# - SpecialCasing.txt data for case mapping
# - Title case offsets
# - Some kind of hinting for composition
-# - Word boundary support
# - ...
#
# NB the generated files DO NOT offer a stable ABI and so are not immediately
# Size of a subtable
#
# This can be varied to trade off the number of subtables against their size.
-our $modulus = 128;
+# 16 gave the smallest results last time I checked (on a Mac with a 32-bit
+# build).
+our $modulus = 16;
+
+if(@ARGV) {
+ $modulus = shift;
+}
+
+# Where to break the table. There is a huge empty section of the Unicode
+# code space and we deal with this by simply leaving it out of the table.
+# This complicates the lookup function a little but should not affect
+# performance in the cases we care about.
+our $break_start = 0x30000;
+our $break_end = 0xE0000;
+
+# Similarly we simply omit the very top of the table and sort it out in the
+# lookup function.
+our $break_top = 0xE0200;
my %cats = (); # known general categories
my %data = (); # mapping of codepoints to information
# Make sure we have our desired input files. We explicitly specify a
# Unicode standard version to make sure that a given version of DisOrder
# supports a given version of Unicode.
-sub need_input {
+sub input {
my $path = shift;
my $lpath = basename($path);
if(!-e $lpath) {
system("wget http://www.unicode.org/Public/5.0.0/ucd/$path");
chmod(0444, $lpath) or die "$lpath: $!\n";
}
+ open(STDIN, "<$lpath") or die "$lpath: $!\n";
+ print STDERR "Reading $lpath...\n";
}
-need_input("UnicodeData.txt");
-need_input("CaseFolding.txt");
-need_input("auxiliary/GraphemeBreakProperty.txt");
# Read the main data file
-open(STDIN, "<UnicodeData.txt") or die "UnicodeData.txt: $!\n";
+input("UnicodeData.txt");
+my ($start, $end);
while(<>) {
my @f = split(/;/, $_);
my $c = hex($f[0]); # codepoint
- next if $c >= 0xE0000; # ignore various high-numbered stuff
- # TODO justify this exclusion!
my $name = $f[1];
+ die "$f[0] $name is in the break\n"
+ if $c >= $break_start && $c < $break_end;
my $gc = $f[2]; # General_Category
+ # Variuos GCs we don't expect to see in UnicodeData.txt
+ $cats{$gc} = 1; # always record all GCs
+ if($name =~ /first>/i) {
+ $start = $c;
+ next;
+ } elsif($name =~ /last>/i) {
+ $end = $c;
+ } else {
+ $start = $end = $c;
+ }
+ die "unexpected Cn" if $gc eq 'Cn';
my $ccc = $f[3]; # Canonical_Combining_Class
my $dm = $f[5]; # Decomposition_Type + Decomposition_Mapping
my $sum = hex($f[12]) || $c; # Simple_Uppercase_Mapping
$maxud = $ud if $ud > $maxud;
$minld = $ld if $ld < $minld;
$maxld = $ld if $ld > $maxld;
- $data{$c} = {
- "gc" => $gc,
- "ccc" => $ccc,
- "ud" => $ud,
- "ld" => $ld,
+ if($start != $end) {
+ printf STDERR "> range %04X-%04X is %s\n", $start, $end, $gc;
+ }
+ for($c = $start; $c <= $end; ++$c) {
+ my $d = {
+ "gc" => $gc,
+ "ccc" => $ccc,
+ "ud" => $ud,
+ "ld" => $ld,
};
- if($dm ne '') {
- if($dm !~ /</) {
- # This is a canonical decomposition
- $data{$c}->{canon} = $dm;
- $data{$c}->{compat} = $dm;
- } else {
- # This is only a compatibility decomposition
- $dm =~ s/^<.*>\s*//;
- $data{$c}->{compat} = $dm;
+ if($dm ne '') {
+ if($dm !~ /</) {
+ # This is a canonical decomposition
+ $d->{canon} = $dm;
+ $d->{compat} = $dm;
+ } else {
+ # This is only a compatibility decomposition
+ $dm =~ s/^<.*>\s*//;
+ $d->{compat} = $dm;
+ }
}
+ $data{$c} = $d;
}
$cats{$gc} = 1;
- $max = $c if $c > $max;
+ $max = $end if $end > $max;
}
-# Grapheme break data
-# NB we do this BEFORE filling in blanks so that the Hangul characters
-# don't get filled in; we can compute their properties mechanically.
-open(STDIN, "<GraphemeBreakProperty.txt") or die "GraphemeBreakProperty.txt: $!\n";
-while(<>) {
- chomp;
- s/\s*\#.*//;
- next if $_ eq '';
- my ($range, $propval) = split(/\s*;\s*/, $_);
- if($range =~ /(.*)\.\.(.*)/) {
- for my $c (hex($1) .. hex($2)) {
- if(exists $data{$c}) {
- $data{$c}->{gbreak} = $propval;
+sub read_prop_with_ranges {
+ my $path = shift;
+ my $propkey = shift;
+ input($path);
+ while(<>) {
+ chomp;
+ s/\s*\#.*//;
+ next if $_ eq '';
+ my ($range, $propval) = split(/\s*;\s*/, $_);
+ if($range =~ /(.*)\.\.(.*)/) {
+ for my $c (hex($1) .. hex($2)) {
+ die "($range)\n" if($c == 0xAC00 and $propkey eq 'gbreak');
+ $data{$c}->{$propkey} = $propval;
}
+ } else {
+ my $c = hex($range);
+ $data{$c}->{$propkey} = $propval;
}
- } else {
- my $c = hex($range);
- if(exists $data{$c}) {
- $data{$c}->{gbreak} = $propval;
+ }
+}
+
+# Grapheme_Break etc
+read_prop_with_ranges("auxiliary/GraphemeBreakProperty.txt", "gbreak");
+read_prop_with_ranges("auxiliary/WordBreakProperty.txt", "wbreak");
+read_prop_with_ranges("auxiliary/SentenceBreakProperty.txt", "sbreak");
+
+# Compute the full list and fill in the Extend category properly
+my %gbreak = ();
+my %wbreak = ();
+my %sbreak = ();
+for my $c (keys %data) {
+ if(!exists $data{$c}->{gbreak}) {
+ $data{$c}->{gbreak} = 'Other';
+ }
+ $gbreak{$data{$c}->{gbreak}} = 1;
+
+ if(!exists $data{$c}->{wbreak}) {
+ if($data{$c}->{gbreak} eq 'Extend') {
+ $data{$c}->{wbreak} = 'Extend';
+ } else {
+ $data{$c}->{wbreak} = 'Other';
+ }
+ }
+ $wbreak{$data{$c}->{wbreak}} = 1;
+
+ if(!exists $data{$c}->{sbreak}) {
+ if($data{$c}->{gbreak} eq 'Extend') {
+ $data{$c}->{sbreak} = 'Extend';
+ } else {
+ $data{$c}->{sbreak} = 'Other';
}
}
+ $sbreak{$data{$c}->{sbreak}} = 1;
}
# Round up the maximum value to a whole number of subtables
$max += ($modulus - 1) - ($max % $modulus);
-# Make sure there are no gaps
+# Private use characters
+# We only fill in values below $max, utf32__unidata()
+my $Co = {
+ "gc" => "Co",
+ "ccc" => 0,
+ "ud" => 0,
+ "ld" => 0
+};
+for(my $c = 0xE000; $c <= 0xF8FF && $c <= $max; ++$c) {
+ $data{$c} = $Co;
+}
+for(my $c = 0xF0000; $c <= 0xFFFFD && $c <= $max; ++$c) {
+ $data{$c} = $Co;
+}
+for(my $c = 0x100000; $c <= 0x10FFFD && $c <= $max; ++$c) {
+ $data{$c} = $Co;
+}
+
+# Anything left is not assigned
+my $Cn = {
+ "gc" => "Cn", # not assigned
+ "ccc" => 0,
+ "ud" => 0,
+ "ld" => 0
+};
for(my $c = 0; $c <= $max; ++$c) {
if(!exists $data{$c}) {
- $data{$c} = {
- "gc" => "Cn", # not assigned
- "ccc" => 0,
- "ud" => 0,
- "ld" => 0
- };
+ $data{$c} = $Cn;
+ }
+ if(!exists $data{$c}->{wbreak}) {
+ $data{$c}->{wbreak} = 'Other';
+ }
+ if(!exists $data{$c}->{gbreak}) {
+ $data{$c}->{gbreak} = 'Other';
+ }
+ if(!exists $data{$c}->{sbreak}) {
+ $data{$c}->{sbreak} = 'Other';
}
}
$cats{'Cn'} = 1;
# Read the casefolding data too
-open(STDIN, "<CaseFolding.txt") or die "CaseFolding.txt: $!\n";
+input("CaseFolding.txt");
while(<>) {
chomp;
next if /^\#/ or $_ eq '';
}
# Generate the header file
+print STDERR "Generating unidata.h...\n";
open(STDOUT, ">unidata.h") or die "unidata.h: $!\n";
out("/* Automatically generated file, see scripts/make-unidata */\n",
"#define UNIDATA_H\n");
# TODO choose stable values for General_Category
-out("enum unicode_gc_cat {\n",
+out("enum unicode_General_Category {\n",
+ join(",\n",
+ map(" unicode_General_Category_$_", sort keys %cats)), "\n};\n");
+
+out("enum unicode_Grapheme_Break {\n",
+ join(",\n",
+ map(" unicode_Grapheme_Break_$_", sort keys %gbreak)),
+ "\n};\n");
+out("extern const char *const unicode_Grapheme_Break_names[];\n");
+
+out("enum unicode_Word_Break {\n",
join(",\n",
- map(" unicode_gc_$_", sort keys %cats)), "\n};\n");
+ map(" unicode_Word_Break_$_", sort keys %wbreak)),
+ "\n};\n");
+out("extern const char *const unicode_Word_Break_names[];\n");
+
+out("enum unicode_Sentence_Break {\n",
+ join(",\n",
+ map(" unicode_Sentence_Break_$_", sort keys %sbreak)),
+ "\n};\n");
+out("extern const char *const unicode_Sentence_Break_names[];\n");
out("enum unicode_flags {\n",
- " unicode_normalize_before_casefold = 1,\n",
- " unicode_grapheme_break_extend = 2\n",
+ " unicode_normalize_before_casefold = 1\n",
"};\n",
"\n");
" const uint32_t *compat;\n",
" const uint32_t *canon;\n",
" const uint32_t *casefold;\n",
- " ".choosetype($minud, $maxud)." upper_offset;\n",
- " ".choosetype($minld, $maxld)." lower_offset;\n",
+# " ".choosetype($minud, $maxud)." upper_offset;\n",
+# " ".choosetype($minld, $maxld)." lower_offset;\n",
" ".choosetype(0, $maxccc)." ccc;\n",
- " char gc;\n",
+ " char general_category;\n",
" uint8_t flags;\n",
+ " char grapheme_break;\n",
+ " char word_break;\n",
+ " char sentence_break;\n",
"};\n");
# compat, canon and casefold do have have non-BMP characters, so we
# can't use a simple 16-bit table. We could use UTF-8 or UTF-16
out("extern const struct unidata *const unidata[];\n");
+out("extern const struct unicode_utf8_row {\n",
+ " uint8_t count;\n",
+ " uint8_t min2, max2;\n",
+ "} unicode_utf8_valid[];\n");
+
out("#define UNICODE_NCHARS ", ($max + 1), "\n");
out("#define UNICODE_MODULUS $modulus\n");
+out("#define UNICODE_BREAK_START $break_start\n");
+out("#define UNICODE_BREAK_END $break_end\n");
+out("#define UNICODE_BREAK_TOP $break_top\n");
out("#endif\n");
close STDOUT or die "unidata.h: $!\n";
+print STDERR "Generating unidata.c...\n";
open(STDOUT, ">unidata.c") or die "unidata.c: $!\n";
out("/* Automatically generated file, see scripts/make-unidata */\n",
"#include \"types.h\"\n",
"#include \"unidata.h\"\n");
-# Short aliases for general category codes
+# Short aliases to keep .c file small
+
+out(map(sprintf("#define %s unicode_General_Category_%s\n", $_, $_),
+ sort keys %cats));
+out(map(sprintf("#define GB%s unicode_Grapheme_Break_%s\n", $_, $_),
+ sort keys %gbreak));
+out(map(sprintf("#define WB%s unicode_Word_Break_%s\n", $_, $_),
+ sort keys %wbreak));
+out(map(sprintf("#define SB%s unicode_Sentence_Break_%s\n", $_, $_),
+ sort keys %sbreak));
-out(map(sprintf("#define %s unicode_gc_%s\n", $_, $_), sort keys %cats));
+# Names for *_Break properties
+out("const char *const unicode_Grapheme_Break_names[] = {\n",
+ join(",\n",
+ map(" \"$_\"", sort keys %gbreak)),
+ "\n};\n");
+out("const char *const unicode_Word_Break_names[] = {\n",
+ join(",\n",
+ map(" \"$_\"", sort keys %wbreak)),
+ "\n};\n");
+out("const char *const unicode_Sentence_Break_names[] = {\n",
+ join(",\n",
+ map(" \"$_\"", sort keys %sbreak)),
+ "\n};\n");
# Generate the decomposition mapping tables. We look out for duplicates
# in order to save space and report this as decompsaved at the end. In
# If canon is set then compat will be too and will be identical.
# If compat is set the canon might be clear. So we use the
# compat version and fix up the symbols after.
- if(exists $data{$c}->{compat}) {
+ if(exists $data{$c} && exists $data{$c}->{compat}) {
my $s = join(",",
(map(hex($_), split(/\s+/, $data{$c}->{compat})), 0));
if(!exists $decompnums{$s}) {
my $cfsaved = 0;
out("static const uint32_t ");
for(my $c = 0; $c <= $max; ++$c) {
- if(exists $data{$c}->{casefold}) {
+ if(exists $data{$c} && exists $data{$c}->{casefold}) {
my $s = join(",",
(map(hex($_), split(/\s+/, $data{$c}->{casefold})), 0));
if(!exists $cfnums{$s}) {
my $subtablecounter = 0; # counter for subtable numbers
my $subtablessaved = 0; # number of tables saved
for(my $base = 0; $base <= $max; $base += $modulus) {
+ next if $base >= $break_start && $base < $break_end;
+ next if $base >= $break_top;
my @t;
for(my $c = $base; $c < $base + $modulus; ++$c) {
my $d = $data{$c};
if($data{$c}->{ypogegrammeni}) {
push(@flags, "unicode_normalize_before_casefold");
}
- # Currently we only store the Extend class, using a bit that would
- # otherwise be wasted. The other classes are readily computable.
- # If there is a conveninet way to compute Extend at runtime I have
- # yet to discover it.
- if(exists $data{$c}->{gbreak} and $data{$c}->{gbreak} eq 'Extend') {
- push(@flags, "unicode_grapheme_break_extend");
- }
my $flags = @flags ? join("|", @flags) : 0;
push(@t, "{".
join(",",
$compatsym,
$canonsym,
$cfsym,
- "$d->{ud}",
- "$d->{ld}",
- "$d->{ccc}",
- "$d->{gc}",
+# $d->{ud},
+# $d->{ld},
+ $d->{ccc},
+ $d->{gc},
$flags,
+ "GB$d->{gbreak}",
+ "WB$d->{wbreak}",
+ "SB$d->{sbreak}",
)."}");
}
my $t = join(",\n", @t);
if(!exists $subtable{$t}) {
+ out(sprintf("/* %04X-%04X */\n", $base, $base + $modulus - 1));
out("static const struct unidata st$subtablecounter\[] = {\n",
"$t\n",
"};\n");
$subtableno{$base} = $subtable{$t};
}
-out("const struct unidata*const unidata[]={\n");
+out("const struct unidata *const unidata[]={\n");
for(my $base = 0; $base <= $max; $base += $modulus) {
+ next if $base >= $break_start && $base < $break_end;
+ next if $base >= $break_top;
+ #out("st$subtableno{$base} /* ".sprintf("%04x", $base)." */,\n");
out("st$subtableno{$base},\n");
}
out("};\n");
+out("const struct unicode_utf8_row unicode_utf8_valid[] = {\n");
+for(my $c = 0; $c <= 0x7F; ++$c) {
+ out(" { 1, 0, 0 }, /* $c */\n");
+}
+for(my $c = 0x80; $c < 0xC2; ++$c) {
+ out(" { 0, 0, 0 }, /* $c */\n");
+}
+for(my $c = 0xC2; $c <= 0xDF; ++$c) {
+ out(" { 2, 0x80, 0xBF }, /* $c */\n");
+}
+for(my $c = 0xE0; $c <= 0xE0; ++$c) {
+ out(" { 3, 0xA0, 0xBF }, /* $c */\n");
+}
+for(my $c = 0xE1; $c <= 0xEC; ++$c) {
+ out(" { 3, 0x80, 0xBF }, /* $c */\n");
+}
+for(my $c = 0xED; $c <= 0xED; ++$c) {
+ out(" { 3, 0x80, 0x9F }, /* $c */\n");
+}
+for(my $c = 0xEE; $c <= 0xEF; ++$c) {
+ out(" { 3, 0x80, 0xBF }, /* $c */\n");
+}
+for(my $c = 0xF0; $c <= 0xF0; ++$c) {
+ out(" { 4, 0x90, 0xBF }, /* $c */\n");
+}
+for(my $c = 0xF1; $c <= 0xF3; ++$c) {
+ out(" { 4, 0x80, 0xBF }, /* $c */\n");
+}
+for(my $c = 0xF4; $c <= 0xF4; ++$c) {
+ out(" { 4, 0x80, 0x8F }, /* $c */\n");
+}
+for(my $c = 0xF5; $c <= 0xFF; ++$c) {
+ out(" { 0, 0, 0 }, /* $c */\n");
+}
+out("};\n");
+
close STDOUT or die "unidata.c: $!\n";
-print STDERR "max=$max, subtables=$subtablecounter, subtablessaved=$subtablessaved\n";
+printf STDERR "modulus=%d\n", $modulus;
+printf STDERR "max=%04X\n", $max;
+print STDERR "subtables=$subtablecounter, subtablessaved=$subtablessaved\n";
print STDERR "decompsaved=$decompsaved cfsaved=$cfsaved\n";