From c621df269d8bc5c9487cab94407a7186253aa590 Mon Sep 17 00:00:00 2001 From: Gengliang Wang Date: Wed, 13 Jul 2022 00:18:23 -0700 Subject: [PATCH] [SPARK-39749][SQL] Always use plain string representation on casting Decimal to String ### What changes were proposed in this pull request? Currently, casting decimal as string type will result in Strings with exponential notations if the adjusted exponent is less than -6. This is consistent with BigDecimal.toString https://docs.oracle.com/javase/8/docs/api/java/math/BigDecimal.html#toString After this PR, the casting always uses plain string representation. ### Why are the changes needed? 1. The current behavior doesn't compliant to the ANSI SQL standard. image image 2. It is different from databases like PostgreSQL/Oracle/MS SQL server/etc. 3. The current behavior may surprise users since it only happens when the adjusted exponent is less than -6. The following query will return `false` by default (when ANSI SQL mode is off) since the `0.0000000123` is converted as `1.23E-7`: ```sql select '0.000000123' in (0.000000123); ``` ### Does this PR introduce _any_ user-facing change? Yes, after changes, Spark SQL always uses plain string representation on casting Decimal to String. To restore the legacy behavior, which uses scientific notation if the adjusted exponent is less than -6, set `spark.sql.legacy.castDecimalToString.enabled` to `true`. ### How was this patch tested? Unit test Closes #37160 from gengliangwang/decimalToString. Authored-by: Gengliang Wang Signed-off-by: Gengliang Wang --- docs/sql-migration-guide.md | 1 + .../apache/spark/sql/catalyst/expressions/Cast.scala | 5 +++++ .../scala/org/apache/spark/sql/internal/SQLConf.scala | 11 +++++++++++ .../scala/org/apache/spark/sql/types/Decimal.scala | 2 ++ .../sql/catalyst/expressions/CastSuiteBase.scala | 8 ++++++++ 5 files changed, 27 insertions(+) diff --git a/docs/sql-migration-guide.md b/docs/sql-migration-guide.md index c4e9d1f2a2569..e37e12f71ebb1 100644 --- a/docs/sql-migration-guide.md +++ b/docs/sql-migration-guide.md @@ -26,6 +26,7 @@ license: | - Since Spark 3.4, Number or Number(\*) from Teradata will be treated as Decimal(38,18). In Spark 3.3 or earlier, Number or Number(\*) from Teradata will be treated as Decimal(38, 0), in which case the fractional part will be removed. - Since Spark 3.4, v1 database, table, permanent view and function identifier will include 'spark_catalog' as the catalog name if database is defined, e.g. a table identifier will be: `spark_catalog.default.t`. To restore the legacy behavior, set `spark.sql.legacy.v1IdentifierNoCatalog` to `true`. + - Since Spark 3.4, the results of casting Decimal values as String type will not contain exponential notations. To restore the legacy behavior, which uses scientific notation if the adjusted exponent is less than -6, set `spark.sql.legacy.castDecimalToString.enabled` to `true`. ## Upgrading from Spark SQL 3.2 to 3.3 diff --git a/sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/expressions/Cast.scala b/sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/expressions/Cast.scala index 45950607e0de9..5dd986e25e7b8 100644 --- a/sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/expressions/Cast.scala +++ b/sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/expressions/Cast.scala @@ -512,6 +512,7 @@ case class Cast( TimestampFormatter.getFractionFormatter(ZoneOffset.UTC) private val legacyCastToStr = SQLConf.get.getConf(SQLConf.LEGACY_COMPLEX_TYPES_TO_STRING) + private val legacyCastDecimalToStr = SQLConf.get.getConf(SQLConf.LEGACY_DECIMAL_TO_STRING) // The brackets that are used in casting structs and maps to strings private val (leftBracket, rightBracket) = if (legacyCastToStr) ("[", "]") else ("{", "}") @@ -625,6 +626,8 @@ case class Cast( case DayTimeIntervalType(startField, endField) => buildCast[Long](_, i => UTF8String.fromString( IntervalUtils.toDayTimeIntervalString(i, ANSI_STYLE, startField, endField))) + case _: DecimalType if !legacyCastDecimalToStr => + buildCast[Decimal](_, d => UTF8String.fromString(d.toPlainString)) case _ => buildCast[Any](_, o => UTF8String.fromString(o.toString)) } @@ -1475,6 +1478,8 @@ case class Cast( $evPrim = UTF8String.fromString($iu.toDayTimeIntervalString($c, $style, (byte)${i.startField}, (byte)${i.endField})); """ + case _: DecimalType if !legacyCastDecimalToStr => + (c, evPrim, _) => code"$evPrim = UTF8String.fromString($c.toPlainString());" case _ => (c, evPrim, evNull) => code"$evPrim = UTF8String.fromString(String.valueOf($c));" } diff --git a/sql/catalyst/src/main/scala/org/apache/spark/sql/internal/SQLConf.scala b/sql/catalyst/src/main/scala/org/apache/spark/sql/internal/SQLConf.scala index 1b7857ead59fa..15bd5cb5e365c 100644 --- a/sql/catalyst/src/main/scala/org/apache/spark/sql/internal/SQLConf.scala +++ b/sql/catalyst/src/main/scala/org/apache/spark/sql/internal/SQLConf.scala @@ -3697,6 +3697,17 @@ object SQLConf { .booleanConf .createWithDefault(false) + val LEGACY_DECIMAL_TO_STRING = + buildConf("spark.sql.legacy.castDecimalToString.enabled") + .internal() + .doc("When true, casting decimal values as string will use scientific notation if an " + + "exponent is needed, which is the same with the method java.math.BigDecimal.toString(). " + + "Otherwise, the casting result won't contain an exponent field, which is compliant to " + + "the ANSI SQL standard.") + .version("3.4.0") + .booleanConf + .createWithDefault(false) + val LEGACY_PATH_OPTION_BEHAVIOR = buildConf("spark.sql.legacy.pathOptionBehavior.enabled") .internal() diff --git a/sql/catalyst/src/main/scala/org/apache/spark/sql/types/Decimal.scala b/sql/catalyst/src/main/scala/org/apache/spark/sql/types/Decimal.scala index 759a5dce967bf..f4f54d2f93bcd 100644 --- a/sql/catalyst/src/main/scala/org/apache/spark/sql/types/Decimal.scala +++ b/sql/catalyst/src/main/scala/org/apache/spark/sql/types/Decimal.scala @@ -225,6 +225,8 @@ final class Decimal extends Ordered[Decimal] with Serializable { override def toString: String = toBigDecimal.toString() + def toPlainString: String = toBigDecimal.bigDecimal.toPlainString + def toDebugString: String = { if (decimalVal.ne(null)) { s"Decimal(expanded, $decimalVal, $precision, $scale)" diff --git a/sql/catalyst/src/test/scala/org/apache/spark/sql/catalyst/expressions/CastSuiteBase.scala b/sql/catalyst/src/test/scala/org/apache/spark/sql/catalyst/expressions/CastSuiteBase.scala index 97cbc781829ef..da9a7dca9f158 100644 --- a/sql/catalyst/src/test/scala/org/apache/spark/sql/catalyst/expressions/CastSuiteBase.scala +++ b/sql/catalyst/src/test/scala/org/apache/spark/sql/catalyst/expressions/CastSuiteBase.scala @@ -1305,4 +1305,12 @@ abstract class CastSuiteBase extends SparkFunSuite with ExpressionEvalHelper { Cast(child, DecimalType.USER_DEFAULT), it) } } + + test("SPARK-39749: cast Decimal to string") { + val input = Literal.create(Decimal(0.000000123), DecimalType(9, 9)) + checkEvaluation(cast(input, StringType), "0.000000123") + withSQLConf(SQLConf.LEGACY_DECIMAL_TO_STRING.key -> "true") { + checkEvaluation(cast(input, StringType), "1.23E-7") + } + } }