rdblue commented on a change in pull request #4262: URL: https://github.com/apache/iceberg/pull/4262#discussion_r830658348
########## File path: python/src/iceberg/expression/literals.py ########## @@ -0,0 +1,601 @@ +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at +# +# http://www.apache.org/licenses/LICENSE-2.0 +# +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. +# +# http://www.apache.org/licenses/LICENSE-2.0 +# +# Unless required by applicable law or agreed to in writing, +# software distributed under the License is distributed on an +# "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY +# KIND, either express or implied. See the License for the +# specific language governing permissions and limitations +# under the License. + +import datetime +import sys +import uuid +from decimal import ROUND_HALF_UP, Decimal +from functools import singledispatch +from typing import Any + +from iceberg.types import PrimitiveType + +if sys.version_info >= (3, 8): + from functools import singledispatchmethod # pragma: no cover +else: + from singledispatch import singledispatchmethod # pragma: no cover + +import pytz + +from iceberg.types import ( + BinaryType, + BooleanType, + DateType, + DecimalType, + DoubleType, + FixedType, + FloatType, + IntegerType, + LongType, + Singleton, + StringType, + TimestampType, + TimestamptzType, + TimeType, + UUIDType, +) + +EPOCH = datetime.datetime.utcfromtimestamp(0) + +""" +Iceberg literal is wrapper class used in expressions, which return unbound predicates +It's being organized as below +Literal +|-- AboveMax +|-- BelowMin +|-- BaseLiteral + |-- FixedLiteral + |-- BinaryLiteral + |-- ComparableLiteral + |-- BooleanLiteral + |-- IntegerLiteral + |-- LongLiteral + |-- FloatLiteral + |-- DoubleLiteral + |-- DateLiteral + |-- StringLiteral + |-- TimeLiteral + |-- TimestampLiteral + |-- DecimalLiteral + |-- UUIDLiteral +""" + + +class Literal: + def to(self, type_var: PrimitiveType): + raise NotImplementedError() + + def __str__(self): + return type(self).__name__ + + +class BaseLiteral(Literal): + """Base literal which has a value and can be converted between types""" + + def __init__(self, repr_string: str, value): + self._repr_string = repr_string + if value is None: + raise TypeError("Cannot set value of BaseLiteral to None") + self._value = value + + def to(self, type_var): + raise NotImplementedError() + + @property + def value(self): + return self._value + + def __repr__(self): + return self._repr_string + + def __str__(self): + return str(self._value) + + +@singledispatch +def literal(value: Any) -> BaseLiteral: + """ + A generic Literal factory to construct an iceberg Literal based on python primitive data type + using dynamic overloading + + Args: + value(python primitive type): the value to be associated with literal + + Example: + from iceberg.expressions.literals import literal + >>> literal(123) + <IntegerLiteral(123)>v + """ + raise TypeError(f"Unimplemented Type Literal for value: {value}") + + [email protected] +def _(value: bool): + return BooleanLiteral(value) + + [email protected] +def _(value: int): + """ + Upgrade to long if python int is outside the JAVA_MIN_INT and JAVA_MAX_INT + """ + if value < IntegerType.min or value > IntegerType.max: + return LongLiteral(value) + return IntegerLiteral(value) + + [email protected] +def _(value: float): + """ + Upgrade to double if python float is outside the JAVA_MIN_FLOAT and JAVA_MAX_FLOAT + """ + if value < FloatType.min or value > FloatType.max: + return DoubleLiteral(value) + return FloatLiteral(value) + + [email protected] +def _(value: str): + return StringLiteral(value) + + [email protected] +def _(value: uuid.UUID): + return UUIDLiteral(value) + + [email protected] +def _(value: bytes): + return FixedLiteral(value) + + [email protected] +def _(value: bytearray): + return BinaryLiteral(value) + + [email protected] +def _(value: Decimal): + return DecimalLiteral(value) + + +class ComparableLiteral(BaseLiteral): + def __init__(self, repr_string: str, value): + super(ComparableLiteral, self).__init__(repr_string, value) + + def to(self, type_var): + raise NotImplementedError() + + def __eq__(self, other): + return self.value == other.value + + def __ne__(self, other): + return not self.__eq__(other) + + def __lt__(self, other): + return self.value < other.value + + def __gt__(self, other): + return self.value > other.value + + def __le__(self, other): + return self.value <= other.value + + def __ge__(self, other): + return self.value >= other.value + + +class AboveMax(Literal, Singleton): + def value(self): + raise RuntimeError("AboveMax has no value") + + def to(self, type_var): + raise RuntimeError("Cannot change the type of AboveMax") + + +class BelowMin(Literal, Singleton): + def value(self): + raise RuntimeError("BelowMin has no value") + + def to(self, type_var): + raise RuntimeError("Cannot change the type of BelowMin") + + +class BooleanLiteral(ComparableLiteral): + def __init__(self, value): + super(BooleanLiteral, self).__init__(f"BooleanLiteral({value})", value) + + @singledispatchmethod + def to(self, type_var): + return None + + @to.register(BooleanType) + def _(self, type_var): + return self + + +class IntegerLiteral(ComparableLiteral): + def __init__(self, value): + super(IntegerLiteral, self).__init__(f"IntegerLiteral({value})", value) + + @singledispatchmethod + def to(self, type_var): + return None + + @to.register(IntegerType) + def _(self, type_var): + return self + + @to.register(LongType) + def _(self, type_var): + return LongLiteral(self.value) + + @to.register(FloatType) + def _(self, type_var): + return FloatLiteral(self.value) + + @to.register(DoubleType) + def _(self, type_var): + return DoubleLiteral(self.value) + + @to.register(DateType) + def _(self, type_var): + return DateLiteral(self.value) + + @to.register(DecimalType) + def _(self, type_var): + if type_var.scale == 0: + return DecimalLiteral(Decimal(self.value)) + else: + return DecimalLiteral( + Decimal(self.value).quantize( + Decimal("." + "".join(["0" for _ in range(1, type_var.scale)]) + "1"), rounding=ROUND_HALF_UP + ) + ) + + +class LongLiteral(ComparableLiteral): + def __init__(self, value): + super(LongLiteral, self).__init__(f"LongLiteral({value})", value) + + @singledispatchmethod + def to(self, type_var): + return None + + @to.register(LongType) + def _(self, type_var): + return self + + @to.register(IntegerType) + def _(self, type_var): + if IntegerType.max < self.value: + return AboveMax() + elif IntegerType.min > self.value: + return BelowMin() + return IntegerLiteral(self.value) + + @to.register(FloatType) + def _(self, type_var): + return FloatLiteral(self.value) + + @to.register(DoubleType) + def _(self, type_var): + return DoubleLiteral(self.value) + + @to.register(TimeType) + def _(self, type_var): + return TimeLiteral(self.value) + + @to.register(TimestampType) + def _(self, type_var): + return TimestampLiteral(self.value) + + @to.register(DecimalType) + def _(self, type_var): + if type_var.scale == 0: + return DecimalLiteral(Decimal(self.value)) + else: + return DecimalLiteral(Decimal(self.value).quantize(Decimal((0, (1,), -type_var.scale)), rounding=ROUND_HALF_UP)) + + +class FloatLiteral(ComparableLiteral): Review comment: I think that floats need special implementations of comparison methods like `__eq__`. The Python VM uses 64-bit double to represent both floating point types. That means that float literals can be more exact than float values that are coming from data files. For example: | Cell | Python code | Result | Note | |:-- |:-- |:-- |:-- | | 1 | `12.34` | `12.34` | Original value | 2 | `struct.pack('<f', _1)` | `b'\xa4pEA'` | 4-byte binary | | 3 | `struct.unpack('<f', _2)[0]` | `12.34000015258789` | Lost precision! | | 4 | `_1 == _3` | `False` | No longer equal | | 5 | `struct.pack('<d', _1)` | `b'\xaeG\xe1z\x14\xae(@'` | 8-byte binary | | 6 | `struct.unpack('<d', _5)[0]` | `12.34` | Trailing 0s omitted by repr | | 7 | `_1 == _6` | `True` | Equal when comparing extra bits | We need to watch out for the problem shown in [3] for 32-bit float data coming from data files. One option to fix this is to round-trip the literal value through `struct.pack` and `struct.unpack`. I think that makes the most sense, but it also raises the question: what should we pass into `DoubleLiteral` when converting? My initial take is to either not allow converting a `FloatLiteral` to `DoubleLiteral` or to pass the original python float (64-bit) that was used to create the `FloatLiteral`. I propose we add `_value32` to this class and use that in overridden comparison methods. That way comparison acts correctly. ```python class FloatLiteral(Literal): def __init__(self, value): super(FloatLiteral, self).__init__(f"FloatLiteral({value})", value) this._value32 = struct.unpack('<f', struct.pack('<f', value))[0] def __eq__(self, value): this._value32 == value ... ``` -- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. To unsubscribe, e-mail: [email protected] For queries about this service, please contact Infrastructure at: [email protected] --------------------------------------------------------------------- To unsubscribe, e-mail: [email protected] For additional commands, e-mail: [email protected]
