_as_str() for 'cp' type macros.
1067 my ( $self, $test, @cond )= @_;
1069 my $item= shift @cond;
1073 sprintf( "( $self->{val_fmt} <= $test && $test <= $self->{val_fmt} )",
1075 $gtv= sprintf "$self->{val_fmt}", $item->[1];
1077 $cstr= sprintf( "$self->{val_fmt} == $test", $item );
1078 $gtv= sprintf "$self->{val_fmt}", $item;
1081 my $combine= $self->_combine( $test, @cond );
1083 return "( $cstr || ( $gtv < $test &&\n"
1084 . $combine . " ) )";
1086 return "( $cstr || $combine )";
1094 # recursively convert an optree to text with reasonably neat formatting
1096 my ( $self, $op, $combine, $brace, $opts_ref )= @_;
1097 return 0 if ! defined $op; # The set is empty
1101 my $cond= $self->_cond_as_str( $op, $combine, $opts_ref );
1102 #no warnings 'recursion'; # This would allow really really inefficient
1103 # code to be generated. See pod
1104 my $yes= $self->_render( $op->{yes}, $combine, 1, $opts_ref );
1105 return $yes if $cond eq '1';
1107 my $no= $self->_render( $op->{no}, $combine, 0, $opts_ref );
1108 return "( $cond )" if $yes eq '1' and $no eq '0';
1109 my ( $lb, $rb )= $brace ? ( "( ", " )" ) : ( "", "" );
1110 return "$lb$cond ? $yes : $no$rb"
1111 if !ref( $op->{yes} ) && !ref( $op->{no} );
1113 my $ind= "\n" . ( $ind1 x $op->{depth} );
1115 if ( ref $op->{yes} ) {
1116 $yes= $ind . $ind1 . $yes;
1121 return "$lb$cond ?$yes$ind: $no$rb";
1124 # $expr=render($op,$combine)
1126 # convert an optree to text with reasonably neat formatting. If $combine
1127 # is true then the condition is created using "fast breakouts" which
1128 # produce uglier expressions that are more efficient for common case,
1129 # longer lists such as that resulting from type 'cp' output.
1130 # Currently only used for type 'cp' macros.
1132 my ( $self, $op, $combine, $opts_ref )= @_;
1133 my $str= "( " . $self->_render( $op, $combine, 0, $opts_ref ) . " )";
1134 return __clean( $str );
1138 # make a macro of a given type.
1139 # calls into make_trie and (generic_|length_)optree as needed
1141 # type : 'cp','cp_high', 'generic','high','low','latin1','utf8','LATIN1','UTF8'
1142 # ret_type : 'cp' or 'len'
1143 # safe : add length guards to macro
1145 # type defaults to 'generic', and ret_type to 'len' unless type is 'cp'
1146 # in which case it defaults to 'cp' as well.
1148 # it is illegal to do a type 'cp' macro on a pattern with multi-codepoint
1149 # sequences in it, as the generated macro will accept only a single codepoint
1152 # returns the macro.
1158 my $type= $opts{type} || 'generic';
1159 die "Can't do a 'cp' on multi-codepoint character class '$self->{op}'"
1161 and $self->{has_multi};
1162 my $ret_type= $opts{ret_type} || ( $opts{type} =~ /^cp/ ? 'cp' : 'len' );
1164 if ( $opts{safe} ) {
1165 $method= 'length_optree';
1166 } elsif ( $type eq 'generic' ) {
1167 $method= 'generic_optree';
1171 my $optree= $self->$method( %opts, type => $type, ret_type => $ret_type );
1172 my $text= $self->render( $optree, ($type =~ /^cp/) ? 1 : 0, \%opts );
1173 my @args= $type =~ /^cp/ ? 'cp' : 's';
1174 push @args, "e" if $opts{safe};
1175 push @args, "is_utf8" if $type eq 'generic';
1176 push @args, "len" if $ret_type eq 'both';
1177 my $pfx= $ret_type eq 'both' ? 'what_len_' :
1178 $ret_type eq 'cp' ? 'what_' : 'is_';
1179 my $ext= $type eq 'generic' ? '' : '_' . lc( $type );
1180 $ext .= "_safe" if $opts{safe};
1181 my $argstr= join ",", @args;
1182 return "/*** GENERATED CODE ***/\n"
1183 . __macro( "#define $pfx$self->{op}$ext($argstr)\n$text" );
1186 # if we arent being used as a module (highly likely) then process
1187 # the __DATA__ below and produce macros in regcharclass.h
1188 # if an argument is provided to the script then it is assumed to
1189 # be the path of the file to output to, if the arg is '-' outputs
1193 my $path= shift @ARGV || "regcharclass.h";
1195 if ( $path eq '-' ) {
1198 $out_fh = open_new( $path );
1200 print $out_fh read_only_top( lang => 'C', by => $0,
1201 file => 'regcharclass.h', style => '*',
1202 copyright => [2007, 2011] );
1203 print $out_fh "\n#ifndef H_REGCHARCLASS /* Guard against nested #includes */\n#define H_REGCHARCLASS 1\n\n";
1205 my ( $op, $title, @txt, @types, %mods );
1209 # Skip if to compile on a different platform.
1210 return if delete $mods{only_ascii_platform} && ! ASCII_PLATFORM;
1211 return if delete $mods{only_ebcdic_platform} && ord 'A' != 193;
1213 print $out_fh "/*\n\t$op: $title\n\n";
1214 print $out_fh join "\n", ( map { "\t$_" } @txt ), "*/", "";
1215 my $obj= __PACKAGE__->new( op => $op, title => $title, txt => \@txt );
1217 #die Dumper(\@types,\%mods);
1220 push @mods, 'safe' if delete $mods{safe};
1221 unshift @mods, 'fast' if delete $mods{fast} || ! @mods; # Default to 'fast'
1226 die "Unknown modifiers: ", join ", ", map { "'$_'" } sort keys %mods;
1229 foreach my $type_spec ( @types ) {
1230 my ( $type, $ret )= split /-/, $type_spec;
1232 foreach my $mod ( @mods ) {
1233 next if $mod eq 'safe' and $type =~ /^cp/;
1235 my $macro= $obj->make_macro(
1238 safe => $mod eq 'safe'
1240 print $out_fh $macro, "\n";
1246 s/^ \s* (?: \# .* ) ? $ //x; # squeeze out comment and blanks
1250 $doit->(); # This starts a new definition; do the previous one
1251 ( $op, $title )= split /\s*:\s*/, $_, 2;
1253 } elsif ( s/^=>// ) {
1254 my ( $type, $modifier )= split /:/, $_;
1255 @types= split ' ', $type;
1257 map { $mods{$_} = 1 } split ' ', $modifier;
1264 print $out_fh "\n#endif /* H_REGCHARCLASS */\n";
1267 print $out_fh "/* ex: set ro: */\n";
1269 read_only_bottom_close_and_rename($out_fh)
1273 # The form of the input is a series of definitions to make macros for.
1274 # The first line gives the base name of the macro, followed by a colon, and
1275 # then text to be used in comments associated with the macro that are its
1276 # title or description. In all cases the first (perhaps only) parameter to
1277 # the macro is a pointer to the first byte of the code point it is to test to
1278 # see if it is in the class determined by the macro. In the case of non-UTF8,
1279 # the code point consists only of a single byte.
1281 # The second line must begin with a '=>' and be followed by the types of
1282 # macro(s) to be generated; these are specified below. A colon follows the
1283 # types, followed by the modifiers, also specified below. At least one
1284 # modifier is required.
1286 # The subsequent lines give what code points go into the class defined by the
1287 # macro. Multiple characters may be specified via a string like "\x0D\x0A",
1288 # enclosed in quotes. Otherwise the lines consist of one of:
1289 # 1) a single Unicode code point, prefaced by 0x
1290 # 2) a single range of Unicode code points separated by a minus (and
1292 # 3) a single Unicode property specified in the standard Perl form
1294 # 4) a line like 'do path'. This will do a 'do' on the file given by
1295 # 'path'. It is assumed that this does nothing but load subroutines
1296 # (See item 5 below). The reason 'require path' is not used instead is
1297 # because 'do' doesn't assume that path is in @INC.
1298 # 5) a subroutine call
1299 # &pkg::foo(arg1, ...)
1300 # where pkg::foo was loaded by a 'do' line (item 4). The subroutine
1301 # returns an array of entries of forms like items 1-3 above. This
1302 # allows more complex inputs than achievable from the other input types.
1304 # A blank line or one whose first non-blank character is '#' is a comment.
1305 # The definition of the macro is terminated by a line unlike those described.
1308 # low generate a macro whose name is 'is_BASE_low' and defines a
1309 # class that includes only ASCII-range chars. (BASE is the
1310 # input macro base name.)
1311 # latin1 generate a macro whose name is 'is_BASE_latin1' and defines a
1312 # class that includes only upper-Latin1-range chars. It is not
1313 # designed to take a UTF-8 input parameter.
1314 # high generate a macro whose name is 'is_BASE_high' and defines a
1315 # class that includes all relevant code points that are above
1316 # the Latin1 range. This is for very specialized uses only.
1317 # It is designed to take only an input UTF-8 parameter.
1318 # utf8 generate a macro whose name is 'is_BASE_utf8' and defines a
1319 # class that includes all relevant characters that aren't ASCII.
1320 # It is designed to take only an input UTF-8 parameter.
1321 # LATIN1 generate a macro whose name is 'is_BASE_latin1' and defines a
1322 # class that includes both ASCII and upper-Latin1-range chars.
1323 # It is not designed to take a UTF-8 input parameter.
1324 # UTF8 generate a macro whose name is 'is_BASE_utf8' and defines a
1325 # class that can include any code point, adding the 'low' ones
1326 # to what 'utf8' works on. It is designed to take only an input
1328 # generic generate a macro whose name is 'is_BASE". It has a 2nd,
1329 # boolean, parameter which indicates if the first one points to
1330 # a UTF-8 string or not. Thus it works in all circumstances.
1331 # cp generate a macro whose name is 'is_BASE_cp' and defines a
1332 # class that returns true if the UV parameter is a member of the
1333 # class; false if not.
1334 # cp_high like cp, but it is assumed that it is known that the UV
1335 # parameter is above Latin1. The name of the generated macro is
1336 # 'is_BASE_cp_high'. This is different from high-cp, derived
1338 # A macro of the given type is generated for each type listed in the input.
1339 # The default return value is the number of octets read to generate the match.
1340 # Append "-cp" to the type to have it instead return the matched codepoint.
1341 # The macro name is changed to 'what_BASE...'. See pod for
1343 # Appending '-both" instead adds an extra parameter to the end of the argument
1344 # list, which is a pointer as to where to store the number of
1345 # bytes matched, while also returning the code point. The macro
1346 # name is changed to 'what_len_BASE...'. See pod for caveats
1349 # safe The input string is not necessarily valid UTF-8. In
1350 # particular an extra parameter (always the 2nd) to the macro is
1351 # required, which points to one beyond the end of the string.
1352 # The macro will make sure not to read off the end of the
1353 # string. In the case of non-UTF8, it makes sure that the
1354 # string has at least one byte in it. The macro name has
1355 # '_safe' appended to it.
1356 # fast The input string is valid UTF-8. No bounds checking is done,
1357 # and the macro can make assumptions that lead to faster
1359 # only_ascii_platform Skip this definition if this program is being run on
1360 # a non-ASCII platform.
1361 # only_ebcdic_platform Skip this definition if this program is being run on
1362 # a non-EBCDIC platform.
1363 # No modifier need be specified; fast is assumed for this case. If both
1364 # 'fast', and 'safe' are specified, two macros will be created for each
1367 # If run on a non-ASCII platform will automatically convert the Unicode input
1368 # to native. The documentation above is slightly wrong in this case. 'low'
1369 # actually refers to code points whose UTF-8 representation is the same as the
1370 # non-UTF-8 version (invariants); and 'latin1' refers to all the rest of the
1371 # code points less than 256.
1373 1; # in the unlikely case we are being used as a module
1376 # This is no longer used, but retained in case it is needed some day.
1377 # TRICKYFOLD: Problematic fold case letters. When adding to this list, also should add them to regcomp.c and fold_grind.t
1378 # => generic cp generic-cp generic-both :fast safe
1379 # 0x00DF # LATIN SMALL LETTER SHARP S
1380 # 0x0390 # GREEK SMALL LETTER IOTA WITH DIALYTIKA AND TONOS
1381 # 0x03B0 # GREEK SMALL LETTER UPSILON WITH DIALYTIKA AND TONOS
1382 # 0x1E9E # LATIN CAPITAL LETTER SHARP S, because maps to same as 00DF
1383 # 0x1FD3 # GREEK SMALL LETTER IOTA WITH DIALYTIKA AND OXIA; maps same as 0390
1384 # 0x1FE3 # GREEK SMALL LETTER UPSILON WITH DIALYTIKA AND OXIA; maps same as 03B0
1386 LNBREAK: Line Break: \R
1387 => generic UTF8 LATIN1 :fast safe
1388 "\x0D\x0A" # CRLF - Network (Windows) line ending
1391 HORIZWS: Horizontal Whitespace: \h \H
1392 => generic UTF8 LATIN1 high cp cp_high :fast safe
1395 VERTWS: Vertical Whitespace: \v \V
1396 => generic UTF8 high LATIN1 cp cp_high :fast safe
1399 XDIGIT: Hexadecimal digits
1400 => UTF8 high cp_high :fast
1403 XPERLSPACE: \p{XPerlSpace}
1404 => generic UTF8 high cp_high :fast
1407 REPLACEMENT: Unicode REPLACEMENT CHARACTER
1411 NONCHAR: Non character code points
1415 SURROGATE: Surrogate characters
1419 GCB_L: Grapheme_Cluster_Break=L
1423 GCB_LV_LVT_V: Grapheme_Cluster_Break=(LV or LVT or V)
1427 GCB_Prepend: Grapheme_Cluster_Break=Prepend
1431 GCB_RI: Grapheme_Cluster_Break=RI
1435 GCB_SPECIAL_BEGIN: Grapheme_Cluster_Break=special_begins
1437 \p{_X_Special_Begin}
1439 GCB_T: Grapheme_Cluster_Break=T
1443 GCB_V: Grapheme_Cluster_Break=V
1447 # This program was run with this enabled, and the results copied to utf8.h;
1448 # then this was commented out because it takes so long to figure out these 2
1449 # million code points. The results would not change unless utf8.h decides it
1450 # wants a maximum other than 4 bytes, or this program creates better
1452 #UTF8_CHAR: Matches utf8 from 1 to 4 bytes
1453 #=> UTF8 :safe only_ascii_platform
1456 # This hasn't been commented out, because we haven't an EBCDIC platform to run
1457 # it on, and the 3 types of EBCDIC allegedly supported by Perl would have
1459 UTF8_CHAR: Matches utf8 from 1 to 5 bytes
1460 => UTF8 :safe only_ebcdic_platform
1463 QUOTEMETA: Meta-characters that \Q should quote
1467 MULTI_CHAR_FOLD: multi-char strings that are folded to by a single character
1469 do regen/regcharclass_multi_char_folds.pl
1472 ®charclass_multi_char_folds::multi_char_folds(1)
1474 MULTI_CHAR_FOLD: multi-char strings that are folded to by a single character
1477 ®charclass_multi_char_folds::multi_char_folds(0)