michael@0: #!/usr/bin/perl
michael@0: #  ********************************************************************
michael@0: #  * COPYRIGHT:
michael@0: #  * Copyright (c) 2002-2007, International Business Machines Corporation and
michael@0: #  * others. All Rights Reserved.
michael@0: #  ********************************************************************
michael@0: #
michael@0: #  regexcst.pl
michael@0: #            Compile the regular expression paser state table data into initialized C data.
michael@0: #            Usage:
michael@0: #                   cd icu/source/i18n
michael@0: #                   perl regexcst.pl < regexcst.txt > regexcst.h
michael@0: #
michael@0: #             The output file, regexcst.h, is included by some of the .cpp regex
michael@0: #             implementation files.   This perl script is NOT run as part
michael@0: #             of a normal ICU build.  It is run by hand when needed, and the
michael@0: #             regexcst.h generated file is put back into cvs.
michael@0: #
michael@0: #             See regexcst.txt for a description of the input format for this script.
michael@0: #
michael@0: #             This script is derived from rbbicst.pl, which peforms the same function
michael@0: #             for the Rule Based Break Iterator Rule Parser.  Perhaps they could be
michael@0: #             merged?
michael@0: #
michael@0: 
michael@0: 
michael@0: $num_states = 1;         # Always the state number for the line being compiled.
michael@0: $line_num  = 0;          # The line number in the input file.
michael@0: 
michael@0: $states{"pop"} = 255;    # Add the "pop"  to the list of defined state names.
michael@0:                          # This prevents any state from being labelled with "pop",
michael@0:                          #  and resolves references to "pop" in the next state field.
michael@0: 
michael@0: line_loop: while (<>) {
michael@0:     chomp();
michael@0:     $line = $_;
michael@0:     @fields = split();
michael@0:     $line_num++;
michael@0: 
michael@0:     # Remove # comments, which are any fields beginning with a #, plus all
michael@0:     #  that follow on the line.
michael@0:     for ($i=0; $i<@fields; $i++) {
michael@0:         if ($fields[$i] =~ /^#/) {
michael@0:             @fields = @fields[0 .. $i-1];
michael@0:             last;
michael@0:         }
michael@0:     }
michael@0:     # ignore blank lines, and those with no fields left after stripping comments..
michael@0:     if (@fields == 0) {
michael@0:         next;
michael@0:     }
michael@0: 
michael@0:     #
michael@0:     # State Label:  handling.
michael@0:     #    Does the first token end with a ":"?  If so, it's the name  of a state.
michael@0:     #    Put in a hash, together with the current state number,
michael@0:     #        so that we can later look up the number from the name.
michael@0:     #
michael@0:     if (@fields[0] =~ /.*:$/) {
michael@0:         $state_name = @fields[0];
michael@0:         $state_name =~ s/://;        # strip off the colon from the state name.
michael@0: 
michael@0:         if ($states{$state_name} != 0) {
michael@0:             print "  rbbicst: at line $line-num duplicate definition of state $state_name\n";
michael@0:         }
michael@0:         $states{$state_name} = $num_states;
michael@0:         $stateNames[$num_states] = $state_name;
michael@0: 
michael@0:         # if the label was the only thing on this line, go on to the next line,
michael@0:         # otherwise assume that a state definition is on the same line and fall through.
michael@0:         if (@fields == 1) {
michael@0:             next line_loop;
michael@0:         }
michael@0:         shift @fields;                       # shift off label field in preparation
michael@0:                                              #  for handling the rest of the line.
michael@0:     }
michael@0: 
michael@0:     #
michael@0:     # State Transition line.
michael@0:     #   syntax is this,
michael@0:     #       character   [n]  target-state  [^push-state]  [function-name]
michael@0:     #   where
michael@0:     #      [something]   is an optional something
michael@0:     #      character     is either a single quoted character e.g. '['
michael@0:     #                       or a name of a character class, e.g. white_space
michael@0:     #
michael@0: 
michael@0:     $state_line_num[$num_states] = $line_num;   # remember line number with each state
michael@0:                                                 #  so we can make better error messages later.
michael@0:     #
michael@0:     # First field, character class or literal character for this transition.
michael@0:     #
michael@0:     if ($fields[0] =~ /^'.'$/) {
michael@0:         # We've got a quoted literal character.
michael@0:         $state_literal_chars[$num_states] = $fields[0];
michael@0:         $state_literal_chars[$num_states] =~ s/'//g;
michael@0:     } else {
michael@0:         # We've got the name of a character class.
michael@0:         $state_char_class[$num_states] = $fields[0];
michael@0:         if ($fields[0] =~ /[\W]/) {
michael@0:             print "  rbbicsts:  at line $line_num, bad character literal or character class name.\n";
michael@0:             print "     scanning $fields[0]\n";
michael@0:             exit(-1);
michael@0:         }
michael@0:     }
michael@0:     shift @fields;
michael@0: 
michael@0:     #
michael@0:     # do the 'n' flag
michael@0:     #
michael@0:     $state_flag[$num_states] = "FALSE";
michael@0:     if ($fields[0] eq "n") {
michael@0:         $state_flag[$num_states] = "TRUE";
michael@0:         shift @fields;
michael@0:     }
michael@0: 
michael@0:     #
michael@0:     # do the destination state.
michael@0:     #
michael@0:     $state_dest_state[$num_states] = $fields[0];
michael@0:     if ($fields[0] eq "") {
michael@0:         print "  rbbicsts:  at line $line_num, destination state missing.\n";
michael@0:         exit(-1);
michael@0:     }
michael@0:     shift @fields;
michael@0: 
michael@0:     #
michael@0:     # do the push state, if present.
michael@0:     #
michael@0:     if ($fields[0] =~ /^\^/) {
michael@0:         $fields[0] =~ s/^\^//;
michael@0:         $state_push_state[$num_states] = $fields[0];
michael@0:         if ($fields[0] eq "" ) {
michael@0:             print "  rbbicsts:  at line $line_num, expected state after ^ (no spaces).\n";
michael@0:             exit(-1);
michael@0:         }
michael@0:         shift @fields;
michael@0:     }
michael@0: 
michael@0:     #
michael@0:     # Lastly, do the optional action name.
michael@0:     #
michael@0:     if ($fields[0] ne "") {
michael@0:         $state_func_name[$num_states] = $fields[0];
michael@0:         shift @fields;
michael@0:     }
michael@0: 
michael@0:     #
michael@0:     #  There should be no fields left on the line at this point.
michael@0:     #
michael@0:     if (@fields > 0) {
michael@0:        print "  rbbicsts:  at line $line_num, unexpected extra stuff on input line.\n";
michael@0:        print "     scanning $fields[0]\n";
michael@0:    }
michael@0:    $num_states++;
michael@0: }
michael@0: 
michael@0: #
michael@0: # We've read in the whole file, now go back and output the
michael@0: #   C source code for the state transition table.
michael@0: #
michael@0: # We read all states first, before writing anything,  so that the state numbers
michael@0: # for the destination states are all available to be written.
michael@0: #
michael@0: 
michael@0: #
michael@0: # Make hashes for the names of the character classes and
michael@0: #      for the names of the actions that appeared.
michael@0: #
michael@0: for ($state=1; $state < $num_states; $state++) {
michael@0:     if ($state_char_class[$state] ne "") {
michael@0:         if ($charClasses{$state_char_class[$state]} == 0) {
michael@0:             $charClasses{$state_char_class[$state]} = 1;
michael@0:         }
michael@0:     }
michael@0:     if ($state_func_name[$state] eq "") {
michael@0:         $state_func_name[$state] = "doNOP";
michael@0:     }
michael@0:     if ($actions{$state_action_name[$state]} == 0) {
michael@0:         $actions{$state_func_name[$state]} = 1;
michael@0:     }
michael@0: }
michael@0: 
michael@0: #
michael@0: # Check that all of the destination states have been defined
michael@0: #
michael@0: #
michael@0: $states{"exit"} = 0;              # Predefined state name, terminates state machine.
michael@0: for ($state=1; $state<$num_states; $state++) {
michael@0:    if ($states{$state_dest_state[$state]} == 0 && $state_dest_state[$state] ne "exit") {
michael@0:        print "Error at line $state_line_num[$state]: target state \"$state_dest_state[$state]\" is not defined.\n";
michael@0:        $errors++;
michael@0:    }
michael@0:    if ($state_push_state[$state] ne "" && $states{$state_push_state[$state]} == 0) {
michael@0:        print "Error at line $state_line_num[$state]: target state \"$state_push_state[$state]\" is not defined.\n";
michael@0:        $errors++;
michael@0:    }
michael@0: }
michael@0: 
michael@0: die if ($errors>0);
michael@0: 
michael@0: print "//---------------------------------------------------------------------------------\n";
michael@0: print "//\n";
michael@0: print "// Generated Header File.  Do not edit by hand.\n";
michael@0: print "//    This file contains the state table for the ICU Regular Expression Pattern Parser\n";
michael@0: print "//    It is generated by the Perl script \"regexcst.pl\" from\n";
michael@0: print "//    the rule parser state definitions file \"regexcst.txt\".\n";
michael@0: print "//\n";
michael@0: print "//   Copyright (C) 2002-2007 International Business Machines Corporation \n";
michael@0: print "//   and others. All rights reserved.  \n";
michael@0: print "//\n";
michael@0: print "//---------------------------------------------------------------------------------\n";
michael@0: print "#ifndef RBBIRPT_H\n";
michael@0: print "#define RBBIRPT_H\n";
michael@0: print "\n";
michael@0: print "U_NAMESPACE_BEGIN\n";
michael@0: 
michael@0: #
michael@0: # Emit the constants for indicies of Unicode Sets
michael@0: #   Define one constant for each of the character classes encountered.
michael@0: #   At the same time, store the index corresponding to the set name back into hash.
michael@0: #
michael@0: print "//\n";
michael@0: print "// Character classes for regex pattern scanning.\n";
michael@0: print "//\n";
michael@0: $i = 128;                   # State Table values for Unicode char sets range from 128-250.
michael@0:                             # Sets "default", "quoted", etc. get special handling.
michael@0:                             #  They have no corresponding UnicodeSet object in the state machine,
michael@0:                             #    but are handled by special case code.  So we emit no reference
michael@0:                             #    to a UnicodeSet object to them here.
michael@0: foreach $setName (keys %charClasses) {
michael@0:     if ($setName eq "default") {
michael@0:         $charClasses{$setName} = 255;}
michael@0:     elsif ($setName eq "quoted") {
michael@0:         $charClasses{$setName} = 254;}
michael@0:     elsif ($setName eq "eof") {
michael@0:         $charClasses{$setName} = 253;}
michael@0:     else {
michael@0:         # Normal character class.  Fill in array with a ptr to the corresponding UnicodeSet in the state machine.
michael@0:        print "    static const uint8_t kRuleSet_$setName = $i;\n";
michael@0:         $charClasses{$setName} = $i;
michael@0:         $i++;
michael@0:     }
michael@0: }
michael@0: print "\n\n";
michael@0: 
michael@0: #
michael@0: # Emit the enum for the actions to be performed.
michael@0: #
michael@0: print "enum Regex_PatternParseAction {\n";
michael@0: foreach $act (keys %actions) {
michael@0:     print "    $act,\n";
michael@0: }
michael@0: print "    rbbiLastAction};\n\n";
michael@0: 
michael@0: #
michael@0: # Emit the struct definition for transtion table elements.
michael@0: #
michael@0: print "//-------------------------------------------------------------------------------\n";
michael@0: print "//\n";
michael@0: print "//  RegexTableEl       represents the structure of a row in the transition table\n";
michael@0: print "//                     for the pattern parser state machine.\n";
michael@0: print "//-------------------------------------------------------------------------------\n";
michael@0: print "struct RegexTableEl {\n";
michael@0: print "    Regex_PatternParseAction      fAction;\n";
michael@0: print "    uint8_t                       fCharClass;       // 0-127:    an individual ASCII character\n";
michael@0: print "                                                    // 128-255:  character class index\n";
michael@0: print "    uint8_t                       fNextState;       // 0-250:    normal next-state numbers\n";
michael@0: print "                                                    // 255:      pop next-state from stack.\n";
michael@0: print "    uint8_t                       fPushState;\n";
michael@0: print "    UBool                         fNextChar;\n";
michael@0: print "};\n\n";
michael@0: 
michael@0: #
michael@0: # emit the state transition table
michael@0: #
michael@0: print "static const struct RegexTableEl gRuleParseStateTable[] = {\n";
michael@0: print "    {doNOP, 0, 0, 0, TRUE}\n";    # State 0 is a dummy.  Real states start with index = 1.
michael@0: for ($state=1; $state < $num_states; $state++) {
michael@0:     print "    , {$state_func_name[$state],";
michael@0:     if ($state_literal_chars[$state] ne "") {
michael@0:         $c = $state_literal_chars[$state];
michael@0:         printf(" %d /* $c */,", ord($c));   #  use numeric value, so EBCDIC machines are ok.
michael@0:     }else {
michael@0:         print " $charClasses{$state_char_class[$state]},";
michael@0:     }
michael@0:     print " $states{$state_dest_state[$state]},";
michael@0: 
michael@0:     # The push-state field is optional.  If omitted, fill field with a zero, which flags
michael@0:     #   the state machine that there is no push state.
michael@0:     if ($state_push_state[$state] eq "") {
michael@0:         print "0, ";
michael@0:     } else {
michael@0:         print " $states{$state_push_state[$state]},";
michael@0:     }
michael@0:     print " $state_flag[$state]} ";
michael@0: 
michael@0:     # Put out a C++ comment showing the number (index) of this state row,
michael@0:     #   and, if this is the first row of the table for this state, the state name.
michael@0:     print "    //  $state ";
michael@0:     if ($stateNames[$state] ne "") {
michael@0:         print "     $stateNames[$state]";
michael@0:     }
michael@0:     print "\n";
michael@0: };
michael@0: print " };\n";
michael@0: 
michael@0: 
michael@0: #
michael@0: # emit a mapping array from state numbers to state names.
michael@0: #
michael@0: #    This array is used for producing debugging output from the pattern parser.
michael@0: #
michael@0: print "static const char * const RegexStateNames[] = {";
michael@0: for ($state=0; $state<$num_states; $state++) {
michael@0:     if ($stateNames[$state] ne "") {
michael@0:         print "     \"$stateNames[$state]\",\n";
michael@0:     } else {
michael@0:         print "    0,\n";
michael@0:     }
michael@0: }
michael@0: print "    0};\n\n";
michael@0: 
michael@0: print "U_NAMESPACE_END\n";
michael@0: print "#endif\n";
michael@0: 
michael@0: 
michael@0: