Type conversions (xitongsys#456)

* add row transform reader/writer

* cleanup

* use sentinel error in transforms to indicate that the output buffer is too short

* add row index to filter function

* add row index to scan function

* remove row index from transforms and filters

* refactor row layout

* simplify setMakeSlice

* simplify transform API

* refactor and enhance conversion API using new row layout

* cleanup debug printf

* remove unused method

* remove debug t.Logf

* use direct function calls in parquet.Type implementations instead of relying on the global variables

* implement type conversions between physical types

* implement type conversions between logical types

* revert changes to intType.Compare to fix preformance regression

* test type conversions

* refuse conversion between enum/json/bson and other types

* add conversions between date/time/string

* add timestampt conversions

* PR feedback: rename functions to be more descriptive

* move isZero to parquet.go so it exists on Go 1.17

deprecated/int96.go

@@ -9,6 +9,29 @@ import (
 // Int96 is an implementation of the deprecated INT96 parquet type.
 type Int96 [3]uint32
 
+// Int32ToInt96 converts a int32 value to a Int96.
+func Int32ToInt96(value int32) (i96 Int96) {
+	if value < 0 {
+		i96[2] = 0xFFFFFFFF
+		i96[1] = 0xFFFFFFFF
+	}
+	i96[0] = uint32(value)
+	return
+}
+
+// Int64ToInt96 converts a int64 value to Int96.
+func Int64ToInt96(value int64) (i96 Int96) {
+	if value < 0 {
+		i96[2] = 0xFFFFFFFF
+	}
+	i96[1] = uint32(value >> 32)
+	i96[0] = uint32(value)
+	return
+}
+
+// IsZero returns true if i is the zero-value.
+func (i Int96) IsZero() bool { return i == Int96{} }
+
 // Negative returns true if i is a negative value.
 func (i Int96) Negative() bool {
 	return (i[2] >> 31) != 0
@@ -48,6 +71,16 @@ func (i Int96) Int() *big.Int {
 	return z
 }
 
+// Int32 converts i to a int32, potentially truncating the value.
+func (i Int96) Int32() int32 {
+	return int32(i[0])
+}
+
+// Int64 converts i to a int64, potentially truncating the value.
+func (i Int96) Int64() int64 {
+	return int64(i[1])<<32 | int64(i[0])
+}
+
 // String returns a string representation of i.
 func (i Int96) String() string {
 	return i.Int().String()

errors.go

@@ -54,6 +54,11 @@ var (
 	// ErrTooManyRowGroups is returned when attempting to generate a parquet
 	// file with more than MaxRowGroups row groups.
 	ErrTooManyRowGroups = errors.New("the limit of 32767 row groups has been reached")
+
+	// ErrConversion is used to indicate that a conversion betwen two values
+	// cannot be done because there are no rules to translate between their
+	// physical types.
+	ErrInvalidConversion = errors.New("invalid conversion between parquet values")
 )
 
 type errno int

null.go

@@ -113,12 +113,3 @@ func nullIndexFuncOfByteArray(n int) nullIndexFunc {
 		}
 	}
 }
-
-func isZero(b []byte) bool {
-	for _, c := range b {
-		if c != 0 {
-			return false
-		}
-	}
-	return true
-}

parquet.go

@@ -42,3 +42,12 @@ func typeNameOf(t reflect.Type) string {
 	}
 	return s1 + " (" + s2 + ")"
 }
+
+func isZero(b []byte) bool {
+	for _, c := range b {
+		if c != 0 {
+			return false
+		}
+	}
+	return true
+}