URL
https://opencores.org/ocsvn/openrisc/openrisc/trunk
Subversion Repositories openrisc
[/] [openrisc/] [trunk/] [gnu-dev/] [or1k-gcc/] [libjava/] [scripts/] [unicode-blocks.pl] - Rev 775
Go to most recent revision | Compare with Previous | Blame | View Log
#!/usr/bin/perl -w # unicode-blocks.pl -- Script to generate java.lang.Character.UnicodeBlock # Copyright (C) 2002 Free Software Foundation, Inc. # # This file is part of GNU Classpath. # # GNU Classpath 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, or (at your option) # any later version. # # GNU Classpath 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 GNU Classpath; see the file COPYING. If not, write to the # Free Software Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA # 02110-1301 USA. # # Linking this library statically or dynamically with other modules is # making a combined work based on this library. Thus, the terms and # conditions of the GNU General Public License cover the whole # combination. # # As a special exception, the copyright holders of this library give you # permission to link this library with independent modules to produce an # executable, regardless of the license terms of these independent # modules, and to copy and distribute the resulting executable under # terms of your choice, provided that you also meet, for each linked # independent module, the terms and conditions of the license of that # module. An independent module is a module which is not derived from # or based on this library. If you modify this library, you may extend # this exception to your version of the library, but you are not # obligated to do so. If you do not wish to do so, delete this # exception statement from your version. # Code for reading Blocks.txt and generating (to standard out) the code for # java.lang.Character.UnicodeBlock, for pasting into java/lang/Character.java. # You should probably check that the results are accurate to the # specification, but I made sure it works OOB for Unicode 3.0.0 and JDK 1.4. # As the grammar for the Blocks.txt file is changing in Unicode 3.2.0, you # will have to tweak this some for future use. For now, the relevant # Unicode definition files are found in libjava/gnu/gcj/convert/. # # author Eric Blake <ebb9@email.byu.edu> # # usage: unicode-blocks.pl <blocks.txt> # where <blocks.txt> is obtained from www.unicode.org (named Blocks-3.txt # for Unicode version 3.0.0). die "Usage: $0 <blocks.txt>" unless @ARGV == 1; open (BLOCKS, $ARGV[0]) || die "Can't open Unicode block file: $!\n"; # A hash of added fields and the JDK they were added in, to automatically # print @since tags. Maintaining this is optional (and tedious), but nice. my %additions = ("SYRIAC" => "1.4", "THAANA" => "1.4", "SINHALA" => "1.4", "MYANMAR" => "1.4", "ETHIOPIC" => "1.4", "CHEROKEE" => "1.4", "UNIFIED_CANADIAN_ABORIGINAL_SYLLABICS" => "1.4", "OGHAM" => "1.4", "RUNIC" => "1.4", "KHMER" => "1.4", "MONGOLIAN" => "1.4", "BRAILLE_PATTERNS" => "1.4", "CJK_RADICALS_SUPPLEMENT" => "1.4", "KANGXI_RADICALS" => "1.4", "IDEOGRAPHIC_DESCRIPTION_CHARACTERS" => "1.4", "BOPOMOFO_EXTENDED" => "1.4", "CJK_UNIFIED_IDEOGRAPHS_EXTENSION_A" => "1.4", "YI_SYLLABLES" => "1.4", "YI_RADICALS" => "1.4", ); print <<'EOF'; /** * A family of character subsets in the Unicode specification. A character * is in at most one of these blocks. * * This inner class was generated automatically from * <code>$ARGV[0]</code>, by some perl scripts. * This Unicode definition file can be found on the * <a href="http://www.unicode.org">http://www.unicode.org</a> website. * JDK 1.4 uses Unicode version 3.0.0. * * @author scripts/unicode-blocks.pl (written by Eric Blake) * @since 1.2 */ public static final class UnicodeBlock extends Subset { /** The start of the subset. */ private final char start; /** The end of the subset. */ private final char end; /** * Constructor for strictly defined blocks. * * @param start the start character of the range * @param end the end character of the range * @param name the block name */ private UnicodeBlock(char start, char end, String name) { super(name); this.start = start; this.end = end; } /** * Returns the Unicode character block which a character belongs to. * * @param ch the character to look up * @return the set it belongs to, or null if it is not in one */ public static UnicodeBlock of(char ch) { // Special case, since SPECIALS contains two ranges. if (ch == '\uFEFF') return SPECIALS; // Simple binary search for the correct block. int low = 0; int hi = sets.length - 1; while (low <= hi) { int mid = (low + hi) >> 1; UnicodeBlock b = sets[mid]; if (ch < b.start) hi = mid - 1; else if (ch > b.end) low = mid + 1; else return b; } return null; } EOF my $seenSpecials = 0; my $seenSurrogates = 0; my $surrogateStart = 0; my @names = (); while (<BLOCKS>) { next if /^\#/; my ($start, $end, $block) = split(/; /); next unless defined $block; chomp $block; $block =~ s/ *$//; if (! $seenSpecials and $block =~ /Specials/) { # Special case SPECIALS, since it is two disjoint ranges $seenSpecials = 1; next; } if ($block =~ /Surrogates/) { # Special case SURROGATES_AREA, since it one range, not three # consecutive, in Java $seenSurrogates++; if ($seenSurrogates == 1) { $surrogateStart = $start; next; } elsif ($seenSurrogates == 2) { next; } else { $start = $surrogateStart; $block = "Surrogates Area"; } } # Special case the name of PRIVATE_USE_AREA. $block =~ s/(Private Use)/$1 Area/; (my $name = $block) =~ tr/a-z -/A-Z__/; push @names, $name; my $since = (defined $additions{$name} ? "\n * \@since $additions{$name}" : ""); my $extra = ($block =~ /Specials/ ? "'\\uFEFF', " : ""); print <<EOF; /** * $block. * $extra'\\u$start' - '\\u$end'.$since */ public static final UnicodeBlock $name = new UnicodeBlock('\\u$start', '\\u$end', "$name"); EOF } print <<EOF; /** * The defined subsets. */ private static final UnicodeBlock sets[] = { EOF foreach (@names) { print " $_,\n"; } print <<EOF; }; } // class UnicodeBlock EOF
Go to most recent revision | Compare with Previous | Blame | View Log