1 # Grind out a lot of combinatoric tests for folding.
3 binmode STDOUT, ":utf8";
9 require Config; import Config;
10 skip_all_if_miniperl("no dynamic loading on miniperl, no Encode nor POSIX");
13 use charnames ":full";
15 my $DEBUG = 0; # Outputs extra information for debugging this .t
22 # Special-cased characters in the .c's that we want to make sure get tested.
23 my %be_sure_to_test = (
24 "\xDF" => 1, # LATIN_SMALL_LETTER_SHARP_S
25 "\x{1E9E}" => 1, # LATIN_CAPITAL_LETTER_SHARP_S
26 "\x{390}" => 1, # GREEK_SMALL_LETTER_IOTA_WITH_DIALYTIKA_AND_TONOS
27 "\x{3B0}" => 1, # GREEK_SMALL_LETTER_UPSILON_WITH_DIALYTIKA_AND_TONOS
28 "\x{1FD3}" => 1, # GREEK SMALL LETTER IOTA WITH DIALYTIKA AND OXIA
29 "\x{1FE3}" => 1, # GREEK SMALL LETTER UPSILON WITH DIALYTIKA AND OXIA
33 # Tests both unicode and not, so make sure not implicitly testing unicode
34 no feature 'unicode_strings';
36 # Case-insensitive matching is a large and complicated issue. Perl does not
37 # implement it fully, properly. For example, it doesn't include normalization
38 # as part of the equation. To test every conceivable combination is clearly
39 # impossible; these tests are mostly drawn from visual inspection of the code
40 # and experience, trying to exercise all areas.
42 # There are three basic ranges of characters that Perl may treat differently:
43 # 1) Invariants under utf8 which on ASCII-ish machines are ASCII, and are
44 # referred to here as ASCII. On EBCDIC machines, the non-ASCII invariants
45 # are all controls that fold to themselves.
48 # 2) Other characters that fit into a byte but are different in utf8 than not;
49 # here referred to, taking some liberties, as Latin1.
52 # 3) Characters that won't fit in a byte; here referred to as Unicode
55 # Within these basic groups are equivalence classes that testing any character
56 # in is likely to lead to the same results as any other character. This is
57 # used to cut down the number of tests needed, unless PERL_RUN_SLOW_TESTS is
59 my $skip_apparently_redundant = ! $ENV{PERL_RUN_SLOW_TESTS};
61 # Additionally parts of this test run a lot of subtests, outputting the
62 # resulting TAP can be expensive so the tests are summarised internally. The
63 # PERL_DEBUG_FULL_TEST environment variable can be set to produce the full
64 # output for debugging purposes.
69 return $ASCII if $ord < 128;
70 return $Latin1 if $ord < 256;
78 my $list_all_tests = $ENV{PERL_DEBUG_FULL_TEST} || $DEBUG;
79 $| = 1 if $list_all_tests;
81 # Significant time is saved by not outputting each test but grouping the
82 # output into subtests
83 my $okays; # Number of ok's in current subtest
84 my $this_iteration; # Number of possible tests in current subtest
85 my $count=0; # Number of subtests = number of total tests
88 my ($test, $todo, $debug) = @_;
90 $debug = "" unless $DEBUG;
93 if (!$res || $list_all_tests) {
94 # Failed or debug; output the result
96 ok($res, "$test; $debug");
98 # Just count the test as passed
104 my %has_test_by_participants; # Makes sure has tests for each range and each
105 # number of characters that fold to the same
107 my %has_test_by_byte_count; # Makes sure has tests for each combination of
108 # n bytes folds to m bytes
110 my %tests; # The set of tests.
111 # Each key is a code point that folds to something else.
112 # Each value is a list of things that the key folds to. If the 'thing' is a
113 # single code point, it is that ordinal. If it is a multi-char fold, it is an
114 # ordered list of the code points in that fold. Here's an example for 'S':
115 # '83' => [ 115, 383 ]
117 # And one for a multi-char fold: \xDF
127 # [ # LATIN SMALL LETTER LONG S
131 # 7838 # LATIN_CAPITAL_LETTER_SHARP_S
134 my %inverse_folds; # keys are strings of the folded-to;
135 # values are lists of characters that fold to them
138 my ($to, @from) = @_;
140 # Called to cause the input to be tested by adding to %tests. @from is
141 # the list of characters that fold to the string $to. @from should be
142 # sorted so the lowest code point is first....
143 # The input is in string form; %tests uses code points, so have to
146 my $to_chars = length $to;
147 my @test_to; # List of tests for $to
149 if ($to_chars == 1) {
153 push @test_to, [ map { ord $_ } split "", $to ];
155 # For multi-char folds, we also test that things that can fold to each
156 # individual character in the fold also work. If we were testing
157 # comprehensively, we would try every combination of upper and lower
158 # case in the fold, but it will have to suffice to avoid running
159 # forever to make sure that each thing that folds to these is tested
160 # at least once. Because of complement matching ([^...]), we need to
161 # do both the folded, and the folded-from.
162 # We first look at each character in the multi-char fold, and save how
163 # many characters fold to it; and also the maximum number of such
165 my @folds_to_count; # 0th char in fold is index 0 ...
166 my $max_folds_to = 0;
168 for (my $i = 0; $i < $to_chars; $i++) {
169 my $to_char = substr($to, $i, 1);
170 if (exists $inverse_folds{$to_char}) {
171 $folds_to_count[$i] = scalar @{$inverse_folds{$to_char}};
172 $max_folds_to = $folds_to_count[$i] if $max_folds_to < $folds_to_count[$i];
175 $folds_to_count[$i] = 0;
179 # We will need to generate as many tests as the maximum number of
180 # folds, so that each fold will have at least one test.
181 # For example, consider character X which folds to the three character
182 # string 'xyz'. If 2 things fold to x (X and x), 4 to y (Y, Y'
183 # (Y-prime), Y'' (Y-prime-prime), and y), and 1 thing to z (itself), 4
184 # tests will be generated:
189 for (my $i = 0; $i < $max_folds_to; $i++) {
190 my @this_test_to; # Assemble a single test
192 # For each character in the multi-char fold ...
193 for (my $j = 0; $j < $to_chars; $j++) {
194 my $this_char = substr($to, $j, 1);
196 # Use its corresponding inverse fold, if available.
197 if ($i < $folds_to_count[$j]) {
198 push @this_test_to, ord $inverse_folds{$this_char}[$i];
200 else { # Or else itself.
201 push @this_test_to, ord $this_char;
205 # Add this test to the list
206 push @test_to, [ @this_test_to ];
209 # Here, have assembled all the tests for the multi-char fold. Sort so
210 # lowest code points are first for consistency and aesthetics in
211 # output. We know there are at least two characters in the fold, but
212 # I haven't bothered to worry about sorting on an optional third
213 # character if the first two are identical.
214 @test_to = sort { ($a->[0] == $b->[0])
215 ? $a->[1] <=> $b->[1]
216 : $a->[0] <=> $b->[0]
221 # This test is from n bytes to m bytes. Record that so won't try to add
222 # another test that does the same.
224 my $to_bytes = length $to;
225 foreach my $from_map (@from) {
226 $has_test_by_byte_count{length $from_map}{$to_bytes} = $to;
230 my $ord_smallest_from = ord shift @from;
231 if (exists $tests{$ord_smallest_from}) {
232 die "There are already tests for $ord_smallest_from"
235 # Add in the fold tests,
236 push @{$tests{$ord_smallest_from}}, @test_to;
238 # Then any remaining froms in the equivalence class.
239 push @{$tests{$ord_smallest_from}}, map { ord $_ } @from;
242 # Get the Unicode rules and construct inverse mappings from them
245 my $file="../lib/unicore/CaseFolding.txt";
247 # Use the Unicode data file if we are on an ASCII platform (which its data is
248 # for), and it is in the modern format (starting in Unicode 3.1.0) and it is
249 # available. This avoids being affected by potential bugs introduced by other
252 && pack("C*", split /\./, Unicode::UCD::UnicodeVersion()) ge v3.1.0
253 && open my $fh, "<", $file)
258 # Lines look like (though without the initial '#')
259 #0130; F; 0069 0307; # LATIN CAPITAL LETTER I WITH DOT ABOVE
261 # Get rid of comments, ignore blank or comment-only lines
262 my $line = $_ =~ s/ (?: \s* \# .* )? $ //rx;
263 next unless length $line;
264 my ($hex_from, $fold_type, @hex_folded) = split /[\s;]+/, $line;
266 next if $fold_type =~ / ^ [IT] $/x; # Perl doesn't do Turkish folding
267 next if $fold_type eq 'S'; # If Unicode's tables are correct, the F
268 # should be a superset of S
270 my $folded_str = pack ("U0U*", map { hex $_ } @hex_folded);
271 push @{$inverse_folds{$folded_str}}, chr hex $hex_from;
274 else { # Here, can't use the .txt file: read the Unicode rules file and
275 # construct inverse mappings from it
277 my ($invlist_ref, $invmap_ref, undef, $default)
278 = Unicode::UCD::prop_invmap('Case_Folding');
279 for my $i (0 .. @$invlist_ref - 1 - 1) {
280 next if $invmap_ref->[$i] == $default;
282 # Make into an array if not so already, so can treat uniformly below
283 $invmap_ref->[$i] = [ $invmap_ref->[$i] ] if ! ref $invmap_ref->[$i];
285 # Each subsequent element of the range requires adjustment of +1 from
286 # the previous element
288 for my $j ($invlist_ref->[$i] .. $invlist_ref->[$i+1] -1) {
291 = pack "U0U*", map { $_ + $adjust } @{$invmap_ref->[$i]};
292 #note (sprintf "%d: %04X: %s", __LINE__, $j, join " ",
293 # map { sprintf "%04X", $_ + $adjust } @{$invmap_ref->[$i]});
294 push @{$inverse_folds{$folded_str}}, chr $j;
299 # Analyze the data and generate tests to get adequate test coverage. We sort
300 # things so that smallest code points are done first.
302 foreach my $to (sort { (length $a == length $b)
304 : length $a <=> length $b
305 } keys %inverse_folds)
308 # Within each fold, sort so that the smallest code points are done first
309 @{$inverse_folds{$to}} = sort { $a cmp $b } @{$inverse_folds{$to}};
310 my @from = @{$inverse_folds{$to}};
312 # Just add it to the tests if doing complete coverage
313 if (! $skip_apparently_redundant) {
314 add_test($to, @from);
318 my $to_chars = length $to;
319 my $to_range_type = range_type(substr($to, 0, 1));
321 # If this is required to be tested, do so. We check for these first, as
322 # they will take up slots of byte-to-byte combinations that we otherwise
323 # would have to have other tests to get.
324 foreach my $from_map (@from) {
325 if (exists $be_sure_to_test{$from_map}) {
326 add_test($to, @from);
331 # If the fold contains heterogeneous range types, is suspect and should be
334 foreach my $char (split "", $to) {
335 if (range_type($char) != $to_range_type) {
336 add_test($to, @from);
342 # If the mapping crosses range types, is suspect and should be tested
343 foreach my $from_map (@from) {
344 if (range_type($from_map) != $to_range_type) {
345 add_test($to, @from);
350 # Here, all components of the mapping are in the same range type. For
351 # single character folds, we test one case in each range type that has 2
352 # particpants, 3 particpants, etc.
353 if ($to_chars == 1) {
354 if (! exists $has_test_by_participants{scalar @from}{$to_range_type}) {
355 add_test($to, @from);
356 $has_test_by_participants{scalar @from}{$to_range_type} = $to;
361 # We also test all combinations of mappings from m to n bytes. This is
362 # because the regex optimizer cares. (Don't bother worrying about that
363 # Latin1 chars will occupy a different number of bytes under utf8, as
364 # there are plenty of other cases that catch these byte numbers.)
366 my $to_bytes = length $to;
367 foreach my $from_map (@from) {
368 if (! exists $has_test_by_byte_count{length $from_map}{$to_bytes}) {
369 add_test($to, @from);
375 # For each range type, test additionally a character that folds to itself
376 add_test(chr 0x3A, chr 0x3A);
377 add_test(chr 0xF7, chr 0xF7);
378 add_test(chr 0x2C7, chr 0x2C7);
380 # To cut down on the number of tests
381 my $has_tested_aa_above_latin1;
382 my $has_tested_latin1_aa;
383 my $has_tested_ascii_aa;
384 my $has_tested_l_above_latin1;
385 my $has_tested_above_latin1_l;
386 my $has_tested_ascii_l;
387 my $has_tested_above_latin1_d;
388 my $has_tested_ascii_d;
389 my $has_tested_non_latin1_d;
390 my $has_tested_above_latin1_a;
391 my $has_tested_ascii_a;
392 my $has_tested_non_latin1_a;
394 # For use by pairs() in generating combinations
400 # Returns all ordered combinations of pairs of elements from the input array.
401 # It doesn't return pairs like (a, a), (b, b). Change the slice to an array
402 # to do that. This was just to have fewer tests.
404 #print __LINE__, ": ", join(" XXX ", map { sprintf "%04X", $_ } @_), "\n";
405 map { prefix $_[$_], @_[0..$_-1, $_+1..$#_] } 0..$#_
408 my @charsets = qw(d u a aa);
409 if($Config{d_setlocale}) {
410 my $current_locale = POSIX::setlocale( &POSIX::LC_ALL, "C") // "";
411 if ($current_locale eq 'C') {
412 require locale; import locale;
414 # Some implementations don't have the 128-255 range characters all
415 # mean nothing under the C locale (an example being VMS). This is
416 # legal, but since we don't know what the right answers should be,
417 # skip the locale tests in that situation.
418 for my $i (128 .. 255) {
420 goto untestable_locale if uc($char) ne $char || lc($char) ne $char;
427 # Finally ready to do the tests
428 foreach my $test (sort { numerically } keys %tests) {
431 my $previous_pattern;
432 my @pairs = pairs(sort numerically $test, @{$tests{$test}});
434 # Each fold can be viewed as a closure of all the characters that
435 # participate in it. Look at each possible pairing from a closure, with the
436 # first member of the pair the target string to match against, and the
437 # second member forming the pattern. Thus each fold member gets tested as
438 # the string, and the pattern with every other member in the opposite role.
439 while (my $pair = shift @pairs) {
440 my ($target, $pattern) = @$pair;
442 # When testing a char that doesn't fold, we can get the same
443 # permutation twice; so skip all but the first.
444 next if $previous_target
445 && $previous_target == $target
446 && $previous_pattern == $pattern;
447 ($previous_target, $previous_pattern) = ($target, $pattern);
449 # Each side may be either a single char or a string. Extract each into an
450 # array (perhaps of length 1)
451 my @target, my @pattern;
452 @target = (ref $target) ? @$target : $target;
453 @pattern = (ref $pattern) ? @$pattern : $pattern;
455 # We are testing just folds to/from a single character. If our pairs
456 # happens to generate multi/multi, skip.
457 next if @target > 1 && @pattern > 1;
459 # Have to convert non-utf8 chars to native char set
460 @target = map { $_ > 255 ? $_ : ord latin1_to_native(chr($_)) } @target;
461 @pattern = map { $_ > 255 ? $_ : ord latin1_to_native(chr($_)) } @pattern;
464 my @x_target = map { sprintf "\\x{%04X}", $_ } @target;
465 my @x_pattern = map { sprintf "\\x{%04X}", $_ } @pattern;
467 my $target_above_latin1 = grep { $_ > 255 } @target;
468 my $pattern_above_latin1 = grep { $_ > 255 } @pattern;
469 my $target_has_ascii = grep { $_ < 128 } @target;
470 my $pattern_has_ascii = grep { $_ < 128 } @pattern;
471 my $target_only_ascii = ! grep { $_ > 127 } @target;
472 my $pattern_only_ascii = ! grep { $_ > 127 } @pattern;
473 my $target_has_latin1 = grep { $_ < 256 } @target;
474 my $target_has_upper_latin1 = grep { $_ < 256 && $_ > 127 } @target;
475 my $pattern_has_upper_latin1 = grep { $_ < 256 && $_ > 127 } @pattern;
476 my $pattern_has_latin1 = grep { $_ < 256 } @pattern;
477 my $is_self = @target == 1 && @pattern == 1 && $target[0] == $pattern[0];
479 # We don't test multi-char folding into other multi-chars. We are testing
480 # a code point that folds to or from other characters. Find the single
481 # code point for diagnostic purposes. (If both are single, choose the
483 my $ord = @target == 1 ? $target[0] : $pattern[0];
484 my $progress = sprintf "%04X: \"%s\" and /%s/",
487 join("", @x_pattern);
490 # Now grind out tests, using various combinations.
491 foreach my $charset (@charsets) {
495 # To cut down somewhat on the enormous quantity of tests this currently
496 # runs, skip some for some of the character sets whose results aren't
497 # likely to differ from others. But run all tests on the code points
498 # that don't fold, plus one other set in each range group.
501 # /aa should only affect things with folds in the ASCII range. But, try
502 # it on one set in the other ranges just to make sure it doesn't break
504 if ($charset eq 'aa') {
505 if (! $target_has_ascii && ! $pattern_has_ascii) {
506 if ($target_above_latin1 || $pattern_above_latin1) {
507 next if defined $has_tested_aa_above_latin1
508 && $has_tested_aa_above_latin1 != $test;
509 $has_tested_aa_above_latin1 = $test;
511 next if defined $has_tested_latin1_aa
512 && $has_tested_latin1_aa != $test;
513 $has_tested_latin1_aa = $test;
515 elsif ($target_only_ascii && $pattern_only_ascii) {
517 # And, except for one set just to make sure, skip tests
518 # where both elements in the pair are ASCII. If one works for
519 # aa, the others are likely too. This skips tests where the
520 # fold is from non-ASCII to ASCII, but this part of the test
521 # is just about the ASCII components.
522 next if defined $has_tested_ascii_l
523 && $has_tested_ascii_l != $test;
524 $has_tested_ascii_l = $test;
527 elsif ($charset eq 'l') {
529 # For l, don't need to test beyond one set those things that are
530 # all above latin1, because unlikely to have different successes
532 if (! $target_has_latin1 && ! $pattern_has_latin1) {
533 next if defined $has_tested_above_latin1_l
534 && $has_tested_above_latin1_l != $test;
535 $has_tested_above_latin1_l = $test;
537 elsif ($target_only_ascii && $pattern_only_ascii) {
539 # And, except for one set just to make sure, skip tests
540 # where both elements in the pair are ASCII. This is
541 # essentially the same reasoning as above for /aa.
542 next if defined $has_tested_ascii_l
543 && $has_tested_ascii_l != $test;
544 $has_tested_ascii_l = $test;
547 elsif ($charset eq 'd') {
548 # Similarly for d. Beyond one test (besides self) each, we don't
549 # test pairs that are both ascii; or both above latin1, or are
550 # combinations of ascii and above latin1.
551 if (! $target_has_upper_latin1 && ! $pattern_has_upper_latin1) {
552 if ($target_has_ascii && $pattern_has_ascii) {
553 next if defined $has_tested_ascii_d
554 && $has_tested_ascii_d != $test;
555 $has_tested_ascii_d = $test
557 elsif (! $target_has_latin1 && ! $pattern_has_latin1) {
558 next if defined $has_tested_above_latin1_d
559 && $has_tested_above_latin1_d != $test;
560 $has_tested_above_latin1_d = $test;
563 next if defined $has_tested_non_latin1_d
564 && $has_tested_non_latin1_d != $test;
565 $has_tested_non_latin1_d = $test;
569 elsif ($charset eq 'a') {
570 # Similarly for a. This should match identically to /u, so wasn't
571 # tested at all until a bug was found that was thereby missed.
572 # As a compromise, beyond one test (besides self) each, we don't
573 # test pairs that are both ascii; or both above latin1, or are
574 # combinations of ascii and above latin1.
575 if (! $target_has_upper_latin1 && ! $pattern_has_upper_latin1) {
576 if ($target_has_ascii && $pattern_has_ascii) {
577 next if defined $has_tested_ascii_a
578 && $has_tested_ascii_a != $test;
579 $has_tested_ascii_a = $test
581 elsif (! $target_has_latin1 && ! $pattern_has_latin1) {
582 next if defined $has_tested_above_latin1_a
583 && $has_tested_above_latin1_a != $test;
584 $has_tested_above_latin1_a = $test;
587 next if defined $has_tested_non_latin1_a
588 && $has_tested_non_latin1_a != $test;
589 $has_tested_non_latin1_a = $test;
595 foreach my $utf8_target (0, 1) { # Both utf8 and not, for
597 my $upgrade_target = "";
599 # These must already be in utf8 because the string to match has
600 # something above latin1. So impossible to test if to not to be in
601 # utf8; and otherwise, no upgrade is needed.
602 next if $target_above_latin1 && ! $utf8_target;
603 $upgrade_target = ' utf8::upgrade($c);' if ! $target_above_latin1 && $utf8_target;
605 foreach my $utf8_pattern (0, 1) {
606 next if $pattern_above_latin1 && ! $utf8_pattern;
608 # Our testing of 'l' uses the POSIX locale, which is ASCII-only
609 my $uni_semantics = $charset ne 'l' && ($utf8_target || $charset eq 'u' || ($charset eq 'd' && $utf8_pattern) || $charset =~ /a/);
610 my $upgrade_pattern = "";
611 $upgrade_pattern = ' utf8::upgrade($p);' if ! $pattern_above_latin1 && $utf8_pattern;
613 my $lhs = join "", @x_target;
614 my $lhs_str = eval qq{"$lhs"}; fail($@) if $@;
615 my @rhs = @x_pattern;
616 my $rhs = join "", @rhs;
617 my $should_fail = (! $uni_semantics && $ord >= 128 && $ord < 256 && ! $is_self)
618 || ($charset eq 'aa' && $target_has_ascii != $pattern_has_ascii)
619 || ($charset eq 'l' && $target_has_latin1 != $pattern_has_latin1);
621 # Do simple tests of referencing capture buffers, named and
624 $op = '!~' if $should_fail;
626 my $todo = 0; # No longer any todo's
627 my $eval = "my \$c = \"$lhs$rhs\"; my \$p = qr/(?$charset:^($rhs)\\1\$)/i;$upgrade_target$upgrade_pattern \$c $op \$p";
628 run_test($eval, $todo, "");
630 $eval = "my \$c = \"$lhs$rhs\"; my \$p = qr/(?$charset:^(?<grind>$rhs)\\k<grind>\$)/i;$upgrade_target$upgrade_pattern \$c $op \$p";
631 run_test($eval, $todo, "");
634 $eval = "my \$c = \"$rhs$lhs\"; my \$p = qr/(?$charset:^($rhs)\\1\$)/i;$upgrade_target$upgrade_pattern \$c $op \$p";
635 run_test($eval, "", "");
637 $eval = "my \$c = \"$rhs$lhs\"; my \$p = qr/(?$charset:^(?<grind>$rhs)\\k<grind>\$)/i;$upgrade_target$upgrade_pattern \$c $op \$p";
638 run_test($eval, "", "");
641 # See if works on what could be a simple trie.
642 $eval = "my \$c = \"$lhs\"; my \$p = qr/$rhs|xyz/i$charset;$upgrade_target$upgrade_pattern \$c $op \$p";
643 run_test($eval, "", "");
645 # Check that works when the folded character follows something that
646 # is quantified. This test knows the regex code internals to the
647 # extent that it knows this is a potential problem, and that there
648 # are three different types of quantifiers generated: 1) The thing
649 # being quantified matches a single character; 2) it matches more
650 # than one character, but is fixed width; 3) it can match a variable
651 # number of characters. (It doesn't know that case 3 shouldn't
652 # matter, since it doesn't do anything special for the character
653 # following the quantifier; nor that some of the different
654 # quantifiers execute the same underlying code, as these tests are
655 # quick, and this insulates these tests from changes in the
657 for my $quantifier ('?', '??', '*', '*?', '+', '+?', '{1,2}', '{1,2}?') {
658 $eval = "my \$c = \"_$lhs\"; my \$p = qr/(?$charset:.$quantifier$rhs)/i;$upgrade_target$upgrade_pattern \$c $op \$p";
659 run_test($eval, "", "");
660 $eval = "my \$c = \"__$lhs\"; my \$p = qr/(?$charset:(?:..)$quantifier$rhs)/i;$upgrade_target$upgrade_pattern \$c $op \$p";
661 run_test($eval, "", "");
662 $eval = "my \$c = \"__$lhs\"; my \$p = qr/(?$charset:(?:.|\\R)$quantifier$rhs)/i;$upgrade_target$upgrade_pattern \$c $op \$p";
663 run_test($eval, "", "");
666 foreach my $bracketed (0, 1) { # Put rhs in [...], or not
667 next if $bracketed && @pattern != 1; # bracketed makes these
668 # or's instead of a sequence
669 foreach my $optimize_bracketed (0, 1) {
670 next if $optimize_bracketed && ! $bracketed;
671 foreach my $inverted (0,1) {
672 next if $inverted && ! $bracketed; # inversion only valid in [^...]
673 next if $inverted && @target != 1; # [perl #89750] multi-char
674 # not valid in [^...]
676 # In some cases, add an extra character that doesn't fold, and
677 # looks ok in the output.
678 my $extra_char = "_";
679 foreach my $prepend ("", $extra_char) {
680 foreach my $append ("", $extra_char) {
682 # Assemble the rhs. Put each character in a separate
683 # bracketed if using charclasses. This creates a stress on
684 # the code to span a match across multiple elements
686 foreach my $rhs_char (@rhs) {
687 $rhs .= '[' if $bracketed;
688 $rhs .= '^' if $inverted;
691 # Add a character to the class, so class doesn't get
692 # optimized out, unless we are testing that optimization
693 $rhs .= '_' if $optimize_bracketed;
694 $rhs .= ']' if $bracketed;
697 # Add one of: no capturing parens
700 # Use quantifiers and extra variable width matches inside
701 # them to keep some optimizations from happening
702 foreach my $parend (0, 1, 2) {
703 my $interior = (! $parend)
708 foreach my $quantifier ("", '?', '*', '+', '{1,3}') {
710 # Perhaps should be TODOs, as are unimplemented, but
711 # maybe will never be implemented
712 next if @pattern != 1 && $quantifier;
714 # A ? or * quantifier normally causes the thing to be
715 # able to match a null string
716 my $quantifier_can_match_null = $quantifier eq '?' || $quantifier eq '*';
718 # But since we only quantify the last character in a
719 # multiple fold, the other characters will have width,
720 # except if we are quantifying the whole rhs
721 my $can_match_null = $quantifier_can_match_null && (@rhs == 1 || $parend);
723 foreach my $l_anchor ("", '^') { # '\A' didn't change result)
724 foreach my $r_anchor ("", '$') { # '\Z', '\z' didn't change result)
726 # The folded part can match the null string if it
727 # isn't required to have width, and there's not
728 # something on one or both sides that force it to.
729 my $both_sides = ($l_anchor && $r_anchor) || ($l_anchor && $append) || ($r_anchor && $prepend) || ($prepend && $append);
730 my $must_match = ! $can_match_null || $both_sides;
731 # for performance, but doing this missed many failures
732 #next unless $must_match;
733 my $quantified = "(?$charset:$l_anchor$prepend$interior${quantifier}$append$r_anchor)";
735 if ($must_match && $should_fail) {
740 $op = ! $op if $must_match && $inverted;
742 if ($inverted && @target > 1) {
743 # When doing an inverted match against a
744 # multi-char target, and there is not something on
745 # the left to anchor the match, if it shouldn't
746 # succeed, skip, as what will happen (when working
747 # correctly) is that it will match the first
748 # position correctly, and then be inverted to not
749 # match; then it will go to the second position
750 # where it won't match, but get inverted to match,
751 # and hence succeeding.
752 next if ! ($l_anchor || $prepend) && ! $op;
754 # Can't ever match for latin1 code points non-uni
755 # semantics that have a inverted multi-char fold
756 # when there is something on both sides and the
757 # quantifier isn't such as to span the required
758 # width, which is 2 or 3.
759 $op = 0 if $ord < 255
762 && ( ! $quantifier || $quantifier eq '?')
765 # Similarly can't ever match when inverting a multi-char
766 # fold for /aa and the quantifier isn't sufficient
767 # to allow it to span to both sides.
768 $op = 0 if $target_has_ascii && $charset eq 'aa' && $both_sides && ( ! $quantifier || $quantifier eq '?') && $parend < 2;
771 $op = 0 if $target_has_latin1 && $charset eq 'l' && $both_sides && ( ! $quantifier || $quantifier eq '?') && $parend < 2;
775 my $desc = "my \$c = \"$prepend$lhs$append\"; "
776 . "my \$p = qr/$quantified/i;"
777 . "$upgrade_target$upgrade_pattern "
778 . "\$c " . ($op ? "=~" : "!~") . " \$p; ";
781 "; uni_semantics=$uni_semantics, "
782 . "should_fail=$should_fail, "
783 . "bracketed=$bracketed, "
784 . "prepend=$prepend, "
787 . "quantifier=$quantifier, "
788 . "l_anchor=$l_anchor, "
789 . "r_anchor=$r_anchor; "
790 . "pattern_above_latin1=$pattern_above_latin1; "
791 . "utf8_pattern=$utf8_pattern"
795 my $c = "$prepend$lhs_str$append";
796 my $p = qr/$quantified/i;
797 utf8::upgrade($c) if length($upgrade_target);
798 utf8::upgrade($p) if length($upgrade_pattern);
799 my $res = $op ? ($c =~ $p): ($c !~ $p);
801 if (!$res || $list_all_tests) {
802 # Failed or debug; output the result
804 ok($res, "test $count - $desc");
806 # Just count the test as passed
821 unless($list_all_tests) {
823 is $okays, $this_iteration, "$okays subtests ok for"
825 . ' target="' . join("", @x_target) . '",'
826 . ' pat="' . join("", @x_pattern) . '"';