// file: $isip/class/pr/LanguageModelBNF/lmbnf_07.cc
//

// system include files
//
#include <typeinfo>

// isip include files
//
#include "LanguageModelBNF.h"


// method: minimizeGraph
//
// arguments: 
//
// return: 
//
//
bool8 LanguageModelBNF::minimizeGraph() {

  RuleModel rm_min;

  Vector<Vector<ProductionRuleSet> > graph_levels;
  
  // loop over levels
  //
  for (int32 i = 0; i < rm_d.first().length(); i++) {

    Vector<ProductionRuleSet> graphs;
    
    // loop over graphs
    //
    for (int32 j = 0; j < rm_d.first()(i).length(); j++) {    

      ProductionRuleSet current_graph = rm_d.first()(i)(j);

      // merge rules
      //
      for (int32 k = 0; k < current_graph.length(); k++) {

	int32 l = 0;

	while (l < current_graph.length()) {

	  // we don't need to compare a rule to itself
	  //
	  if (k != l) {
	    
	    // since all rules below the merged rule will be
	    // shifted up, we have to adjust the indices
	    // appropriately
	    //
	    if (!mergeRules(current_graph, k, l)) {
	      l++;
	    }
	    else {
	      if (k > l) {
		k--;
	      }	    
	    }
	  }
	  else {
	    l++;
	  }
	}
      }
      graphs.concat(current_graph);
    }
    graph_levels.concat(graphs);
  }

  rm_min.assign(graph_levels, rm_d.second());
  
  // return the minimized graph
  //
  return rm_d.assign(rm_min);
  
}

// method: mergeRules
//
// arguments: 
//
// return: bool8 indicating whether or not the two rules were merged
//
bool8 LanguageModelBNF::mergeRules(ProductionRuleSet& rules_a,
				     int32 index1_a, int32 index2_a) {
  
  ProductionRule rule1, rule2;
  String non_terminal1, non_terminal2;

  rule1.assign(rules_a(index1_a));
  rule2.assign(rules_a(index2_a));

  // get the terminal symbol from each rule (1st token).  if they're not the
  // same, these two rules cannot be merged.
  //
  rule1.gotoFirst();
  rule2.gotoFirst();  

  // make sure that these rules are valid BNF rules
  //
  if (!isValidBNF(rule1) || !isValidBNF(rule2)) {

    return false;
  }

  // if both of these are start rules, we need to check to see if they
  // are of the form:
  //
  // (start rule)->(non_terminal)
  //
  // and if they are, we need to see if the terminal referenced by
  // R0 is the same for both rules.  if it is, we can merge these
  // start rules
  //
  if (rule1.getRuleType() == ProductionRule::START &&
      rule1.length() == 1 &&
      rule2.getRuleType() == ProductionRule::START &&
      rule2.length() == 1) {

    Vector<SearchSymbol> terminals1, terminals2;    

    non_terminal1.assign(rule1.getValue());
    non_terminal2.assign(rule2.getValue());
    
    // get all the terminals referenced by the non_terminal
    //
    terminals1.assign(findTerminals(rules_a, non_terminal1));
    terminals2.assign(findTerminals(rules_a, non_terminal2));

    // if either start rule references more than one terminal, then
    // we can't merge them
    //
    if (terminals1.length() != 1 || terminals2.length() != 1) {

      return false;
    }

    // if the terminals are different, then we cant merge the start rules
    //
    if (!terminals1(0).eq(terminals2(0))) {

      return false;
    }
    
  }
  else {
    
    // verify that both rules are mergeable types.  that is, make sure that
    // the rules match this form:
    //
    // (rule)->(terminal),(non_terminal)
    //
    if (!isMergeType(rule1) || !isMergeType(rule2)) {
      
      return false;
    }
    
    if (!rule1.getValue().eq(rule2.getValue()) ||
	!rule1.getRuleName().eq(rule2.getRuleName())) {
      
      return false;
    }
    
    // get the terminal referenced by the non_terminal (last token) for
    // each rule.
    //
    Vector<SearchSymbol> terminals1, terminals2;
    
    rule1.gotoPosition(MERGE_TYPE_LEN - 1);
    rule2.gotoPosition(MERGE_TYPE_LEN - 1);
    
    non_terminal1.assign(rule1.getValue());
    non_terminal2.assign(rule2.getValue());
    
    // goto the concatenation symbol
    //
    rule1.gotoPrev();
    rule2.gotoPrev();
    
    // we must make sure that the weights on each concatenation
    // are the same.  if not, the rules cannot be merged.
    //
    if (rule1.getWeight() != rule2.getWeight()) {
      
      return false;
    }
    
    // get all the terminals referenced by the non_terminal
    //
    terminals1.assign(findTerminals(rules_a, non_terminal1));
    terminals2.assign(findTerminals(rules_a, non_terminal2));
    
    // check to see that the non_terminal for each node only references
    // one terminal. all sets of BNF rules generated by the IFC LM classes
    // ensure that each non_terminal only references one terminal.  however,
    // this is not a requirement of BNF, and we can't make this assumption
    // when minimizing the graphs.  if the non_terminal for either rule
    // references more than one terminal, then the two rules cannot be merged
    //
    if (terminals1.length() != 1 || terminals2.length() != 1) {
      
      return false;
    }
    
    // if the terminal node isn't the same, we can't merge
    //
    if (!terminals1(0).eq(terminals2(0))) {
      
      return false;
    }
  }
  
  // check to see if rule1 directly references rule2.  if so, we can only
  // merge the two rules if one of the rules references itself, i.e. has
  // a self loop.
  //
  if (detectFalseLoop(non_terminal1, non_terminal2, rules_a)) {

    return false;
  }

  // if we reach this point, we should be able to merge the two rules.  Find
  // all non_terminal tokens and rules that match the name of the non_terminal
  // in the second rule, and rename them to the non_terminal name of the first
  // rule
  //
  for (int32 i = 0; i < rules_a.length(); i++) {    

    if (isValidBNF(rules_a(i))) {
      
      if (rules_a(i).getRuleName().eq(non_terminal2)) {
	rules_a(i).setRuleName(non_terminal1);
      }
      
      rules_a(i).gotoFirst();
      
      if (rules_a(i).getType() == ProductionRuleTokenType::NON_TERMINAL &&
	  rules_a(i).getValue().eq(non_terminal2)) {
	rules_a(i).setValue(non_terminal1);
      }
      else if (rules_a(i).length() == MERGE_TYPE_LEN) {
	rules_a(i).gotoPosition(MERGE_TYPE_LEN - 1);
	if ( rules_a(i).getValue().eq(non_terminal2)) {
	  rules_a(i).setValue(non_terminal1);
	}
      }
    }
  }
  
  // finally, remove the 2nd rule from the rule set and return true
  //
  rules_a.deleteRange(index2_a, 1);
  
  return true;
}

// method: isValidBNF
//
// arguments: ProductionRule rule
//
// return: bool8 indicating whether or not the rule is a valid BNF Rule
//
bool8 LanguageModelBNF::isValidBNF(ProductionRule rule_a) {

  rule_a.gotoFirst();

  // check the types:
  //
  // (rule)->(epsilon)
  // (rule)->(non_terminal)
  //
  if (rule_a.length() == 1) {

    if (rule_a.getType() == ProductionRuleTokenType::EPSILON ||
	rule_a.getType() == ProductionRuleTokenType::NON_TERMINAL) {

      return true;
    }
  }  
  else {

    // check the type:
    //
    // (rule)->(terminal),(non_terminal)
    //
    if (!isMergeType(rule_a)) {	
      return false;
    }
  }
  
  // exit gracefully
  //
  return true;
}

// method: isMergeType
//
// arguments: ProductionRule rule
//
// return: bool8 indicating whether or not this is the type of rule that
//         can be merged
//
bool8 LanguageModelBNF::isMergeType(ProductionRule rule_a) {

  rule_a.gotoFirst();
  
  // check the type:
  //
  // (rule)->(terminal),(non_terminal)
  //
  if (rule_a.length() == 3) {
    
    if (rule_a.getType() != ProductionRuleTokenType::TERMINAL) {
      
      return false;
    }
    
    rule_a.gotoNext();
    
    if (rule_a.getType() != ProductionRuleTokenType::CONCATENATION) {
      
      return false;
    }
    
    rule_a.gotoNext();
    
    if (rule_a.getType() != ProductionRuleTokenType::NON_TERMINAL) {
      
      return false;
    }
  }
  else {
    return false;
  }

  return true;
}

// method: findTerminal
//
// arguments: ProductionRuleSet rules
//            String rule_name
//
// return: a Vector<SearchSymbol> containing all the terminal tokens.
//         these correspond to unique terminal nodes
//
// finds all unique terminal symbols given a non_terminal symbol
//
Vector<SearchSymbol> LanguageModelBNF::findTerminals(ProductionRuleSet rules_a,
						     String rule_name_a) {

  Vector<SearchSymbol> terminal_nodes;
  
  for (int32 i = 0; i < rules_a.length(); i++) {
    
    if (rules_a(i).getRuleName().eq(rule_name_a)){
      
      rules_a(i).gotoFirst();
      
      if (rules_a(i).getType() == ProductionRuleTokenType::TERMINAL) {
	
	SearchSymbol terminal = rules_a(i).getValue();

	if (!terminal_nodes.contains(&terminal)) {
	  terminal_nodes.concat(terminal);
	}
      }
      else if (rules_a(i).getType() == ProductionRuleTokenType::NON_TERMINAL) {
	
	terminal_nodes.concat(findTerminals(rules_a, rules_a(i).getValue()));
      }
    }
  }
  
  return terminal_nodes;
}

// method: detectFalseLoop
//
// arguments:  String non_terminal1_a
//	       String non_terminal2_a
//
// return: bool8 indicating whether or not the two non_terminals
//         (rule references) result in a false loop.  A false loop
//         occurs when rule1 references rule2, but not vice versa.
//
bool8 LanguageModelBNF::detectFalseLoop(String non_terminal1_a,
					  String non_terminal2_a,
					  ProductionRuleSet rules_a) {

  ProductionRuleSet rules1, rules2, rules;  
  bool8 reference1 = false, reference2 = false, selfloop = false;
  bool8 falseloop = false;

  rules1.assign(getMergeTypeRules(non_terminal1_a, rules_a));
  rules2.assign(getMergeTypeRules(non_terminal2_a, rules_a));

  rules.concat(rules1);
  rules.concat(rules2);
  
  for (int32 i = 0; i < rules.length(); i++) {

    rules(i).gotoPosition(MERGE_TYPE_LEN - 1);

    if (rules(i).getRuleName().eq(non_terminal1_a) &&
	rules(i).getValue().eq(non_terminal2_a)) {
      reference1 = true;
    }
    if (rules(i).getRuleName().eq(non_terminal2_a) &&
	     rules(i).getValue().eq(non_terminal1_a)) {
      reference2 = true;
    }
    if ((rules(i).getRuleName().eq(non_terminal2_a) &&
	  rules(i).getValue().eq(non_terminal2_a)) ||
	 (rules(i).getRuleName().eq(non_terminal1_a) &&
	  rules(i).getValue().eq(non_terminal1_a))) {

      selfloop = true;
    }
  }
  
  if (selfloop) {
    
    Vector<String> non_terms1, non_terms2;
    
    for (int32 j = 0; j < rules1.length(); j++) {
      
      rules1(j).gotoPosition(MERGE_TYPE_LEN - 1);
      
      if (!rules1(j).getValue().eq(non_terminal1_a) &&
	  !rules1(j).getValue().eq(non_terminal2_a)) {
	non_terms1.concat(rules1(j).getValue());
      }
    }
    
    for (int32 j = 0; j < rules2.length(); j++) {
      
      rules2(j).gotoPosition(MERGE_TYPE_LEN - 1);
      
      if (!rules2(j).getValue().eq(non_terminal2_a) &&
	  !rules2(j).getValue().eq(non_terminal1_a)) {
	non_terms2.concat(rules2(j).getValue());
      }      
    }
    
    non_terms1.sort();
    non_terms2.sort();
    
    if (!non_terms1.eq(non_terms2) &&
	(non_terms1.length() > 0 && non_terms2.length() > 0)) {
      falseloop = true;
    }
  }

  if (falseloop) {
    return true;
  }
  else if (selfloop || (reference1 && reference2)) {
    return false;
  }
  else if (reference1 || reference2) {
    return true;
  }
  
  return false;
}

// method: getMergeTypeRules
//
// arguments:  
//
// return: 
//
ProductionRuleSet LanguageModelBNF::getMergeTypeRules(String non_terminal_a,
						      ProductionRuleSet rules_a) {

  ProductionRuleSet merge_type_rules;

  for (int32 i = 0; i < rules_a.length(); i++) {

    if (rules_a(i).getRuleName().eq(non_terminal_a)) {

      if (isMergeType(rules_a(i))) {

	merge_type_rules.concat(rules_a(i));
      }
      else if (rules_a(i).length() == 1) {

	ProductionRuleSet temp_rules;

	temp_rules.concat(getMergeTypeRules(rules_a(i).getValue(), rules_a));

	for (int32 j = 0; j < temp_rules.length(); j++) {

	  temp_rules(j).setRuleName(non_terminal_a);
	}

	merge_type_rules.concat(temp_rules);        
	
      }
    }
  }
  
  return merge_type_rules;  
}