// Convert converts the val with type tp. func Convert(val interface{}, target *FieldType) (v interface{}, err error) { //NTYPE tp := target.Tp if val == nil { return nil, nil } switch tp { // TODO: implement mysql types convert when "CAST() AS" syntax are supported. case mysql.TypeFloat: x, err := ToFloat64(val) if err != nil { return invConv(val, tp) } if target.Flen != UnspecifiedLength { x, err = TruncateFloat(x, target.Flen, target.Decimal) if err != nil { return nil, errors.Trace(err) } } return float32(x), nil case mysql.TypeDouble: x, err := ToFloat64(val) if err != nil { return invConv(val, tp) } if target.Flen != UnspecifiedLength { x, err = TruncateFloat(x, target.Flen, target.Decimal) if err != nil { return nil, errors.Trace(err) } } return float64(x), nil case mysql.TypeBlob, mysql.TypeTinyBlob, mysql.TypeMediumBlob, mysql.TypeLongBlob, mysql.TypeString, mysql.TypeVarchar, mysql.TypeVarString: x, err := ToString(val) if err != nil { return invConv(val, tp) } // TODO: consider target.Charset/Collate x = truncateStr(x, target.Flen) if target.Charset == charset.CharsetBin { return []byte(x), nil } return x, nil case mysql.TypeDuration: fsp := mysql.DefaultFsp if target.Decimal != UnspecifiedLength { fsp = target.Decimal } switch x := val.(type) { case mysql.Duration: return x.RoundFrac(fsp) case mysql.Time: t, err := x.ConvertToDuration() if err != nil { return nil, errors.Trace(err) } return t.RoundFrac(fsp) case string: return mysql.ParseDuration(x, fsp) default: return invConv(val, tp) } case mysql.TypeTimestamp, mysql.TypeDatetime, mysql.TypeDate: fsp := mysql.DefaultFsp if target.Decimal != UnspecifiedLength { fsp = target.Decimal } switch x := val.(type) { case mysql.Time: t, err := x.Convert(tp) if err != nil { return nil, errors.Trace(err) } return t.RoundFrac(fsp) case mysql.Duration: t, err := x.ConvertToTime(tp) if err != nil { return nil, errors.Trace(err) } return t.RoundFrac(fsp) case string: return mysql.ParseTime(x, tp, fsp) case int64: return mysql.ParseTimeFromNum(x, tp, fsp) default: return invConv(val, tp) } case mysql.TypeTiny, mysql.TypeShort, mysql.TypeInt24, mysql.TypeLong, mysql.TypeLonglong: unsigned := mysql.HasUnsignedFlag(target.Flag) if unsigned { return convertToUint(val, target) } return convertToInt(val, target) case mysql.TypeBit: x, err := convertToUint(val, target) if err != nil { return x, errors.Trace(err) } // check bit boundary, if bit has n width, the boundary is // in [0, (1 << n) - 1] width := target.Flen if width == 0 { width = mysql.MinBitWidth } else if width == mysql.UnspecifiedBitWidth { width = mysql.MaxBitWidth } maxValue := uint64(1)<<uint64(width) - 1 if x > maxValue { return maxValue, overflow(val, tp) } return x, nil case mysql.TypeDecimal, mysql.TypeNewDecimal: x, err := ToDecimal(val) if err != nil { return invConv(val, tp) } if target.Decimal != UnspecifiedLength { x = x.Round(int32(target.Decimal)) } // TODO: check Flen return x, nil case mysql.TypeYear: var ( intVal int64 err error ) switch x := val.(type) { case string: intVal, err = StrToInt(x) case mysql.Time: return int64(x.Year()), nil case mysql.Duration: return int64(time.Now().Year()), nil default: intVal, err = ToInt64(x) } if err != nil { return invConv(val, tp) } y, err := mysql.AdjustYear(int(intVal)) if err != nil { return invConv(val, tp) } return int64(y), nil default: panic("should never happen") } }
func (c *Col) castIntegerValue(val int64, errCode int) (casted interface{}, err error) { unsigned := mysql.HasUnsignedFlag(c.Flag) var overflow bool switch c.Tp { case mysql.TypeTiny: if unsigned { if val > math.MaxUint8 || errCode == errCodeOverflowUpper { overflow = true casted = uint8(math.MaxUint8) } else if val < 0 || errCode == errCodeOverflowLower { overflow = true casted = uint8(0) } else { casted = uint8(val) } } else { if val > math.MaxInt8 || errCode == errCodeOverflowUpper { overflow = true casted = int8(math.MaxInt8) } else if val < math.MinInt8 || errCode == errCodeOverflowLower { overflow = true casted = int8(math.MinInt8) } else { casted = int8(val) } } case mysql.TypeShort: if unsigned { if val > math.MaxUint16 || errCode == errCodeOverflowUpper { overflow = true casted = uint16(math.MaxUint16) } else if val < 0 || errCode == errCodeOverflowLower { overflow = true casted = uint16(0) } else { casted = uint16(val) } } else { if val > math.MaxInt16 || errCode == errCodeOverflowUpper { overflow = true casted = int16(math.MaxInt16) } else if val < math.MinInt16 || errCode == errCodeOverflowLower { overflow = true casted = int16(math.MinInt16) } else { casted = int16(val) } } case mysql.TypeYear: if val > int64(mysql.MaxYear) || errCode == errCodeOverflowUpper { overflow = true casted = mysql.MaxYear } else if val < int64(mysql.MinYear) { overflow = true casted = mysql.MinYear } else { casted, _ = mysql.AdjustYear(int(val)) } case mysql.TypeInt24: if unsigned { if val < 0 || errCode == errCodeOverflowLower { overflow = true casted = uint32(0) } else if val > 1<<24-1 || errCode == errCodeOverflowUpper { overflow = true casted = uint32(1<<24 - 1) } else { casted = uint32(val) } } else { if val > 1<<23-1 || errCode == errCodeOverflowUpper { overflow = true casted = int32(1<<23 - 1) } else if val < -1<<23 || errCode == errCodeOverflowLower { overflow = true casted = int32(-1 << 23) } else { casted = int32(val) } } case mysql.TypeLong: if unsigned { if val > math.MaxUint32 || errCode == errCodeOverflowUpper { overflow = true casted = uint32(math.MaxUint32) } else if (val < 0 && errCode != errCodeOverflowMaxInt64) || errCode == errCodeOverflowLower { overflow = true casted = uint32(0) } else { casted = uint32(val) } } else { if val > math.MaxInt32 || errCode == errCodeOverflowUpper { overflow = true casted = int32(math.MaxInt32) } else if val < math.MinInt32 || errCode == errCodeOverflowLower { overflow = true casted = int32(math.MinInt32) } else { casted = int32(val) } } case mysql.TypeLonglong: // TypeLonglong overflow has already been handled by normalizeInteger if unsigned { if errCode == errCodeOverflowUpper { overflow = true casted = uint64(math.MaxUint64) } else if (val < 0 && errCode != errCodeOverflowMaxInt64) || errCode == errCodeOverflowLower { overflow = true casted = uint64(0) } else { casted = uint64(val) } } else { if errCode == errCodeOverflowUpper { overflow = true casted = int64(math.MaxInt64) } else if errCode == errCodeOverflowLower { overflow = true casted = int64(math.MinInt64) } else { casted = int64(val) } } case mysql.TypeBit: // Convert bit as uint64 if errCode == errCodeOverflowUpper { overflow = true casted = uint64(math.MaxUint64) } else if (val < 0 && errCode != errCodeOverflowMaxInt64) || errCode == errCodeOverflowLower { overflow = true casted = uint64(0) } else { casted = uint64(val) } } if overflow { err = types.Overflow(val, c.Tp) } return }
// Convert converts the val with type tp. func Convert(val interface{}, target *FieldType) (v interface{}, err error) { //NTYPE tp := target.Tp if val == nil { return nil, nil } switch tp { // TODO: implement mysql types convert when "CAST() AS" syntax are supported. case mysql.TypeFloat: x, err := ToFloat64(val) if err != nil { return InvConv(val, tp) } if target.Flen != UnspecifiedLength { x, err = TruncateFloat(x, target.Flen, target.Decimal) if err != nil { return nil, errors.Trace(err) } } return float32(x), nil case mysql.TypeDouble: x, err := ToFloat64(val) if err != nil { return InvConv(val, tp) } if target.Flen != UnspecifiedLength { x, err = TruncateFloat(x, target.Flen, target.Decimal) if err != nil { return nil, errors.Trace(err) } } return float64(x), nil case mysql.TypeString: x, err := ToString(val) if err != nil { return InvConv(val, tp) } // TODO: consider target.Charset/Collate x = truncateStr(x, target.Flen) if target.Charset == charset.CharsetBin { return []byte(x), nil } return x, nil case mysql.TypeBlob: x, err := ToString(val) if err != nil { return InvConv(val, tp) } x = truncateStr(x, target.Flen) if target.Charset == charset.CharsetBin { return []byte(x), nil } return x, nil case mysql.TypeDuration: fsp := mysql.DefaultFsp if target.Decimal != UnspecifiedLength { fsp = target.Decimal } switch x := val.(type) { case mysql.Duration: return x.RoundFrac(fsp) case mysql.Time: t, err := x.ConvertToDuration() if err != nil { return nil, errors.Trace(err) } return t.RoundFrac(fsp) case string: return mysql.ParseDuration(x, fsp) default: return InvConv(val, tp) } case mysql.TypeTimestamp, mysql.TypeDatetime, mysql.TypeDate: fsp := mysql.DefaultFsp if target.Decimal != UnspecifiedLength { fsp = target.Decimal } switch x := val.(type) { case mysql.Time: t, err := x.Convert(tp) if err != nil { return nil, errors.Trace(err) } return t.RoundFrac(fsp) case mysql.Duration: t, err := x.ConvertToTime(tp) if err != nil { return nil, errors.Trace(err) } return t.RoundFrac(fsp) case string: return mysql.ParseTime(x, tp, fsp) default: return InvConv(val, tp) } case mysql.TypeLonglong: x, err := ToInt64(val) if err != nil { return InvConv(val, tp) } // TODO: We should first convert to uint64 then check unsigned flag. if mysql.HasUnsignedFlag(target.Flag) { return uint64(x), nil } return x, nil case mysql.TypeNewDecimal: x, err := ToDecimal(val) if err != nil { return InvConv(val, tp) } if target.Decimal != UnspecifiedLength { x = x.Round(int32(target.Decimal)) } // TODO: check Flen return x, nil case mysql.TypeYear: var ( intVal int64 err error ) switch x := val.(type) { case string: intVal, err = StrToInt(x) case mysql.Time: return int16(x.Year()), nil case mysql.Duration: return int16(time.Now().Year()), nil default: intVal, err = ToInt64(x) } if err != nil { return InvConv(val, tp) } y, err := mysql.AdjustYear(int(intVal)) if err != nil { return InvConv(val, tp) } return int16(y), nil default: panic("should never happen") } }
// Convert converts the val with type tp. func Convert(val interface{}, target *FieldType) (v interface{}, err error) { tp := target.Tp if val == nil { return nil, nil } switch tp { // TODO: implement mysql types convert when "CAST() AS" syntax are supported. case mysql.TypeFloat: x, err := ToFloat64(val) if err != nil { return invConv(val, tp) } // For float and following double type, we will only truncate it for float(M, D) format. // If no D is set, we will handle it like origin float whether M is set or not. if target.Flen != UnspecifiedLength && target.Decimal != UnspecifiedLength { x, err = TruncateFloat(x, target.Flen, target.Decimal) if err != nil { return nil, errors.Trace(err) } } return float32(x), nil case mysql.TypeDouble: x, err := ToFloat64(val) if err != nil { return invConv(val, tp) } if target.Flen != UnspecifiedLength && target.Decimal != UnspecifiedLength { x, err = TruncateFloat(x, target.Flen, target.Decimal) if err != nil { return nil, errors.Trace(err) } } return float64(x), nil case mysql.TypeBlob, mysql.TypeTinyBlob, mysql.TypeMediumBlob, mysql.TypeLongBlob, mysql.TypeString, mysql.TypeVarchar, mysql.TypeVarString: x, err := ToString(val) if err != nil { return invConv(val, tp) } // TODO: consider target.Charset/Collate x = truncateStr(x, target.Flen) if target.Charset == charset.CharsetBin { return []byte(x), nil } return x, nil case mysql.TypeDuration: fsp := mysql.DefaultFsp if target.Decimal != UnspecifiedLength { fsp = target.Decimal } switch x := val.(type) { case mysql.Duration: return x.RoundFrac(fsp) case mysql.Time: t, err := x.ConvertToDuration() if err != nil { return nil, errors.Trace(err) } return t.RoundFrac(fsp) case string: return mysql.ParseDuration(x, fsp) default: return invConv(val, tp) } case mysql.TypeTimestamp, mysql.TypeDatetime, mysql.TypeDate: fsp := mysql.DefaultFsp if target.Decimal != UnspecifiedLength { fsp = target.Decimal } switch x := val.(type) { case mysql.Time: t, err := x.Convert(tp) if err != nil { return nil, errors.Trace(err) } return t.RoundFrac(fsp) case mysql.Duration: t, err := x.ConvertToTime(tp) if err != nil { return nil, errors.Trace(err) } return t.RoundFrac(fsp) case string: return mysql.ParseTime(x, tp, fsp) case int64: return mysql.ParseTimeFromNum(x, tp, fsp) default: return invConv(val, tp) } case mysql.TypeTiny, mysql.TypeShort, mysql.TypeInt24, mysql.TypeLong, mysql.TypeLonglong: unsigned := mysql.HasUnsignedFlag(target.Flag) if unsigned { return convertToUint(val, target) } return convertToInt(val, target) case mysql.TypeBit: x, err := convertToUint(val, target) if err != nil { return x, errors.Trace(err) } // check bit boundary, if bit has n width, the boundary is // in [0, (1 << n) - 1] width := target.Flen if width == 0 || width == mysql.UnspecifiedBitWidth { width = mysql.MinBitWidth } maxValue := uint64(1)<<uint64(width) - 1 if x > maxValue { return maxValue, overflow(val, tp) } return mysql.Bit{Value: x, Width: width}, nil case mysql.TypeDecimal, mysql.TypeNewDecimal: x, err := ToDecimal(val) if err != nil { return invConv(val, tp) } if target.Decimal != UnspecifiedLength { x = x.Round(int32(target.Decimal)) } // TODO: check Flen return x, nil case mysql.TypeYear: var ( intVal int64 err error ) switch x := val.(type) { case string: intVal, err = StrToInt(x) case mysql.Time: return int64(x.Year()), nil case mysql.Duration: return int64(time.Now().Year()), nil default: intVal, err = ToInt64(x) } if err != nil { return invConv(val, tp) } y, err := mysql.AdjustYear(int(intVal)) if err != nil { return invConv(val, tp) } return int64(y), nil case mysql.TypeEnum: var ( e mysql.Enum err error ) switch x := val.(type) { case string: e, err = mysql.ParseEnumName(target.Elems, x) case []byte: e, err = mysql.ParseEnumName(target.Elems, string(x)) default: var number uint64 number, err = ToUint64(x) if err != nil { return nil, errors.Trace(err) } e, err = mysql.ParseEnumValue(target.Elems, number) } if err != nil { return invConv(val, tp) } return e, nil case mysql.TypeSet: var ( s mysql.Set err error ) switch x := val.(type) { case string: s, err = mysql.ParseSetName(target.Elems, x) case []byte: s, err = mysql.ParseSetName(target.Elems, string(x)) default: var number uint64 number, err = ToUint64(x) if err != nil { return nil, errors.Trace(err) } s, err = mysql.ParseSetValue(target.Elems, number) } if err != nil { return invConv(val, tp) } return s, nil default: panic("should never happen") } }