Skip to content

Commit

Permalink
refactor(deps): remove multipledispatch as a dependency (#8332)
Browse files Browse the repository at this point in the history 
This PR removes `multipledispatch` as a dependency of Ibis. After
`the-epic-split` was merged our largest consumer of `multipledispatch`
was removed, now there is this remaining bit of code to port over to use
non-multipledispatch-style checks.
  • Loading branch information
@cpcloud
cpcloud authored Feb 13, 2024
1 parent dd12198 commit d587166
Show file tree
Hide file tree
Showing 8 changed files with 142 additions and 200 deletions.
172 changes: 3 additions & 169 deletions ibis/expr/datatypes/cast.py
View file
Original file line number Diff line number Diff line change
Expand Up @@ -3,7 +3,6 @@
import functools
from typing import TYPE_CHECKING

from multipledispatch import Dispatcher
from public import public

import ibis.expr.datatypes.core as dt
Expand All @@ -12,15 +11,13 @@
if TYPE_CHECKING:
from collections.abc import Iterator

castable = Dispatcher("castable")


@public
def cast(source: str | dt.DataType, target: str | dt.DataType, **kwargs) -> dt.DataType:
"""Attempts to implicitly cast from source dtype to target dtype."""
source, target = dt.dtype(source), dt.dtype(target)

if not castable(source, target, **kwargs):
if not source.castable(target, **kwargs):
raise IbisTypeError(
f"Datatype {source} cannot be implicitly casted to {target}"
)
Expand All @@ -31,9 +28,9 @@ def cast(source: str | dt.DataType, target: str | dt.DataType, **kwargs) -> dt.D
def higher_precedence(left: dt.DataType, right: dt.DataType) -> dt.DataType:
nullable = left.nullable or right.nullable

if castable(left, right, upcast=True):
if left.castable(right, upcast=True):
return right.copy(nullable=nullable)
elif castable(right, left, upcast=True):
elif right.castable(left, upcast=True):
return left.copy(nullable=nullable)

raise IbisTypeError(f"Cannot compute precedence for `{left}` and `{right}` types")
Expand All @@ -46,166 +43,3 @@ def highest_precedence(dtypes: Iterator[dt.DataType]) -> dt.DataType:
return functools.reduce(higher_precedence, collected)
else:
return dt.null


@castable.register(dt.DataType, dt.DataType)
def can_cast_subtype(source: dt.DataType, target: dt.DataType, **kwargs) -> bool:
return isinstance(target, source.__class__)


@castable.register(dt.Integer, (dt.Floating, dt.Decimal))
@castable.register(dt.Floating, dt.Decimal)
@castable.register((dt.Date, dt.Timestamp), (dt.Date, dt.Timestamp))
@castable.register(dt.String, dt.JSON)
@castable.register(dt.JSON, dt.String)
def can_cast_any(source: dt.DataType, target: dt.DataType, **kwargs) -> bool:
return True


@castable.register(dt.Null, dt.DataType)
def can_cast_null(source: dt.DataType, target: dt.DataType, **kwargs) -> bool:
# The null type is castable to any type, even if the target type is *not*
# nullable.
#
# We handle the promotion of `null + !T -> T` at the `castable` call site.
#
# It might be possible to build a system with a single function that tries
# to promote types and use the exception to indicate castability, but that
# is a deeper refactor to be tackled later.
#
# See https://github.com/ibis-project/ibis/issues/2891 for the bug report
return True


@castable.register(dt.SignedInteger, dt.UnsignedInteger)
@castable.register(dt.UnsignedInteger, dt.SignedInteger)
def can_cast_to_differently_signed_integer_type(
source: dt.Integer, target: dt.Integer, value: int | None = None, **kwargs
) -> bool:
if value is not None:
# TODO(kszucs): we may not need to actually check the value since the
# literal construction now checks for bounds and doesn't use castable()
# anymore
return target.bounds.lower <= value <= target.bounds.upper
else:
return (target.bounds.upper - target.bounds.lower) >= (
source.bounds.upper - source.bounds.lower
)


@castable.register(dt.SignedInteger, dt.SignedInteger)
@castable.register(dt.UnsignedInteger, dt.UnsignedInteger)
def can_cast_integers(source: dt.Integer, target: dt.Integer, **kwargs) -> bool:
return target.nbytes >= source.nbytes


@castable.register(dt.Floating, dt.Floating)
def can_cast_floats(
source: dt.Floating, target: dt.Floating, upcast: bool = False, **kwargs
) -> bool:
if upcast:
return target.nbytes >= source.nbytes

# double -> float must be allowed because
# float literals are inferred as doubles
return True


@castable.register(dt.Decimal, dt.Decimal)
def can_cast_decimals(source: dt.Decimal, target: dt.Decimal, **kwargs) -> bool:
target_prec = target.precision
source_prec = source.precision
target_sc = target.scale
source_sc = source.scale
return (
# If either sides precision and scale are both `None`, return `True`.
target_prec is None
and target_sc is None
or source_prec is None
and source_sc is None
# Otherwise, return `True` unless we are downcasting precision or scale.
or (
target_prec is None
or (source_prec is not None and target_prec >= source_prec)
)
and (target_sc is None or (source_sc is not None and target_sc >= source_sc))
)


@castable.register(dt.Interval, dt.Interval)
def can_cast_intervals(source: dt.Interval, target: dt.Interval, **kwargs) -> bool:
return source.unit == target.unit


@castable.register(dt.Integer, dt.Boolean)
def can_cast_integer_to_boolean(
source: dt.Integer, target: dt.Boolean, value: int | None = None, **kwargs
) -> bool:
return value is not None and 0 <= value <= 1


@castable.register(dt.Integer, dt.Interval)
def can_cast_integer_to_interval(
source: dt.Integer, target: dt.Interval, **kwargs
) -> bool:
return True


@castable.register(dt.String, (dt.Date, dt.Time, dt.Timestamp))
def can_cast_string_to_temporal(
source: dt.String,
target: dt.Date | dt.Time | dt.Timestamp,
value: str | None = None,
**kwargs,
) -> bool:
import pandas as pd

if value is None:
return False
try:
pd.Timestamp(value)
except ValueError:
return False
else:
return True


@castable.register(dt.Map, dt.Map)
def can_cast_map(source, target, **kwargs):
return castable(source.key_type, target.key_type) and castable(
source.value_type, target.value_type
)


@castable.register(dt.Struct, dt.Struct)
def can_cast_struct(source, target, **kwargs):
return all(castable(source[field], target[field]) for field in target.names)


@castable.register(dt.Array, dt.Array)
def can_cast_array_or_set(source: dt.Array, target: dt.Array, **kwargs) -> bool:
return castable(source.value_type, target.value_type)


@castable.register(dt.JSON, dt.JSON)
@castable.register(dt.Floating, dt.JSON)
@castable.register(dt.Integer, dt.JSON)
def can_cast_json(source, target, **kwargs):
return True


@castable.register(dt.Array, dt.GeoSpatial)
def can_cast_geospatial(source, target, **kwargs):
return True


@castable.register(dt.UUID, dt.UUID)
@castable.register(dt.UUID, dt.String)
@castable.register(dt.String, dt.UUID)
@castable.register(dt.MACADDR, dt.MACADDR)
@castable.register(dt.INET, dt.INET)
def can_cast_special_string(source, target, **kwargs):
return True


public(castable=castable)
Loading

0 comments on commit d587166

Please sign in to comment.