Child: [913dff] (diff)

Download this file

builder.in    500 lines (470 with data), 15.6 kB

#
# Copyright (C) 2000, 2001, 2002 Loic Dachary <loic@senga.org>
#
# 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 a table mapping UTF-16 characters to their unaccented
# equivalent. Some characters such as fi (one character) are expanded
# into two letters : f and i. In Unicode jargon it means that the table
# map each character to its compatibility decomposition in which marks
# were stripped.
#
# The format of the $base file can be found at:
# http://www.unicode.org/Public/3.2-Update/UnicodeData-3.2.0.html
#
use strict;

use Getopt::Long;

sub main {
    my($base) = "UnicodeData-@UNICODE_VERSION@.txt";
    my($cfbase) = "CaseFolding-@UNICODE_VERSION@.txt";
    my($verbose);
    my($source);
    my($reference);

    GetOptions("verbose+" => \$verbose,
	       "database=s" => \$base,
	       "source!" => \$source,
               "reference!" => \$reference);

    my(%decomposition, %mark, %name);
    my(%ranges);
    open(FILE, "<$base") or die "cannot open $base for reading : $!";
    while(<FILE>) {
	next if(/^\s*#/);    # Skip comments
	my($code_value,
	   $character_name,
	   $general_category,
	   $canonical_combining_classes,
	   $bidirectional_category,
	   $character_decomposition_mapping,
	   $decimal_digit_value,
	   $digit_value,
	   $numeric_value,
	   $mirrored,
	   $unicode_1_0_name,
	   $_10646_comment_field,
	   $uppercase_mapping,
	   $lowercase_mapping,
	   $titlecase_mapping) = split(/;/, $_);

        # Basic plane only
	last if (hex $code_value > 0xffff); 

	#
	# Generate ranges of values that are not explicitly listed.
	# CJK ideographs for instance.
	#
	if($character_name =~ /^<(.*), (First|Last)>/) {
	    $ranges{$1}{$2} = $code_value;
	}

	# For kana japanese characters, we don't want to strip accents as I'm
	# told that they are essential and stripping them does not
	# make sense. Wonder why Unicode does these decompositions
	# then...  Problem: the first solution used was to decompose
	# the japanese accented kana and not remove accents. But then
	# the unaccented character would match the string with
	# accent. So now we don't decompose at all, but this means
	# that, if the original text was decomposed, things don't work
	# as intended as we should actually recombine the
	# letter+accents in this case for data to be unified.
	if($character_decomposition_mapping =~ /(<.*>)?\s*(.+)/) {
	    # Not for Hiragana + Katakana 
	    if (!(hex $code_value >= 0x3040 && hex $code_value <= 0x30ff) &&
		# and Halfwidth katakana
		!(hex $code_value >= 0xff65 && hex $code_value <= 0xff9f) ) {
		$decomposition{$code_value} = $2;
	    }
	}
	if($general_category =~ /^M/) {
	    $mark{$code_value} = 1;
            # For mark caracters, we generate a 0 entry in the
            # decomposition table. This signals to the c code that no
            # output should be generated. Slightly hacky but ok. The
            # original code left mark character go through (generating
            # still accented output if the input was in decomposed
            # form). Decomposed text is rare, but, for example, macosx file
            # names have separate combining accent characters.
	    $decomposition{$code_value} = "0000";
	}
	$name{$code_value} = $character_name;
    }
    close(FILE);
    
    # Generate compatibility decomposition and strip marks
    # (marks == diacritics == accents)
    #
    # We also forbid any excursion out of the basic plane. 
    my($from, $to);
    while(($from, $to) = each(%decomposition)) {
	my(@code_values) = split(' ', $to);
	my($code_value);
	my(@decomposition);

	while(@code_values) {
	    my($code_value) = shift(@code_values);
	    if (hex $code_value > 0xffff) {
		undef @decomposition;
		last;
	    }
            # marks also have entries in the decomposition table (so that 
            # they can be suppressed when found in input), but no output
            # component should be generated for them.
            if (!exists($mark{$code_value})) {
                if(exists($decomposition{$code_value})) {
                    push(@code_values, split(' ', $decomposition{$code_value}));
                } else {
                    push(@decomposition, $code_value);
                }
            }
	}
	if(@decomposition) {
	    $decomposition{$from} = "@decomposition";
	} else {
	    delete($decomposition{$from});
	}
    }

    # Read in the casefolding file
    my(%casefold);
    open(FILE, "<$cfbase") or die "cannot open $cfbase for reading : $!";
    while(<FILE>) {
	next if(/^\s*#/);    # Skip comments
	my($code_value,
	   $foldstatus,
	   $folded) = split(/;/, $_);
	if ($foldstatus =~ /C|F/) {
	    $casefold{$code_value} = $folded;
	}
    }
    close(FILE);

    #showcasefold(\%casefold);
    reference(\%decomposition, $verbose) if($reference);
    source(\%decomposition, \%name, \%casefold, $verbose) if($source);
}

sub showcasefold {
    my($casefold) = @_;

    my($code_value);
    foreach $code_value (0 .. 0xFFFF) {
	$code_value = uc(sprintf("%04x", $code_value));
	print "$code_value";
	if(exists($casefold->{$code_value})) {
	    print " => $casefold->{$code_value}\n";
	} else {
	    print "\n";
	}
    }
}

#
# Generate machine readable file mapping all UTF-16 codes
# to their unaccented replacement. This file can be compared
# with the output of a program doing the same mapping using the
# libunac library.
#
sub reference {
    my($decomposition, $verbose) = @_;

    my($code_value);
    foreach $code_value (0 .. 0xFFFF) {
	$code_value = uc(sprintf("%04x", $code_value));
	print "$code_value";
	if(exists($decomposition->{$code_value})) {
	    print " => $decomposition->{$code_value}\n";
	} else {
	    print "\n";
	}
    }
}

#
# Read input file into hash table and return it.
#
# The input is divided in chuncks according to special markers. For
# instance:
#
# before
# /* Generated by builder. Do not modify. Start a_tag */
# bla bla
# /* Generated by builder. Do not modify. End a_tag */
# after
# /* Generated by builder. Do not modify. Start b_tag */
# more stuff
# /* Generated by builder. Do not modify. End b_tag */
# still something
#
# Will generate the following hash:
#
# {
#   'list' => [ 1, a_tag, 2, b_tag, 3 ],
#   '1' => "before\n",
#   'a_tag' => undef,
#   '2' => "after\n";
#   'b_tag' => undef,
#   '3' => "still something\n"
# }
#
# The caller may then assign a string to the a_tag and b_tag entries
# and then call the spit function to rebuild the file.
#
sub slurp {
    my($file) = @_;
    my(%content);
    my($count) = 1;
    my(@lines);
    open(FILE, "<$file") or die "cannot open $file for reading : $!";
    while(<FILE>) {
	if(/Do not modify. Start\s+(\w+)/i) {
	    push(@{$content{'list'}}, $count);
	    $content{$count} = join("", @lines); 
	    $count++;
	    push(@{$content{'list'}}, $1);
	    @lines = ();
	}
	next if(/Do not modify. Start/i .. /Do not modify. End/i);
	push(@lines, $_);
    }
    if(@lines) {
	push(@{$content{'list'}}, $count);
	$content{$count} = join("", @lines); 
    }
    close(FILE);
    return \%content;
}

#
# Write the $file with the content of the $content hash table. 
# See the slurp function for a description of the $content format.
#
sub spit {
    my($file, $content) = @_;
    open(FILE, ">$file") or die "cannot open $file for writing : $!";
    my($tag);
    foreach $tag (@{$content->{'list'}}) {
	print(FILE "/* Generated by builder. Do not modify. Start $tag */\n") if($tag !~ /^\d+$/);
	print FILE $content->{$tag};
	print(FILE "/* Generated by builder. Do not modify. End $tag */\n") if($tag !~ /^\d+$/);
    }
    close(FILE);
}

#
# Generate tables, defines and code in the unac.c and unac.h files.
# The unac.c and unac.h files are substituted in place.
#
sub source {
    my($decomposition, $name, $casefold, $verbose) = @_;

    my($csource) = slurp("unac.c");
    my($hsource) = slurp("unac.h");
    #
    # Human readable table
    #
    my(@comment);
    push(@comment, "/*\n");
    my($from);
    foreach $from (sort(keys(%$decomposition))) {
	my($character_name) = $name->{$from};
	$character_name = "??" if(!$character_name);
	push(@comment, " * $from $character_name\n");
	my($code_value);
	foreach $code_value (split(' ', $decomposition->{$from})) {
	    $character_name = $name->{$code_value} || "??";
	    push(@comment, " * \t$code_value $character_name\n");
	}
    }
    push(@comment, "*/\n");
    my($comment) = join("", @comment);

    #
    # Select the best block size (the one that takes less space)
    #
    # result: $best_blocks (array of blocks that contain exactly
    #                       $block_count replacements. Each block
    #                       is a string containing replacements 
    #                       separated by |)
    #         $best_indexes (array mapping block number to a block
    #                        in the $best_blocks array)
    #         $best_block_shift (the size of the block)
    # 
    # Within a block, if the character has no replacement the 0xFFFF 
    # placeholder is inserted.
    #
    my($best_blocks);
    my($best_indexes);
    my($best_block_shift);
    my($best_total_size) = 10 * 1024 * 1024;
    my($block_shift);
    foreach $block_shift (2 .. 10) {
	my($block_count) = 1 << $block_shift;
	my(@blocks, @indexes);
	my($duplicate) = 0;
	my(@values);
	my($code_value);
	foreach $code_value (0 .. 0x10000) {
	    if($code_value > 0 && $code_value % $block_count == 0) {
		my($block) = join("|", @values);
		my($existing_block);
		my($index) = 0;
		my($found);
		foreach $existing_block (@blocks) {
		    if($block eq $existing_block) {
			push(@indexes, $index);
			$found = 1;
			$duplicate++;
			last;
		    }
		    $index++;
		}
		if(!$found) {
		    push(@indexes, $index);
		    push(@blocks, $block);
		}
		@values = ();
	    }
	    $code_value = uc(sprintf("%04x", $code_value));
	    if(exists($decomposition->{$code_value})) {
		push(@values, $decomposition->{$code_value});
	    } else {
		push(@values, "FFFF");
	    }
	    # We also push the case-folded version of the unaccented char
	    # Note that by pushing the case-folded version of the original
	    # char, we'd have the possibility of independant unaccenting and
	    # case folding, but with less performance. 
	    # We could also keep the three chunks, using a little more memory
	    if(exists($decomposition->{$code_value})) {
	      my($cv);
	      my(@vl);
	      foreach $cv (split(' ', $decomposition->{$code_value})) {
		if(exists($casefold->{$cv})) {
		  push(@vl, $casefold->{$cv});
		} else {
		  push(@vl, $cv);
		}
	      }
	      #print STDERR "Pushing " . join(" ", @vl) . " for " . 
	      #$code_value . "\n";
	      push(@values, join(" ", @vl));
	    } else {
	      if(exists($casefold->{$code_value})) {
		push(@values, $casefold->{$code_value});
	      } else {
		push(@values, "FFFF");
	      }
	    }
	}
	print STDERR scalar(@blocks) . " blocks of " . $block_count . " entries, factorized $duplicate blocks\n\t" if($verbose);
	my($block_size) = 0;
	my($block);
	foreach $block (@blocks) {
	    my(@tmp) = split(/[| ]/, $block);
	    $block_size += scalar(@tmp) * 2;
	}
	#
	# Pointer to the block array
	#
	$block_size += scalar(@blocks) * 4;
	#
	# Positions of the entries in the block
	#
	$block_size += $block_count * scalar(@blocks) * 2;
	print STDERR "total block size = $block_size, " if($verbose);
	my($index_size) = (1 << (16 - $block_shift)) * 2;
	print STDERR "index size = " . $index_size . "\n\t" if($verbose);
	my($total_size) = $block_size + $index_size;
	print STDERR "total size = $total_size\n" if($verbose);

	if($total_size < $best_total_size) {
	    $best_total_size = $total_size;
	    $best_blocks = \@blocks;
	    $best_indexes = \@indexes;
	    $best_block_shift = $block_shift;
	}
    }

    my($block_count) = scalar(@$best_blocks);
    my($block_size) = 1 << $best_block_shift;

    #
    # Constants that depend on the block size.
    # result : $defines
    #
    my($defines) = <<EOF;
#define UNAC_BLOCK_SHIFT $best_block_shift
#define UNAC_BLOCK_MASK ((1 << UNAC_BLOCK_SHIFT) - 1)
#define UNAC_BLOCK_SIZE (1 << UNAC_BLOCK_SHIFT)
#define UNAC_BLOCK_COUNT $block_count
#define UNAC_INDEXES_SIZE (0x10000 >> UNAC_BLOCK_SHIFT)
EOF
    #
    # Mapping block number to index in data_table or position table.
    # result : $index_out
    #
    my($count) = 0;
    my($index);
    my($index_out) = "unsigned short unac_indexes[UNAC_INDEXES_SIZE] = {\n";
    foreach $index (@$best_indexes) {
	$count++;
	$index_out .= sprintf("%4s,", $index);
	if($count % 15 == 0) {
	    $index_out .= "\n";
	}
    }
    $index_out =~ s/,\s*\Z/\n/s;
    $index_out .= "};\n";

    #
    # Generate the position table (map character number in block to
    # position in the data string), the data_table that maps a block
    # to a unsigned short array that contains the character (aka the
    # data array) and the data arrays themselves that is a pure concatenation
    # of all the unsigned short in a block. 
    # result : $position_out, $data_table_out, $data_out
    #
    my(@positions_out);
    my($highest_position) = 0;
    my(@data_table_out);
    my($data_table_out) = "unsigned short* unac_data_table[UNAC_BLOCK_COUNT] = {\n";
    my(@data_out);
    my($block_number) = 0;
    my($block);
    foreach $block (@$best_blocks) {
	my(@index);
	my($position) = 0;
	my($entry);
	my(@data);
	foreach $entry (split('\|', $block)) {
	    push(@index, $position);
	    my(@tmp) = split(' ', $entry);
	    push(@data, @tmp);
	    $position += scalar(@tmp);
	}
	push(@index, $position);
	$highest_position = $position if($position > $highest_position);
	push(@positions_out, "/* $block_number */ { " . join(", ", @index) . " }");
	push(@data_table_out, "unac_data$block_number");
	push(@data_out, "unsigned short unac_data$block_number" . "[] = { 0x" . join(", 0x", @data) . " };\n");
	$block_number++;
    }
    my($position_type) = $highest_position >= 256 ? "short" : "char";
    my($positions_out) = "unsigned $position_type unac_positions[UNAC_BLOCK_COUNT][2*UNAC_BLOCK_SIZE + 1] = {\n";

    $positions_out .= join(",\n", @positions_out);
    $positions_out .= "\n};\n";
    my($data_out) = join("", @data_out);
    $data_table_out .= join(",\n", @data_table_out);
    $data_table_out .= "\n};\n";

    #
    # Tables declarations
    # result : $declarations
    #
    my($declarations);
    $declarations = <<EOF;
extern unsigned short unac_indexes[UNAC_INDEXES_SIZE];
extern unsigned $position_type unac_positions[UNAC_BLOCK_COUNT][2*UNAC_BLOCK_SIZE + 1];
extern unsigned short* unac_data_table[UNAC_BLOCK_COUNT];
EOF
    for($block_number = 0; $block_number < $block_count; $block_number++) {
	$declarations .= "extern unsigned short unac_data$block_number" . "[];\n";
    }

    $csource->{'tables'} = "$comment\n$index_out\n$positions_out\n$data_out\n$data_table_out";
    $hsource->{'defines'} = $defines;
    $hsource->{'declarations'} = $declarations;

    spit("unac.c", $csource);
    spit("unac.h", $hsource);
}

main();