Add int32 support and change format of castVarchar (#26)

* add support to int32 type * fix fortmat of cast varchar
oap-project · Feb 28, 2022 · 9a4876e · 9a4876e
1 parent 6799b38
commit 9a4876e
Show file tree

Hide file tree

Showing 3 changed files with 32 additions and 4 deletions.
diff --git a/cpp/src/gandiva/function_registry_datetime.cc b/cpp/src/gandiva/function_registry_datetime.cc
@@ -164,12 +164,21 @@ std::vector<NativeFunction> GetDateTimeFunctionRegistry() {
       NativeFunction("seconds_to_timestamp", {}, DataTypeVector{int64()}, timestampusutc(),
                      kResultNullIfNull, "seconds_to_timestamp_int64"),
 
+      NativeFunction("seconds_to_timestamp", {}, DataTypeVector{int32()}, timestampusutc(),
+                     kResultNullIfNull, "seconds_to_timestamp_int32"),
+
       NativeFunction("millis_to_timestamp", {}, DataTypeVector{int64()}, timestampusutc(),
                      kResultNullIfNull, "millis_to_timestamp_int64"),
+
+      NativeFunction("millis_to_timestamp", {}, DataTypeVector{int32()}, timestampusutc(),
+                     kResultNullIfNull, "millis_to_timestamp_int32"),
 
       NativeFunction("micros_to_timestamp", {}, DataTypeVector{int64()}, timestampusutc(),
                      kResultNullIfNull, "micros_to_timestamp_int64"),
 
+      NativeFunction("micros_to_timestamp", {}, DataTypeVector{int32()}, timestampusutc(),
+                     kResultNullIfNull, "micros_to_timestamp_int32"),
+
       NativeFunction("date_diff", {}, DataTypeVector{date32(), date32()}, int32(),
                      kResultNullIfNull, "micros_to_timestamp_date32_date32"),
 

diff --git a/cpp/src/gandiva/precompiled/arithmetic_ops.cc b/cpp/src/gandiva/precompiled/arithmetic_ops.cc
@@ -300,10 +300,14 @@ CAST_UNARY_UTF8(castVARCHAR, int8, utf8, "%d", NOFMT, nothing)
 CAST_UNARY_UTF8(castVARCHAR, int16, utf8, "%d", NOFMT, nothing)
 CAST_UNARY_UTF8(castVARCHAR, int32, utf8, "%d", NOFMT, nothing)
 CAST_UNARY_UTF8(castVARCHAR, int64, utf8, "%ld", NOFMT, nothing)
-// CAST_UNARY_UTF8(castVARCHAR, float32, utf8, "%.*f", FMT, FLT_DIG)
-// CAST_UNARY_UTF8(castVARCHAR, float64, utf8, "%.*f", FMT, DBL_DIG)
-CAST_UNARY_UTF8(castVARCHAR, float32, utf8, "%g", NOFMT, nothing)
-CAST_UNARY_UTF8(castVARCHAR, float64, utf8, "%g", NOFMT, nothing)
+// We set to display the full precision of Float/Double, but this may still behave 
+// differently with vanilla spark.
+CAST_UNARY_UTF8(castVARCHAR, float32, utf8, "%.*f", FMT, FLT_DIG)
+CAST_UNARY_UTF8(castVARCHAR, float64, utf8, "%.*f", FMT, DBL_DIG)
+// %g means converting floating-point number depending on the value and the precision. 
+// It bahaves similar to the shortest representation.
+// CAST_UNARY_UTF8(castVARCHAR, float32, utf8, "%g", NOFMT, nothing)
+// CAST_UNARY_UTF8(castVARCHAR, float64, utf8, "%g", NOFMT, nothing)
 
 #undef CAST_UNARY_UTF8
 

diff --git a/cpp/src/gandiva/precompiled/time.cc b/cpp/src/gandiva/precompiled/time.cc
@@ -990,16 +990,31 @@ gdv_timestamp seconds_to_timestamp_int64(gdv_int64 in) {
   return in * 1000000;
 }
 
+FORCE_INLINE
+gdv_timestamp seconds_to_timestamp_int32(gdv_int32 in) {
+  return (int64_t)in * 1000000;
+}
+
 FORCE_INLINE
 gdv_timestamp millis_to_timestamp_int64(gdv_int64 in) {
   return in * 1000;
 }
 
+FORCE_INLINE
+gdv_timestamp millis_to_timestamp_int32(gdv_int32 in) {
+  return (int64_t)in * 1000;
+}
+
 FORCE_INLINE
 gdv_timestamp micros_to_timestamp_int64(gdv_int64 in) {
   return in;
 }
 
+FORCE_INLINE
+gdv_timestamp micros_to_timestamp_int32(gdv_int32 in) {
+  return (int64_t)in;
+}
+
 FORCE_INLINE
 gdv_int32 micros_to_timestamp_date32_date32(gdv_date32 left, gdv_date32 right) {
   return left - right;