}//method
#region some explanations
//Computes non-canonical lookaheads and jump lookaheads - those that cause jump
// to non-canonical state
// We are doing it top-down way, starting from most reduced lookaheads - they are not conflicting.
// (If there were conflicting reduced lookaheads in a state initially, the grammar transformation algorithm
// should have already wrapped them into non-conflicting "tail" non-terminals.)
// We want to eliminate reduced lookaheads as much as possible, and replace them with expanded "child"
// terms, to have only those non-canonical lookaheads that are absolutely necessary.
// So for each reduced lookahead we check if we can replace it with its expanded, "child" terms
// (from DirectFirsts set). We do it only if lookaheads child terms are all non-conflicting as lookaheads in
// the state. If however, at least one child is conflicting, the reduced parent should stay.
// What if we have some children conflicting and some not? We leave the parent reduced lookahead in state,
// to cover (hide) the conflicting children, but we also add non-conflicting children as well, to allow
// the parser automaton to use them (in canonical state) as soon as they are recognized, without need
// to reduce the parent and switch back to canonical state.
#endregion
private void ComputeStateNonCanonicalLookaheads(ParserState state) {
var stateData = state.BuilderData;
//rename for shorter code
var jumps = stateData.JumpLookaheads; // conflicting lookaheads, that must result in jump to non-canonical state
var valids = stateData.NonCanonicalLookaheads; // valid non-canonical lookaheads, non-terminals only
jumps.Clear();
valids.Clear();
var alreadyChecked = new BnfTermSet();
var toCheck = new BnfTermSet(); //terms to check for expansion
//1. precompute initial set to check
foreach (var reduceItem in stateData.ReduceItems)
toCheck.UnionWith(reduceItem.ReducedLookaheads);
toCheck.RemoveWhere(t => t is Terminal); //we are interested in non-terminals only
//2. Try to expand all initial (reduced) lookaheads, and replace original lookaheads with expanded versions
while (toCheck.Count > 0) { // do until no terms to check left
var lkhInCheck = toCheck.First() as NonTerminal;
toCheck.Remove(lkhInCheck);
//to prevent repeated checking of mutually recursive terms
if (alreadyChecked.Contains(lkhInCheck)) continue;
alreadyChecked.Add(lkhInCheck);
//Now check children for conflicts; go through all direct firsts of lkhInCheck and check them for conflicts
bool hasJumpChild = false;
foreach (var lkhChild in lkhInCheck.DirectFirsts) {
if (lkhChild == lkhInCheck) continue;
if (jumps.Contains(lkhChild)) {
hasJumpChild = true;
continue;
}
var ntChild = lkhChild as NonTerminal;
if (ntChild != null && valids.Contains(ntChild)) continue;
//the child has not been tested yet; check if it is a conflict in current state
var occurCount = GetLookaheadOccurenceCount(state, lkhChild);
if (occurCount > 1) {
//possible conflict, check precedence
if (lkhChild.IsSet(TermOptions.UsePrecedence)) {
if (ntChild != null) {
valids.Add(ntChild); //if it is terminal, it is valid;
if (!alreadyChecked.Contains(lkhChild)) toCheck.Add(ntChild);
} //if ntChild
} else {
//conflict!
hasJumpChild = true;
jumps.Add(lkhChild);
//if it is non-terminal, add its Firsts to conflict as well
if (ntChild != null) {
jumps.UnionWith(ntChild.Firsts);
//valids.ExceptWith(ntChild.Firsts);
}
}//if IsSet... else...
} else { //occurCount == 1
//no conflict: if it is non-terminal, add it to toCheck set to check in the future
if (ntChild != null && !alreadyChecked.Contains(ntChild))
toCheck.Add(ntChild); //if nonterminal and not checked yet, add it to toCheck for further checking
}//if ...else...
}//foreach lkhChild
//Ok, we finished checking all direct children; if at least one of them has conflict,
// then lkhInCheck (parent) must stay as a lookahead - we cannot fully expand it replacing by all children
if (hasJumpChild)
valids.Add(lkhInCheck);
}//while toCheck.Count > 0
//remove conflicts
stateData.Conflicts.Clear();
}//method
}//method
#region some explanations
//Computes non-canonical lookaheads and jump lookaheads - those that cause jump
// to non-canonical state
// We are doing it top-down way, starting from most reduced lookaheads - they are not conflicting.
// (If there were conflicting reduced lookaheads in a state initially, the grammar transformation algorithm
// should have already wrapped them into non-conflicting "tail" non-terminals.)
// We want to eliminate reduced lookaheads as much as possible, and replace them with expanded "child"
// terms, to have only those non-canonical lookaheads that are absolutely necessary.
// So for each reduced lookahead we check if we can replace it with its expanded, "child" terms
// (from DirectFirsts set). We do it only if lookaheads child terms are all non-conflicting as lookaheads in
// the state. If however, at least one child is conflicting, the reduced parent should stay.
// What if we have some children conflicting and some not? We leave the parent reduced lookahead in state,
// to cover (hide) the conflicting children, but we also add non-conflicting children as well, to allow
// the parser automaton to use them (in canonical state) as soon as they are recognized, without need
// to reduce the parent and switch back to canonical state.
#endregion
private void ComputeStateNonCanonicalLookaheads(ParserState state)
{
var stateData = state.BuilderData;
//rename for shorter code
var jumps = stateData.JumpLookaheads; // conflicting lookaheads, that must result in jump to non-canonical state
var valids = stateData.NonCanonicalLookaheads; // valid non-canonical lookaheads, non-terminals only
jumps.Clear();
valids.Clear();
var alreadyChecked = new BnfTermSet();
var toCheck = new BnfTermSet(); //terms to check for expansion
//1. precompute initial set to check
foreach (var reduceItem in stateData.ReduceItems)
{
toCheck.UnionWith(reduceItem.ReducedLookaheads);
}
toCheck.RemoveWhere(t => t is Terminal); //we are interested in non-terminals only
//2. Try to expand all initial (reduced) lookaheads, and replace original lookaheads with expanded versions
while (toCheck.Count > 0) // do until no terms to check left
{
var lkhInCheck = toCheck.First() as NonTerminal;
toCheck.Remove(lkhInCheck);
//to prevent repeated checking of mutually recursive terms
if (alreadyChecked.Contains(lkhInCheck))
{
continue;
}
alreadyChecked.Add(lkhInCheck);
//Now check children for conflicts; go through all direct firsts of lkhInCheck and check them for conflicts
bool hasJumpChild = false;
foreach (var lkhChild in lkhInCheck.DirectFirsts)
{
if (lkhChild == lkhInCheck)
{
continue;
}
if (jumps.Contains(lkhChild))
{
hasJumpChild = true;
continue;
}
var ntChild = lkhChild as NonTerminal;
if (ntChild != null && valids.Contains(ntChild))
{
continue;
}
//the child has not been tested yet; check if it is a conflict in current state
var occurCount = GetLookaheadOccurenceCount(state, lkhChild);
if (occurCount > 1)
{
//possible conflict, check precedence
if (lkhChild.IsSet(TermOptions.UsePrecedence))
{
if (ntChild != null)
{
valids.Add(ntChild); //if it is terminal, it is valid;
if (!alreadyChecked.Contains(lkhChild))
{
toCheck.Add(ntChild);
}
} //if ntChild
}
else
{
//conflict!
hasJumpChild = true;
jumps.Add(lkhChild);
//if it is non-terminal, add its Firsts to conflict as well
if (ntChild != null)
{
jumps.UnionWith(ntChild.Firsts);
//valids.ExceptWith(ntChild.Firsts);
}
} //if IsSet... else...
}
else //occurCount == 1
//no conflict: if it is non-terminal, add it to toCheck set to check in the future
{
if (ntChild != null && !alreadyChecked.Contains(ntChild))
{
toCheck.Add(ntChild); //if nonterminal and not checked yet, add it to toCheck for further checking
}
}//if ...else...
}//foreach lkhChild
//Ok, we finished checking all direct children; if at least one of them has conflict,
// then lkhInCheck (parent) must stay as a lookahead - we cannot fully expand it replacing by all children
if (hasJumpChild)
{
valids.Add(lkhInCheck);
}
}//while toCheck.Count > 0
//remove conflicts
stateData.Conflicts.Clear();
}//method