// DecodeTableValue decodes a value encoded by EncodeTableValue.
func DecodeTableValue(a *DatumAlloc, valType parser.Type, b []byte) (parser.Datum, []byte, error) {
_, dataOffset, _, typ, err := encoding.DecodeValueTag(b)
if err != nil {
return nil, b, err
}
if typ == encoding.Null {
return parser.DNull, b[dataOffset:], nil
}
switch valType {
case parser.TypeBool:
var x bool
b, x, err = encoding.DecodeBoolValue(b)
// No need to chunk allocate DBool as MakeDBool returns either
// parser.DBoolTrue or parser.DBoolFalse.
return parser.MakeDBool(parser.DBool(x)), b, err
case parser.TypeInt:
var i int64
b, i, err = encoding.DecodeIntValue(b)
return a.NewDInt(parser.DInt(i)), b, err
case parser.TypeFloat:
var f float64
b, f, err = encoding.DecodeFloatValue(b)
return a.NewDFloat(parser.DFloat(f)), b, err
case parser.TypeDecimal:
var d *inf.Dec
b, d, err = encoding.DecodeDecimalValue(b)
dd := a.NewDDecimal(parser.DDecimal{})
dd.Set(d)
return dd, b, err
case parser.TypeString:
var data []byte
b, data, err = encoding.DecodeBytesValue(b)
return a.NewDString(parser.DString(data)), b, err
case parser.TypeBytes:
var data []byte
b, data, err = encoding.DecodeBytesValue(b)
return a.NewDBytes(parser.DBytes(data)), b, err
case parser.TypeDate:
var i int64
b, i, err = encoding.DecodeIntValue(b)
return a.NewDDate(parser.DDate(i)), b, err
case parser.TypeTimestamp:
var t time.Time
b, t, err = encoding.DecodeTimeValue(b)
return a.NewDTimestamp(parser.DTimestamp{Time: t}), b, err
case parser.TypeTimestampTZ:
var t time.Time
b, t, err = encoding.DecodeTimeValue(b)
return a.NewDTimestampTZ(parser.DTimestampTZ{Time: t}), b, err
case parser.TypeInterval:
var d duration.Duration
b, d, err = encoding.DecodeDurationValue(b)
return a.NewDInterval(parser.DInterval{Duration: d}), b, err
default:
return nil, nil, errors.Errorf("TODO(pmattis): decoded index value: %s", valType)
}
}
// GenerateRandomArg generates a random, valid, SQL function argument of
// the spcified type.
func (r *RSG) GenerateRandomArg(typ parser.Type) string {
if r.Intn(10) == 0 {
return "NULL"
}
var v interface{}
switch typ {
case parser.TypeInt:
i := r.Int63()
i -= r.Int63()
v = i
case parser.TypeFloat, parser.TypeDecimal:
v = r.Float64()
case parser.TypeString:
v = `'string'`
case parser.TypeBytes:
v = `b'bytes'`
case parser.TypeTimestamp, parser.TypeTimestampTZ:
t := time.Unix(0, r.Int63())
v = fmt.Sprintf(`'%s'`, t.Format(time.RFC3339Nano))
case parser.TypeBool:
if r.Intn(2) == 0 {
v = "false"
} else {
v = "true"
}
case parser.TypeDate:
i := r.Int63()
i -= r.Int63()
d := parser.NewDDate(parser.DDate(i))
v = fmt.Sprintf(`'%s'`, d)
case parser.TypeInterval:
d := duration.Duration{Nanos: r.Int63()}
v = fmt.Sprintf(`'%s'`, &parser.DInterval{Duration: d})
case parser.TypeIntArray,
parser.TypeStringArray,
parser.TypeAnyArray,
parser.TypeAny:
v = "NULL"
default:
switch typ.(type) {
case parser.TTuple:
v = "NULL"
default:
panic(fmt.Errorf("unknown arg type: %s (%T)", typ, typ))
}
}
return fmt.Sprintf("%v::%s", v, typ.String())
}
// RandDatum generates a random Datum of the given type.
// If null is true, the datum can be DNull.
func RandDatum(rng *rand.Rand, typ ColumnType_Kind, null bool) parser.Datum {
if null && rng.Intn(10) == 0 {
return parser.DNull
}
switch typ {
case ColumnType_BOOL:
return parser.MakeDBool(rng.Intn(2) == 1)
case ColumnType_INT:
return parser.NewDInt(parser.DInt(rng.Int63()))
case ColumnType_FLOAT:
return parser.NewDFloat(parser.DFloat(rng.NormFloat64()))
case ColumnType_DECIMAL:
d := &parser.DDecimal{}
d.Dec.SetScale(inf.Scale(rng.Intn(40) - 20))
d.Dec.SetUnscaled(rng.Int63())
return d
case ColumnType_DATE:
return parser.NewDDate(parser.DDate(rng.Intn(10000)))
case ColumnType_TIMESTAMP:
return &parser.DTimestamp{Time: time.Unix(rng.Int63n(1000000), rng.Int63n(1000000))}
case ColumnType_INTERVAL:
return &parser.DInterval{Duration: duration.Duration{Months: rng.Int63n(1000),
Days: rng.Int63n(1000),
Nanos: rng.Int63n(1000000),
}}
case ColumnType_STRING:
// Generate a random ASCII string.
p := make([]byte, rng.Intn(10))
for i := range p {
p[i] = byte(1 + rng.Intn(127))
}
return parser.NewDString(string(p))
case ColumnType_BYTES:
p := make([]byte, rng.Intn(10))
_, _ = rng.Read(p)
return parser.NewDBytes(parser.DBytes(p))
case ColumnType_TIMESTAMPTZ:
return &parser.DTimestampTZ{Time: time.Unix(rng.Int63n(1000000), rng.Int63n(1000000))}
case ColumnType_INT_ARRAY:
// TODO(cuongdo): we don't support for persistence of arrays yet
return parser.DNull
default:
panic(fmt.Sprintf("invalid type %s", typ))
}
}
// DecodeTableKey decodes a table key/value.
func DecodeTableKey(
a *DatumAlloc, valType parser.Type, key []byte, dir encoding.Direction,
) (parser.Datum, []byte, error) {
if (dir != encoding.Ascending) && (dir != encoding.Descending) {
return nil, nil, errors.Errorf("invalid direction: %d", dir)
}
var isNull bool
if key, isNull = encoding.DecodeIfNull(key); isNull {
return parser.DNull, key, nil
}
var rkey []byte
var err error
switch valType {
case parser.TypeBool:
var i int64
if dir == encoding.Ascending {
rkey, i, err = encoding.DecodeVarintAscending(key)
} else {
rkey, i, err = encoding.DecodeVarintDescending(key)
}
// No need to chunk allocate DBool as MakeDBool returns either
// parser.DBoolTrue or parser.DBoolFalse.
return parser.MakeDBool(parser.DBool(i != 0)), rkey, err
case parser.TypeInt:
var i int64
if dir == encoding.Ascending {
rkey, i, err = encoding.DecodeVarintAscending(key)
} else {
rkey, i, err = encoding.DecodeVarintDescending(key)
}
return a.NewDInt(parser.DInt(i)), rkey, err
case parser.TypeFloat:
var f float64
if dir == encoding.Ascending {
rkey, f, err = encoding.DecodeFloatAscending(key)
} else {
rkey, f, err = encoding.DecodeFloatDescending(key)
}
return a.NewDFloat(parser.DFloat(f)), rkey, err
case parser.TypeDecimal:
var d *inf.Dec
if dir == encoding.Ascending {
rkey, d, err = encoding.DecodeDecimalAscending(key, nil)
} else {
rkey, d, err = encoding.DecodeDecimalDescending(key, nil)
}
dd := a.NewDDecimal(parser.DDecimal{})
dd.Set(d)
return dd, rkey, err
case parser.TypeString:
var r string
if dir == encoding.Ascending {
rkey, r, err = encoding.DecodeUnsafeStringAscending(key, nil)
} else {
rkey, r, err = encoding.DecodeUnsafeStringDescending(key, nil)
}
return a.NewDString(parser.DString(r)), rkey, err
case parser.TypeBytes:
var r []byte
if dir == encoding.Ascending {
rkey, r, err = encoding.DecodeBytesAscending(key, nil)
} else {
rkey, r, err = encoding.DecodeBytesDescending(key, nil)
}
return a.NewDBytes(parser.DBytes(r)), rkey, err
case parser.TypeDate:
var t int64
if dir == encoding.Ascending {
rkey, t, err = encoding.DecodeVarintAscending(key)
} else {
rkey, t, err = encoding.DecodeVarintDescending(key)
}
return a.NewDDate(parser.DDate(t)), rkey, err
case parser.TypeTimestamp:
var t time.Time
if dir == encoding.Ascending {
rkey, t, err = encoding.DecodeTimeAscending(key)
} else {
rkey, t, err = encoding.DecodeTimeDescending(key)
}
return a.NewDTimestamp(parser.DTimestamp{Time: t}), rkey, err
case parser.TypeTimestampTZ:
var t time.Time
if dir == encoding.Ascending {
rkey, t, err = encoding.DecodeTimeAscending(key)
} else {
rkey, t, err = encoding.DecodeTimeDescending(key)
}
return a.NewDTimestampTZ(parser.DTimestampTZ{Time: t}), rkey, err
case parser.TypeInterval:
var d duration.Duration
if dir == encoding.Ascending {
rkey, d, err = encoding.DecodeDurationAscending(key)
} else {
rkey, d, err = encoding.DecodeDurationDescending(key)
}
return a.NewDInterval(parser.DInterval{Duration: d}), rkey, err
default:
return nil, nil, errors.Errorf("TODO(pmattis): decoded index key: %s", valType)
}
}
// UnmarshalColumnValue decodes the value from a key-value pair using the type
// expected by the column. An error is returned if the value's type does not
// match the column's type.
func UnmarshalColumnValue(
a *DatumAlloc, kind ColumnType_Kind, value *roachpb.Value,
) (parser.Datum, error) {
if value == nil {
return parser.DNull, nil
}
switch kind {
case ColumnType_BOOL:
v, err := value.GetBool()
if err != nil {
return nil, err
}
return parser.MakeDBool(parser.DBool(v)), nil
case ColumnType_INT:
v, err := value.GetInt()
if err != nil {
return nil, err
}
return a.NewDInt(parser.DInt(v)), nil
case ColumnType_FLOAT:
v, err := value.GetFloat()
if err != nil {
return nil, err
}
return a.NewDFloat(parser.DFloat(v)), nil
case ColumnType_DECIMAL:
v, err := value.GetDecimal()
if err != nil {
return nil, err
}
dd := a.NewDDecimal(parser.DDecimal{})
dd.Set(v)
return dd, nil
case ColumnType_STRING:
v, err := value.GetBytes()
if err != nil {
return nil, err
}
return a.NewDString(parser.DString(v)), nil
case ColumnType_BYTES:
v, err := value.GetBytes()
if err != nil {
return nil, err
}
return a.NewDBytes(parser.DBytes(v)), nil
case ColumnType_DATE:
v, err := value.GetInt()
if err != nil {
return nil, err
}
return a.NewDDate(parser.DDate(v)), nil
case ColumnType_TIMESTAMP:
v, err := value.GetTime()
if err != nil {
return nil, err
}
return a.NewDTimestamp(parser.DTimestamp{Time: v}), nil
case ColumnType_TIMESTAMPTZ:
v, err := value.GetTime()
if err != nil {
return nil, err
}
return a.NewDTimestampTZ(parser.DTimestampTZ{Time: v}), nil
case ColumnType_INTERVAL:
d, err := value.GetDuration()
if err != nil {
return nil, err
}
return a.NewDInterval(parser.DInterval{Duration: d}), nil
default:
return nil, errors.Errorf("unsupported column type: %s", kind)
}
}
// decodeOidDatum decodes bytes with specified Oid and format code into
// a datum.
func decodeOidDatum(id oid.Oid, code formatCode, b []byte) (parser.Datum, error) {
var d parser.Datum
switch id {
case oid.T_bool:
switch code {
case formatText:
v, err := strconv.ParseBool(string(b))
if err != nil {
return d, err
}
d = parser.MakeDBool(parser.DBool(v))
case formatBinary:
switch b[0] {
case 0:
d = parser.MakeDBool(false)
case 1:
d = parser.MakeDBool(true)
default:
return d, errors.Errorf("unsupported binary bool: %q", b)
}
default:
return d, errors.Errorf("unsupported bool format code: %d", code)
}
case oid.T_int2:
switch code {
case formatText:
i, err := strconv.ParseInt(string(b), 10, 64)
if err != nil {
return d, err
}
d = parser.NewDInt(parser.DInt(i))
case formatBinary:
if len(b) < 2 {
return d, errors.Errorf("int2 requires 2 bytes for binary format")
}
i := int16(binary.BigEndian.Uint16(b))
d = parser.NewDInt(parser.DInt(i))
default:
return d, errors.Errorf("unsupported int2 format code: %d", code)
}
case oid.T_int4:
switch code {
case formatText:
i, err := strconv.ParseInt(string(b), 10, 64)
if err != nil {
return d, err
}
d = parser.NewDInt(parser.DInt(i))
case formatBinary:
if len(b) < 4 {
return d, errors.Errorf("int4 requires 4 bytes for binary format")
}
i := int32(binary.BigEndian.Uint32(b))
d = parser.NewDInt(parser.DInt(i))
default:
return d, errors.Errorf("unsupported int4 format code: %d", code)
}
case oid.T_int8:
switch code {
case formatText:
i, err := strconv.ParseInt(string(b), 10, 64)
if err != nil {
return d, err
}
d = parser.NewDInt(parser.DInt(i))
case formatBinary:
if len(b) < 8 {
return d, errors.Errorf("int8 requires 8 bytes for binary format")
}
i := int64(binary.BigEndian.Uint64(b))
d = parser.NewDInt(parser.DInt(i))
default:
return d, errors.Errorf("unsupported int8 format code: %d", code)
}
case oid.T_float4:
switch code {
case formatText:
f, err := strconv.ParseFloat(string(b), 64)
if err != nil {
return d, err
}
d = parser.NewDFloat(parser.DFloat(f))
case formatBinary:
if len(b) < 4 {
return d, errors.Errorf("float4 requires 4 bytes for binary format")
}
f := math.Float32frombits(binary.BigEndian.Uint32(b))
d = parser.NewDFloat(parser.DFloat(f))
default:
return d, errors.Errorf("unsupported float4 format code: %d", code)
}
case oid.T_float8:
switch code {
case formatText:
f, err := strconv.ParseFloat(string(b), 64)
if err != nil {
return d, err
}
d = parser.NewDFloat(parser.DFloat(f))
case formatBinary:
if len(b) < 8 {
return d, errors.Errorf("float8 requires 8 bytes for binary format")
}
f := math.Float64frombits(binary.BigEndian.Uint64(b))
d = parser.NewDFloat(parser.DFloat(f))
default:
return d, errors.Errorf("unsupported float8 format code: %d", code)
}
case oid.T_numeric:
switch code {
case formatText:
dd := &parser.DDecimal{}
if _, ok := dd.SetString(string(b)); !ok {
return nil, errors.Errorf("could not parse string %q as decimal", b)
}
d = dd
case formatBinary:
r := bytes.NewReader(b)
alloc := struct {
pgNum pgNumeric
i16 int16
dd parser.DDecimal
}{}
for _, ptr := range []interface{}{
&alloc.pgNum.ndigits,
&alloc.pgNum.weight,
&alloc.pgNum.sign,
&alloc.pgNum.dscale,
} {
if err := binary.Read(r, binary.BigEndian, ptr); err != nil {
return d, err
}
}
if alloc.pgNum.ndigits > 0 {
decDigits := make([]byte, 0, alloc.pgNum.ndigits*pgDecDigits)
nextDigit := func() error {
if err := binary.Read(r, binary.BigEndian, &alloc.i16); err != nil {
return err
}
numZeroes := pgDecDigits
for i16 := alloc.i16; i16 > 0; i16 /= 10 {
numZeroes--
}
for ; numZeroes > 0; numZeroes-- {
decDigits = append(decDigits, '0')
}
return nil
}
for i := int16(0); i < alloc.pgNum.ndigits-1; i++ {
if err := nextDigit(); err != nil {
return d, err
}
if alloc.i16 > 0 {
decDigits = strconv.AppendUint(decDigits, uint64(alloc.i16), 10)
}
}
// The last digit may contain padding, which we need to deal with.
if err := nextDigit(); err != nil {
return d, err
}
dscale := (alloc.pgNum.ndigits - (alloc.pgNum.weight + 1)) * pgDecDigits
if overScale := dscale - alloc.pgNum.dscale; overScale > 0 {
dscale -= overScale
for i := int16(0); i < overScale; i++ {
alloc.i16 /= 10
}
}
decDigits = strconv.AppendUint(decDigits, uint64(alloc.i16), 10)
decString := string(decDigits)
if _, ok := alloc.dd.UnscaledBig().SetString(decString, 10); !ok {
return nil, errors.Errorf("could not parse string %q as decimal", decString)
}
alloc.dd.SetScale(inf.Scale(dscale))
}
switch alloc.pgNum.sign {
case pgNumericPos:
case pgNumericNeg:
alloc.dd.Neg(&alloc.dd.Dec)
default:
return d, errors.Errorf("unsupported numeric sign: %d", alloc.pgNum.sign)
}
d = &alloc.dd
default:
return d, errors.Errorf("unsupported numeric format code: %d", code)
}
case oid.T_text, oid.T_varchar:
switch code {
case formatText, formatBinary:
d = parser.NewDString(string(b))
default:
return d, errors.Errorf("unsupported text format code: %d", code)
}
case oid.T_bytea:
switch code {
case formatText:
// http://www.postgresql.org/docs/current/static/datatype-binary.html#AEN5667
// Code cribbed from github.com/lib/pq.
// We only support hex encoding.
if len(b) >= 2 && bytes.Equal(b[:2], []byte("\\x")) {
b = b[2:] // trim off leading "\\x"
result := make([]byte, hex.DecodedLen(len(b)))
_, err := hex.Decode(result, b)
if err != nil {
return d, err
}
d = parser.NewDBytes(parser.DBytes(result))
} else {
return d, errors.Errorf("unsupported bytea encoding: %q", b)
}
case formatBinary:
d = parser.NewDBytes(parser.DBytes(b))
default:
return d, errors.Errorf("unsupported bytea format code: %d", code)
}
case oid.T_timestamp:
switch code {
case formatText:
ts, err := parseTs(string(b))
if err != nil {
return d, errors.Errorf("could not parse string %q as timestamp", b)
}
d = parser.MakeDTimestamp(ts, time.Microsecond)
case formatBinary:
if len(b) < 8 {
return d, errors.Errorf("timestamp requires 8 bytes for binary format")
}
i := int64(binary.BigEndian.Uint64(b))
d = parser.MakeDTimestamp(pgBinaryToTime(i), time.Microsecond)
default:
return d, errors.Errorf("unsupported timestamp format code: %d", code)
}
case oid.T_timestamptz:
switch code {
case formatText:
ts, err := parseTs(string(b))
if err != nil {
return d, errors.Errorf("could not parse string %q as timestamp", b)
}
d = parser.MakeDTimestampTZ(ts, time.Microsecond)
case formatBinary:
if len(b) < 8 {
return d, errors.Errorf("timestamptz requires 8 bytes for binary format")
}
i := int64(binary.BigEndian.Uint64(b))
d = parser.MakeDTimestampTZ(pgBinaryToTime(i), time.Microsecond)
default:
return d, errors.Errorf("unsupported timestamptz format code: %d", code)
}
case oid.T_date:
switch code {
case formatText:
ts, err := parseTs(string(b))
if err != nil {
res, err := parser.ParseDDate(string(b), time.UTC)
if err != nil {
return d, errors.Errorf("could not parse string %q as date", b)
}
d = res
} else {
daysSinceEpoch := ts.Unix() / secondsInDay
d = parser.NewDDate(parser.DDate(daysSinceEpoch))
}
case formatBinary:
if len(b) < 4 {
return d, errors.Errorf("date requires 4 bytes for binary format")
}
i := int32(binary.BigEndian.Uint32(b))
d = pgBinaryToDate(i)
default:
return d, errors.Errorf("unsupported date format code: %d", code)
}
case oid.T_interval:
switch code {
case formatText:
d, err := parser.ParseDInterval(string(b))
if err != nil {
return d, errors.Errorf("could not parse string %q as interval", b)
}
return d, nil
default:
return d, errors.Errorf("unsupported interval format code: %d", code)
}
default:
return d, errors.Errorf("unsupported OID: %v", id)
}
return d, nil
}
// pgBinaryToDate takes an int32 and interprets it as the Postgres binary format
// for a date. To create a date from this value, it takes the day delta and adds
// it to pgEpochJDate.
func pgBinaryToDate(i int32) *parser.DDate {
daysSinceEpoch := pgEpochJDateFromUnix + i
return parser.NewDDate(parser.DDate(daysSinceEpoch))
}
// RandDatum generates a random Datum of the given type.
// If null is true, the datum can be DNull.
func RandDatum(rng *rand.Rand, typ ColumnType, null bool) parser.Datum {
if null && rng.Intn(10) == 0 {
return parser.DNull
}
switch typ.Kind {
case ColumnType_BOOL:
return parser.MakeDBool(rng.Intn(2) == 1)
case ColumnType_INT:
return parser.NewDInt(parser.DInt(rng.Int63()))
case ColumnType_FLOAT:
return parser.NewDFloat(parser.DFloat(rng.NormFloat64()))
case ColumnType_DECIMAL:
d := &parser.DDecimal{}
d.Dec.SetScale(inf.Scale(rng.Intn(40) - 20))
d.Dec.SetUnscaled(rng.Int63())
return d
case ColumnType_DATE:
return parser.NewDDate(parser.DDate(rng.Intn(10000)))
case ColumnType_TIMESTAMP:
return &parser.DTimestamp{Time: time.Unix(rng.Int63n(1000000), rng.Int63n(1000000))}
case ColumnType_INTERVAL:
return &parser.DInterval{Duration: duration.Duration{Months: rng.Int63n(1000),
Days: rng.Int63n(1000),
Nanos: rng.Int63n(1000000),
}}
case ColumnType_STRING:
// Generate a random ASCII string.
p := make([]byte, rng.Intn(10))
for i := range p {
p[i] = byte(1 + rng.Intn(127))
}
return parser.NewDString(string(p))
case ColumnType_BYTES:
p := make([]byte, rng.Intn(10))
_, _ = rng.Read(p)
return parser.NewDBytes(parser.DBytes(p))
case ColumnType_TIMESTAMPTZ:
return &parser.DTimestampTZ{Time: time.Unix(rng.Int63n(1000000), rng.Int63n(1000000))}
case ColumnType_COLLATEDSTRING:
if typ.Locale == nil {
panic("locale is required for COLLATEDSTRING")
}
// Generate a random Unicode string.
var buf bytes.Buffer
n := rng.Intn(10)
for i := 0; i < n; i++ {
var r rune
for {
r = rune(rng.Intn(unicode.MaxRune + 1))
if !unicode.Is(unicode.C, r) {
break
}
}
buf.WriteRune(r)
}
return parser.NewDCollatedString(buf.String(), *typ.Locale, &parser.CollationEnvironment{})
case ColumnType_INT_ARRAY:
// TODO(cuongdo): we don't support for persistence of arrays yet
return parser.DNull
default:
panic(fmt.Sprintf("invalid type %s", typ.String()))
}
}