//------------------------------------------------------------------------- // // findNodes() Locate all the nodes of the specified type, starting // at the specified root. // //------------------------------------------------------------------------- internal virtual void FindNodes(IList <RBBINode> dest, int kind) { if (fType == kind) { dest.Add(this); } if (fLeftChild != null) { fLeftChild.FindNodes(dest, kind); } if (fRightChild != null) { fRightChild.FindNodes(dest, kind); } }
//----------------------------------------------------------------------------- // // calcChainedFollowPos. Modify the previously calculated followPos sets // to implement rule chaining. NOT described by Aho // //----------------------------------------------------------------------------- internal virtual void CalcChainedFollowPos(RBBINode tree) { IList <RBBINode> endMarkerNodes = new JCG.List <RBBINode>(); IList <RBBINode> leafNodes = new JCG.List <RBBINode>(); // get a list of all endmarker nodes. tree.FindNodes(endMarkerNodes, RBBINode.endMark); // get a list all leaf nodes tree.FindNodes(leafNodes, RBBINode.leafChar); // Collect all leaf nodes that can start matches for rules // with inbound chaining enabled, which is the union of the // firstPosition sets from each of the rule root nodes. IList <RBBINode> ruleRootNodes = new JCG.List <RBBINode>(); AddRuleRootNodes(ruleRootNodes, tree); ISet <RBBINode> matchStartNodes = new JCG.HashSet <RBBINode>(); foreach (RBBINode node in ruleRootNodes) { if (node.fChainIn) { matchStartNodes.UnionWith(node.fFirstPosSet); } } // Iterate over all leaf nodes, // foreach (RBBINode tNode in leafNodes) { RBBINode endNode = null; // Identify leaf nodes that correspond to overall rule match positions. // These include an endMarkerNode in their followPos sets. foreach (RBBINode endMarkerNode in endMarkerNodes) { if (tNode.fFollowPos.Contains(endMarkerNode)) { endNode = tNode; break; } } if (endNode == null) { // node wasn't an end node. Try again with the next. continue; } // We've got a node that can end a match. // Line Break Specific hack: If this node's val correspond to the $CM char class, // don't chain from it. // TODO: Add rule syntax for this behavior, get specifics out of here and // into the rule file. if (fRB.fLBCMNoChain) { int c = this.fRB.fSetBuilder.GetFirstChar(endNode.fVal); if (c != -1) { // c == -1 occurs with sets containing only the {eof} marker string. int cLBProp = UChar.GetIntPropertyValue(c, UProperty.Line_Break); if (cLBProp == LineBreak.CombiningMark) { continue; } } } // Now iterate over the nodes that can start a match, looking for ones // with the same char class as our ending node. foreach (RBBINode startNode in matchStartNodes) { if (startNode.fType != RBBINode.leafChar) { continue; } if (endNode.fVal == startNode.fVal) { // The end val (character class) of one possible match is the // same as the start of another. // Add all nodes from the followPos of the start node to the // followPos set of the end node, which will have the effect of // letting matches transition from a match state at endNode // to the second char of a match starting with startNode. endNode.fFollowPos.UnionWith(startNode.fFollowPos); } } } }