5 use Unicode::UCD qw(prop_invlist prop_invmap);
6 require 'regen/regen_lib.pl';
8 # This program outputs charclass_invlists.h, which contains various inversion
9 # lists in the form of C arrays that are to be used as-is for inversion lists.
10 # Thus, the lists it contains are essentially pre-compiled, and need only a
11 # light-weight fast wrapper to make them usable at run-time.
13 # As such, this code knows about the internal structure of these lists, and
14 # any change made to that has to be done here as well. A random number stored
15 # in the headers is used to minimize the possibility of things getting
16 # out-of-sync, or the wrong data structure being passed. Currently that
18 my $VERSION_DATA_STRUCTURE_TYPE = 148565664;
20 my $out_fh = open_new('charclass_invlists.h', '>',
21 {style => '*', by => $0,
22 from => "Unicode::UCD"});
24 print $out_fh "/* See the generating file for comments */\n\n";
26 my %include_in_ext_re = ( NonL1_Perl_Non_Final_Folds => 1 );
28 sub output_invlist ($$) {
30 my $invlist = shift; # Reference to inversion list array
32 die "No inversion list for $name" unless defined $invlist
33 && ref $invlist eq 'ARRAY'
36 # Output the inversion list $invlist using the name $name for it.
37 # It is output in the exact internal form for inversion lists.
39 # Is the last element of the header 0, or 1 ?
41 if ($invlist->[0] != 0) {
45 my $count = @$invlist;
47 print $out_fh "\n#ifndef PERL_IN_XSUB_RE\n" unless exists $include_in_ext_re{$name};
48 print $out_fh "\nstatic const UV ${name}_invlist[] = {\n";
50 print $out_fh "\t$count,\t/* Number of elements */\n";
51 print $out_fh "\t$VERSION_DATA_STRUCTURE_TYPE, /* Version and data structure type */\n";
52 print $out_fh "\t", $zero_or_one,
53 ",\t/* 0 if the list starts at 0;",
54 "\n\t\t 1 if it starts at the element beyond 0 */\n";
56 # The main body are the UVs passed in to this routine. Do the final
58 for my $i (0 .. @$invlist - 1 - 1) {
59 print $out_fh "\t$invlist->[$i],\n";
62 # The final element does not have a trailing comma, as C can't handle it.
63 print $out_fh "\t$invlist->[-1]\n";
66 print $out_fh "\n#endif\n" unless exists $include_in_ext_re{$name};
70 sub mk_invlist_from_cp_list {
72 # Returns an inversion list constructed from the sorted input array of
77 # Initialize to just the first element
78 my @invlist = ( $list_ref->[0], $list_ref->[0] + 1);
80 # For each succeeding element, if it extends the previous range, adjust
81 # up, otherwise add it.
82 for my $i (1 .. @$list_ref - 1) {
83 if ($invlist[-1] == $list_ref->[$i]) {
87 push @invlist, $list_ref->[$i], $list_ref->[$i] + 1;
93 # Read in the Case Folding rules, and construct arrays of code points for the
95 my ($cp_ref, $folds_ref, $format) = prop_invmap("Case_Folding");
96 die "Could not find inversion map for Case_Folding" unless defined $format;
97 die "Incorrect format '$format' for Case_Folding inversion map"
98 unless $format eq 'al';
99 my @has_multi_char_fold;
100 my @is_non_final_fold;
102 for my $i (0 .. @$folds_ref - 1) {
103 next unless ref $folds_ref->[$i]; # Skip single-char folds
104 push @has_multi_char_fold, $cp_ref->[$i];
106 # Add to the non-finals list each code point that is in a non-final
108 for my $j (0 .. @{$folds_ref->[$i]} - 2) {
109 push @is_non_final_fold, $folds_ref->[$i][$j]
110 unless grep { $folds_ref->[$i][$j] == $_ } @is_non_final_fold;
114 sub _Perl_Multi_Char_Folds {
115 @has_multi_char_fold = sort { $a <=> $b } @has_multi_char_fold;
116 return mk_invlist_from_cp_list(\@has_multi_char_fold);
119 sub _Perl_Non_Final_Folds {
120 @is_non_final_fold = sort { $a <=> $b } @is_non_final_fold;
121 return mk_invlist_from_cp_list(\@is_non_final_fold);
126 for my $i (0 .. 255) { # Complicated because of EBCDIC
127 push @upper_latin1, $i if chr($i) =~ /[[:^ascii:]]/;
129 return mk_invlist_from_cp_list(\@upper_latin1);
132 output_invlist("Latin1", [ 0, 256 ]);
133 output_invlist("AboveLatin1", [ 256 ]);
135 # We construct lists for all the POSIX and backslash sequence character
136 # classes in two forms:
137 # 1) ones which match only in the ASCII range
138 # 2) ones which match either in the Latin1 range, or the entire Unicode range
140 # These get compiled in, and hence affect the memory footprint of every Perl
141 # program, even those not using Unicode. To minimize the size, currently
142 # the Latin1 version is generated for the beyond ASCII range except for those
143 # lists that are quite small for the entire range, such as for \s, which is 22
144 # UVs long plus 4 UVs (currently) for the header.
146 # To save even more memory, the ASCII versions could be derived from the
147 # larger ones at runtime, saving some memory (minus the expense of the machine
148 # instructions to do so), but these are all small anyway, so their total is
151 # In the list of properties below that get generated, the L1 prefix is a fake
152 # property that means just the Latin1 range of the full property (whose name
153 # has an X prefix instead of L1).
155 # An initial & means to use the subroutine from this file instead of an
156 # official inversion list.
189 &NonL1_Perl_Non_Final_Folds
190 &_Perl_Multi_Char_Folds
195 # For the Latin1 properties, we change to use the eXtended version of the
196 # base property, then go through the result and get rid of everything not
197 # in Latin1 (above 255). Actually, we retain the element for the range
198 # that crosses the 255/256 boundary if it is one that matches the
199 # property. For example, in the Word property, there is a range of code
200 # points that start at U+00F8 and goes through U+02C1. Instead of
201 # artificially cutting that off at 256 because 256 is the first code point
202 # above Latin1, we let the range go to its natural ending. That gives us
203 # extra information with no added space taken. But if the range that
204 # crosses the boundary is one that doesn't match the property, we don't
205 # start a new range above 255, as that could be construed as going to
206 # infinity. For example, the Upper property doesn't include the character
207 # at 255, but does include the one at 256. We don't include the 256 one.
208 my $prop_name = $prop;
209 my $is_local_sub = $prop_name =~ s/^&//;
210 my $lookup_prop = $prop_name;
211 my $l1_only = ($lookup_prop =~ s/^L1Posix/XPosix/ or $lookup_prop =~ s/^L1//);
213 $nonl1_only = $lookup_prop =~ s/^NonL1// unless $l1_only;
217 @invlist = eval $lookup_prop;
220 @invlist = prop_invlist($lookup_prop, '_perl_core_internal_ok');
222 die "Could not find inversion list for '$lookup_prop'" unless @invlist;
225 for my $i (0 .. @invlist - 1 - 1) {
226 if ($invlist[$i] > 255) {
228 # In an inversion list, even-numbered elements give the code
229 # points that begin ranges that match the property;
230 # odd-numbered give ones that begin ranges that don't match.
231 # If $i is odd, we are at the first code point above 255 that
232 # doesn't match, which means the range it is ending does
233 # match, and crosses the 255/256 boundary. We want to include
234 # this ending point, so increment $i, so the splice below
235 # includes it. Conversely, if $i is even, it is the first
236 # code point above 255 that matches, which means there was no
237 # matching range that crossed the boundary, and we don't want
238 # to include this code point, so splice before it.
241 # Remove everything past this.
247 elsif ($nonl1_only) {
249 for my $i (0 .. @invlist - 1 - 1) {
250 next if $invlist[$i] < 256;
252 # Here, we have the first element in the array that indicates an
253 # element above Latin1. Get rid of all previous ones.
254 splice @invlist, 0, $i;
256 # If this one's index is not divisible by 2, it means that this
257 # element is inverting away from being in the list, which means
258 # all code points from 256 to this one are in this list.
259 unshift @invlist, 256 if $i % 2 != 0;
263 die "No non-Latin1 code points in $lookup_prop" unless $found_nonl1;
266 output_invlist($prop_name, \@invlist);
269 read_only_bottom_close_and_rename($out_fh)