[CALCITE-7270] Add support for a DIVIDE_0_NULL operation #4653

mihaibudiu · 2025-11-27T19:42:40Z

https://issues.apache.org/jira/browse/CALCITE-7270

Signed-off-by: Mihai Budiu <mbudiu@feldera.com>

sonarqubecloud · 2025-11-28T14:21:13Z

Quality Gate passed

Issues
5 New issues
0 Accepted issues

Measures
0 Security Hotspots
43.5% Coverage on New Code
0.0% Duplication on New Code

See analysis details on SonarQube Cloud

xiedeyantu

I'm not familiar with this particular module, so I'm not entirely clear on the correct approach. I've only performed a basic check—it might require someone with expertise in this area to review it.

xiedeyantu · 2025-11-29T13:37:36Z

core/src/main/java/org/apache/calcite/runtime/SqlFunctions.java

  }

+  // nullable divide
+


We can remove this empty line.

xiedeyantu · 2025-11-29T13:51:32Z

core/src/test/resources/sql/misc.iq

 !use post
 !set outputformat mysql

+select CAST(5 AS DOUBLE) / 0 as a;


Mysql result is here. Was this done intentionally here?

+------+ | a | +------+ | NULL | +------+

this is terrible, it's against the IEEE 754 rules. we can probably emulate this behavior, but this means that there is yet another DIVISION operator besides the ones I have introduced...

I'm not quite sure I understand. Are you suggesting that MySQL's behavior might be problematic? However, when this SQL is executed in PostgreSQL, it directly throws an error (ERROR: division by zero).

Yes, MySQL is really weird.
5e0 / 0e0 does not conform to the IEEE 754 rules, where it should produce Infinity

I think you are correct. DuckDB also returns inf. You can use this link to check it.

duckdb> select CAST(5 AS DOUBLE) / 0 as a; ┌─────┐ │ a │ ╞═════╡ │ inf │ └─────┘ duckdb> select CAST(-5 AS DOUBLE) / 0 as a; ┌──────┐ │ a │ ╞══════╡ │ -inf │ └──────┘

For the record, Apache Hive also returns NULL whenever you divide by zero no matter the data type.

CREATE TABLE types ( cint INT, cbint BIGINT, cfloat FLOAT, cdouble DOUBLE, cdecimal DECIMAL(10, 2) ); INSERT INTO types VALUES (5, 5, 5.5, 5.5, 5.5); SELECT cint / 0 AS cint_div_zero, cbint / 0 AS cbint_div_zero, cfloat / 0 AS cfloat_div_zero, cdouble / 0 AS cdouble_div_zero, cdecimal / 0 AS cdecimal_div_zero

NULL NULL NULL NULL NULL

xiedeyantu · 2025-11-29T13:53:40Z

core/src/test/resources/sql/misc.iq

 !use scott-mysql

+# [CALCITE-7270] Add support for a DIVIDE_0_NULL operator
+select 5 / 0 as a;


This is not same as mysql result.

The link you submitted does not execute the same operation.

Sorry, I have fixed it. it also return NULL.

The duckdb return Inf, maybe I didn't understand what DIVIDE_0_NULL means.

It's supposed to return NULL for integer or decimal division by 0.
For double values it's supposed to follow IEEE 794.
But you see, there are actually lots of DIVISION operators.
Even MySQL has strict and non-strict modes.

core/src/test/resources/sql/misc.iq

xiedeyantu · 2025-12-02T00:52:08Z

core/src/main/java/org/apache/calcite/rex/RexSimplify.java

    case MINUS:
    case TIMES:
    case DIVIDE:
+    case DIVIDE_0_NULL:


Would it be better to use DIVIDE_ZERO_NULL?

silundong · 2025-12-02T10:12:37Z

core/src/main/java/org/apache/calcite/adapter/enumerable/RexImpTable.java

+      if (CHECKED_OPERATORS.contains(op)
+          || op.kind == SqlKind.DIVIDE_0_NULL
+          || op.kind == SqlKind.MOD_0_NULL
+          || op.kind == SqlKind.CHECKED_DIVIDE_0_NULL) {


This seems to target checked arithmetic. Are DIVIDE_0_NULL and MOD_0_NULL part of that? Can we add them to CHECKED_OPERATORS?

xiedeyantu · 2025-12-02T12:31:46Z

core/src/main/java/org/apache/calcite/sql/validate/SqlConformanceEnum.java

+    // sqlite, postgres, duckdb, mysql and MariaDB (non-strict mode).  Note that MYSQL is actually
+    // dynamically-typed, so Calcite cannot implement its behavior accurately, since Calcite is
+    // statically-typed.
+    case MYSQL_5:


Should we make LENIENT support this new division behavior as Calcite's default? Because the current behavior is inconsistent with MySQL (although MySQL might be incorrect). I tried to verify this across several databases: PostgreSQL throws an error directly, while MariaDB, SQLite and MySQL behave consistently. DuckDB actually differs from both MySQL and the current implementation as well.

zabetak · 2025-12-02T12:39:20Z

core/src/test/resources/sql/misc.iq

 !use post
 !set outputformat mysql

+select CAST(5 AS DOUBLE) / 0 as a;


For the record, Apache Hive also returns NULL whenever you divide by zero no matter the data type.

CREATE TABLE types ( cint INT, cbint BIGINT, cfloat FLOAT, cdouble DOUBLE, cdecimal DECIMAL(10, 2) ); INSERT INTO types VALUES (5, 5, 5.5, 5.5, 5.5); SELECT cint / 0 AS cint_div_zero, cbint / 0 AS cbint_div_zero, cfloat / 0 AS cfloat_div_zero, cdouble / 0 AS cdouble_div_zero, cdecimal / 0 AS cdecimal_div_zero

NULL NULL NULL NULL NULL

zabetak · 2025-12-02T12:49:44Z

core/src/main/java/org/apache/calcite/sql/SqlKind.java

+  /**
+   * Unchecked nullable version of DIVIDE, which produces NULL when dividing by zero.
+   */
+  DIVIDE_0_NULL,
+
+  /**
+   * Checked nullable version of DIVIDE, which produces NULL when dividing by zero.
+   */
+  CHECKED_DIVIDE_0_NULL,
+


Do we need to introduce a new SqlKind entry for each division variant? It's true that we have used this pattern before but maybe its not necessary in every case. In some cases the exact division peculiarities may not be that important so it could be fine to keep the existing SqlKind.DIVIDE.

zabetak · 2025-12-02T12:54:05Z

core/src/main/java/org/apache/calcite/sql/validate/SqlConformance.java

+   * Most SQL dialects will produce a runtime exception on division by zero,
+   * but some dialects return NULL instead (e.g. sqlite).
+   */
+  boolean nullableDivide();


Should we make this a default method otherwise this becomes a breaking change.

zabetak · 2025-12-02T13:01:58Z

core/src/main/java/org/apache/calcite/sql/type/SqlTypeTransforms.java

+   */
+  public static final SqlTypeTransform FORCE_NULLABLE_NON_FP =
+      (opBinding, typeToTransform) -> {
+        boolean nullable = !SqlTypeName.APPROX_TYPES.contains(typeToTransform.getSqlTypeName());


For approximate types nullable becomes false which will end up making the resulting type non-nullable which I don't think its correct. I have the impression that the result of the division is always nullable and this is also unrelated to the fact that we divide by zero or not.
Example

2.5/null null/2.5 (float) null/ (float) null

all these results inherit the nullability of the operands, but divide_0_null is always nullable

zabetak · 2025-12-02T13:04:54Z

core/src/main/java/org/apache/calcite/rex/RexSimplify.java

+    case DIVIDE_0_NULL:
+    case CHECKED_DIVIDE_0_NULL:


The fact that we don't have separate handling for DIVIDE_0_NULL and CHECKED_DIVIDE_0_NULL anywhere in the code is a hint that maybe we don't need separate entries in SqlKind.

zabetak · 2025-12-02T14:33:47Z

core/src/main/java/org/apache/calcite/adapter/enumerable/RexImpTable.java

      defineBinary(CHECKED_DIVIDE_INTEGER, DivideChecked, NullPolicy.STRICT, "checkedDivide");
      defineUnary(CHECKED_UNARY_MINUS, NegateChecked, NullPolicy.STRICT, "checkedUnaryMinus");
+      // nullable division
+      defineBinary(DIVIDE_0_NULL, Divide0Null, NullPolicy.SEMI_STRICT, "nullableDivide");


DIVIDE_0_NULL is already defined in line 836. Why do we need two implementors for the same operator?

zabetak · 2025-12-02T14:38:33Z

core/src/main/java/org/apache/calcite/sql/fun/SqlStdOperatorTable.java

+      // Unfortunately there cannot exist two functions in the standard operator
+      // table with the same exact name, so we need to use a different name,


Is this statement really true? How come we don't have the same problem for division?

zabetak · 2025-12-02T14:41:22Z

core/src/main/java/org/apache/calcite/sql/fun/SqlStdOperatorTable.java

+      // Unfortunately there cannot exist two functions in the standard operator
+      // table with the same exact name, so we need to use a different name,
+      // although in SQL this would be shown just as MOD
+      SqlBasicFunction.create("NULLABLE_MOD", SqlKind.MOD_0_NULL,


There is also PERCENT_REMAINDER with name % why did we pick "NULLABLE_MOD" as a name?

zabetak · 2025-12-02T14:43:16Z

core/src/main/java/org/apache/calcite/sql/fun/SqlStdOperatorTable.java

      // Return type is same as divisor (2nd operand)
      // SQL2003 Part2 Section 6.27, Syntax Rules 9
      SqlBasicFunction.create(SqlKind.MOD,
+          // A rather unfortunate name for this return type strategy


What's the problem with the name?

zabetak · 2025-12-02T14:49:48Z

core/src/main/java/org/apache/calcite/plan/Strong.java

+    map.put(SqlKind.DIVIDE_0_NULL, Policy.AS_IS);
+    map.put(SqlKind.MOD_0_NULL, Policy.AS_IS);


We could override the SqlOperator.getStrongPolicyInference for the new operators and avoid the need for introducing a new SqlKind entry.

Should we also do something about CHECKED_DIVIDE_0_NULL?

github-actions · 2026-01-02T03:55:00Z

This pull request has been marked as stale due to 30 days of inactivity. It will be closed in 90 days if no further activity occurs. If you think that’s incorrect or this pull request requires a review, please simply write any comment. If closed, you can revive the PR at any time and @mention a reviewer or discuss it on the dev@calcite.apache.org list. Thank you for your contributions.

mihaibudiu force-pushed the issue7270 branch 8 times, most recently from bb01644 to 7fb33a6 Compare November 28, 2025 03:51

[CALCITE-7270] Add support for a DIVIDE_0_NULL operation

5d687ce

Signed-off-by: Mihai Budiu <mbudiu@feldera.com>

mihaibudiu force-pushed the issue7270 branch from 7fb33a6 to 5d687ce Compare November 28, 2025 04:31

xiedeyantu reviewed Nov 29, 2025

View reviewed changes

xiedeyantu reviewed Dec 2, 2025

View reviewed changes

silundong reviewed Dec 2, 2025

View reviewed changes

xiedeyantu reviewed Dec 2, 2025

View reviewed changes

zabetak reviewed Dec 2, 2025

View reviewed changes

github-actions bot added the stale label Jan 2, 2026

		// Unfortunately there cannot exist two functions in the standard operator
		// table with the same exact name, so we need to use a different name,

		map.put(SqlKind.DIVIDE_0_NULL, Policy.AS_IS);
		map.put(SqlKind.MOD_0_NULL, Policy.AS_IS);

[CALCITE-7270] Add support for a DIVIDE_0_NULL operation #4653

Are you sure you want to change the base?

[CALCITE-7270] Add support for a DIVIDE_0_NULL operation #4653

Uh oh!

Conversation

mihaibudiu commented Nov 27, 2025

Uh oh!

sonarqubecloud bot commented Nov 28, 2025

Quality Gate passed

Uh oh!

xiedeyantu left a comment

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

xiedeyantu Dec 2, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

mihaibudiu Dec 2, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

github-actions bot commented Jan 2, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

xiedeyantu Dec 2, 2025 •

edited

Loading

mihaibudiu Dec 2, 2025 •

edited

Loading