[SPARK-52235][SQL] Add implicit cast to DefaultValue V2 Expressions passed to DSV2 #50959
Conversation
| @@ -432,18 +432,34 @@ object ResolveDefaultColumns extends QueryErrorsBase | |||
| targetType: DataType, | |||
| colName: String): Option[Expression] = { | |||
| expr match { | |||
| case l: Literal if !Seq(targetType, l.dataType).exists(_ match { | |||
| case l: Literal => defaultValueFromWiderType(l, targetType, colName) | |||
There was a problem hiding this comment.
To explain this, the analyze() function in this class for strings does the following
- analyze/ checkAnalyze
- constant fold
- cast
However, because we add the rule in TypeCoercion (Analyzer phase), it runs before constant folding. Hence, relaxing the cast code to also take Expression instead of Literal.
|
|
||
| val alterExecCol3 = executeAndKeepPhysicalPlan[AlterTableExec] { | ||
| sql(s"ALTER TABLE $tableName ALTER COLUMN salary SET DEFAULT CAST(0 AS BOOLEAN)") | ||
| sql( |
There was a problem hiding this comment.
Some test cleanup, I didnt realize multiple ALTER COLUMN can be the same statement, so it was unnecessarily long
sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/analysis/TypeCoercionBase.scala
Show resolved
Hide resolved
sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/analysis/TypeCoercionBase.scala
Show resolved
Hide resolved
| column.newCurrentDefault() == expectedDefault, | ||
| s"Default value mismatch for column '${column.toString}': " + | ||
| s"expected $expectedDefault but found ${column.newCurrentDefault()}") | ||
| private def checkDefaultValues( |
There was a problem hiding this comment.
I guess you could define new checkDefaultValues using checkDefaultValue to minimize changes.
There was a problem hiding this comment.
i think i prefer to use the new method in all the place, as it makes the test significantly shorter (no need to check each change individually)
szehon-ho
force-pushed
the
cast_default_value
branch
from
a6750dd to
38717a6
Compare
|
rebased |
|
@cloud-fan can you take a look at this one? Thanks ! test failure is unrelated, re-running. |
sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/util/ResolveDefaultColumnsUtil.scala
Show resolved
Hide resolved
szehon-ho
force-pushed
the
cast_default_value
branch
from
38717a6 to
b1995d7
Compare
| try { | ||
| Option(casted.eval(EmptyRow)).map(Literal(_, targetType)) | ||
| } catch { | ||
| case e @ ( _: SparkThrowable | _: RuntimeException) => | ||
| logWarning(log"Failed to cast default value '${MDC(COLUMN_DEFAULT_VALUE, l)}' " + | ||
| case ex @ ( _: SparkThrowable | _: RuntimeException) => |
There was a problem hiding this comment.
to be safe, can we match all exceptions here? If anything gets wrong we just fail and say the type is incompatible.
There was a problem hiding this comment.
makes sense, thanks
| try { | ||
| Option(casted.eval(EmptyRow)).map(Literal(_, targetType)) | ||
| } catch { | ||
| case e @ ( _: SparkThrowable | _: RuntimeException) => | ||
| logWarning(log"Failed to cast default value '${MDC(COLUMN_DEFAULT_VALUE, l)}' " + | ||
| case ex: Exception => |
There was a problem hiding this comment.
| case ex: Exception => | |
| case NonFatal(e) => |
What changes were proposed in this pull request?
Add implicit cast to DefaultValue V2 expressions passed to DSV2, by adding rule to TypeCoercion.
Why are the changes needed?
Now default values are passed as V2 Expressions to DSV2 in Create/Replace/Alter table DDL statements.
We should match what the normal default value analysis path does for sql strings (ie, implicit cast).
Does this PR introduce any user-facing change?
No
How was this patch tested?
Add unit test to DatasourceV2DataFrameSuite
Was this patch authored or co-authored using generative AI tooling?
No