X-Git-Url: http://www.chiark.greenend.org.uk/ucgi/~mdw/git/disorder/blobdiff_plain/1452363583a176aafcb00a17bf76c223e3a1f31c..05b75f8d50b83e943af3be4071449304d82dbdcd:/scripts/make-unidata

diff --git a/scripts/make-unidata b/scripts/make-unidata
index 81f347d..5611b0d 100755
--- a/scripts/make-unidata
+++ b/scripts/make-unidata
@@ -31,7 +31,6 @@
 #  - 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
@@ -60,7 +59,24 @@ sub key {
 # 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
@@ -82,18 +98,35 @@ sub input {
 	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);
+my $maxcompat = 0;
+my $maxcanon = 0;
+my $hangul_syllable_decomps = 0;
+my $hangul_choseong_decomps = 0;
 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
@@ -107,25 +140,42 @@ while(<>) {
     $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 '') {
+	    my $maxref;
+	    if($dm =~ /</) {
+		# This is a compatibility decomposition
+		$dm =~ s/^<.*>\s*//;
+		$d->{compat} = 1;
+		$maxref = \$maxcompat;
+	    } else {
+		$maxref = \$maxcanon;
+	    }
+	    $d->{decomp} = [map(hex($_), split(/\s+/, $dm))];
+	    my $len = scalar @{$d->{decomp}};
+	    $$maxref = $len if $len > $$maxref;
+	    if(!$d->{compat}) {
+		if(${$d->{decomp}}[0] >= 0xAC00 && ${$d->{decomp}}[0] <= 0xD7A3) {
+		    ++$hangul_syllable_decomps;
+		}
+		if(${$d->{decomp}}[0] >= 0x1100 && ${$d->{decomp}}[0] <= 0x115F) {
+		    ++$hangul_choseong_decomps;
+		}
+	    }
 	}
+	$data{$c} = $d;
     }
     $cats{$gc} = 1;
-    $max = $c if $c > $max;
+    $max = $end if $end > $max;
 }
 
 sub read_prop_with_ranges {
@@ -139,22 +189,16 @@ sub read_prop_with_ranges {
 	my ($range, $propval) = split(/\s*;\s*/, $_);
 	if($range =~ /(.*)\.\.(.*)/) {
 	    for my $c (hex($1) .. hex($2)) {
-		if(exists $data{$c}) {
-		    $data{$c}->{$propkey} = $propval;
-		}
+		$data{$c}->{$propkey} = $propval;
 	    }
 	} else {
 	    my $c = hex($range);
-	    if(exists $data{$c}) {
-		$data{$c}->{$propkey} = $propval;
-	    }
+	    $data{$c}->{$propkey} = $propval;
 	}
     }
 }
 
 # Grapheme_Break etc
-# NB we do this BEFORE filling in blanks so that the Hangul characters
-# don't get filled in; we can compute their properties mechanically.
 read_prop_with_ranges("auxiliary/GraphemeBreakProperty.txt", "gbreak");
 read_prop_with_ranges("auxiliary/WordBreakProperty.txt", "wbreak");
 read_prop_with_ranges("auxiliary/SentenceBreakProperty.txt", "sbreak");
@@ -188,21 +232,67 @@ for my $c (keys %data) {
     $sbreak{$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);
 
-# 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,
-	    "wbreak" => 'Other',
-	    "gbreak" => 'Other',
-	    "sbreak" => 'Other',
-	    };
+	$data{$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;
@@ -245,6 +335,7 @@ while(<>) {
 }
 
 # 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",
@@ -275,7 +366,8 @@ out("enum unicode_Sentence_Break {\n",
 out("extern const char *const unicode_Sentence_Break_names[];\n");
 
 out("enum unicode_flags {\n",
-    "  unicode_normalize_before_casefold = 1\n",
+    "  unicode_normalize_before_casefold = 1,\n",
+    "  unicode_compatibility_decomposition = 2\n",
     "};\n",
     "\n");
 
@@ -296,37 +388,58 @@ sub choosetype {
 }
 
 out("struct unidata {\n",
-    "  const uint32_t *compat;\n",
-    "  const uint32_t *canon;\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",
-    "  ".choosetype($minud, $maxud)." upper_offset;\n",
-    "  ".choosetype($minld, $maxld)." lower_offset;\n",
+
+    # composed characters that start with this code point.  This only
+    # includes primary composites, i.e. the decomposition mapping is
+    # canonical and this code point is not in the exclusion table.
+    "  const uint32_t *composed;\n",
+
+#    "  ".choosetype($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");
-# compat, canon and casefold do have have non-BMP characters, so we
+# 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 \"common.h\"\n",
     "#include \"unidata.h\"\n");
 
 # Short aliases to keep .c file small
@@ -339,6 +452,8 @@ 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",
@@ -354,56 +469,71 @@ out("const char *const unicode_Sentence_Break_names[] = {\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
-# Unicode 5.0.0 this saves 1795 entries, which is at least 14Kbytes.
-my $decompnum = 0;
-my %decompnums = ();
-my $decompsaved = 0;
+our $ddnum = 0;
+our $ddsaved = 0;
+our %ddnums = ();
+my $ddfirst = 1;
 out("static const uint32_t ");
-for(my $c = 0; $c <= $max; ++$c) {
-    # 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}) {
-	my $s = join(",",
-		     (map(hex($_), split(/\s+/, $data{$c}->{compat})), 0));
-	if(!exists $decompnums{$s}) {
-	    out(",\n") if $decompnum != 0;
-	    out("cd$decompnum\[]={$s}");
-	    $decompnums{$s} = $decompnum++;
+sub dedupe {
+    my $s = join(",", @_);
+    if(!exists $ddnums{$s}) {
+	if($ddfirst) {
+	    $ddfirst = 0;
 	} else {
-	    ++$decompsaved;
-	}
-	$data{$c}->{compatsym} = "cd$decompnums{$s}";
-	if(exists $data{$c}->{canon}) {
-	    $data{$c}->{canonsym} = "cd$decompnums{$s}";
+	    out(",\n");
 	}
+	out("dd$ddnum\[]={$s}");
+	$ddnums{$s} = $ddnum++;
+    } else {
+	++$ddsaved;
     }
+    return "dd$ddnums{$s}";
 }
-out(";\n");
 
-# ...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.
-my $cfnum = 0;
-my %cfnums = ();
-my $cfsaved = 0;
-out("static const uint32_t ");
+# Generate the decomposition mapping tables.
+print STDERR "> decomposition mappings\n";
 for(my $c = 0; $c <= $max; ++$c) {
-    if(exists $data{$c}->{casefold}) {
-	my $s = join(",",
-		     (map(hex($_), split(/\s+/, $data{$c}->{casefold})), 0));
-	if(!exists $cfnums{$s}) {
-	    out(",\n") if $cfnum != 0;
-	    out("cf$cfnum\[]={$s}");
-	    $cfnums{$s} = $cfnum++;
+    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} = [$c];
 	} else {
-	    ++$cfsaved;
+	    push(@{$data{$first}->{compose}}, $c);
 	}
-	$data{$c}->{cfsym} = "cf$cfnums{$s}";
     }
 }
+# Then we can generate the tables.
+for(my $c = 0; $c <= $max; ++$c) {
+    if(exists $data{$c} && exists $data{$c}->{compose}) {
+	$data{$c}->{compsym} = dedupe(@{$data{$c}->{compose}}, 0);
+    }
+}
+
+# The case folding table.
+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
@@ -411,30 +541,35 @@ out(";\n");
 # 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 $canonsym = ($data{$c}->{canonsym} or "0");
-	my $compatsym = ($data{$c}->{compatsym} or "0");
+	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{$c}->{ypogegrammeni}) {
-	    push(@flags, "unicode_normalize_before_casefold");
+	    push(@flags, "NBC");
+	}
+	if($data{$c}->{compat}) {
+	    push(@flags, "CD");
 	}
 	my $flags = @flags ? join("|", @flags) : 0;
 	push(@t, "{".
 	     join(",",
-		  $compatsym,
-		  $canonsym,
+		  $decompsym,
 		  $cfsym,
-		  $d->{ud},
-		  $d->{ld},
+		  $compsym,
+#		  $d->{ud},
+#		  $d->{ld},
 		  $d->{ccc},
 		  $d->{gc},
 		  $flags,
@@ -445,6 +580,7 @@ for(my $base = 0; $base <= $max; $base += $modulus) {
     }
     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");
@@ -455,13 +591,66 @@ for(my $base = 0; $base <= $max; $base += $modulus) {
     $subtableno{$base} = $subtable{$t};
 }
 
-out("const struct unidata*const unidata[]={\n");
+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 "max=$max, subtables=$subtablecounter, subtablessaved=$subtablessaved\n";
-print STDERR "decompsaved=$decompsaved cfsaved=$cfsaved\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";
+print STDERR "maxcompat=$maxcompat maxcanon=$maxcanon\n";
+print STDERR "$hangul_syllable_decomps canonical decompositions to Hangul syllables\n";
+print STDERR "$hangul_choseong_decomps canonical decompositions to Hangul Choseong\n";
+
+die "We assumed that canonical decompositions were never more than 2 long!\n"
+    if $maxcanon > 2;
+
+die "We assumed no canonical decompositions to Hangul syllables/Choseong!\n"
+    if $hangul_syllable_decomps || $hangul_choseong_decomps;