Skip to content

Commit

Permalink
Merge pull request duckdb#10553 from Tmonster/fuzzer_column_binding_i…
Browse files Browse the repository at this point in the history 
…ssues

Do not replace filters that evaluate to always true
  • Loading branch information
@Mytherin
Mytherin authored Feb 12, 2024
2 parents 6bbc083 + af9bd25 commit 3603860
Show file tree
Hide file tree
Showing 3 changed files with 41 additions and 8 deletions.
3 changes: 1 addition & 2 deletions src/optimizer/statistics/operator/propagate_filter.cpp
View file
Original file line number Diff line number Diff line change
Expand Up @@ -234,8 +234,7 @@ unique_ptr<NodeStatistics> StatisticsPropagator::PropagateStatistics(LogicalFilt
filter.expressions.erase(filter.expressions.begin() + i);
i--;
if (filter.expressions.empty()) {
// all conditions have been erased: remove the entire filter
*node_ptr = std::move(filter.children[0]);
// just break. The physical filter planner will plan a projection instead
break;
}
} else if (ExpressionIsConstant(*condition, Value::BOOLEAN(false)) ||
Expand Down
29 changes: 29 additions & 0 deletions test/fuzzer/duckfuzz/duck_fuzz_column_binding_tests.test
View file
Original file line number Diff line number Diff line change
@@ -0,0 +1,29 @@
# name: test/fuzzer/duckfuzz/duck_fuzz_column_binding_tests.test
# description: test why column bindings fail.
# group: [duckfuzz]

statement ok
create table all_types as select * exclude(small_enum, medium_enum, large_enum) from test_all_types() limit 0;

# https://github.com/duckdb/duckdb-fuzzer/issues/1357
# simplified query
statement ok
SELECT ref_8.uint AS c0,
Argmin(Cast(ref_8."timestamp" AS TIMESTAMP), Cast(ref_10."timestamp" AS TIMESTAMP)) OVER (partition BY ref_10."nested_int_array", ref_10."timestamp" ORDER BY ref_8.hugeint) AS c1
FROM main.all_types AS ref_8
INNER JOIN main.all_types AS ref_9
INNER JOIN main.all_types AS ref_10
ON ( ref_10.dec38_10 IS NOT NULL)
OR EXISTS
( SELECT ref_9."float" FROM main.all_types)
ON ((ref_9."smallint" = ref_8."smallint"))


# original query from fuzzer
statement ok
SELECT ref_8.uint AS c0, CASE WHEN ((min_by(CAST(ref_8."varchar" AS VARCHAR), CAST(ref_3."varchar" AS VARCHAR)) OVER (PARTITION BY subq_3.c1, ref_10.double_array ORDER BY ref_7."varchar") ~~* ref_10."varchar")) THEN (argmin(CAST(ref_6."timestamp" AS TIMESTAMP), CAST(ref_10."timestamp" AS TIMESTAMP)) OVER (PARTITION BY ref_10.nested_int_array, ref_10."timestamp" ORDER BY ref_8.hugeint)) ELSE argmin(CAST(ref_6."timestamp" AS TIMESTAMP), CAST(ref_10."timestamp" AS TIMESTAMP)) OVER (PARTITION BY ref_10.nested_int_array, ref_10."timestamp" ORDER BY ref_8.hugeint) END AS c1 FROM main.all_types AS ref_0 INNER JOIN (SELECT ref_1.timestamp_array AS c1 FROM main.all_types AS ref_1 INNER JOIN main.all_types AS ref_2 ON (NULL) WHERE (((CAST(NULL AS VARCHAR) ~~~ CAST(NULL AS VARCHAR)) OR (ref_1.blob IS NOT NULL) OR 0) AND (ref_1.timestamp_ms IS NULL) AND ((SELECT histogram("varchar") FROM main.all_types) IS NOT NULL) AND ((ref_2.timestamp_s IS NULL) OR (((ref_2."varchar" ~~~ ref_1."varchar") OR 1 OR (ref_2."varchar" ~~ (SELECT "varchar" FROM main.all_types LIMIT 1 OFFSET 5))) AND (ref_2."varchar" ~~ ref_2."varchar")) OR ((SELECT "varchar" FROM main.all_types LIMIT 1 OFFSET 6) !~~* ref_1."varchar")))) AS subq_0 ON (ref_0."varchar") LEFT JOIN main.all_types AS ref_3 LEFT JOIN (SELECT ref_4.ubigint AS c0, 13 AS c1, ref_4."time" AS c2, ref_4."float" AS c3, (SELECT double_array FROM main.all_types LIMIT 1 OFFSET 6) AS c4 FROM main.all_types AS ref_4 WHERE NULL) AS subq_1 ON ((ref_3."float" = subq_1.c3)) ON (NULL) INNER JOIN (SELECT ref_5.timestamp_ms AS c4 FROM main.all_types AS ref_5) AS subq_2 LEFT JOIN main.all_types AS ref_6 INNER JOIN main.all_types AS ref_7 ON (((SELECT "varchar" FROM main.all_types LIMIT 1 OFFSET 1) ~~* ref_7."varchar")) INNER JOIN main.all_types AS ref_8 INNER JOIN main.all_types AS ref_9 INNER JOIN main.all_types AS ref_10 ON (((ref_9."varchar" !~~* ref_9."varchar") OR (ref_10.dec38_10 IS NOT NULL) OR EXISTS(SELECT ref_9."float" AS c0, ref_9.usmallint AS c1, ref_10."bigint" AS c2, (SELECT bool FROM main.all_types LIMIT 1 OFFSET 5) AS c3, ref_10.nested_int_array AS c4, ref_9.timestamp_ms AS c5, 4 AS c6, ref_11."map" AS c7, (SELECT uint FROM main.all_types LIMIT 1 OFFSET 3) AS c8, ref_9.dec_4_1 AS c9 FROM main.all_types AS ref_11 WHERE (1 AND EXISTS(SELECT ref_10.varchar_array AS c0 FROM main.all_types AS ref_12 WHERE 0 LIMIT 149) AND 1) LIMIT 180))) ON (EXISTS(SELECT ref_8.double_array AS c0, ref_10."timestamp" AS c1, ref_8.uuid AS c2, ref_13.dec_9_4 AS c3, ref_13."int" AS c4, ref_13.timestamp_ns AS c5, ref_8."float" AS c6, 63 AS c7 FROM main.all_types AS ref_13 WHERE (ref_13."varchar" ~~ ref_9."varchar"))) ON ((ref_7."smallint" = ref_8."smallint")) INNER JOIN (SELECT ref_14."interval" AS c1 FROM main.all_types AS ref_14 WHERE ref_14."varchar") AS subq_3 ON (EXISTS(SELECT ref_15.utinyint AS c0 FROM main.all_types AS ref_15 WHERE (ref_9."varchar" ~~* (SELECT "varchar" FROM main.all_types LIMIT 1 OFFSET 4)))) ON ((ref_9."varchar" !~~ ref_6."varchar")) ON ((ref_3.timestamp_array = ref_6.timestamp_array))


# https://github.com/duckdb/duckdb-fuzzer/issues/1358
statement ok
SELECT subq_0.c6 AS c1, subq_0.c14 AS c2, subq_0.c7 AS c4, subq_0.c4 AS c5 FROM (SELECT (SELECT date FROM main.all_types LIMIT 1 OFFSET 6) AS c3, ref_2."time" AS c4, (SELECT uuid FROM main.all_types LIMIT 1 OFFSET 1) AS c5, ref_3.array_of_structs AS c6, CASE WHEN (((ref_0."varchar" !~~* ref_1."varchar") OR (ref_5."varchar" ~~~ ref_6."varchar"))) THEN (ref_2."bigint") ELSE ref_2."bigint" END AS c7, rtrim(CAST(CASE WHEN ((((ref_8."varchar" ~~~ ref_2."varchar") AND (ref_2."varchar" ~~~ ref_0."varchar")) OR (1 AND (ref_7."varchar" ~~ ref_2."varchar")))) THEN (ref_8."varchar") ELSE ref_8."varchar" END AS VARCHAR), CAST(ref_2."varchar" AS VARCHAR)) AS c9, ref_8.ubigint AS c14 FROM main.all_types AS ref_0 INNER JOIN main.all_types AS ref_1 INNER JOIN main.all_types AS ref_2 ON ((ref_1.int_array = ref_2.int_array)) ON (((ref_2."varchar" !~~ ref_1."varchar") OR (ref_1.blob IS NOT NULL))) INNER JOIN main.all_types AS ref_3 INNER JOIN main.all_types AS ref_4 RIGHT JOIN main.all_types AS ref_5 ON ((ref_4.dec_18_6 = ref_5.dec_18_6)) ON (ref_5."varchar") LEFT JOIN main.all_types AS ref_6 RIGHT JOIN main.all_types AS ref_7 INNER JOIN main.all_types AS ref_8 ON ((ref_7.timestamp_array = ref_8.timestamp_array)) ON (1) ON (NULL) ON ((ref_0.dec_18_6 = ref_3.dec_18_6)) WHERE (ref_5."varchar" ^@ (SELECT "varchar" FROM main.all_types LIMIT 1 OFFSET 6)) LIMIT 96) AS subq_0 WHERE subq_0.c9
17 changes: 11 additions & 6 deletions test/optimizer/statistics/statistics_between.test
View file
Original file line number Diff line number Diff line change
Expand Up @@ -62,12 +62,6 @@ EXPLAIN SELECT * FROM (SELECT * FROM integers LIMIT 10) integers(i) WHERE i BETW
----
logical_opt <REGEX>:.*EMPTY_RESULT.*

# wide between: both are always true, entire filter can be pruned
query II
EXPLAIN SELECT * FROM (SELECT * FROM integers LIMIT 10) integers(i) WHERE i BETWEEN 0 AND 10;
----
logical_opt <!REGEX>:.*FILTER.*

# lower clause is always true: between should be converted into i <= 2
query II
EXPLAIN SELECT * FROM (SELECT * FROM integers LIMIT 10) integers(i) WHERE i BETWEEN 0 AND 2;
Expand Down Expand Up @@ -178,3 +172,14 @@ SELECT * FROM (SELECT * FROM integers LIMIT 10) integers(i) WHERE i BETWEEN 0 AN
1
2
3

statement ok
PRAGMA explain_output = PHYSICAL_ONLY;

# wide between: both are always true, entire filter can be pruned. (happens during physical planning).
# see https://github.com/duckdb/duckdb-fuzzer/issues/1357
# https://github.com/duckdb/duckdb-fuzzer/issues/1358
query II
EXPLAIN SELECT * FROM (SELECT * FROM integers LIMIT 10) integers(i) WHERE i BETWEEN 0 AND 10;
----
physical_plan <!REGEX>:.*FILTER*

0 comments on commit 3603860

Please sign in to comment.