org.apache.spark.sql.catalyst.analysis.HiveTypeCoercion
DecimalPrecision
Name for this rule, automatically inferred based on class name.
Name for this rule, automatically inferred based on class name.
Calculates and propagates precision for fixed-precision decimals. Hive has a number of rules for this based on the SQL standard and MS SQL: https://cwiki.apache.org/confluence/download/attachments/27362075/Hive_Decimal_Precision_Scale_Support.pdf
In particular, if we have expressions e1 and e2 with precision/scale p1/s2 and p2/s2 respectively, then the following operations have the following precision / scale:
Operation Result Precision Result Scale ------------------------------------------------------------------------ e1 + e2 max(s1, s2) + max(p1-s1, p2-s2) + 1 max(s1, s2) e1 - e2 max(s1, s2) + max(p1-s1, p2-s2) + 1 max(s1, s2) e1 * e2 p1 + p2 + 1 s1 + s2 e1 / e2 p1 - s1 + s2 + max(6, s1 + p2 + 1) max(6, s1 + p2 + 1) e1 % e2 min(p1-s1, p2-s2) + max(s1, s2) max(s1, s2) sum(e1) p1 + 10 s1 avg(e1) p1 + 4 s1 + 4
Catalyst also has unlimited-precision decimals. For those, all ops return unlimited precision.
To implement the rules for fixed-precision types, we introduce casts to turn them to unlimited precision, do the math on unlimited-precision numbers, then introduce casts back to the required fixed precision. This allows us to do all rounding and overflow handling in the cast-to-fixed-precision operator.
In addition, when mixing non-decimal types with decimals, we use the following rules: - BYTE gets turned into DECIMAL(3, 0) - SHORT gets turned into DECIMAL(5, 0) - INT gets turned into DECIMAL(10, 0) - LONG gets turned into DECIMAL(20, 0) - FLOAT and DOUBLE cause fixed-length decimals to turn into DOUBLE (this is the same as Hive, but note that unlimited decimals are considered bigger than doubles in WidenTypes)