Generate, but do not use, composition mappings.

[disorder] / scripts / make-unidata

#! /usr/bin/perl -w
#
# This file is part of DisOrder.
# Copyright (C) 2007 Richard Kettlewell
#
# This program is free software; you can redistribute it and/or modify
# it under the terms of the GNU General Public License as published by
# the Free Software Foundation; either version 2 of the License, or
# (at your option) any later version.
#
# This program is distributed in the hope that it will be useful, but
# WITHOUT ANY WARRANTY; without even the implied warranty of
# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
# General Public License for more details.
#
# You should have received a copy of the GNU General Public License
# along with this program; if not, write to the Free Software
# Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307
# USA
#
#
# Generate Unicode support tables
#
# This script will download data from unicode.org if the required files
# aren't in the current directory.
#
# After modifying this script you should run:
#  make -C lib rebuild-unicode check
#
# Things not supported yet:
#  - SpecialCasing.txt data for case mapping
#  - Title case offsets
#  - Some kind of hinting for composition
#  - ...
#
# NB the generated files DO NOT offer a stable ABI and so are not immediately
# suitable for use in a general-purpose library.  Things that would need to
# be done:
#  - Hide unidata.h from applications; it will never be ABI- or even API-stable.
#  - Stablized General_Category values
#  - Extend the unicode.h API to general utility rather than just what
#    DisOrder needs.
#  - ...
#
use strict;
use File::Basename;

sub out {
    print @_ or die "$!\n";
}

sub key {
    my $d = shift;
    local $_;

    return join("-", map($d->{$_}, sort keys %$d));
}

# Size of a subtable
#
# This can be varied to trade off the number of subtables against their size.
# 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
my $max = 0;			# maximum codepoint
my $maxccc = 0;			# maximum combining class
my $maxud = 0;
my $minud = 0;			# max/min upper case offset
my $maxld = 0;
my $minld = 0;			# max/min lower case offset

# 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 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";
}


# Read the main data file
input("UnicodeData.txt");
my ($start, $end);
while(<>) {
    my @f = split(/;/, $_);
    my $c = hex($f[0]);		# codepoint
    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
    my $slm = hex($f[13]) || $c; # Simple_Lowercase_Mapping
    # recalculate the upper/lower case mappings as offsets
    my $ud = $sum - $c;
    my $ld = $slm - $c;
    # update bounds on various values
    $maxccc = $ccc if $ccc > $maxccc; # assumed never to be -ve
    $minud = $ud if $ud < $minud;
    $maxud = $ud if $ud > $maxud;
    $minld = $ld if $ld < $minld;
    $maxld = $ld if $ld > $maxld;
    if($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 compatibility decomposition
		$dm =~ s/^<.*>\s*//;
		$d->{compat} = 1;
	    }
	    $d->{decomp} = [map(hex($_), split(/\s+/, $dm))];
	}
	$data{$c} = $d;
    }
    $cats{$gc} = 1;
    $max = $end if $end > $max;
}

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)) {
		$data{$c}->{$propkey} = $propval;
	    }
	} else {
	    my $c = hex($range);
	    $data{$c}->{$propkey} = $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;
}

# Various derived properties
input("DerivedNormalizationProps.txt");
while(<>) {
    chomp;
    s/\s*\#.*//;
    next if $_ eq '';
    my @f = split(/\s*;\s*/, $_);
    if(@f == 2) {
	push(@f, 1);
    }
    my ($range, $propkey, $propval) = @f;
    if($range =~ /(.*)\.\.(.*)/) {
	for my $c (hex($1) .. hex($2)) {
	    $data{$c}->{$propkey} = $propval
	}
    } else {
	my $c = hex($range);
	$data{$c}->{$propkey} = $propval
    }
}

# Round up the maximum value to a whole number of subtables
$max += ($modulus - 1) - ($max % $modulus);

# 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} = $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
input("CaseFolding.txt");
while(<>) {
    chomp;
    next if /^\#/ or $_ eq '';
    my @f = split(/\s*;\s*/, $_);
    # Full case folding means use status C and F.
    # We discard status T, Turkish users may wish to change this.
    if($f[1] eq 'C' or $f[1] eq 'F') {
	my $c = hex($f[0]);
	$data{$c}->{casefold} = $f[2];
	# We are particularly interest in combining characters that
	# case-fold to non-combining characters, or characters that
	# case-fold to sequences with combining characters in non-initial
	# positions, as these required decomposiiton before case-folding
	my @d = map(hex($_), split(/\s+/, $data{$c}->{casefold}));
	if($data{$c}->{ccc} != 0) {
	    # This is a combining character
	    if($data{$d[0]}->{ccc} == 0) {
		# The first character of its case-folded form is NOT
		# a combining character.  The field name is the example
		# explicitly mentioned in the spec.
		$data{$c}->{ypogegrammeni} = 1;
	    }
	} else {
	    # This is a non-combining character; inspect the non-initial
	    # code points of the case-folded sequence
	    shift(@d);
	    if(grep($data{$_}->{ccc} != 0, @d)) {
		# Some non-initial code point in the case-folded for is NOT a
		# a combining character.
		$data{$c}->{ypogegrammeni} = 1;
	    }
	}
    }
}

# 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",
    "#ifndef UNIDATA_H\n",
    "#define UNIDATA_H\n");

# TODO choose stable values for General_Category
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_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_compatibility_decomposition = 2\n",
    "};\n",
    "\n");

# Choose the narrowest type that will fit the required values
sub choosetype {
    my ($min, $max) = @_;
    if($min >= 0) {
	return "char" if $max <= 127;
	return "unsigned char" if $max <= 255;
	return "int16_t" if $max < 32767;
	return "uint16_t" if $max < 65535;
	return "int32_t";
    } else {
	return "char" if $min >= -127 && $max <= 127;
	return "int16_t" if $min >= -32767 && $max <= 32767;
	return "int32_t";
    }
}

out("struct unidata {\n",
    # decomposition (canonical or compatibility;
    # unicode_compatibility_decomposition distinguishes) or NULL
    "  const uint32_t *decomp;\n",

    # case-folded string or NULL
    "  const uint32_t *casefold;\n",

    # composed characters that start with this code point.  This only
    # includes primary composites, i.e. the decomposition mapping is
    # canonical and this code point is not in the exclusion table.
    "  const uint32_t *composed;\n",

#    "  ".choosetype($minud, $maxud)." upper_offset;\n",
#    "  ".choosetype($minld, $maxld)." lower_offset;\n",

    # canonical combining class
    "  ".choosetype(0, $maxccc)." ccc;\n",
    "  char general_category;\n",

    # see unicode_flags enum
    "  uint8_t flags;\n",
    "  char grapheme_break;\n",
    "  char word_break;\n",
    "  char sentence_break;\n",
    "};\n");
# decomp 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
# though, saving a bit of space (probably not that much...) at the
# cost of marginally reduced performance and additional complexity

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 <config.h>\n",
    "#include \"types.h\"\n",
    "#include \"unidata.h\"\n");

# 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("#define NBC unicode_normalize_before_casefold\n");
out("#define CD unicode_compatibility_decomposition\n");

# 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");

our $ddnum = 0;
our $ddsaved = 0;
our %ddnums = ();
my $ddfirst = 1;
out("static const uint32_t ");
sub dedupe {
    my $s = join(",", @_);
    if(!exists $ddnums{$s}) {
	if($ddfirst) {
	    $ddfirst = 0;
	} else {
	    out(",\n");
	}
	out("dd$ddnum\[]={$s}");
	$ddnums{$s} = $ddnum++;
    } else {
	++$ddsaved;
    }
    return "dd$ddnums{$s}";
}

# Generate the decomposition mapping tables.  We look out for duplicates
# in order to save space and report this as decompsaved at the end.  In
# Unicode 5.0.0 this saves 1795 entries, which is at least 14Kbytes.
print STDERR "> decomposition mappings\n";
for(my $c = 0; $c <= $max; ++$c) {
    if(exists $data{$c} && exists $data{$c}->{decomp}) {
	$data{$c}->{decompsym} = dedupe(@{$data{$c}->{decomp}}, 0);
    }
}

print STDERR "> composition mappings\n";
# First we must generate the mapping of each code point to possible
# compositions
for(my $c = 0; $c <= $max; ++$c) {
    if(exists $data{$c}
       && exists $data{$c}->{decomp}
       && !exists $data{$c}->{compat}
       && !$data{$c}->{Full_Composition_Exclusion}) {
	# $c has a non-excluded canonical decomposition, i.e. it is
        # a primary composite.  Find the first code point of the decomposition
	my $first = ${$data{$c}->{decomp}}[0];
	if(!exists $data{$first}->{compose}) {
	    $data{$first}->{compose} = [$first];
	} else {
	    push(@{$data{$first}->{compose}}, $first);
	}
    }
}
for(my $c = 0; $c <= $max; ++$c) {
    if(exists $data{$c} && exists $data{$c}->{compose}) {
	$data{$c}->{compsym} = dedupe(@{$data{$c}->{compose}}, 0);
    }
}

# ...and the case folding table.  Again we compress equal entries to save
# space.  In Unicode 5.0.0 this saves 51 entries or at least 408 bytes.
# This doesns't seem as worthwhile as the decomposition mapping saving above.
print STDERR "> case-fold mappings\n";
for(my $c = 0; $c <= $max; ++$c) {
    if(exists $data{$c} && exists $data{$c}->{casefold}) {
	$data{$c}->{cfsym} = dedupe(map(hex($_), split(/\s+/,
						       $data{$c}->{casefold})),
				    0);
    }
}

# End of de-dupable arrays
out(";\n");

# Visit all the $modulus-character blocks in turn and generate the
# required subtables.  As above we spot duplicates to save space.  In
# Unicode 5.0.0 with $modulus=128 and current table data this saves
# 1372 subtables or at least three and a half megabytes on 32-bit
# platforms.
print STDERR "> subtables\n";
my %subtable = ();		# base->subtable number
my %subtableno = ();		# subtable number -> content
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};
	my $decompsym = ($data{$c}->{decompsym} or "0");
	my $cfsym = ($data{$c}->{cfsym} or "0");
	my $compsym = ($data{$c}->{compsym} or "0");
	my @flags = ();
	if($data{$c}->{ypogegrammeni}) {
	    push(@flags, "NBC");
	}
	if($data{$c}->{compat}) {
	    push(@flags, "CD");
	}
	my $flags = @flags ? join("|", @flags) : 0;
	push(@t, "{".
	     join(",",
		  $decompsym,
		  $cfsym,
		  $compsym,
#		  $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");
	$subtable{$t} = $subtablecounter++;
    } else {
	++$subtablessaved;
    }
    $subtableno{$base} = $subtable{$t};
}

print STDERR "> main table\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");

print STDERR "> UTF-8 table\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 "Done.\n\n";
printf STDERR "modulus=%d\n", $modulus;
printf STDERR "max=%04X\n", $max;
print STDERR "subtables=$subtablecounter, subtablessaved=$subtablessaved\n";
print STDERR "ddsaved=$ddsaved\n";