v41-0004-Row-pattern-recognition-patch-planner.patch
application/octet-stream
Filename: v41-0004-Row-pattern-recognition-patch-planner.patch
Type: application/octet-stream
Part: 3
Message:
Re: Row pattern recognition
Patch
Same data as JSON:
GET /api/v1/attachments/:id/patch
the parsed metadata as JSON — format, series position, per-file stats; never the diff bytes.
API reference →
Format: format-patch
Series: patch v41-0004
Subject: Row pattern recognition patch (planner).
| File | + | − |
|---|---|---|
| src/backend/optimizer/plan/createplan.c | 46 | 17 |
| src/backend/optimizer/plan/Makefile | 1 | 0 |
| src/backend/optimizer/plan/meson.build | 1 | 0 |
| src/backend/optimizer/plan/planner.c | 3 | 0 |
| src/backend/optimizer/plan/rpr.c | 1556 | 0 |
| src/backend/optimizer/plan/setrefs.c | 26 | 1 |
| src/backend/optimizer/prep/prepjointree.c | 9 | 0 |
| src/include/nodes/plannodes.h | 74 | 0 |
| src/include/optimizer/rpr.h | 50 | 0 |
From df615c387741de9554bdeeaafa74e0d9605f81fc Mon Sep 17 00:00:00 2001
From: Tatsuo Ishii <ishii@postgresql.org>
Date: Tue, 20 Jan 2026 19:00:01 +0900
Subject: [PATCH v41 4/8] Row pattern recognition patch (planner).
---
src/backend/optimizer/plan/Makefile | 1 +
src/backend/optimizer/plan/createplan.c | 63 +-
src/backend/optimizer/plan/meson.build | 1 +
src/backend/optimizer/plan/planner.c | 3 +
src/backend/optimizer/plan/rpr.c | 1556 +++++++++++++++++++++
src/backend/optimizer/plan/setrefs.c | 27 +-
src/backend/optimizer/prep/prepjointree.c | 9 +
src/include/nodes/plannodes.h | 74 +
src/include/optimizer/rpr.h | 50 +
9 files changed, 1766 insertions(+), 18 deletions(-)
create mode 100644 src/backend/optimizer/plan/rpr.c
create mode 100644 src/include/optimizer/rpr.h
diff --git a/src/backend/optimizer/plan/Makefile b/src/backend/optimizer/plan/Makefile
index 80ef162e484..7e0167789d8 100644
--- a/src/backend/optimizer/plan/Makefile
+++ b/src/backend/optimizer/plan/Makefile
@@ -19,6 +19,7 @@ OBJS = \
planagg.o \
planmain.o \
planner.o \
+ rpr.o \
setrefs.o \
subselect.o
diff --git a/src/backend/optimizer/plan/createplan.c b/src/backend/optimizer/plan/createplan.c
index eb4806b084a..9a4a0ca7ac7 100644
--- a/src/backend/optimizer/plan/createplan.c
+++ b/src/backend/optimizer/plan/createplan.c
@@ -35,6 +35,7 @@
#include "optimizer/prep.h"
#include "optimizer/restrictinfo.h"
#include "optimizer/subselect.h"
+#include "optimizer/rpr.h"
#include "optimizer/tlist.h"
#include "parser/parse_clause.h"
#include "parser/parsetree.h"
@@ -287,7 +288,10 @@ static Memoize *make_memoize(Plan *lefttree, Oid *hashoperators,
static WindowAgg *make_windowagg(List *tlist, WindowClause *wc,
int partNumCols, AttrNumber *partColIdx, Oid *partOperators, Oid *partCollations,
int ordNumCols, AttrNumber *ordColIdx, Oid *ordOperators, Oid *ordCollations,
- List *runCondition, List *qual, bool topWindow,
+ List *runCondition, RPSkipTo rpSkipTo,
+ RPRPattern *compiledPattern,
+ List *defineClause,
+ List *qual, bool topWindow,
Plan *lefttree);
static Group *make_group(List *tlist, List *qual, int numGroupCols,
AttrNumber *grpColIdx, Oid *grpOperators, Oid *grpCollations,
@@ -2528,21 +2532,37 @@ create_windowagg_plan(PlannerInfo *root, WindowAggPath *best_path)
ordNumCols++;
}
- /* And finally we can make the WindowAgg node */
- plan = make_windowagg(tlist,
- wc,
- partNumCols,
- partColIdx,
- partOperators,
- partCollations,
- ordNumCols,
- ordColIdx,
- ordOperators,
- ordCollations,
- best_path->runCondition,
- best_path->qual,
- best_path->topwindow,
- subplan);
+ /* Build defineVariableList from defineClause for pattern compilation */
+ {
+ List *defineVariableList = NIL;
+
+ foreach(lc, wc->defineClause)
+ {
+ TargetEntry *te = (TargetEntry *) lfirst(lc);
+
+ defineVariableList = lappend(defineVariableList,
+ makeString(pstrdup(te->resname)));
+ }
+
+ /* And finally we can make the WindowAgg node */
+ plan = make_windowagg(tlist,
+ wc,
+ partNumCols,
+ partColIdx,
+ partOperators,
+ partCollations,
+ ordNumCols,
+ ordColIdx,
+ ordOperators,
+ ordCollations,
+ best_path->runCondition,
+ wc->rpSkipTo,
+ buildRPRPattern(wc->rpPattern, defineVariableList, wc->rpSkipTo, wc->frameOptions),
+ wc->defineClause,
+ best_path->qual,
+ best_path->topwindow,
+ subplan);
+ }
copy_generic_path_info(&plan->plan, (Path *) best_path);
@@ -6608,7 +6628,10 @@ static WindowAgg *
make_windowagg(List *tlist, WindowClause *wc,
int partNumCols, AttrNumber *partColIdx, Oid *partOperators, Oid *partCollations,
int ordNumCols, AttrNumber *ordColIdx, Oid *ordOperators, Oid *ordCollations,
- List *runCondition, List *qual, bool topWindow, Plan *lefttree)
+ List *runCondition, RPSkipTo rpSkipTo,
+ RPRPattern *compiledPattern,
+ List *defineClause,
+ List *qual, bool topWindow, Plan *lefttree)
{
WindowAgg *node = makeNode(WindowAgg);
Plan *plan = &node->plan;
@@ -6635,6 +6658,12 @@ make_windowagg(List *tlist, WindowClause *wc,
node->inRangeAsc = wc->inRangeAsc;
node->inRangeNullsFirst = wc->inRangeNullsFirst;
node->topWindow = topWindow;
+ node->rpSkipTo = rpSkipTo;
+
+ /* Store compiled pattern for NFA execution */
+ node->rpPattern = compiledPattern;
+
+ node->defineClause = defineClause;
plan->targetlist = tlist;
plan->lefttree = lefttree;
diff --git a/src/backend/optimizer/plan/meson.build b/src/backend/optimizer/plan/meson.build
index c565b2adbcc..5b2381cd774 100644
--- a/src/backend/optimizer/plan/meson.build
+++ b/src/backend/optimizer/plan/meson.build
@@ -7,6 +7,7 @@ backend_sources += files(
'planagg.c',
'planmain.c',
'planner.c',
+ 'rpr.c',
'setrefs.c',
'subselect.c',
)
diff --git a/src/backend/optimizer/plan/planner.c b/src/backend/optimizer/plan/planner.c
index 7a6b8b749f2..2a93feee25b 100644
--- a/src/backend/optimizer/plan/planner.c
+++ b/src/backend/optimizer/plan/planner.c
@@ -964,6 +964,9 @@ subquery_planner(PlannerGlobal *glob, Query *parse, char *plan_name,
EXPRKIND_LIMIT);
wc->endOffset = preprocess_expression(root, wc->endOffset,
EXPRKIND_LIMIT);
+ wc->defineClause = (List *) preprocess_expression(root,
+ (Node *) wc->defineClause,
+ EXPRKIND_TARGET);
}
parse->limitOffset = preprocess_expression(root, parse->limitOffset,
diff --git a/src/backend/optimizer/plan/rpr.c b/src/backend/optimizer/plan/rpr.c
new file mode 100644
index 00000000000..add757a1b98
--- /dev/null
+++ b/src/backend/optimizer/plan/rpr.c
@@ -0,0 +1,1556 @@
+/*-------------------------------------------------------------------------
+ *
+ * rpr.c
+ * Row Pattern Recognition pattern compilation for planner
+ *
+ * This file contains functions for optimizing RPR pattern AST and
+ * compiling it to bytecode for execution by WindowAgg.
+ *
+ * Key components:
+ * 1. Pattern Optimization: Simplifies patterns before compilation
+ * (e.g., flatten nested SEQ/ALT, merge consecutive vars)
+ * 2. Pattern Compilation: Converts AST to flat element array for NFA
+ * 3. Absorption Analysis: Computes flags for O(n^2)->O(n) optimization
+ *
+ * Context Absorption Optimization:
+ * When a pattern starts with an unbounded element (e.g., A+ or (A B)+),
+ * newer contexts cannot produce longer matches than older contexts.
+ * By absorbing (eliminating) redundant newer contexts, we reduce
+ * complexity from O(n^2) to O(n) for patterns like A+ B.
+ *
+ * The absorption analysis uses two element flags:
+ * - RPR_ELEM_ABSORBABLE: marks WHERE to compare (judgment point)
+ * - RPR_ELEM_ABSORBABLE_BRANCH: marks the absorbable region
+ *
+ * See computeAbsorbability() and the detailed comments before
+ * isUnboundedStart() for the full design explanation.
+ *
+ * Portions Copyright (c) 1996-2026, PostgreSQL Global Development Group
+ * Portions Copyright (c) 1994, Regents of the University of California
+ *
+ *
+ * IDENTIFICATION
+ * src/backend/optimizer/plan/rpr.c
+ *
+ *-------------------------------------------------------------------------
+ */
+
+#include "postgres.h"
+
+#include "nodes/makefuncs.h"
+#include "optimizer/rpr.h"
+
+/* Forward declarations - pattern comparison */
+static bool rprPatternEqual(RPRPatternNode *a, RPRPatternNode *b);
+static bool rprPatternChildrenEqual(List *a, List *b);
+
+/* Forward declarations - pattern optimization (shared) */
+static RPRPatternNode *tryUnwrapSingleChild(RPRPatternNode *pattern);
+
+/* Forward declarations - SEQ optimization */
+static List *flattenSeqChildren(List *children);
+static List *mergeConsecutiveVars(List *children);
+static List *mergeConsecutiveGroups(List *children);
+static List *mergeGroupPrefixSuffix(List *children);
+static RPRPatternNode *optimizeSeqPattern(RPRPatternNode *pattern);
+
+/* Forward declarations - ALT optimization */
+static List *flattenAltChildren(List *children);
+static List *removeDuplicateAlternatives(List *children);
+static RPRPatternNode *optimizeAltPattern(RPRPatternNode *pattern);
+
+/* Forward declarations - GROUP optimization */
+static RPRPatternNode *tryMultiplyQuantifiers(RPRPatternNode *pattern);
+static RPRPatternNode *tryUnwrapGroup(RPRPatternNode *pattern);
+static RPRPatternNode *optimizeGroupPattern(RPRPatternNode *pattern);
+
+/* Forward declarations - optimization dispatcher */
+static RPRPatternNode *optimizeRPRPattern(RPRPatternNode *pattern);
+
+/* Forward declarations - pattern compilation */
+static int collectDefineVariables(List *defineVariableList, char **varNames);
+static void scanRPRPatternRecursive(RPRPatternNode *node, char **varNames,
+ int *numVars, int *numElements,
+ RPRDepth depth, RPRDepth *maxDepth);
+static void scanRPRPattern(RPRPatternNode *node, char **varNames, int *numVars,
+ int *numElements, RPRDepth *maxDepth);
+static RPRPattern *allocateRPRPattern(int numVars, int numElements,
+ RPRDepth maxDepth, char **varNamesStack);
+static RPRVarId getVarIdFromPattern(RPRPattern *pat, const char *varName);
+static void fillRPRPatternVar(RPRPatternNode *node, RPRPattern *pat,
+ int *idx, RPRDepth depth);
+static void fillRPRPatternGroup(RPRPatternNode *node, RPRPattern *pat,
+ int *idx, RPRDepth depth);
+static void fillRPRPatternAlt(RPRPatternNode *node, RPRPattern *pat,
+ int *idx, RPRDepth depth);
+static void fillRPRPattern(RPRPatternNode *node, RPRPattern *pat,
+ int *idx, RPRDepth depth);
+static void finalizeRPRPattern(RPRPattern *result);
+
+/* Forward declarations - context absorption */
+static bool isUnboundedStart(RPRPattern *pattern, RPRElemIdx idx);
+static void computeAbsorbability(RPRPattern *pattern);
+
+/*
+ * rprPatternEqual
+ * Compare two RPRPatternNode trees for equality.
+ *
+ * Returns true if the trees are structurally identical.
+ */
+static bool
+rprPatternEqual(RPRPatternNode *a, RPRPatternNode *b)
+{
+ if (a == NULL && b == NULL)
+ return true;
+ if (a == NULL || b == NULL)
+ return false;
+
+ /* Must have same node type and quantifiers */
+ if (a->nodeType != b->nodeType)
+ return false;
+ if (a->min != b->min || a->max != b->max)
+ return false;
+ if (a->reluctant != b->reluctant)
+ return false;
+
+ switch (a->nodeType)
+ {
+ case RPR_PATTERN_VAR:
+ return strcmp(a->varName, b->varName) == 0;
+
+ case RPR_PATTERN_SEQ:
+ case RPR_PATTERN_ALT:
+ case RPR_PATTERN_GROUP:
+ return rprPatternChildrenEqual(a->children, b->children);
+ }
+
+ return false; /* keep compiler quiet */
+}
+
+/*
+ * rprPatternChildrenEqual
+ * Compare children lists of two pattern nodes for equality.
+ *
+ * Returns true if the children lists are structurally identical.
+ */
+static bool
+rprPatternChildrenEqual(List *a, List *b)
+{
+ ListCell *lca,
+ *lcb;
+
+ if (list_length(a) != list_length(b))
+ return false;
+
+ forboth(lca, a, lcb, b)
+ {
+ if (!rprPatternEqual((RPRPatternNode *) lfirst(lca),
+ (RPRPatternNode *) lfirst(lcb)))
+ return false;
+ }
+
+ return true;
+}
+
+/*
+ * tryUnwrapSingleChild
+ * Try to unwrap pattern node with single child.
+ *
+ * Examples (internal node representation):
+ * SEQ[A] -> A (single-element sequence becomes the element)
+ * ALT[A] -> A (single-alternative becomes the alternative)
+ *
+ * If pattern has exactly one child, return the child directly.
+ * Otherwise returns the pattern unchanged.
+ * Used by both SEQ and ALT optimization.
+ */
+static RPRPatternNode *
+tryUnwrapSingleChild(RPRPatternNode *pattern)
+{
+ if (list_length(pattern->children) == 1)
+ return (RPRPatternNode *) linitial(pattern->children);
+
+ return pattern;
+}
+
+/*
+ * flattenSeqChildren
+ * Recursively optimize children and flatten nested SEQ/GROUP{1,1}.
+ *
+ * Examples:
+ * A ((B) (C)) -> A B C (flatten nested GROUP{1,1})
+ * A (B) -> A B (flatten GROUP{1,1})
+ *
+ * Returns a new list with optimized children, with nested SEQ and
+ * GROUP{1,1} children flattened into the parent list.
+ */
+static List *
+flattenSeqChildren(List *children)
+{
+ ListCell *lc;
+ List *newChildren = NIL;
+
+ foreach(lc, children)
+ {
+ RPRPatternNode *child = (RPRPatternNode *) lfirst(lc);
+ RPRPatternNode *opt = optimizeRPRPattern(child);
+
+ /* Flatten GROUP{1,1} or nested SEQ */
+ if ((opt->nodeType == RPR_PATTERN_GROUP &&
+ opt->min == 1 && opt->max == 1 && !opt->reluctant) ||
+ opt->nodeType == RPR_PATTERN_SEQ)
+ {
+ newChildren = list_concat(newChildren,
+ list_copy(opt->children));
+ }
+ else
+ {
+ newChildren = lappend(newChildren, opt);
+ }
+ }
+
+ return newChildren;
+}
+
+/*
+ * mergeConsecutiveVars
+ * Merge consecutive identical VAR nodes.
+ *
+ * A{m1,M1} A{m2,M2} -> A{m1+m2, M1+M2} where INF + x = INF.
+ * Only merges non-reluctant VAR nodes with the same variable name.
+ */
+static List *
+mergeConsecutiveVars(List *children)
+{
+ ListCell *lc;
+ List *mergedChildren = NIL;
+ RPRPatternNode *prev = NULL;
+
+ foreach(lc, children)
+ {
+ RPRPatternNode *child = (RPRPatternNode *) lfirst(lc);
+
+ if (child->nodeType == RPR_PATTERN_VAR && !child->reluctant)
+ {
+ if (prev != NULL &&
+ prev->nodeType == RPR_PATTERN_VAR &&
+ strcmp(prev->varName, child->varName) == 0 &&
+ !prev->reluctant &&
+ prev->min <= RPR_QUANTITY_INF - child->min &&
+ (prev->max == RPR_QUANTITY_INF ||
+ child->max == RPR_QUANTITY_INF ||
+ prev->max <= RPR_QUANTITY_INF - child->max))
+ {
+ /*
+ * Merge: accumulate min/max into prev.
+ */
+ prev->min += child->min;
+
+ if (prev->max == RPR_QUANTITY_INF ||
+ child->max == RPR_QUANTITY_INF)
+ prev->max = RPR_QUANTITY_INF;
+ else
+ prev->max += child->max;
+ }
+ else
+ {
+ /* Flush previous and start new */
+ if (prev != NULL)
+ mergedChildren = lappend(mergedChildren, prev);
+ prev = child;
+ }
+ }
+ else
+ {
+ /* Non-mergeable - flush previous */
+ if (prev != NULL)
+ mergedChildren = lappend(mergedChildren, prev);
+ mergedChildren = lappend(mergedChildren, child);
+ prev = NULL;
+ }
+ }
+
+ /* Flush remaining */
+ if (prev != NULL)
+ mergedChildren = lappend(mergedChildren, prev);
+
+ return mergedChildren;
+}
+
+/*
+ * mergeConsecutiveGroups
+ * Merge consecutive identical GROUP nodes.
+ *
+ * Example:
+ * (A B)+ (A B)+ -> (A B){2,}
+ *
+ * Only merges non-reluctant GROUP nodes with identical children.
+ */
+static List *
+mergeConsecutiveGroups(List *children)
+{
+ ListCell *lc;
+ List *mergedChildren = NIL;
+ RPRPatternNode *prev = NULL;
+
+ foreach(lc, children)
+ {
+ RPRPatternNode *child = (RPRPatternNode *) lfirst(lc);
+
+ if (child->nodeType == RPR_PATTERN_GROUP && !child->reluctant)
+ {
+ if (prev != NULL &&
+ prev->nodeType == RPR_PATTERN_GROUP &&
+ !prev->reluctant &&
+ rprPatternChildrenEqual(prev->children, child->children) &&
+ prev->min <= RPR_QUANTITY_INF - child->min &&
+ (prev->max == RPR_QUANTITY_INF ||
+ child->max == RPR_QUANTITY_INF ||
+ prev->max <= RPR_QUANTITY_INF - child->max))
+ {
+ /*
+ * Merge: accumulate min/max into prev.
+ */
+ prev->min += child->min;
+
+ if (prev->max == RPR_QUANTITY_INF ||
+ child->max == RPR_QUANTITY_INF)
+ prev->max = RPR_QUANTITY_INF;
+ else
+ prev->max += child->max;
+ }
+ else
+ {
+ /* Flush previous and start new */
+ if (prev != NULL)
+ mergedChildren = lappend(mergedChildren, prev);
+ prev = child;
+ }
+ }
+ else
+ {
+ /* Non-mergeable - flush previous */
+ if (prev != NULL)
+ mergedChildren = lappend(mergedChildren, prev);
+ mergedChildren = lappend(mergedChildren, child);
+ prev = NULL;
+ }
+ }
+
+ /* Flush remaining */
+ if (prev != NULL)
+ mergedChildren = lappend(mergedChildren, prev);
+
+ return mergedChildren;
+}
+
+/*
+ * mergeGroupPrefixSuffix
+ * Merge sequence prefix/suffix into GROUP with matching children.
+ *
+ * Examples:
+ * A B (A B)+ -> (A B){2,}
+ * (A B)+ A B -> (A B){2,}
+ * A B (A B)+ A B -> (A B){3,}
+ */
+static List *
+mergeGroupPrefixSuffix(List *children)
+{
+ List *result = NIL;
+ int numChildren = list_length(children);
+ int i;
+ int skipUntil = -1; /* skip suffix elements already absorbed */
+
+ for (i = 0; i < numChildren; i++)
+ {
+ RPRPatternNode *child = (RPRPatternNode *) list_nth(children, i);
+
+ /* Skip elements that were absorbed as suffix */
+ if (i < skipUntil)
+ continue;
+
+ /*
+ * If this is a GROUP, see if preceding/following elements match its
+ * children. GROUP's content may be wrapped in a SEQ - unwrap for
+ * comparison.
+ */
+ if (child->nodeType == RPR_PATTERN_GROUP && !child->reluctant)
+ {
+ List *groupContent = child->children;
+ int groupChildCount;
+ int prefixLen = list_length(result);
+
+ /*
+ * If GROUP has single SEQ child, compare with SEQ's children.
+ * e.g., (A B)+ internally contains sequence A B; compare against
+ * that.
+ */
+ if (list_length(groupContent) == 1)
+ {
+ RPRPatternNode *inner = (RPRPatternNode *) linitial(groupContent);
+
+ if (inner->nodeType == RPR_PATTERN_SEQ)
+ groupContent = inner->children;
+ }
+
+ groupChildCount = list_length(groupContent);
+
+ /*
+ * PREFIX MERGE: Check if preceding elements match. Keep absorbing
+ * as long as we have matching prefixes.
+ */
+ while (prefixLen >= groupChildCount && groupChildCount > 0)
+ {
+ List *prefixElements = NIL;
+ int j;
+
+ /* Extract last groupChildCount elements from prefix */
+ for (j = prefixLen - groupChildCount; j < prefixLen; j++)
+ {
+ prefixElements = lappend(prefixElements,
+ list_nth(result, j));
+ }
+
+ /* Compare with GROUP's (possibly unwrapped) children */
+ if (rprPatternChildrenEqual(prefixElements, groupContent) &&
+ child->min < RPR_QUANTITY_INF)
+ {
+ /*
+ * Match! Merge by incrementing GROUP's min. Remove the
+ * prefix elements from output.
+ */
+ child->min += 1;
+
+ /* Rebuild result without matched prefix */
+ {
+ List *trimmed = NIL;
+
+ for (j = 0; j < prefixLen - groupChildCount; j++)
+ {
+ trimmed = lappend(trimmed,
+ list_nth(result, j));
+ }
+ result = trimmed;
+ prefixLen = list_length(result);
+ }
+ }
+ else
+ {
+ list_free(prefixElements);
+ break;
+ }
+
+ list_free(prefixElements);
+ }
+
+ /*
+ * SUFFIX MERGE: Check if following elements match. Keep absorbing
+ * as long as we have matching suffixes.
+ */
+ while (i + groupChildCount < numChildren && groupChildCount > 0)
+ {
+ List *suffixElements = NIL;
+ int j;
+ int suffixStart = i + 1;
+
+ /* Adjust for already absorbed elements */
+ if (skipUntil > suffixStart)
+ break;
+
+ /* Extract next groupChildCount elements as suffix */
+ for (j = 0; j < groupChildCount; j++)
+ {
+ int idx = suffixStart + j;
+
+ if (idx >= numChildren)
+ break;
+ suffixElements = lappend(suffixElements,
+ list_nth(children, idx));
+ }
+
+ /* Compare with GROUP's children */
+ if (list_length(suffixElements) == groupChildCount &&
+ rprPatternChildrenEqual(suffixElements, groupContent) &&
+ child->min < RPR_QUANTITY_INF)
+ {
+ /*
+ * Match! Absorb suffix by incrementing min and skipping.
+ */
+ child->min += 1;
+ skipUntil = suffixStart + groupChildCount;
+
+ /*
+ * Update i to continue suffix check after absorbed
+ * elements
+ */
+ i = skipUntil - 1;
+ }
+ else
+ {
+ list_free(suffixElements);
+ break;
+ }
+
+ list_free(suffixElements);
+ }
+ }
+
+ result = lappend(result, child);
+ }
+
+ return result;
+}
+
+/*
+ * optimizeSeqPattern
+ * Optimize SEQ pattern node.
+ *
+ * Optimizations:
+ * 1. Flatten nested SEQ and GROUP{1,1}
+ * 2. Merge consecutive identical VAR nodes
+ * 3. Merge consecutive identical GROUP nodes
+ * 4. Merge prefix/suffix into GROUP with matching children
+ * 5. Unwrap single-item SEQ
+ */
+static RPRPatternNode *
+optimizeSeqPattern(RPRPatternNode *pattern)
+{
+ /* Recursively optimize children and flatten nested SEQ/GROUP{1,1} */
+ pattern->children = flattenSeqChildren(pattern->children);
+
+ /* Merge consecutive identical VAR nodes */
+ pattern->children = mergeConsecutiveVars(pattern->children);
+
+ /* Merge consecutive identical GROUP nodes */
+ pattern->children = mergeConsecutiveGroups(pattern->children);
+
+ /* Merge prefix/suffix into GROUP with matching children */
+ pattern->children = mergeGroupPrefixSuffix(pattern->children);
+
+ /* Unwrap single-item SEQ */
+ return tryUnwrapSingleChild(pattern);
+}
+
+/*
+ * flattenAltChildren
+ * Recursively optimize children and flatten nested ALT nodes.
+ *
+ * Example:
+ * (A | (B | C)) -> (A | B | C)
+ *
+ * Returns a new list with optimized children, with nested ALT children
+ * flattened into the parent list.
+ */
+static List *
+flattenAltChildren(List *children)
+{
+ ListCell *lc;
+ List *newChildren = NIL;
+
+ foreach(lc, children)
+ {
+ RPRPatternNode *child = (RPRPatternNode *) lfirst(lc);
+ RPRPatternNode *opt = optimizeRPRPattern(child);
+
+ if (opt->nodeType == RPR_PATTERN_ALT)
+ newChildren = list_concat(newChildren, list_copy(opt->children));
+ else
+ newChildren = lappend(newChildren, opt);
+ }
+
+ return newChildren;
+}
+
+/*
+ * removeDuplicateAlternatives
+ * Remove duplicate alternatives from a list.
+ *
+ * Returns a new list with only unique children (first occurrence kept).
+ */
+static List *
+removeDuplicateAlternatives(List *children)
+{
+ ListCell *lc;
+ ListCell *lc2;
+ List *uniqueChildren = NIL;
+
+ foreach(lc, children)
+ {
+ RPRPatternNode *child = (RPRPatternNode *) lfirst(lc);
+ bool isDuplicate = false;
+
+ foreach(lc2, uniqueChildren)
+ {
+ if (rprPatternEqual((RPRPatternNode *) lfirst(lc2), child))
+ {
+ isDuplicate = true;
+ break;
+ }
+ }
+
+ if (!isDuplicate)
+ uniqueChildren = lappend(uniqueChildren, child);
+ }
+
+ return uniqueChildren;
+}
+
+/*
+ * optimizeAltPattern
+ * Optimize ALT pattern node.
+ *
+ * Optimizations:
+ * 1. Flatten nested ALT
+ * 2. Remove duplicate alternatives
+ * 3. Unwrap single-item ALT
+ */
+static RPRPatternNode *
+optimizeAltPattern(RPRPatternNode *pattern)
+{
+ /* Recursively optimize children and flatten nested ALT */
+ pattern->children = flattenAltChildren(pattern->children);
+
+ /* Remove duplicate alternatives */
+ pattern->children = removeDuplicateAlternatives(pattern->children);
+
+ /* Unwrap single-item ALT */
+ return tryUnwrapSingleChild(pattern);
+}
+
+/*
+ * tryMultiplyQuantifiers
+ * Try to multiply quantifiers: (A{m}){n} -> A{m*n}
+ *
+ * Returns the child node with multiplied quantifiers if successful,
+ * otherwise returns the original pattern unchanged.
+ */
+static RPRPatternNode *
+tryMultiplyQuantifiers(RPRPatternNode *pattern)
+{
+ RPRPatternNode *child;
+ int64 new_min_64;
+ int new_max;
+
+ if (list_length(pattern->children) != 1 || pattern->reluctant)
+ return pattern;
+
+ child = (RPRPatternNode *) linitial(pattern->children);
+
+ if (child->nodeType != RPR_PATTERN_VAR || child->reluctant)
+ return pattern;
+
+ /* Can't multiply if child has infinite max */
+ if (child->max == RPR_QUANTITY_INF)
+ return pattern;
+
+ /* Calculate new min (must not overflow) */
+ new_min_64 = (int64) pattern->min * child->min;
+ if (new_min_64 >= RPR_QUANTITY_INF)
+ return pattern;
+
+ /* Calculate new max */
+ if (pattern->max == RPR_QUANTITY_INF)
+ {
+ new_max = RPR_QUANTITY_INF;
+ }
+ else
+ {
+ int64 new_max_64 = (int64) pattern->max * child->max;
+
+ if (new_max_64 >= RPR_QUANTITY_INF)
+ return pattern;
+ new_max = (int) new_max_64;
+ }
+
+ child->min = (int) new_min_64;
+ child->max = new_max;
+ return child;
+}
+
+/*
+ * tryUnwrapGroup
+ * Try to unwrap GROUP{1,1} node.
+ *
+ * Examples:
+ * (A){1,1} -> A
+ * (A B){1,1} -> A B (converts to SEQ internally)
+ *
+ * If GROUP has min=1, max=1, and is not reluctant:
+ * - Single child: return the child directly
+ * - Multiple children: convert GROUP to SEQ
+ * Otherwise returns the pattern unchanged.
+ */
+static RPRPatternNode *
+tryUnwrapGroup(RPRPatternNode *pattern)
+{
+ if (pattern->min != 1 || pattern->max != 1 || pattern->reluctant)
+ return pattern;
+
+ if (list_length(pattern->children) == 1)
+ return (RPRPatternNode *) linitial(pattern->children);
+
+ /* Multiple children: convert to SEQ */
+ pattern->nodeType = RPR_PATTERN_SEQ;
+ return pattern;
+}
+
+/*
+ * optimizeGroupPattern
+ * Optimize GROUP pattern node.
+ *
+ * Optimizations:
+ * 1. Quantifier multiplication: (A{m}){n} -> A{m*n}
+ * 2. Unwrap GROUP{1,1}
+ */
+static RPRPatternNode *
+optimizeGroupPattern(RPRPatternNode *pattern)
+{
+ ListCell *lc;
+ List *newChildren;
+ RPRPatternNode *result;
+
+ /* Recursively optimize children */
+ newChildren = NIL;
+ foreach(lc, pattern->children)
+ {
+ RPRPatternNode *child = (RPRPatternNode *) lfirst(lc);
+
+ newChildren = lappend(newChildren, optimizeRPRPattern(child));
+ }
+ pattern->children = newChildren;
+
+ /* Try quantifier multiplication */
+ result = tryMultiplyQuantifiers(pattern);
+ if (result != pattern)
+ return result;
+
+ /* Try unwrapping GROUP{1,1} */
+ result = tryUnwrapGroup(pattern);
+
+ /* If converted to SEQ (not just unwrapped child), optimize as SEQ */
+ if (result == pattern && result->nodeType == RPR_PATTERN_SEQ)
+ return optimizeSeqPattern(result);
+
+ return result;
+}
+
+/*
+ * optimizeRPRPattern
+ * Optimize RPRPatternNode tree (dispatcher).
+ *
+ * Dispatches to type-specific optimization functions.
+ * Returns the optimized pattern (may be a different node).
+ */
+static RPRPatternNode *
+optimizeRPRPattern(RPRPatternNode *pattern)
+{
+ if (pattern == NULL)
+ return NULL;
+
+ switch (pattern->nodeType)
+ {
+ case RPR_PATTERN_VAR:
+ return pattern;
+ case RPR_PATTERN_SEQ:
+ return optimizeSeqPattern(pattern);
+ case RPR_PATTERN_ALT:
+ return optimizeAltPattern(pattern);
+ case RPR_PATTERN_GROUP:
+ return optimizeGroupPattern(pattern);
+ }
+
+ return pattern; /* keep compiler quiet */
+}
+
+/*
+ * collectDefineVariables
+ * Collect variable names from DEFINE clause.
+ *
+ * Populates varNames array with variable names in DEFINE order.
+ * This ensures varId == defineIdx, eliminating runtime mapping.
+ * Returns the number of variables collected.
+ */
+static int
+collectDefineVariables(List *defineVariableList, char **varNames)
+{
+ ListCell *lc;
+ int numVars = 0;
+
+ foreach(lc, defineVariableList)
+ {
+ if (numVars >= RPR_VARID_MAX)
+ ereport(ERROR,
+ (errcode(ERRCODE_PROGRAM_LIMIT_EXCEEDED),
+ errmsg("too many pattern variables"),
+ errdetail("Maximum is %d.", RPR_VARID_MAX)));
+
+ varNames[numVars++] = strVal(lfirst(lc));
+ }
+
+ return numVars;
+}
+
+/*
+ * scanRPRPatternRecursive
+ * Recursively scan pattern AST (pass 1 internal).
+ *
+ * Collects unique variable names and counts elements while tracking depth.
+ * Variables from DEFINE clause are already in varNames; this adds any
+ * additional variables found in the pattern.
+ */
+static void
+scanRPRPatternRecursive(RPRPatternNode *node, char **varNames, int *numVars,
+ int *numElements, RPRDepth depth, RPRDepth *maxDepth)
+{
+ ListCell *lc;
+ int i;
+
+ if (node == NULL)
+ return;
+
+ /* Track maximum depth */
+ if (depth > *maxDepth)
+ *maxDepth = depth;
+
+ switch (node->nodeType)
+ {
+ case RPR_PATTERN_VAR:
+ /* Count element */
+ (*numElements)++;
+
+ /* Collect variable name if not already present */
+ for (i = 0; i < *numVars; i++)
+ {
+ if (strcmp(varNames[i], node->varName) == 0)
+ return; /* Already have this variable */
+ }
+
+ /* New variable */
+ if (*numVars >= RPR_VARID_MAX)
+ ereport(ERROR,
+ (errcode(ERRCODE_PROGRAM_LIMIT_EXCEEDED),
+ errmsg("too many pattern variables"),
+ errdetail("Maximum is %d.", RPR_VARID_MAX)));
+
+ varNames[*numVars] = node->varName;
+ (*numVars)++;
+ break;
+
+ case RPR_PATTERN_SEQ:
+ /* Sequence: just recurse into children */
+ foreach(lc, node->children)
+ {
+ scanRPRPatternRecursive((RPRPatternNode *) lfirst(lc), varNames,
+ numVars, numElements, depth, maxDepth);
+ }
+ break;
+
+ case RPR_PATTERN_GROUP:
+ /* Recurse into children at increased depth */
+ foreach(lc, node->children)
+ {
+ scanRPRPatternRecursive((RPRPatternNode *) lfirst(lc), varNames,
+ numVars, numElements, depth + 1, maxDepth);
+ }
+
+ /* Add END element if group has non-trivial quantifier */
+ if (node->min != 1 || node->max != 1)
+ (*numElements)++;
+ break;
+
+ case RPR_PATTERN_ALT:
+ /* Count ALT start element */
+ (*numElements)++;
+
+ /* Recurse into children at increased depth */
+ foreach(lc, node->children)
+ {
+ scanRPRPatternRecursive((RPRPatternNode *) lfirst(lc), varNames,
+ numVars, numElements, depth + 1, maxDepth);
+ }
+ break;
+ }
+}
+
+/*
+ * scanRPRPattern
+ * Scan pattern AST (pass 1 entry point).
+ *
+ * Collects unique variable names (appending to those from DEFINE clause),
+ * counts total elements (including FIN marker), and tracks maximum depth.
+ * Reports error if element count exceeds RPR_ELEMIDX_MAX.
+ */
+static void
+scanRPRPattern(RPRPatternNode *node, char **varNames, int *numVars,
+ int *numElements, RPRDepth *maxDepth)
+{
+ *numElements = 0;
+ *maxDepth = 0;
+
+ scanRPRPatternRecursive(node, varNames, numVars, numElements, 0, maxDepth);
+
+ (*numElements)++; /* +1 for FIN marker */
+
+ if (*numElements > RPR_ELEMIDX_MAX)
+ ereport(ERROR,
+ (errcode(ERRCODE_PROGRAM_LIMIT_EXCEEDED),
+ errmsg("pattern too complex"),
+ errdetail("Pattern has %d elements, maximum is %d.",
+ *numElements, RPR_ELEMIDX_MAX)));
+}
+
+/*
+ * allocateRPRPattern
+ * Allocate and initialize RPRPattern structure.
+ *
+ * Creates the pattern structure, copies variable names, and allocates
+ * the elements array. The elements array is zero-initialized.
+ */
+static RPRPattern *
+allocateRPRPattern(int numVars, int numElements, RPRDepth maxDepth,
+ char **varNamesStack)
+{
+ RPRPattern *result;
+ int i;
+
+ result = makeNode(RPRPattern);
+ result->numVars = numVars;
+ result->maxDepth = maxDepth + 1; /* +1: depth is 0-based */
+ result->numElements = numElements;
+
+ /* Copy varNames */
+ if (numVars > 0)
+ {
+ result->varNames = palloc(numVars * sizeof(char *));
+ for (i = 0; i < numVars; i++)
+ result->varNames[i] = pstrdup(varNamesStack[i]);
+ }
+ else
+ {
+ result->varNames = NULL;
+ }
+
+ /* Allocate elements array (zero-init for reserved fields) */
+ result->elements = palloc0(numElements * sizeof(RPRPatternElement));
+
+ return result;
+}
+
+/*
+ * getVarIdFromPattern
+ * Get variable ID for a variable name from RPRPattern.
+ *
+ * Returns the index of the variable in the varNames array.
+ */
+static RPRVarId
+getVarIdFromPattern(RPRPattern *pat, const char *varName)
+{
+ for (int i = 0; i < pat->numVars; i++)
+ {
+ if (strcmp(pat->varNames[i], varName) == 0)
+ return (RPRVarId) i;
+ }
+
+ /* Should not happen - variable should already be collected */
+ elog(ERROR, "pattern variable \"%s\" not found", varName);
+ pg_unreachable();
+}
+
+/*
+ * fillRPRPatternVar
+ * Fill a VAR pattern element.
+ */
+static void
+fillRPRPatternVar(RPRPatternNode *node, RPRPattern *pat, int *idx, RPRDepth depth)
+{
+ RPRPatternElement *elem = &pat->elements[*idx];
+
+ memset(elem, 0, sizeof(RPRPatternElement));
+ elem->varId = getVarIdFromPattern(pat, node->varName);
+ elem->depth = depth;
+ elem->min = node->min;
+ elem->max = (node->max == INT_MAX) ? RPR_QUANTITY_INF : node->max;
+ elem->next = RPR_ELEMIDX_INVALID;
+ elem->jump = RPR_ELEMIDX_INVALID;
+ if (node->reluctant)
+ elem->flags |= RPR_ELEM_RELUCTANT;
+ (*idx)++;
+}
+
+/*
+ * fillRPRPatternGroup
+ * Fill a GROUP pattern and its children.
+ *
+ * Creates elements for group content at increased depth, plus an END marker
+ * if the group has a non-trivial quantifier.
+ */
+static void
+fillRPRPatternGroup(RPRPatternNode *node, RPRPattern *pat, int *idx, RPRDepth depth)
+{
+ ListCell *lc;
+ int groupStartIdx = *idx;
+ bool altOnlyChild;
+ RPRDepth childDepth;
+
+ /*
+ * Fill group content at increased depth. Exception: if the only child is
+ * ALT, don't increase depth since GROUP's parens already provide visual
+ * grouping. This avoids output like "((a | b))+" instead of "(a | b)+".
+ */
+ altOnlyChild = (list_length(node->children) == 1 &&
+ ((RPRPatternNode *) linitial(node->children))->nodeType == RPR_PATTERN_ALT);
+ childDepth = altOnlyChild ? depth : depth + 1;
+
+ foreach(lc, node->children)
+ {
+ fillRPRPattern((RPRPatternNode *) lfirst(lc), pat, idx, childDepth);
+ }
+
+ /* Add group end marker if group has non-trivial quantifier */
+ if (node->min != 1 || node->max != 1)
+ {
+ RPRPatternElement *elem = &pat->elements[*idx];
+
+ memset(elem, 0, sizeof(RPRPatternElement));
+ elem->varId = RPR_VARID_END;
+ elem->depth = depth;
+ elem->min = node->min;
+ elem->max = (node->max == INT_MAX) ? RPR_QUANTITY_INF : node->max;
+ elem->next = RPR_ELEMIDX_INVALID;
+ elem->jump = groupStartIdx;
+ if (node->reluctant)
+ elem->flags |= RPR_ELEM_RELUCTANT;
+ (*idx)++;
+ }
+}
+
+/*
+ * fillRPRPatternAlt
+ * Fill an ALT pattern and its alternatives.
+ *
+ * Creates ALT_START marker, fills each alternative at increased depth,
+ * and sets up jump pointers between alternatives.
+ */
+static void
+fillRPRPatternAlt(RPRPatternNode *node, RPRPattern *pat, int *idx, RPRDepth depth)
+{
+ ListCell *lc;
+ RPRPatternElement *elem;
+ int altStartIdx = *idx;
+ List *altBranchStarts = NIL;
+ List *altEndPositions = NIL;
+
+ /* Add alternation start marker */
+ elem = &pat->elements[*idx];
+ memset(elem, 0, sizeof(RPRPatternElement));
+ elem->varId = RPR_VARID_ALT;
+ elem->depth = depth;
+ elem->min = 1;
+ elem->max = 1;
+ elem->next = RPR_ELEMIDX_INVALID;
+ elem->jump = RPR_ELEMIDX_INVALID;
+ (*idx)++;
+
+ /* Fill each alternative */
+ foreach(lc, node->children)
+ {
+ RPRPatternNode *alt = (RPRPatternNode *) lfirst(lc);
+ int branchStart = *idx;
+
+ altBranchStarts = lappend_int(altBranchStarts, branchStart);
+ fillRPRPattern(alt, pat, idx, depth + 1);
+
+ if (*idx > branchStart)
+ altEndPositions = lappend_int(altEndPositions, *idx - 1);
+ }
+
+ /* Set ALT_START.next to first alternative */
+ if (list_length(altBranchStarts) > 0)
+ pat->elements[altStartIdx].next = linitial_int(altBranchStarts);
+
+ /* Set jump on first element of each alternative to next alternative */
+ foreach(lc, altBranchStarts)
+ {
+ int firstElemIdx = lfirst_int(lc);
+
+ if (lnext(altBranchStarts, lc) != NULL)
+ pat->elements[firstElemIdx].jump = lfirst_int(lnext(altBranchStarts, lc));
+ }
+
+ /* Set next on last element of each alternative to after the alternation */
+ {
+ int afterAltIdx = *idx;
+
+ foreach(lc, altEndPositions)
+ {
+ int endPos = lfirst_int(lc);
+
+ if (pat->elements[endPos].next == RPR_ELEMIDX_INVALID)
+ pat->elements[endPos].next = afterAltIdx;
+ }
+ }
+
+ list_free(altBranchStarts);
+ list_free(altEndPositions);
+}
+
+/*
+ * fillRPRPattern
+ * Fill pattern elements array from AST (pass 2).
+ *
+ * Recursively traverses AST and populates pre-allocated elements array.
+ * Dispatches to type-specific fill functions.
+ */
+static void
+fillRPRPattern(RPRPatternNode *node, RPRPattern *pat, int *idx, RPRDepth depth)
+{
+ ListCell *lc;
+
+ if (node == NULL)
+ return;
+
+ switch (node->nodeType)
+ {
+ case RPR_PATTERN_SEQ:
+ foreach(lc, node->children)
+ {
+ fillRPRPattern((RPRPatternNode *) lfirst(lc), pat, idx, depth);
+ }
+ break;
+
+ case RPR_PATTERN_VAR:
+ fillRPRPatternVar(node, pat, idx, depth);
+ break;
+
+ case RPR_PATTERN_GROUP:
+ fillRPRPatternGroup(node, pat, idx, depth);
+ break;
+
+ case RPR_PATTERN_ALT:
+ fillRPRPatternAlt(node, pat, idx, depth);
+ break;
+ }
+}
+
+/*
+ * finalizeRPRPattern
+ * Finalize pattern structure after filling elements.
+ *
+ * This performs:
+ * 1. Set up next pointers for sequential flow
+ * 2. Add FIN marker at the end
+ */
+static void
+finalizeRPRPattern(RPRPattern *result)
+{
+ int finIdx = result->numElements - 1;
+ int i;
+ RPRPatternElement *finElem;
+
+ /* Set up next pointers for elements that don't have one */
+ for (i = 0; i < finIdx; i++)
+ {
+ RPRPatternElement *elem = &result->elements[i];
+
+ if (elem->next == RPR_ELEMIDX_INVALID)
+ elem->next = (i < finIdx - 1) ? i + 1 : finIdx;
+ }
+
+ /* Add FIN marker at the end */
+ finElem = &result->elements[finIdx];
+ memset(finElem, 0, sizeof(RPRPatternElement));
+ finElem->varId = RPR_VARID_FIN;
+ finElem->depth = 0;
+ finElem->min = 1;
+ finElem->max = 1;
+ finElem->next = RPR_ELEMIDX_INVALID;
+ finElem->jump = RPR_ELEMIDX_INVALID;
+}
+
+/*-------------------------------------------------------------------------
+ * CONTEXT ABSORPTION: TWO-FLAG DESIGN
+ *-------------------------------------------------------------------------
+ *
+ * Context absorption eliminates redundant match searches by absorbing newer
+ * contexts that cannot produce longer matches than older contexts. This
+ * achieves O(n^2) -> O(n) performance improvement for patterns like A+ B.
+ *
+ * Core Insight:
+ * For pattern A+ B, if Ctx1 starts at row 0 and Ctx2 starts at row 1,
+ * both matching A continuously, Ctx1 will always have more A matches.
+ * When B finally appears, Ctx1's match (0 to current) is always longer
+ * than Ctx2's match (1 to current). So Ctx2 can be safely eliminated.
+ *
+ * Two Flags:
+ * 1. RPR_ELEM_ABSORBABLE - "Absorption judgment point"
+ * WHERE contexts can be compared for absorption.
+ * - Simple unbounded VAR (A+): the VAR element itself
+ * - Unbounded GROUP ((A B)+): the END element only
+ *
+ * 2. RPR_ELEM_ABSORBABLE_BRANCH - "Absorbable region marker"
+ * ALL elements within the absorbable region.
+ * - Used for tracking state.isAbsorbable at runtime
+ * - States leaving this region become non-absorbable permanently
+ *
+ * Why Two Flags?
+ * For pattern "(A B)+", contexts at different positions (one at A,
+ * another at B) cannot be compared - they must synchronize at END.
+ *
+ * Example: "(A B)+" with input A B A B A B...
+ * Row 0 (A): Ctx1 starts, matches A
+ * Row 1 (B): Ctx1 matches B -> END (count=1)
+ * Row 2 (A): Ctx1 loops to A, Ctx2 starts at A
+ * Row 3 (B): Ctx1 at END (count=2), Ctx2 at END (count=1)
+ * -> Both at END, comparable! Ctx1 absorbs Ctx2.
+ *
+ * Contexts synchronize at END every group-length rows. Therefore:
+ * - ABSORBABLE marks END as judgment point (where to compare)
+ * - ABSORBABLE_BRANCH keeps state.isAbsorbable=true through A->B->END
+ *
+ * Pattern Examples:
+ *
+ * Pattern: A+ B
+ * Element 0 (A): ABSORBABLE | ABSORBABLE_BRANCH <- judgment point
+ * Element 1 (B): (none)
+ * -> Compare at A every row. When contexts move to B, absorption stops.
+ *
+ * Pattern: (A B)+ C
+ * Element 0 (A): ABSORBABLE_BRANCH
+ * Element 1 (B): ABSORBABLE_BRANCH
+ * Element 2 (END): ABSORBABLE | ABSORBABLE_BRANCH <- judgment point
+ * Element 3 (C): (none)
+ * -> Compare at END every 2 rows. When contexts move to C, absorption stops.
+ *
+ * Pattern: (A+ B+)+ C
+ * Element 0 (A): ABSORBABLE | ABSORBABLE_BRANCH <- only first A+ flagged
+ * Element 1 (B): (none)
+ * Element 2 (END): (none)
+ * Element 3 (C): (none)
+ * -> Only first unbounded portion (A+) gets flags. Absorption happens
+ * at A during first iteration. After moving to B+, absorption stops.
+ *
+ * First Unbounded Portion Strategy:
+ * The algorithm only flags the FIRST unbounded portion starting from
+ * element 0. This is sufficient because:
+ * - Absorption in first portion already achieves O(n) complexity
+ * - Later portions have different synchronization characteristics
+ * - Nested unbounded patterns are too complex for simple absorption
+ * - Complex patterns (nested groups, etc.) naturally die from mismatch
+ *
+ * Runtime Usage (in nodeWindowAgg.c):
+ * - state.isAbsorbable = (previous && elem.ABSORBABLE_BRANCH)
+ * - Monotonic: once false, stays false (cannot re-enter region)
+ * - context.hasAbsorbableState: can absorb others (>=1 absorbable state)
+ * - context.allStatesAbsorbable: can be absorbed (ALL states absorbable)
+ * - Absorption check: if Ctx1.hasAbsorbable && Ctx2.allAbsorbable,
+ * compare counts at same elemIdx, absorb if Ctx1.count >= Ctx2.count
+ *
+ *-------------------------------------------------------------------------
+ */
+
+/*
+ * isUnboundedStart
+ * Check if the element at idx starts an unbounded sequence.
+ *
+ * For context absorption to work, the sequence starting at idx must be
+ * unbounded (max = infinity) so that we can "shift" by decrementing count.
+ *
+ * Algorithm:
+ * - Descend through ALT/GROUP structures to find first actual VAR
+ * - For GROUP: must be unbounded AND contain only simple {1,1} VARs
+ * - Check if that VAR is unbounded
+ *
+ * Two cases are handled:
+ * 1. Simple var: A+ B C - first element A has max=INF
+ * 2. Group: (A B){2,} C - group END has max=INF, and all elements
+ * inside the group must be simple {1,1} vars (no nested complexity)
+ *
+ * Returns false for patterns where absorption cannot work:
+ * - A B+ (unbounded not at start)
+ * - (A | B){2,} (ALT inside group)
+ * - (A B+){2,} (unbounded element inside group)
+ * - ((A B){2,} C){3,} (nested groups)
+ */
+static bool
+isUnboundedStart(RPRPattern *pattern, RPRElemIdx idx)
+{
+ RPRPatternElement *elem = &pattern->elements[idx];
+ RPRDepth startDepth = elem->depth;
+ RPRPatternElement *lastElem = NULL;
+ bool allSimpleInside = true;
+
+ /* Traverse elements until we exit the current scope */
+ for (RPRElemIdx i = idx;; i++)
+ {
+ RPRPatternElement *e;
+
+ Assert(i >= 0 && i < pattern->numElements);
+ e = &pattern->elements[i];
+
+ /* Exit when depth drops (left the group) or reached FIN */
+ if (e->depth < startDepth || RPRElemIsFin(e))
+ {
+ lastElem = e;
+ break;
+ }
+
+ /* Track whether all elements at this depth are simple {1,1} vars */
+ if (e->depth == startDepth)
+ {
+ if (!RPRElemIsVar(e) || e->min != 1 || e->max != 1)
+ allSimpleInside = false;
+ }
+ else
+ {
+ /* Any nested structure disqualifies */
+ allSimpleInside = false;
+ }
+ }
+
+ Assert(lastElem != NULL);
+
+ /* Case 2: Unbounded group - END element points back to idx */
+ if (RPRElemIsEnd(lastElem) &&
+ lastElem->jump == idx &&
+ lastElem->max == RPR_QUANTITY_INF)
+ {
+ /*
+ * Check for outer nesting: if there's another END at a lower depth
+ * before FIN, we're inside a nested structure that doesn't support
+ * simple absorption. For example, ((A B){2,} C){3,} has an inner END
+ * at depth=1 and an outer END at depth=0.
+ */
+ RPRElemIdx lastIdx = (RPRElemIdx) (lastElem - pattern->elements);
+ RPRElemIdx j;
+
+ for (j = lastIdx + 1; j < pattern->numElements; j++)
+ {
+ RPRPatternElement *e = &pattern->elements[j];
+
+ if (RPRElemIsFin(e))
+ break; /* Reached end, no outer nesting */
+ if (RPRElemIsEnd(e) && e->depth < lastElem->depth)
+ return false; /* Found outer END, nested structure */
+ }
+
+ return allSimpleInside;
+ }
+
+ /* Case 1: Simple unbounded var at start */
+ return RPRElemIsVar(elem) && elem->max == RPR_QUANTITY_INF;
+}
+
+/*
+ * setAbsorbabilityFlagsForBranch
+ * Set absorption flags for a single branch.
+ *
+ * For unbounded GROUP (e.g., (A B)+):
+ * - ABSORBABLE_BRANCH on all elements (A, B, END)
+ * - ABSORBABLE on END only (judgment point)
+ *
+ * For simple unbounded VAR (e.g., A+):
+ * - Both flags on first element only
+ */
+static void
+setAbsorbabilityFlagsForBranch(RPRPattern *pattern, RPRElemIdx branchIdx)
+{
+ RPRPatternElement *branchFirst = &pattern->elements[branchIdx];
+ RPRDepth startDepth = branchFirst->depth;
+ bool isUnboundedGroup = false;
+ RPRElemIdx endIdx = RPR_ELEMIDX_INVALID;
+ RPRElemIdx i;
+
+ /*
+ * First pass: detect if this is an unbounded GROUP pattern. Look for END
+ * element with unbounded max that jumps back to branch start. Note: END
+ * is at parent depth (less than content depth), so check END before the
+ * depth-based break condition.
+ */
+ for (i = branchIdx; i < pattern->numElements; i++)
+ {
+ RPRPatternElement *e = &pattern->elements[i];
+
+ /* Check for unbounded END first (END is at lower depth than content) */
+ if (RPRElemIsEnd(e) &&
+ e->jump == branchIdx &&
+ e->max == RPR_QUANTITY_INF)
+ {
+ isUnboundedGroup = true;
+ endIdx = i;
+ break;
+ }
+
+ if (e->depth < startDepth || RPRElemIsFin(e))
+ break;
+ }
+
+ /*
+ * Set flags based on pattern type.
+ */
+ if (isUnboundedGroup)
+ {
+ /*
+ * Unbounded GROUP: set ABSORBABLE_BRANCH on all elements from
+ * branchIdx to endIdx, and ABSORBABLE on END only.
+ */
+ for (i = branchIdx; i <= endIdx; i++)
+ {
+ RPRPatternElement *e = &pattern->elements[i];
+
+ e->flags |= RPR_ELEM_ABSORBABLE_BRANCH;
+ if (i == endIdx)
+ e->flags |= RPR_ELEM_ABSORBABLE;
+ }
+ }
+ else
+ {
+ /*
+ * Simple unbounded VAR: set both flags on first element only.
+ */
+ branchFirst->flags |= RPR_ELEM_ABSORBABLE_BRANCH | RPR_ELEM_ABSORBABLE;
+ }
+}
+
+/*
+ * computeAbsorbability
+ * Determine if pattern supports context absorption optimization.
+ *
+ * Context absorption eliminates redundant match searches by absorbing
+ * newer contexts that cannot produce longer matches than older contexts.
+ * This achieves O(n²) → O(n) performance improvement.
+ *
+ * This function sets two flags:
+ * RPR_ELEM_ABSORBABLE: Absorption judgment point
+ * - Simple unbounded VAR: the VAR itself (e.g., A in A+)
+ * - Unbounded GROUP: the END element (e.g., END in (A B)+)
+ * RPR_ELEM_ABSORBABLE_BRANCH: All elements in absorbable region
+ * - All elements at same depth as unbounded start
+ *
+ * Examples:
+ * A+ B C - absorbable (A gets both flags)
+ * (A B)+ C - absorbable (A,B,END get BRANCH, END gets ABSORBABLE)
+ * A B+ - NOT absorbable (unbounded not at start)
+ * (A+ B+)+ - only first A+ on first iteration (nested unbounded not supported)
+ * A+ | B+ - both branches absorbable independently
+ * A+ | C D - only A+ branch absorbable (C D branch not absorbable)
+ */
+static void
+computeAbsorbability(RPRPattern *pattern)
+{
+ RPRPatternElement *first = &pattern->elements[0];
+
+ pattern->isAbsorbable = false;
+
+ if (pattern->numElements < 2) /* At minimum: one element + FIN */
+ return;
+
+ if (RPRElemIsAlt(first))
+ {
+ /* ALT: check each branch */
+ RPRElemIdx branchIdx = first->next;
+ bool hasAbsorbableBranch = false;
+
+ while (branchIdx != RPR_ELEMIDX_INVALID &&
+ branchIdx < pattern->numElements - 1)
+ {
+ RPRPatternElement *branchFirst = &pattern->elements[branchIdx];
+
+ if (isUnboundedStart(pattern, branchIdx))
+ {
+ pattern->isAbsorbable = true;
+ hasAbsorbableBranch = true;
+ setAbsorbabilityFlagsForBranch(pattern, branchIdx);
+ }
+
+ branchIdx = branchFirst->jump;
+ }
+
+ /*
+ * If any branch is absorbable, mark ALT element too. This allows
+ * efficient branch-level flag management.
+ */
+ if (hasAbsorbableBranch)
+ first->flags |= RPR_ELEM_ABSORBABLE_BRANCH;
+ }
+ else
+ {
+ /* Non-ALT: check first element */
+ if (isUnboundedStart(pattern, 0))
+ {
+ pattern->isAbsorbable = true;
+ setAbsorbabilityFlagsForBranch(pattern, 0);
+ }
+ }
+}
+
+/*
+ * buildRPRPattern
+ * Compile pattern AST to flat bytecode array.
+ *
+ * Compilation phases:
+ * 1. Optimize AST (flatten, merge, deduplicate)
+ * 2. Scan: collect variables, count elements (pass 1)
+ * 3. Allocate result structure
+ * 4. Fill elements from AST (pass 2)
+ * 5. Finalize: set up next pointers, add FIN marker
+ * 6. Compute context absorption eligibility
+ *
+ * Called from createplan.c during plan creation.
+ * Returns NULL if pattern is NULL.
+ */
+RPRPattern *
+buildRPRPattern(RPRPatternNode *pattern, List *defineVariableList,
+ RPSkipTo rpSkipTo, int frameOptions)
+{
+ RPRPattern *result;
+ RPRPatternNode *optimized;
+ char *varNamesStack[RPR_VARID_MAX + 1];
+ int numVars;
+ int numElements;
+ RPRDepth maxDepth;
+ int idx;
+ bool hasLimitedFrame;
+
+ if (pattern == NULL)
+ return NULL;
+
+ /* Optimize the pattern tree (planner phase) */
+ optimized = optimizeRPRPattern(pattern);
+
+ /* Collect variable names from DEFINE clause */
+ numVars = collectDefineVariables(defineVariableList, varNamesStack);
+
+ /* Scan pattern: collect variables, count elements, validate limits */
+ scanRPRPattern(optimized, varNamesStack, &numVars, &numElements, &maxDepth);
+
+ /* Allocate result structure */
+ result = allocateRPRPattern(numVars, numElements, maxDepth, varNamesStack);
+
+ /* Fill elements (pass 2) */
+ idx = 0;
+ fillRPRPattern(optimized, result, &idx, 0);
+
+ /* Finalize: set up next pointers, flags, and add FIN marker */
+ finalizeRPRPattern(result);
+
+ /*
+ * Compute context absorption eligibility. Absorption requires both
+ * structural absorbability and runtime conditions. Check runtime
+ * conditions first to avoid unnecessary pattern analysis.
+ */
+ hasLimitedFrame = (frameOptions & FRAMEOPTION_ROWS) &&
+ !(frameOptions & FRAMEOPTION_END_UNBOUNDED_FOLLOWING);
+
+ if (rpSkipTo == ST_PAST_LAST_ROW && !hasLimitedFrame)
+ {
+ /* Runtime conditions met - check structural absorbability */
+ computeAbsorbability(result);
+ }
+ else
+ {
+ /* Runtime conditions not met - skip pattern analysis */
+ result->isAbsorbable = false;
+ }
+
+ return result;
+}
diff --git a/src/backend/optimizer/plan/setrefs.c b/src/backend/optimizer/plan/setrefs.c
index 16d200cfb46..2efeec22102 100644
--- a/src/backend/optimizer/plan/setrefs.c
+++ b/src/backend/optimizer/plan/setrefs.c
@@ -211,7 +211,6 @@ static List *set_windowagg_runcondition_references(PlannerInfo *root,
List *runcondition,
Plan *plan);
-
/*****************************************************************************
*
* SUBPLAN REFERENCES
@@ -2576,6 +2575,32 @@ set_upper_references(PlannerInfo *root, Plan *plan, int rtoffset)
NRM_EQUAL,
NUM_EXEC_QUAL(plan));
+ /*
+ * Modifies an expression tree in each DEFINE clause so that all Var
+ * nodes's varno refers to OUTER_VAR.
+ */
+ if (IsA(plan, WindowAgg))
+ {
+ WindowAgg *wplan = (WindowAgg *) plan;
+
+ if (wplan->defineClause != NIL)
+ {
+ foreach(l, wplan->defineClause)
+ {
+ TargetEntry *tle = (TargetEntry *) lfirst(l);
+
+ tle->expr = (Expr *)
+ fix_upper_expr(root,
+ (Node *) tle->expr,
+ subplan_itlist,
+ OUTER_VAR,
+ rtoffset,
+ NRM_EQUAL,
+ NUM_EXEC_QUAL(plan));
+ }
+ }
+ }
+
pfree(subplan_itlist);
}
diff --git a/src/backend/optimizer/prep/prepjointree.c b/src/backend/optimizer/prep/prepjointree.c
index c80bfc88d82..63d872b029d 100644
--- a/src/backend/optimizer/prep/prepjointree.c
+++ b/src/backend/optimizer/prep/prepjointree.c
@@ -2510,6 +2510,15 @@ perform_pullup_replace_vars(PlannerInfo *root,
parse->returningList = (List *)
pullup_replace_vars((Node *) parse->returningList, rvcontext);
+ foreach(lc, parse->windowClause)
+ {
+ WindowClause *wc = lfirst_node(WindowClause, lc);
+
+ if (wc->defineClause != NIL)
+ wc->defineClause = (List *)
+ pullup_replace_vars((Node *) wc->defineClause, rvcontext);
+ }
+
if (parse->onConflict)
{
parse->onConflict->onConflictSet = (List *)
diff --git a/src/include/nodes/plannodes.h b/src/include/nodes/plannodes.h
index 4bc6fb5670e..499cf3c1385 100644
--- a/src/include/nodes/plannodes.h
+++ b/src/include/nodes/plannodes.h
@@ -20,6 +20,7 @@
#include "lib/stringinfo.h"
#include "nodes/bitmapset.h"
#include "nodes/lockoptions.h"
+#include "nodes/parsenodes.h"
#include "nodes/primnodes.h"
@@ -1223,6 +1224,70 @@ typedef struct Agg
List *chain;
} Agg;
+/* ----------------
+ * Row Pattern Recognition compiled pattern types
+ * ----------------
+ */
+
+/* Type definitions for RPR pattern elements */
+typedef uint8 RPRVarId; /* pattern variable ID */
+typedef uint8 RPRElemFlags; /* element flags */
+typedef uint8 RPRDepth; /* group nesting depth */
+typedef int32 RPRQuantity; /* quantifier min/max */
+typedef int16 RPRElemIdx; /* element array index */
+
+/*
+ * RPRPatternElement - flat element for NFA pattern matching (16 bytes)
+ *
+ * Layout optimized for alignment (no padding holes):
+ * varId(1) + depth(1) + flags(1) + reserved(1) + min(4) + max(4) + next(2) + jump(2)
+ */
+typedef struct RPRPatternElement
+{
+ RPRVarId varId; /* variable ID, or special value for control */
+ RPRDepth depth; /* group nesting depth */
+ RPRElemFlags flags; /* flags (reluctant, etc.) */
+ uint8 reserved; /* reserved padding byte */
+ RPRQuantity min; /* quantifier minimum */
+ RPRQuantity max; /* quantifier maximum */
+ RPRElemIdx next; /* next element index */
+ RPRElemIdx jump; /* jump target (for ALT/GROUP) */
+} RPRPatternElement;
+
+/*
+ * RPRPattern - compiled pattern for NFA execution
+ *
+ * Requires custom copy/out/read functions due to elements array.
+ */
+typedef struct RPRPattern
+{
+ pg_node_attr(custom_copy_equal, custom_read_write)
+
+ NodeTag type; /* T_RPRPattern */
+ int numVars; /* number of pattern variables */
+ char **varNames; /* array of variable names (DEFINE order
+ * first) */
+ RPRDepth maxDepth; /* maximum group nesting depth */
+ int numElements; /* number of elements */
+ RPRPatternElement *elements; /* array of pattern elements */
+
+ /*----------------
+ * Context absorption optimization.
+ *
+ * Absorption is only safe when later matches are guaranteed to be
+ * suffixes of earlier matches. This requires simple pattern structure:
+ *
+ * Case 1: No ALT, single unbounded element (A+, (A B)+)
+ * Case 2: Top-level ALT with each branch being single unbounded (A+ | B+)
+ *
+ * Complex patterns like A B (A B)+ could theoretically be transformed to
+ * (A B){2,} for absorption, but this changes lexical order and is not
+ * implemented. Similarly, (A|B)+ cannot be absorbed because different
+ * start positions produce different match contents (not suffix relation).
+ */
+ bool isAbsorbable; /* true if pattern supports context absorption */
+} RPRPattern;
+
/* ----------------
* window aggregate node
* ----------------
@@ -1293,6 +1358,15 @@ typedef struct WindowAgg
/* nulls sort first for in_range tests? */
bool inRangeNullsFirst;
+ /* Row Pattern Recognition AFTER MATCH SKIP clause */
+ RPSkipTo rpSkipTo; /* Row Pattern Skip To type */
+
+ /* Compiled Row Pattern for NFA execution */
+ struct RPRPattern *rpPattern;
+
+ /* Row Pattern DEFINE clause (list of TargetEntry) */
+ List *defineClause;
+
/*
* false for all apart from the WindowAgg that's closest to the root of
* the plan
diff --git a/src/include/optimizer/rpr.h b/src/include/optimizer/rpr.h
new file mode 100644
index 00000000000..2b215188e7a
--- /dev/null
+++ b/src/include/optimizer/rpr.h
@@ -0,0 +1,50 @@
+/*-------------------------------------------------------------------------
+ *
+ * rpr.h
+ * Row Pattern Recognition pattern compilation for planner
+ *
+ * Portions Copyright (c) 1996-2026, PostgreSQL Global Development Group
+ * Portions Copyright (c) 1994, Regents of the University of California
+ *
+ * src/include/optimizer/rpr.h
+ *
+ *-------------------------------------------------------------------------
+ */
+#ifndef OPTIMIZER_RPR_H
+#define OPTIMIZER_RPR_H
+
+#include "nodes/parsenodes.h"
+#include "nodes/plannodes.h"
+
+/* Limits and special values */
+#define RPR_VARID_MAX 252 /* max pattern variables: 252 */
+#define RPR_QUANTITY_INF INT32_MAX /* unbounded quantifier */
+#define RPR_COUNT_MAX INT32_MAX /* max runtime count (NFA state) */
+#define RPR_ELEMIDX_MAX INT16_MAX /* max pattern elements */
+#define RPR_ELEMIDX_INVALID ((RPRElemIdx) -1) /* invalid index */
+
+/* Special varId values for control elements (253-255) */
+#define RPR_VARID_ALT ((RPRVarId) 253) /* alternation start */
+#define RPR_VARID_END ((RPRVarId) 254) /* group end */
+#define RPR_VARID_FIN ((RPRVarId) 255) /* pattern finish */
+
+/* Element flags */
+#define RPR_ELEM_RELUCTANT 0x01 /* reluctant (non-greedy)
+ * quantifier */
+#define RPR_ELEM_ABSORBABLE_BRANCH 0x02 /* element in absorbable region */
+#define RPR_ELEM_ABSORBABLE 0x04 /* absorption judgment point */
+
+/* Accessor macros for RPRPatternElement */
+#define RPRElemIsReluctant(e) ((e)->flags & RPR_ELEM_RELUCTANT)
+#define RPRElemIsAbsorbableBranch(e) ((e)->flags & RPR_ELEM_ABSORBABLE_BRANCH)
+#define RPRElemIsAbsorbable(e) ((e)->flags & RPR_ELEM_ABSORBABLE)
+#define RPRElemIsVar(e) ((e)->varId <= RPR_VARID_MAX)
+#define RPRElemIsAlt(e) ((e)->varId == RPR_VARID_ALT)
+#define RPRElemIsEnd(e) ((e)->varId == RPR_VARID_END)
+#define RPRElemIsFin(e) ((e)->varId == RPR_VARID_FIN)
+#define RPRElemCanSkip(e) ((e)->min == 0)
+
+extern RPRPattern *buildRPRPattern(RPRPatternNode *pattern, List *defineVariableList,
+ RPSkipTo rpSkipTo, int frameOptions);
+
+#endif /* OPTIMIZER_RPR_H */
--
2.43.0