clean

src/include/processor/operator/recursive_extend/path_scanner.h

@@ -7,14 +7,27 @@
 namespace kuzu {
 namespace processor {
 
+/*
+ * PathScanner Design
+ *
+ * RecursiveJoin first compute frontier from src to dst using BFS.
+ *
+ * PathScanner then scans frontiers from dst to src using DFS. We use stack-based DFS to track
+ * prefix. We scan from dst to src because semi mask might be available and we only need to scan a
+ * subset of all dst nodes.
+ *
+ * Since variable-length join requires union of path with different length (e.g. *2..3), PathScanner
+ * scans length by length, i.e. we first scan all length 2 path from second frontier, then scan all
+ * length 3 path from third frontier.
+ */
+
 // FixedSizePathScanner scans all paths of a fixed length.
 struct FixedLengthPathScanner {
     using nbrs_t = std::vector<common::offset_t>*;
-
+    // FixedLengthPathScanner scans from bfsMorsel frontiers and semi mask.
     const BaseBFSMorsel& bfsMorsel;
     // Number of extension performed during recursive join
-    size_t depth;
-    // Current path.
+    size_t numRels;
     std::vector<common::offset_t> path;
 
     // DFS states
@@ -23,9 +36,9 @@ struct FixedLengthPathScanner {
 
     size_t lastFrontierCursor;
 
-    FixedLengthPathScanner(const BaseBFSMorsel& bfsMorsel, size_t depth)
-        : bfsMorsel{bfsMorsel}, depth{depth} {
-        path.resize(depth + 1);
+    FixedLengthPathScanner(const BaseBFSMorsel& bfsMorsel, size_t numRels)
+        : bfsMorsel{bfsMorsel}, numRels{numRels} {
+        path.resize(numRels + 1);
         resetState();
     }
 

src/processor/operator/recursive_extend/path_scanner.cpp

@@ -7,24 +7,27 @@ size_t FixedLengthPathScanner::scanPaths(common::table_id_t tableID,
     common::ValueVector* pathOffsetVector, common::ValueVector* pathDataVector,
     common::ValueVector* dstNodeIDVector, common::sel_t& offsetVectorPos,
     common::sel_t& dataVectorPos) {
-    if (depth >= bfsMorsel.frontiers.size()) {
+    if (numRels >= bfsMorsel.frontiers.size()) {
         // BFS terminate before current depth. No need to scan.
         return 0;
     }
     auto offsetVectorPosBeforeScanning = offsetVectorPos;
-    auto frontier = bfsMorsel.frontiers[depth].get();
+    auto lastFrontier = bfsMorsel.frontiers[numRels].get();
     while (offsetVectorPos < common::DEFAULT_VECTOR_CAPACITY) {
         if (!continueDfs()) {
-            if (lastFrontierCursor == frontier->nodeOffsets.size()) {
+            // All paths from current dst have been scanned. Fetch next dst.
+            if (lastFrontierCursor == lastFrontier->nodeOffsets.size()) {
+                // All nodes from last frontier have been scanned.
                 break;
             }
-            auto currentDstOffset = frontier->nodeOffsets[lastFrontierCursor++];
+            auto currentDstOffset = lastFrontier->nodeOffsets[lastFrontierCursor++];
             // Skip nodes that is not in semi mask.
             if (!bfsMorsel.isAllDstTarget() &&
                 !bfsMorsel.targetDstNodeOffsets.contains(currentDstOffset)) {
                 continue;
             }
-            initDfs(currentDstOffset, depth);
+            // Prepare dfs stack.
+            initDfs(currentDstOffset, numRels);
         }
         writePathToVector(tableID, pathOffsetVector, pathDataVector, dstNodeIDVector,
             offsetVectorPos, dataVectorPos);
@@ -37,16 +40,17 @@ bool FixedLengthPathScanner::continueDfs() {
     while (!nbrsStack.empty()) {
         auto& cursor = cursorStack.top();
         cursor++;
-        if (cursor < nbrsStack.top()->size()) {
+        if (cursor < nbrsStack.top()->size()) { // Found a new nbr
             auto offset = nbrsStack.top()->at(cursor);
             path[level] = offset;
-            if (level == 0) {
+            if (level == 0) { // Found a new nbr at level 0. Found a new path.
                 return true;
             }
+            // Push new stack.
             cursorStack.push(-1);
             nbrsStack.push(&bfsMorsel.frontiers[level]->bwdEdges.at(offset));
             level--;
-        } else {
+        } else { // Failed to find a nbr. Pop stack.
             cursorStack.pop();
             nbrsStack.pop();
             level++;