GIN (JSONPath) checks equality only on the first node of each operand
GIN (JSONPath) checks equality only on the first node of each operand
От:
Chinmay Kanchi <cgkanchi@gmail.com>
Дата:
When checking equality with an operand that has additional nodes ("2".double()), GIN only considers the first node ("2"). So $.a == "2".integer() looks up a match for the json string "2" instead of the number 2, and silently drops rows that a sequential scan would have returned.
```
CREATE TABLE t (j jsonb);
INSERT INTO t VALUES ('{"a": 2}'), ('{"a": "2"}'), ('{"a": 3}');
CREATE INDEX t_ops ON t USING gin (j jsonb_ops);
CREATE INDEX t_path_ops ON t USING gin (j jsonb_path_ops);
SET enable_seqscan = off;
SET enable_indexscan = off; -- leave bitmap scans only
SELECT 'seq', count(*) FROM (SELECT j FROM t OFFSET 0) s WHERE j @@ '$.a == "2".double()'; -- 1
CREATE TABLE t (j jsonb);
INSERT INTO t VALUES ('{"a": 2}'), ('{"a": "2"}'), ('{"a": 3}');
CREATE INDEX t_ops ON t USING gin (j jsonb_ops);
CREATE INDEX t_path_ops ON t USING gin (j jsonb_path_ops);
SET enable_seqscan = off;
SET enable_indexscan = off; -- leave bitmap scans only
SELECT 'seq', count(*) FROM (SELECT j FROM t OFFSET 0) s WHERE j @@ '$.a == "2".double()'; -- 1
DROP INDEX t_path_ops;
SELECT 'jsonb_ops', count(*) FROM t WHERE j @@ '$.a == "2".double()'; -- 0, should be 1
SELECT 'jsonb_ops', count(*) FROM t WHERE j @? '$.a ? (@ == (-2).abs())'; -- 0, should be 1
SELECT 'jsonb_ops', count(*) FROM t WHERE j @@ '$.a == "2".double()'; -- 0, should be 1
SELECT 'jsonb_ops', count(*) FROM t WHERE j @? '$.a ? (@ == (-2).abs())'; -- 0, should be 1
CREATE INDEX t_path_ops ON t USING gin (j jsonb_path_ops);
DROP INDEX t_ops;
SELECT 'jsonb_path_ops', count(*) FROM t WHERE j @@ '$.a == "2".double()'; -- 0, should be 1
DROP INDEX t_ops;
SELECT 'jsonb_path_ops', count(*) FROM t WHERE j @@ '$.a == "2".double()'; -- 0, should be 1
```
Re: GIN (JSONPath) checks equality only on the first node of each operand
От:
Tom Lane <tgl@sss.pgh.pa.us>
Дата:
Chinmay Kanchi writes:
> When checking equality with an operand that has additional nodes
> ("2".double()), GIN only considers the first node ("2"). So $.a ==
> "2".integer() looks up a match for the json string "2" instead of the
> number 2, and silently drops rows that a sequential scan would have
> returned.
Hmph, another bug in the same function as your previous report :-(.
Here's a less-WIP patch for both.
regards, tom lane
From 327a2ad9221f146e9fe323e6f121db980a46ad8b Mon Sep 17 00:00:00 2001
From: Tom Lane
Date: Sun, 11 Oct 2026 14:03:28 -0400
Subject: [PATCH v1] Fix two errors in extraction of GIN indexquals from
jsonpath queries.
extract_jsp_bool_expr applied De Morgan's laws incorrectly when
considering an AND operator underneath a NOT. If one AND input
was implementable as a GIN indexqual while the other was not,
we derived an incomplete indexqual, leading to some rows not
being returned that should be.
extract_jsp_bool_expr also failed to check for methods being
applied to literals, so that for example '$.x == "42".double()'
was treated as '$.x == "42"' not '$.x == 42'. The indexqual
thus selected the wrong rows, all of which would fail qual
recheck, leading to no rows being returned.
While we're here, improve the comments a trifle.
Reported-by: Chinmay Kanchi
Author: Tom Lane
Discussion: https://postgr.es/m/CAJqPDh9Cm=x8eYZ1Jf=_avd2WEqc=0t_f7_SO4igo-CKPi25_Q@mail.gmail.com
Discussion: https://postgr.es/m/CAJqPDh-tu5KY66KU8cvAEg3DQrZLzDZwjkLLa42zD+_Xm1WJAg@mail.gmail.com
Backpatch-through: 14
---
src/backend/utils/adt/jsonb_gin.c | 28 +++++++++++++++++++++------
src/test/regress/expected/jsonb.out | 30 +++++++++++++++++++++++++++++
src/test/regress/sql/jsonb.sql | 5 +++++
3 files changed, 57 insertions(+), 6 deletions(-)
diff --git a/src/backend/utils/adt/jsonb_gin.c b/src/backend/utils/adt/jsonb_gin.c
index f5dbd5589d3..96506661cb0 100644
--- a/src/backend/utils/adt/jsonb_gin.c
+++ b/src/backend/utils/adt/jsonb_gin.c
@@ -558,6 +558,7 @@ extract_jsp_path_expr_nodes(JsonPathGinContext *cxt, JsonPathGinPath path,
* Extract an expression node from one of following jsonpath path expressions:
* EXISTS(jsp) (when 'scalar' is NULL)
* jsp == scalar (when 'scalar' is not NULL).
+ * Returns NULL if no usable filter expression can be extracted.
*
* The current path (@) is passed in 'path'.
*/
@@ -579,7 +580,10 @@ extract_jsp_path_expr(JsonPathGinContext *cxt, JsonPathGinPath path,
return make_jsp_expr_node_args(JSP_GIN_AND, nodes);
}
-/* Recursively extract nodes from the boolean jsonpath expression. */
+/*
+ * Recursively extract a GIN qual from the boolean jsonpath expression.
+ * Returns NULL if no usable indexqual can be extracted.
+ */
static JsonPathGinNode *
extract_jsp_bool_expr(JsonPathGinContext *cxt, JsonPathGinPath path,
JsonPathItem *jsp, bool not)
@@ -591,6 +595,8 @@ extract_jsp_bool_expr(JsonPathGinContext *cxt, JsonPathGinPath path,
case jpiAnd: /* expr && expr */
case jpiOr: /* expr || expr */
{
+ /* We use De Morgan's laws to handle a passed-down NOT */
+ bool is_and = (not ^ (jsp->type == jpiAnd));
JsonPathItem arg;
JsonPathGinNode *larg;
JsonPathGinNode *rarg;
@@ -604,13 +610,19 @@ extract_jsp_bool_expr(JsonPathGinContext *cxt, JsonPathGinPath path,
if (!larg || !rarg)
{
- if (jsp->type == jpiOr)
+ /*
+ * For OR, we have to fail if either arm is unsupported.
+ * For AND, we can just ignore the unsupported arm and
+ * apply the supported one, relying on indexscan qual
+ * recheck to reject rows that don't match the other arm.
+ */
+ if (!is_and)
return NULL;
return larg ? larg : rarg;
}
- type = not ^ (jsp->type == jpiAnd) ? JSP_GIN_AND : JSP_GIN_OR;
+ type = is_and ? JSP_GIN_AND : JSP_GIN_OR;
return make_jsp_expr_node_binary(type, larg, rarg);
}
@@ -662,24 +674,28 @@ extract_jsp_bool_expr(JsonPathGinContext *cxt, JsonPathGinPath path,
JsonPathItem *scalar_item;
JsonbValue scalar;
+ /* Can't support !(x == y) */
if (not)
return NULL;
jspGetLeftArg(jsp, &left_item);
jspGetRightArg(jsp, &right_item);
- if (jspIsScalar(left_item.type))
+ /* To build an indexqual, one side must be a simple scalar */
+ if (jspIsScalar(left_item.type) &&
+ !jspHasNext(&left_item))
{
scalar_item = &left_item;
path_item = &right_item;
}
- else if (jspIsScalar(right_item.type))
+ else if (jspIsScalar(right_item.type) &&
+ !jspHasNext(&right_item))
{
scalar_item = &right_item;
path_item = &left_item;
}
else
- return NULL; /* at least one operand should be a scalar */
+ return NULL;
switch (scalar_item->type)
{
diff --git a/src/test/regress/expected/jsonb.out b/src/test/regress/expected/jsonb.out
index b1d2ce5f03a..82c31e2962b 100644
--- a/src/test/regress/expected/jsonb.out
+++ b/src/test/regress/expected/jsonb.out
@@ -3053,6 +3053,12 @@ SELECT count(*) FROM testjsonb WHERE j @@ '$.age == 25.0';
2
(1 row)
+SELECT count(*) FROM testjsonb WHERE j @@ '$.age == "25".double()';
+ count
+-------
+ 2
+(1 row)
+
SELECT count(*) FROM testjsonb WHERE j @@ 'exists($)';
count
-------
@@ -3260,6 +3266,12 @@ SELECT count(*) FROM testjsonb WHERE j @@ '$.age == 25.0';
2
(1 row)
+SELECT count(*) FROM testjsonb WHERE j @@ '$.age == "25".double()';
+ count
+-------
+ 2
+(1 row)
+
SELECT count(*) FROM testjsonb WHERE j @@ '$.array[*] == "foo"';
count
-------
@@ -3272,6 +3284,12 @@ SELECT count(*) FROM testjsonb WHERE j @@ '$.array[*] == "bar"';
3
(1 row)
+SELECT count(*) FROM testjsonb WHERE j @@ '!(!($.bad == false) && $.coauthors == 98)';
+ count
+-------
+ 850
+(1 row)
+
SELECT count(*) FROM testjsonb WHERE j @@ 'exists($ ? (@.array[*] == "bar"))';
count
-------
@@ -3589,6 +3607,12 @@ SELECT count(*) FROM testjsonb WHERE j @@ '$.age == 25.0';
2
(1 row)
+SELECT count(*) FROM testjsonb WHERE j @@ '$.age == "25".double()';
+ count
+-------
+ 2
+(1 row)
+
SELECT count(*) FROM testjsonb WHERE j @@ '$.array[*] == "foo"';
count
-------
@@ -3601,6 +3625,12 @@ SELECT count(*) FROM testjsonb WHERE j @@ '$.array[*] == "bar"';
3
(1 row)
+SELECT count(*) FROM testjsonb WHERE j @@ '!(!($.bad == false) && $.coauthors == 98)';
+ count
+-------
+ 850
+(1 row)
+
SELECT count(*) FROM testjsonb WHERE j @@ 'exists($ ? (@.array[*] == "bar"))';
count
-------
diff --git a/src/test/regress/sql/jsonb.sql b/src/test/regress/sql/jsonb.sql
index d4e8d7d8c9e..610ee28b97f 100644
--- a/src/test/regress/sql/jsonb.sql
+++ b/src/test/regress/sql/jsonb.sql
@@ -843,6 +843,7 @@ SELECT count(*) FROM testjsonb WHERE j @@ '"CC" == $.wait';
SELECT count(*) FROM testjsonb WHERE j @@ '$.wait == "CC" && true == $.public';
SELECT count(*) FROM testjsonb WHERE j @@ '$.age == 25';
SELECT count(*) FROM testjsonb WHERE j @@ '$.age == 25.0';
+SELECT count(*) FROM testjsonb WHERE j @@ '$.age == "25".double()';
SELECT count(*) FROM testjsonb WHERE j @@ 'exists($)';
SELECT count(*) FROM testjsonb WHERE j @@ 'exists($.public)';
SELECT count(*) FROM testjsonb WHERE j @@ 'exists($.bar)';
@@ -884,8 +885,10 @@ SELECT count(*) FROM testjsonb WHERE j @@ '"CC" == $.wait';
SELECT count(*) FROM testjsonb WHERE j @@ '$.wait == "CC" && true == $.public';
SELECT count(*) FROM testjsonb WHERE j @@ '$.age == 25';
SELECT count(*) FROM testjsonb WHERE j @@ '$.age == 25.0';
+SELECT count(*) FROM testjsonb WHERE j @@ '$.age == "25".double()';
SELECT count(*) FROM testjsonb WHERE j @@ '$.array[*] == "foo"';
SELECT count(*) FROM testjsonb WHERE j @@ '$.array[*] == "bar"';
+SELECT count(*) FROM testjsonb WHERE j @@ '!(!($.bad == false) && $.coauthors == 98)';
SELECT count(*) FROM testjsonb WHERE j @@ 'exists($ ? (@.array[*] == "bar"))';
SELECT count(*) FROM testjsonb WHERE j @@ 'exists($.array ? (@[*] == "bar"))';
SELECT count(*) FROM testjsonb WHERE j @@ 'exists($.array[*] ? (@ == "bar"))';
@@ -964,8 +967,10 @@ SELECT count(*) FROM testjsonb WHERE j @@ '"CC" == $.wait';
SELECT count(*) FROM testjsonb WHERE j @@ '$.wait == "CC" && true == $.public';
SELECT count(*) FROM testjsonb WHERE j @@ '$.age == 25';
SELECT count(*) FROM testjsonb WHERE j @@ '$.age == 25.0';
+SELECT count(*) FROM testjsonb WHERE j @@ '$.age == "25".double()';
SELECT count(*) FROM testjsonb WHERE j @@ '$.array[*] == "foo"';
SELECT count(*) FROM testjsonb WHERE j @@ '$.array[*] == "bar"';
+SELECT count(*) FROM testjsonb WHERE j @@ '!(!($.bad == false) && $.coauthors == 98)';
SELECT count(*) FROM testjsonb WHERE j @@ 'exists($ ? (@.array[*] == "bar"))';
SELECT count(*) FROM testjsonb WHERE j @@ 'exists($.array ? (@[*] == "bar"))';
SELECT count(*) FROM testjsonb WHERE j @@ 'exists($.array[*] ? (@ == "bar"))';
--
2.52.0
Re: GIN (JSONPath) checks equality only on the first node of each operand
Hi Tom,
Thanks for the fixes. I think v1 covers both reported problems.
While reading v1, I saw that extract_jsp_query() sends @@ expressions
to extract_jsp_bool_expr(). That function switches on node types such
as jpiAnd and jpiEqual. I wondered how it handles a predicate followed
by a method, so I tried ($.age == 25).size() in GDB. jsp->type was still
jpiEqual, with $.age on the left, 25 on the right, and size() as its
next item.
In an SQL expression tree, a function call would have a FuncExpr node
around its argument. Here, jsonpath_gram.y builds a list of the
predicate and accessor:
accessor_expr:
path_primary { $$ = list_make1($1); }
| '(' expr ')' accessor_op { $$ = list_make2($2, $4); }
| '(' predicate ')' accessor_op { $$ = list_make2($2, $4); }
| accessor_expr accessor_op { $$ = lappend($1, $2); }
;
expr:
accessor_expr { $$ = makeItemList($1); }
makeItemList() links ($.age == 25) and size() through next. The head
remains jpiEqual, but extract_jsp_bool_expr() ignores its next item.
V1 checks for following items on the literal operand, but not on the
predicate itself.
Using the existing testjsonb data from jsonb.sql:
SELECT count(*) FROM testjsonb
WHERE j @@ '($.age == 25).size().boolean()';
v1 v1 + 0001
Sequential scan 1012 1012
GIN jsonb_ops 2 1012
GIN jsonb_path_ops 2 1012
The equality takes this branch:
else if (jspIsScalar(right_item.type) &&
!jspHasNext(&right_item))
{
scalar_item = &right_item;
path_item = &left_item;
}
left_item is $.age, and right_item is the constant 25 with no next item.
The function therefore extracts $.age == 25 as the index condition,
ignoring the methods attached to the equality itself.
In this case, .size() returns 1 for both true and false, and .boolean()
turns that into true. All 1012 rows should match, but GIN returns only
the two rows with age = 25.
To be honest, this is a corner case, and I haven't found a more common
way to trigger it. It seems unlikely to occur in normal queries.
0001 adds this check at the start of extract_jsp_bool_expr():
if (jspHasNext(jsp))
return NULL;
It is on top of v1 and adds regression tests.
All 239 core regression tests passed.
Regards,
Rui