Failing test for GiST indexes on jsonb arrays

Even though in some contexts jsonb arrays of strings have as keys (or at
least indexable values) their string elements, this only works with
seqscan plans.  Both GiST and GIN are broken here, which I'm inclined to
consider a submission-blocking issue.

This presumably breaks non-utf8 tests for now, but currently it isn't
clear to me how I ought to deal with that, so pending an answer I'll go
ahead and commit this.

src/test/regress/data/jsonb.data

@@ -988,6 +988,7 @@
 {"world":"CAB", "org":21, "indexed":true, "line":988, "abstract":"ABC"}
 {"title":"CBC", "status":66, "line":989}
 {}
+{"array":["foo", "bar", "baz"]}
 {"line":991, "abstract":"BA", "node":"BBB"}
 {"line":992, "disabled":true, "pos":29, "public":false}
 {"state":53, "wait":"CB", "subtitle":"CCC", "line":993, "date":"CAC", "public":false, "coauthors":"BB"}
@@ -1001,4 +1002,4 @@
 {"wait":null, "line":1000}
 {"age":25}
 {"age":25.0}
-{}
+{}

src/test/regress/expected/jsonb.out

@@ -424,6 +424,8 @@ SELECT '{"x":"y"}'::jsonb <> '{"x":"z"}'::jsonb;
  t
 (1 row)
 
+CREATE TABLE testjsonb (j jsonb);
+\copy testjsonb FROM 'data/jsonb.data'
 -- containment
 SELECT jsonb_contains('{"a":"b", "b":1, "c":null}', '{"a":"b"}');
  jsonb_contains 
@@ -737,6 +739,13 @@ SELECT jsonb '{"a":"null", "b":"qq"}' ? 'a';
  t
 (1 row)
 
+-- array exists - array elements should behave as keys
+SELECT 1 from testjsonb  WHERE j->'array' ? 'bar';
+ ?column? 
+----------
+        1
+(1 row)
+
 SELECT jsonb_exists_any('{"a":null, "b":"qq"}', ARRAY['a','b']);
  jsonb_exists_any 
 ------------------
@@ -1324,8 +1333,6 @@ SELECT jsonb '{ "a":  "null \u0000 escape" }' ->> 'a' AS not_unescaped;
 (1 row)
 
 -- indexing
-CREATE TABLE testjsonb (j jsonb);
-\copy testjsonb FROM 'data/jsonb.data'
 SELECT count(*) FROM testjsonb WHERE j @> '{"wait":null}';
  count 
 -------
@@ -1424,8 +1431,17 @@ SELECT count(*) FROM testjsonb WHERE j ?& ARRAY['public','disabled'];
     42
 (1 row)
 
+-- array exists - array elements should behave as keys (for index scans too)
+CREATE INDEX jidx_array ON testjsonb USING gist((j->'array'));
+SELECT 1 from testjsonb  WHERE j->'array' ? 'bar';
+ ?column? 
+----------
+        1
+(1 row)
+
 RESET enable_seqscan;
 DROP INDEX jidx;
+DROP INDEX jidx_array;
 CREATE INDEX jidx ON testjsonb USING gin (j);
 SET enable_seqscan = off;
 SELECT count(*) FROM testjsonb WHERE j @> '{"wait":null}';
@@ -1480,7 +1496,7 @@ RESET enable_seqscan;
 SELECT count(*) FROM (SELECT (jsonb_each(j)).key FROM testjsonb) AS wow;
  count 
 -------
-  4783
+  4784
 (1 row)
 
 SELECT key, count(*) FROM (SELECT (jsonb_each(j)).key FROM testjsonb) AS wow GROUP BY key ORDER BY count DESC, key;
@@ -1509,28 +1525,29 @@ SELECT key, count(*) FROM (SELECT (jsonb_each(j)).key FROM testjsonb) AS wow GRO
  auth      |   168
  abstract  |   161
  age       |     2
-(23 rows)
+ array     |     1
+(24 rows)
 
 -- sort/hash
 SELECT count(distinct j) FROM testjsonb;
  count 
 -------
-   886
+   887
 (1 row)
 
 SET enable_hashagg = off;
 SELECT count(*) FROM (SELECT j FROM (SELECT * FROM testjsonb UNION ALL SELECT * FROM testjsonb) js GROUP BY j) js2;
  count 
 -------
-   886
+   887
 (1 row)
 
 SET enable_hashagg = on;
 SET enable_sort = off;
 SELECT count(*) FROM (SELECT j FROM (SELECT * FROM testjsonb UNION ALL SELECT * FROM testjsonb) js GROUP BY j) js2;
  count 
 -------
-   886
+   887
 (1 row)
 
 SELECT distinct * FROM (values (jsonb '{}' || ''),('{}')) v(j);

src/test/regress/sql/jsonb.sql

@@ -108,6 +108,9 @@ SELECT '{"x":"y"}'::jsonb = '{"x":"z"}'::jsonb;
 SELECT '{"x":"y"}'::jsonb <> '{"x":"y"}'::jsonb;
 SELECT '{"x":"y"}'::jsonb <> '{"x":"z"}'::jsonb;
 
+CREATE TABLE testjsonb (j jsonb);
+\copy testjsonb FROM 'data/jsonb.data'
+
 -- containment
 SELECT jsonb_contains('{"a":"b", "b":1, "c":null}', '{"a":"b"}');
 SELECT jsonb_contains('{"a":"b", "b":1, "c":null}', '{"a":"b", "c":null}');
@@ -165,6 +168,8 @@ SELECT jsonb '{"a":null, "b":"qq"}' ? 'a';
 SELECT jsonb '{"a":null, "b":"qq"}' ? 'b';
 SELECT jsonb '{"a":null, "b":"qq"}' ? 'c';
 SELECT jsonb '{"a":"null", "b":"qq"}' ? 'a';
+-- array exists - array elements should behave as keys
+SELECT 1 from testjsonb  WHERE j->'array' ? 'bar';
 
 SELECT jsonb_exists_any('{"a":null, "b":"qq"}', ARRAY['a','b']);
 SELECT jsonb_exists_any('{"a":null, "b":"qq"}', ARRAY['b','a']);
@@ -294,9 +299,6 @@ SELECT jsonb '{ "a":  "dollar \u0024 character" }' ->> 'a' AS correct_everyWHERE
 SELECT jsonb '{ "a":  "null \u0000 escape" }' ->> 'a' AS not_unescaped;
 
 -- indexing
-CREATE TABLE testjsonb (j jsonb);
-\copy testjsonb FROM 'data/jsonb.data'
-
 SELECT count(*) FROM testjsonb WHERE j @> '{"wait":null}';
 SELECT count(*) FROM testjsonb WHERE j @> '{"wait":"CC"}';
 SELECT count(*) FROM testjsonb WHERE j @> '{"wait":"CC", "public":true}';
@@ -317,10 +319,14 @@ SELECT count(*) FROM testjsonb WHERE j @> '{"age":25.0}';
 SELECT count(*) FROM testjsonb WHERE j ? 'public';
 SELECT count(*) FROM testjsonb WHERE j ?| ARRAY['public','disabled'];
 SELECT count(*) FROM testjsonb WHERE j ?& ARRAY['public','disabled'];
+-- array exists - array elements should behave as keys (for index scans too)
+CREATE INDEX jidx_array ON testjsonb USING gist((j->'array'));
+SELECT 1 from testjsonb  WHERE j->'array' ? 'bar';
 
 RESET enable_seqscan;
 
 DROP INDEX jidx;
+DROP INDEX jidx_array;
 CREATE INDEX jidx ON testjsonb USING gin (j);
 SET enable_seqscan = off;