Repository navigation
Update REPLACE scalar function to support Utf8View #11913
Description
Currently, a call to REPLACE with a Utf8View datatypes induces a cast. After the change that fixes this issue, it should not.
query TT
EXPLAIN SELECT
REPLACE(column1_utf8view, 'foo', 'bar') as c1,
REPLACE(column1_utf8view, column2_utf8view, 'bar') as c2
FROM test;
----
logical_plan
01)Projection: replace(__common_expr_1, Utf8("foo"), Utf8("bar")) AS c1, replace(__common_expr_1, CAST(test.column2_utf8view AS Utf8), Utf8("bar")) AS c2
02)--Projection: CAST(test.column1_utf8view AS Utf8) AS __common_expr_1, test.column2_utf8view
03)----TableScan: test projection=[column1_utf8view, column2_utf8view]REPLACE is defined here: https://github.com/apache/datafusion/blob/e088945c38b74bb1d86dcbb88a69dfc21d59e375/datafusion/functions/src/string/replace.rs
casting tests are in: https://github.com/apache/datafusion/blob/main/datafusion/sqllogictest/test_files/string_view.slt
Is your feature request related to a problem or challenge?
We are working to add complete StringView support in DataFusion, which permits potentially much faster processing of string data. See #10918 for more background.
Today, most DataFusion string functions support DataType::Utf8 and DataType::LargeUtf8 and when called with a StringView argument DataFusion will cast the argument back to DataType::Utf8 which is expensive.
To realize the full speed of StringView, we need to ensure that all string functions support the DataType::Utf8View directly.
Describe the solution you'd like
Update the function to support DataType::Utf8View directly
Describe alternatives you've considered
The typical steps are:
- Write some tests showing the function doesn't support Utf8View (see the tests in
string_view.sltto ensure the arguments are not being cast - Change the
Signatureof the function to acceptUtf8Viewin addition toUtf8/LargeUtf8 - Update the implementation of the function to operate on
Utf8View
Example PRs
- Update to use an arrow kernel that already supports StringView: feat: support
Utf8Viewtype instarts_withfunction #11787 - Change the implementation to support StringView directly: Implement native support StringView for character length #11676
- Change implementation (option 2): Initial support for regex_replace on
StringViewArray#11556
Additional context
The documentation of string functions can be found here: https://datafusion.apache.org/user-guide/sql/scalar_functions.html#string-functions
To test a function with StringView with datafusion-cli you can use an example like this (replacing starts_with with the relevant function)
> create table foo as values (arrow_cast('foo', 'Utf8View'), arrow_cast('bar', 'Utf8View'));
0 row(s) fetched.
Elapsed 0.043 seconds.
> select starts_with(column1, column2) from foo;
+--------------------------------------+
| starts_with(foo.column1,foo.column2) |
+--------------------------------------+
| false |
+--------------------------------------+
1 row(s) fetched.
Elapsed 0.015 seconds.To see if it is using utf8 view, use EXPLAIN to see the plan and verify there is no CAST. In this example the CAST(column1@0 AS Utf8) indicates that the function is not using Utf8View natively
> explain select starts_with(column1, column2) from foo;
+---------------+------------------------------------------------------------------------------------------------------------------------------+
| plan_type | plan |
+---------------+------------------------------------------------------------------------------------------------------------------------------+
| logical_plan | Projection: starts_with(CAST(foo.column1 AS Utf8), CAST(foo.column2 AS Utf8)) |
| | TableScan: foo projection=[column1, column2] |
| physical_plan | ProjectionExec: expr=[starts_with(CAST(column1@0 AS Utf8), CAST(column2@1 AS Utf8)) as starts_with(foo.column1,foo.column2)] |
| | MemoryExec: partitions=1, partition_sizes=[1] |
| | |
+---------------+------------------------------------------------------------------------------------------------------------------------------+
2 row(s) fetched.
Elapsed 0.006 seconds.It is also often good to test with a constant as well (likewise there should be no cast):
> explain select starts_with(column1, 'foo') from foo;
+---------------+----------------------------------------------------------------------------------------------------------+
| plan_type | plan |
+---------------+----------------------------------------------------------------------------------------------------------+
| logical_plan | Projection: starts_with(CAST(foo.column1 AS Utf8), Utf8("foo")) |
| | TableScan: foo projection=[column1] |
| physical_plan | ProjectionExec: expr=[starts_with(CAST(column1@0 AS Utf8), foo) as starts_with(foo.column1,Utf8("foo"))] |
| | MemoryExec: partitions=1, partition_sizes=[1] |
| | |
+---------------+----------------------------------------------------------------------------------------------------------+
2 row(s) fetched.
Elapsed 0.002 seconds.
### Additional context
_No response_
Activity
thinh2 commentedon Aug 10, 2024
take
thinh2 commentedon Aug 15, 2024
Hi @tshauck, after updating the replace.rs to process Utf8View, the sql logic test still use the CAST in its logical plan. Do you have any suggestion on how to debug this issue? As I am a newbie, any documents, tips will be helpful. I also created a draft pull request above.