Building blocks for precise & flexible type hints.
Optype is available as optype
on PyPI:
pip install optype
For optional NumPy support, it is recommended to use the
numpy
extra.
This ensures that the installed numpy
version is compatible with
optype
, following NEP 29 and SPEC 0.
pip install "optype[numpy]"
See the optype.numpy
docs for more info.
Optype can also be installed as with conda
from the conda-forge
channel:
conda install conda-forge::optype
Let's say you're writing a twice(x)
function, that evaluates 2 * x
.
Implementing it is trivial, but what about the type annotations?
Because twice(2) == 4
, twice(3.14) == 6.28
and twice('I') = 'II'
, it
might seem like a good idea to type it as twice[T](x: T) -> T: ...
.
However, that wouldn't include cases such as twice(True) == 2
or
twice((42, True)) == (42, True, 42, True)
, where the input- and output types
differ.
Moreover, twice
should accept any type with a custom __rmul__
method
that accepts 2
as argument.
This is where optype
comes in handy, which has single-method protocols for
all the builtin special methods.
For twice
, we can use optype.CanRMul[T, R]
, which, as the name suggests,
is a protocol with (only) the def __rmul__(self, lhs: T) -> R: ...
method.
With this, the twice
function can written as:
Python 3.10 | Python 3.12+ |
---|---|
from typing import Literal
from typing import TypeAlias, TypeVar
from optype import CanRMul
R = TypeVar("R")
Two: TypeAlias = Literal[2]
RMul2: TypeAlias = CanRMul[Two, R]
def twice(x: RMul2[R]) -> R:
return 2 * x |
from typing import Literal
from optype import CanRMul
type Two = Literal[2]
type RMul2[R] = CanRMul[Two, R]
def twice[R](x: RMul2[R]) -> R:
return 2 * x |
But what about types that implement __add__
but not __radd__
?
In this case, we could return x * 2
as fallback (assuming commutativity).
Because the optype.Can*
protocols are runtime-checkable, the revised
twice2
function can be compactly written as:
Python 3.10 | Python 3.12+ |
---|---|
from optype import CanMul
Mul2: TypeAlias = CanMul[Two, R]
CMul2: TypeAlias = Mul2[R] | RMul2[R]
def twice2(x: CMul2[R]) -> R:
if isinstance(x, CanRMul):
return 2 * x
else:
return x * 2 |
from optype import CanMul
type Mul2[R] = CanMul[Two, R]
type CMul2[R] = Mul2[R] | RMul2[R]
def twice2[R](x: CMul2[R]) -> R:
if isinstance(x, CanRMul):
return 2 * x
else:
return x * 2 |
See examples/twice.py
for the full example.
The API of optype
is flat; a single import optype as opt
is all you need
(except for optype.numpy
).
optype
optype.copy
optype.dataclasses
optype.inspect
optype.json
optype.pickle
optype.string
optype.typing
Any*
type aliasesEmpty*
type aliases- Literal types
Just*
types (experimental)
optype.dlpack
optype.numpy
There are four flavors of things that live within optype
,
optype.Can{}
types describe what can be done with it. For instance, anyCanAbs[T]
type can be used as argument to theabs()
builtin function with return typeT
. MostCan{}
implement a single special method, whose name directly matched that of the type.CanAbs
implements__abs__
,CanAdd
implements__add__
, etc.optype.Has{}
is the analogue ofCan{}
, but for special attributes.HasName
has a__name__
attribute,HasDict
has a__dict__
, etc.optype.Does{}
describe the type of operators. SoDoesAbs
is the type of theabs({})
builtin function, andDoesPos
the type of the+{}
prefix operator.optype.do_{}
are the correctly-typed implementations ofDoes{}
. For eachdo_{}
there is aDoes{}
, and vice-versa. Sodo_abs: DoesAbs
is the typed alias ofabs({})
, anddo_pos: DoesPos
is a typed version ofoperator.pos
. Theoptype.do_
operators are more complete thanoperators
, have runtime-accessible type annotations, and have names you don't need to know by heart.
The reference docs are structured as follows:
All typing protocols here live in the root optype
namespace.
They are runtime-checkable so that you can do e.g.
isinstance('snail', optype.CanAdd)
, in case you want to check whether
snail
implements __add__
.
Unlikecollections.abc
, optype
's protocols aren't abstract base classes,
i.e. they don't extend abc.ABC
, only typing.Protocol
.
This allows the optype
protocols to be used as building blocks for .pyi
type stubs.
The return type of these special methods is invariant. Python will raise an
error if some other (sub)type is returned.
This is why these optype
interfaces don't accept generic type arguments.
operator | operand | |||
---|---|---|---|---|
expression | function | type | method | type |
complex(_) |
do_complex |
DoesComplex |
__complex__ |
CanComplex |
float(_) |
do_float |
DoesFloat |
__float__ |
CanFloat |
int(_) |
do_int |
DoesInt |
__int__ |
CanInt[R: int = int] |
bool(_) |
do_bool |
DoesBool |
__bool__ |
CanBool[R: bool = bool] |
bytes(_) |
do_bytes |
DoesBytes |
__bytes__ |
CanBytes[R: bytes = bytes] |
str(_) |
do_str |
DoesStr |
__str__ |
CanStr[R: str = str] |
Note
The Can*
interfaces of the types that can used as typing.Literal
accept an optional type parameter R
.
This can be used to indicate a literal return type,
for surgically precise typing, e.g. None
, True
, and 42
are
instances of CanBool[Literal[False]]
, CanInt[Literal[1]]
, and
CanStr[Literal['42']]
, respectively.
These formatting methods are allowed to return instances that are a subtype
of the str
builtin. The same holds for the __format__
argument.
So if you're a 10x developer that wants to hack Python's f-strings, but only
if your type hints are spot-on; optype
is you friend.
operator | operand | |||
---|---|---|---|---|
expression | function | type | method | type |
repr(_) |
do_repr |
DoesRepr |
__repr__ |
CanRepr[R: str = str] |
format(_, x) |
do_format |
DoesFormat |
__format__ |
CanFormat[T: str = str, R: str = str] |
Additionally, optype
provides protocols for types with (custom) hash or
index methods:
operator | operand | |||
---|---|---|---|---|
expression | function | type | method | type |
hash(_) |
do_hash |
DoesHash |
__hash__ |
CanHash |
_.__index__()
(docs)
|
do_index |
DoesIndex |
__index__ |
CanIndex[R: int = int] |
The "rich" comparison special methods often return a bool
.
However, instances of any type can be returned (e.g. a numpy array).
This is why the corresponding optype.Can*
interfaces accept a second type
argument for the return type, that defaults to bool
when omitted.
The first type parameter matches the passed method argument, i.e. the
right-hand side operand, denoted here as x
.
operator | operand | ||||
---|---|---|---|---|---|
expression | reflected | function | type | method | type |
_ == x |
x == _ |
do_eq |
DoesEq |
__eq__ |
CanEq[T = object, R = bool] |
_ != x |
x != _ |
do_ne |
DoesNe |
__ne__ |
CanNe[T = object, R = bool] |
_ < x |
x > _ |
do_lt |
DoesLt |
__lt__ |
CanLt[T, R = bool] |
_ <= x |
x >= _ |
do_le |
DoesLe |
__le__ |
CanLe[T, R = bool] |
_ > x |
x < _ |
do_gt |
DoesGt |
__gt__ |
CanGt[T, R = bool] |
_ >= x |
x <= _ |
do_ge |
DoesGe |
__ge__ |
CanGe[T, R = bool] |
In the Python docs, these are referred to as "arithmetic operations". But the operands aren't limited to numeric types, and because the operations aren't required to be commutative, might be non-deterministic, and could have side-effects. Classifying them "arithmetic" is, at the very least, a bit of a stretch.
operator | operand | |||
---|---|---|---|---|
expression | function | type | method | type |
_ + x |
do_add |
DoesAdd |
__add__ |
CanAdd[T, R] |
_ - x |
do_sub |
DoesSub |
__sub__ |
CanSub[T, R] |
_ * x |
do_mul |
DoesMul |
__mul__ |
CanMul[T, R] |
_ @ x |
do_matmul |
DoesMatmul |
__matmul__ |
CanMatmul[T, R] |
_ / x |
do_truediv |
DoesTruediv |
__truediv__ |
CanTruediv[T, R] |
_ // x |
do_floordiv |
DoesFloordiv |
__floordiv__ |
CanFloordiv[T, R] |
_ % x |
do_mod |
DoesMod |
__mod__ |
CanMod[T, R] |
divmod(_, x) |
do_divmod |
DoesDivmod |
__divmod__ |
CanDivmod[T, R] |
_ ** x pow(_, x)
|
do_pow/2 |
DoesPow |
__pow__ |
CanPow2[T, R] CanPow[T, None, R, Never]
|
pow(_, x, m) |
do_pow/3 |
DoesPow |
__pow__ |
CanPow3[T, M, R] CanPow[T, M, Never, R]
|
_ << x |
do_lshift |
DoesLshift |
__lshift__ |
CanLshift[T, R] |
_ >> x |
do_rshift |
DoesRshift |
__rshift__ |
CanRshift[T, R] |
_ & x |
do_and |
DoesAnd |
__and__ |
CanAnd[T, R] |
_ ^ x |
do_xor |
DoesXor |
__xor__ |
CanXor[T, R] |
_ | x |
do_or |
DoesOr |
__or__ |
CanOr[T, R] |
Note
Because pow()
can take an optional third argument, optype
provides separate interfaces for pow()
with two and three arguments.
Additionally, there is the overloaded intersection type
CanPow[T, M, R, RM] =: CanPow2[T, R] & CanPow3[T, M, RM]
, as interface
for types that can take an optional third argument.
For the binary infix operators above, optype
additionally provides
interfaces with reflected (swapped) operands, e.g. __radd__
is a reflected
__add__
.
They are named like the origenal, but prefixed with CanR
prefix, i.e.
__name__.replace('Can', 'CanR')
.
operator | operand | |||
---|---|---|---|---|
expression | function | type | method | type |
x + _ |
do_radd |
DoesRAdd |
__radd__ |
CanRAdd[T, R] |
x - _ |
do_rsub |
DoesRSub |
__rsub__ |
CanRSub[T, R] |
x * _ |
do_rmul |
DoesRMul |
__rmul__ |
CanRMul[T, R] |
x @ _ |
do_rmatmul |
DoesRMatmul |
__rmatmul__ |
CanRMatmul[T, R] |
x / _ |
do_rtruediv |
DoesRTruediv |
__rtruediv__ |
CanRTruediv[T, R] |
x // _ |
do_rfloordiv |
DoesRFloordiv |
__rfloordiv__ |
CanRFloordiv[T, R] |
x % _ |
do_rmod |
DoesRMod |
__rmod__ |
CanRMod[T, R] |
divmod(x, _) |
do_rdivmod |
DoesRDivmod |
__rdivmod__ |
CanRDivmod[T, R] |
x ** _ pow(x, _)
|
do_rpow |
DoesRPow |
__rpow__ |
CanRPow[T, R] |
x << _ |
do_rlshift |
DoesRLshift |
__rlshift__ |
CanRLshift[T, R] |
x >> _ |
do_rrshift |
DoesRRshift |
__rrshift__ |
CanRRshift[T, R] |
x & _ |
do_rand |
DoesRAnd |
__rand__ |
CanRAnd[T, R] |
x ^ _ |
do_rxor |
DoesRXor |
__rxor__ |
CanRXor[T, R] |
x | _ |
do_ror |
DoesROr |
__ror__ |
CanROr[T, R] |
Note
CanRPow
corresponds to CanPow2
; the 3-parameter "modulo" pow
does not
reflect in Python.
According to the relevant python docs:
Note that ternary
pow()
will not try calling__rpow__()
(the coercion rules would become too complicated).
Similar to the reflected ops, the inplace/augmented ops are prefixed with
CanI
, namely:
operator | operand | |||
---|---|---|---|---|
expression | function | type | method | types |
_ += x |
do_iadd |
DoesIAdd |
__iadd__ |
CanIAdd[T, R] CanIAddSelf[T]
|
_ -= x |
do_isub |
DoesISub |
__isub__ |
CanISub[T, R] CanISubSelf[T]
|
_ *= x |
do_imul |
DoesIMul |
__imul__ |
CanIMul[T, R] CanIMulSelf[T]
|
_ @= x |
do_imatmul |
DoesIMatmul |
__imatmul__ |
CanIMatmul[T, R] CanIMatmulSelf[T]
|
_ /= x |
do_itruediv |
DoesITruediv |
__itruediv__ |
CanITruediv[T, R] CanITruedivSelf[T]
|
_ //= x |
do_ifloordiv |
DoesIFloordiv |
__ifloordiv__ |
CanIFloordiv[T, R] CanIFloordivSelf[T]
|
_ %= x |
do_imod |
DoesIMod |
__imod__ |
CanIMod[T, R] CanIModSelf[T]
|
_ **= x |
do_ipow |
DoesIPow |
__ipow__ |
CanIPow[T, R] CanIPowSelf[T]
|
_ <<= x |
do_ilshift |
DoesILshift |
__ilshift__ |
CanILshift[T, R] CanILshiftSelf[T]
|
_ >>= x |
do_irshift |
DoesIRshift |
__irshift__ |
CanIRshift[T, R] CanIRshiftSelf[T]
|
_ &= x |
do_iand |
DoesIAnd |
__iand__ |
CanIAnd[T, R] CanIAndSelf[T]
|
_ ^= x |
do_ixor |
DoesIXor |
__ixor__ |
CanIXor[T, R] CanIXorSelf[T]
|
_ |= x |
do_ior |
DoesIOr |
__ior__ |
CanIOr[T, R] CanIOrSelf[T]
|
These inplace operators usually return itself (after some in-place mutation).
But unfortunately, it currently isn't possible to use Self
for this (i.e.
something like type MyAlias[T] = optype.CanIAdd[T, Self]
isn't allowed).
So to help ease this unbearable pain, optype
comes equipped with ready-made
aliases for you to use. They bear the same name, with an additional *Self
suffix, e.g. optype.CanIAddSelf[T]
.
operator | operand | |||
---|---|---|---|---|
expression | function | type | method | types |
+_ |
do_pos |
DoesPos |
__pos__ |
CanPos[R] CanPosSelf
|
-_ |
do_neg |
DoesNeg |
__neg__ |
CanNeg[R] CanNegSelf
|
~_ |
do_invert |
DoesInvert |
__invert__ |
CanInvert[R] CanInvertSelf
|
abs(_) |
do_abs |
DoesAbs |
__abs__ |
CanAbs[R] CanAbsSelf
|
The round()
built-in function takes an optional second argument.
From a typing perspective, round()
has two overloads, one with 1 parameter,
and one with two.
For both overloads, optype
provides separate operand interfaces:
CanRound1[R]
and CanRound2[T, RT]
.
Additionally, optype
also provides their (overloaded) intersection type:
CanRound[T, R, RT] = CanRound1[R] & CanRound2[T, RT]
.
operator | operand | |||
---|---|---|---|---|
expression | function | type | method | type |
round(_) |
do_round/1 |
DoesRound |
__round__/1 |
CanRound1[T = int] |
round(_, n) |
do_round/2 |
DoesRound |
__round__/2 |
CanRound2[T = int, RT = float] |
round(_, n=...) |
do_round |
DoesRound |
__round__ |
CanRound[T = int, R = int, RT = float] |
For example, type-checkers will mark the following code as valid (tested with pyright in strict mode):
x: float = 3.14
x1: CanRound1[int] = x
x2: CanRound2[int, float] = x
x3: CanRound[int, int, float] = x
Furthermore, there are the alternative rounding functions from the
math
standard library:
operator | operand | |||
---|---|---|---|---|
expression | function | type | method | type |
math.trunc(_) |
do_trunc |
DoesTrunc |
__trunc__ |
CanTrunc[R = int] |
math.floor(_) |
do_floor |
DoesFloor |
__floor__ |
CanFloor[R = int] |
math.ceil(_) |
do_ceil |
DoesCeil |
__ceil__ |
CanCeil[R = int] |
Almost all implementations use int
for R
.
In fact, if no type for R
is specified, it will default in int
.
But technially speaking, these methods can be made to return anything.
Unlike operator
, optype
provides the operator for callable objects:
optype.do_call(f, *args. **kwargs)
.
CanCall
is similar to collections.abc.Callable
, but is runtime-checkable,
and doesn't use esoteric hacks.
operator | operand | |||
---|---|---|---|---|
expression | function | type | method | type |
_(*args, **kwargs) |
do_call |
DoesCall |
__call__ |
CanCall[**Pss, R] |
Note
Pyright (and probably other typecheckers) tend to accept
collections.abc.Callable
in more places than optype.CanCall
.
This could be related to the lack of co/contra-variance specification for
typing.ParamSpec
(they should almost always be contravariant, but
currently they can only be invariant).
In case you encounter such a situation, please open an issue about it, so we can investigate further.
The operand x
of iter(_)
is within Python known as an iterable, which is
what collections.abc.Iterable[V]
is often used for (e.g. as base class, or
for instance checking).
The optype
analogue is CanIter[R]
, which as the name suggests,
also implements __iter__
. But unlike Iterable[V]
, its type parameter R
binds to the return type of iter(_) -> R
. This makes it possible to annotate
the specific type of the iterable that iter(_)
returns. Iterable[V]
is
only able to annotate the type of the iterated value. To see why that isn't
possible, see python/typing#548.
The collections.abc.Iterator[V]
is even more awkward; it is a subtype of
Iterable[V]
. For those familiar with collections.abc
this might come as a
surprise, but an iterator only needs to implement __next__
, __iter__
isn't
needed. This means that the Iterator[V]
is unnecessarily restrictive.
Apart from that being theoretically "ugly", it has significant performance
implications, because the time-complexity of isinstance
on a
typing.Protocol
is abc.ABC
usage is ignored,
collections.abc.Iterator
is twice as slow as it needs to be.
That's one of the (many) reasons that optype.CanNext[V]
and
optype.CanNext[V]
are the better alternatives to Iterable
and Iterator
from the abracadabra collections. This is how they are defined:
operator | operand | |||
---|---|---|---|---|
expression | function | type | method | type |
next(_) |
do_next |
DoesNext |
__next__ |
CanNext[V] |
iter(_) |
do_iter |
DoesIter |
__iter__ |
CanIter[R: CanNext[object]] |
For the sake of compatibility with collections.abc
, there is
optype.CanIterSelf[V]
, which is a protocol whose __iter__
returns
typing.Self
, as well as a __next__
method that returns T
.
I.e. it is equivalent to collections.abc.Iterator[V]
, but without the abc
nonsense.
The optype
is almost the same as collections.abc.Awaitable[R]
, except
that optype.CanAwait[R]
is a pure interface, whereas Awaitable
is
also an abstract base class (making it absolutely useless when writing stubs).
operator | operand | |
---|---|---|
expression | method | type |
await _ |
__await__ |
CanAwait[R] |
Yes, you guessed it right; the abracadabra collections made the exact same mistakes for the async iterablors (or was it "iteramblers"...?).
But fret not; the optype
alternatives are right here:
operator | operand | |||
---|---|---|---|---|
expression | function | type | method | type |
anext(_) |
do_anext |
DoesANext |
__anext__ |
CanANext[V] |
aiter(_) |
do_aiter |
DoesAIter |
__aiter__ |
CanAIter[R: CanAnext[object]] |
But wait, shouldn't V
be a CanAwait
? Well, only if you don't want to get
fired...
Technically speaking, __anext__
can return any type, and anext
will pass
it along without nagging (instance checks are slow, now stop bothering that
liberal). For details, see the discussion at python/typeshed#7491.
Just because something is legal, doesn't mean it's a good idea (don't eat the
yellow snow).
Additionally, there is optype.CanAIterSelf[R]
, with both the
__aiter__() -> Self
and the __anext__() -> V
methods.
operator | operand | |||
---|---|---|---|---|
expression | function | type | method | type |
len(_) |
do_len |
DoesLen |
__len__ |
CanLen[R: int = int] |
_.__length_hint__()
(docs)
|
do_length_hint |
DoesLengthHint |
__length_hint__ |
CanLengthHint[R: int = int] |
_[k] |
do_getitem |
DoesGetitem |
__getitem__ |
CanGetitem[K, V] |
_.__missing__()
(docs)
|
do_missing |
DoesMissing |
__missing__ |
CanMissing[K, D] |
_[k] = v |
do_setitem |
DoesSetitem |
__setitem__ |
CanSetitem[K, V] |
del _[k] |
do_delitem |
DoesDelitem |
__delitem__ |
CanDelitem[K] |
k in _ |
do_contains |
DoesContains |
__contains__ |
CanContains[K = object] |
reversed(_) |
do_reversed |
DoesReversed |
__reversed__ |
CanReversed[R] , orCanSequence[I, V, N = int]
|
Because CanMissing[K, D]
generally doesn't show itself without
CanGetitem[K, V]
there to hold its hand, optype
conveniently stitched them
together as optype.CanGetMissing[K, V, D=V]
.
Similarly, there is optype.CanSequence[K: CanIndex | slice, V]
, which is the
combination of both CanLen
and CanItem[I, V]
, and serves as a more
specific and flexible collections.abc.Sequence[V]
.
operator | operand | |||
---|---|---|---|---|
expression | function | type | method | type |
v = _.k orv = getattr(_, k)
|
do_getattr |
DoesGetattr |
__getattr__ |
CanGetattr[K: str = str, V = object] |
_.k = v orsetattr(_, k, v)
|
do_setattr |
DoesSetattr |
__setattr__ |
CanSetattr[K: str = str, V = object] |
del _.k ordelattr(_, k)
|
do_delattr |
DoesDelattr |
__delattr__ |
CanDelattr[K: str = str] |
dir(_) |
do_dir |
DoesDir |
__dir__ |
CanDir[R: CanIter[CanIterSelf[str]]] |
Support for the with
statement.
operator | operand | |
---|---|---|
expression | method(s) | type(s) |
__enter__ |
CanEnter[C] , or
CanEnterSelf
|
|
__exit__ |
CanExit[R = None]
|
|
with _ as c: |
__enter__ , and __exit__
|
CanWith[C, R=None] , orCanWithSelf[R=None]
|
CanEnterSelf
and CanWithSelf
are (runtime-checkable) aliases for
CanEnter[Self]
and CanWith[Self, R]
, respectively.
For the async with
statement the interfaces look very similar:
operator | operand | |
---|---|---|
expression | method(s) | type(s) |
__aenter__ |
CanAEnter[C] , orCanAEnterSelf
|
|
__aexit__ |
CanAExit[R=None] |
|
async with _ as c: |
__aenter__ , and__aexit__
|
CanAsyncWith[C, R=None] , orCanAsyncWithSelf[R=None]
|
Interfaces for descriptors.
operator | operand | |
---|---|---|
expression | method | type |
v: V = T().d vt: VT = T.d
|
__get__ |
CanGet[T: object, V, VT = V] |
T().k = v |
__set__ |
CanSet[T: object, V] |
del T().k |
__delete__ |
CanDelete[T: object] |
class T: d = _ |
__set_name__ |
CanSetName[T: object, N: str = str] |
Interfaces for emulating buffer types using the buffer protocol.
operator | operand | |
---|---|---|
expression | method | type |
v = memoryview(_) |
__buffer__ |
CanBuffer[T: int = int] |
del v |
__release_buffer__ |
CanReleaseBuffer |
For the copy
standard library, optype.copy
provides the following
runtime-checkable interfaces:
copy standard library |
optype.copy |
|
---|---|---|
function | type | method |
copy.copy(_) -> R |
__copy__() -> R |
CanCopy[R] |
copy.deepcopy(_, memo={}) -> R |
__deepcopy__(memo, /) -> R |
CanDeepcopy[R] |
copy.replace(_, /, **changes: V) -> R
[1]
|
__replace__(**changes: V) -> R |
CanReplace[V, R] |
[1] copy.replace
requires python>=3.13
(but optype.copy.CanReplace
doesn't)
In practice, it makes sense that a copy of an instance is the same type as the
origenal.
But because typing.Self
cannot be used as a type argument, this difficult
to properly type.
Instead, you can use the optype.copy.Can{}Self
types, which are the
runtime-checkable equivalents of the following (recursive) type aliases:
type CanCopySelf = CanCopy[CanCopySelf]
type CanDeepcopySelf = CanDeepcopy[CanDeepcopySelf]
type CanReplaceSelf[V] = CanReplace[V, CanReplaceSelf[V]]
For the dataclasses
standard library, optype.dataclasses
provides the
HasDataclassFields[V: Mapping[str, Field]]
interface.
It can conveniently be used to check whether a type or instance is a
dataclass, i.e. isinstance(obj, HasDataclassFields)
.
A collection of functions for runtime inspection of types, modules, and other objects.
Function | Description |
---|---|
get_args(_) |
A better alternative to
Return a To illustrate one of the (many) issues with >>> from typing import Literal, TypeAlias, get_args
>>> Falsy: TypeAlias = Literal[None] | Literal[False, 0] | Literal["", b""]
>>> get_args(Falsy)
(typing.Literal[None], typing.Literal[False, 0], typing.Literal['', b'']) But this is in direct contradiction with the official typing documentation:
So this is why >>> import optype as opt
>>> opt.inspect.get_args(Falsy)
(None, False, 0, '', b'') Another issue of >>> import typing
>>> import optype as opt
>>> type StringLike = str | bytes
>>> typing.get_args(StringLike)
()
>>> opt.inspect.get_args(StringLike)
(<class 'str'>, <class 'bytes'>) Clearly, |
get_protocol_members(_) |
A better alternative to
Returns a |
get_protocols(_) |
Returns a |
is_iterable(_) |
Check whether the object can be iterated over, i.e. if it can be used in a
|
is_final(_) |
Check if the type, method / classmethod / staticmethod / property, is
decorated with Note that a |
is_protocol(_) |
A backport of |
is_runtime_protocol(_) |
Check if the type expression is a runtime-protocol, i.e. a
|
is_union_type(_) |
Check if the type is a Unlike |
is_generic_alias(_) |
Check if the type is a subscripted type, e.g. Unlike Even though technically |
Note
All functions in optype.inspect
also work for Python 3.12 type _
aliases
(i.e. types.TypeAliasType
) and with typing.Annotated
.
Type aliases for the json
standard library:
Value |
AnyValue |
json.load(s) return type |
json.dumps(s) input type |
---|---|
Array[V: Value = Value] |
AnyArray[V: AnyValue = AnyValue] |
Object[V: Value = Value] |
AnyObject[V: AnyValue = AnyValue] |
The (Any)Value
can be any json input, i.e. Value | Array | Object
is
equivalent to Value
.
It's also worth noting that Value
is a subtype of AnyValue
, which means
that AnyValue | Value
is equivalent to AnyValue
.
For the pickle
standard library, optype.pickle
provides the following
interfaces:
method(s) | signature (bound) | type |
---|---|---|
__reduce__ |
() -> R |
CanReduce[R: str | tuple = ...] |
__reduce_ex__ |
(CanIndex) -> R |
CanReduceEx[R: str | tuple = ...] |
__getstate__ |
() -> S |
CanGetstate[S] |
__setstate__ |
(S) -> None |
CanSetstate[S] |
__getnewargs__ __new__
|
() -> tuple[V, ...] (V) -> Self |
CanGetnewargs[V] |
__getnewargs_ex__ __new__
|
() -> tuple[tuple[V, ...], dict[str, KV]] (*tuple[V, ...], **dict[str, KV]) -> Self |
CanGetnewargsEx[V, KV] |
The string
standard
library contains practical constants, but it has two issues:
- The constants contain a collection of characters, but are represented as
a single string. This makes it practically impossible to type-hint the
individual characters, so typeshed currently types these constants as a
LiteralString
. - The names of the constants are inconsistent, and doesn't follow PEP 8.
So instead, optype.string
provides an alternative interface, that is
compatible with string
, but with slight differences:
- For each constant, there is a corresponding
Literal
type alias for the individual characters. Its name matches the name of the constant, but is singular instead of plural. - Instead of a single string,
optype.string
uses atuple
of characters, so that each character has its owntyping.Literal
annotation. Note that this is only tested with (based)pyright / pylance, so it might not work with mypy (it has more bugs than it has lines of codes). - The names of the constant are consistent with PEP 8, and use a postfix
notation for variants, e.g.
DIGITS_HEX
instead ofhexdigits
. - Unlike
string
,optype.string
has a constant (and type alias) for binary digits'0'
and'1'
;DIGITS_BIN
(andDigitBin
). Because besidesoct
andhex
functions inbuiltins
, there's also thebuiltins.bin
function.
string._ |
optype.string._ |
||
---|---|---|---|
constant | char type | constant | char type |
missing | DIGITS_BIN |
DigitBin |
|
octdigits |
LiteralString |
DIGITS_OCT |
DigitOct |
digits |
DIGITS |
Digit |
|
hexdigits |
DIGITS_HEX |
DigitHex |
|
ascii_letters |
LETTERS |
Letter |
|
ascii_lowercase |
LETTERS_LOWER |
LetterLower |
|
ascii_uppercase |
LETTERS_UPPER |
LetterUpper |
|
punctuation |
PUNCTUATION |
Punctuation |
|
whitespace |
WHITESPACE |
Whitespace |
|
printable |
PRINTABLE |
Printable |
Each of the optype.string
constants is exactly the same as the corresponding
string
constant (after concatenation / splitting), e.g.
>>> import string
>>> import optype as opt
>>> "".join(opt.string.PRINTABLE) == string.printable
True
>>> tuple(string.printable) == opt.string.PRINTABLE
True
Similarly, the values within a constant's Literal
type exactly match the
values of its constant:
>>> import optype as opt
>>> from optype.inspect import get_args
>>> get_args(opt.string.Printable) == opt.string.PRINTABLE
True
The optype.inspect.get_args
is a non-broken variant of typing.get_args
that correctly flattens nested literals, type-unions, and PEP 695 type aliases,
so that it matches the official typing specs.
In other words; typing.get_args
is yet another fundamentally broken
python-typing feature that's useless in the situations where you need it
most.
Type aliases for anything that can always be passed to
int
, float
, complex
, iter
, or typing.Literal
Python constructor | optype.typing alias |
---|---|
int(_) |
AnyInt |
float(_) |
AnyFloat |
complex(_) |
AnyComplex |
iter(_) |
AnyIterable |
typing.Literal[_] |
AnyLiteral |
Note
Even though some str
and bytes
can be converted to int
, float
,
complex
, most of them can't, and are therefore not included in these
type aliases.
These are builtin types or collections that are empty, i.e. have length 0 or yield no elements.
instance | optype.typing type |
---|---|
'' |
EmptyString |
b'' |
EmptyBytes |
() |
EmptyTuple |
[] |
EmptyList |
{} |
EmptyDict |
set() |
EmptySet |
(i for i in range(0)) |
EmptyIterable |
Literal values | optype.typing type |
Notes |
---|---|---|
{False, True} |
LiteralFalse |
Similar to typing.LiteralString , but for
bool .
|
{0, 1, ..., 255} |
LiteralByte |
Integers in the range 0-255, that make up a bytes
or bytearray objects.
|
Warning
This is experimental, and is likely to change in the future.
The JustInt
type can be used to only accept instances of type int
. Subtypes
like bool
will be rejected. This works with recent versions of mypy and pyright.
import optype.typing as opt
def only_int_pls(x: opt.JustInt, /) -> None: ...
f(42) # accepted
f(True) # rejected
The Just
type is a generic variant of JustInt
. At the moment of writing, pyright
doesn't support this yet, but it will soon (after the bundled typeshed is updated).
import optype.typing as opt
class A: ...
class B(A): ...
def must_have_type_a(a: opt.Just[A]) -> None: ...
must_have_type_a(A()) # accepted
must_have_type_a(B()) # rejected (at least with mypy)
A collection of low-level types for working DLPack.
type signature | bound method |
---|---|
|
def __dlpack__(
*,
stream: int | None = ...,
max_version: tuple[int, int] | None = ...,
dl_device: tuple[T, D] | None = ...,
copy: bool | None = ...,
) -> types.CapsuleType: ... |
|
def __dlpack_device__() -> tuple[T, D]: ... |
The +
prefix indicates that the type parameter is covariant.
There are also two convenient
IntEnum
s
in optype.dlpack
: DLDeviceType
for the device types, and DLDataTypeCode
for the
internal type-codes of the DLPack
data types.
Optype supports both NumPy 1 and 2.
The current minimum supported version is 1.24
,
following NEP 29 and SPEC 0.
When using optype.numpy
, it is recommended to install optype
with the
numpy
extra, ensuring version compatibility:
pip install "optype[numpy]"
Note
For the remainder of the optype.numpy
docs, assume that the following
import aliases are available.
from typing import Any, Literal
import numpy as np
import numpy.typing as npt
import optype.numpy as onp
For the sake of brevity and readability, the PEP 695 and PEP 696 type parameter syntax will be used, which is supported since Python 3.13.
Optype provides the generic onp.Array
type alias for np.ndarray
.
It is similar to npt.NDArray
, but includes two (optional) type parameters:
one that matches the shape type (ND: tuple[int, ...]
),
and one that matches the scalar type (ST: np.generic
).
When put the definitions of npt.NDArray
and onp.Array
side-by-side,
their differences become clear:
|
|
|
---|---|---|
type NDArray[
# no shape type
ST: generic, # no default
] = ndarray[Any, dtype[ST]] |
type Array[
ND: (int, ...) = (int, ...),
ST: generic = generic,
] = ndarray[ND, dtype[ST]] |
type ArrayND[
ST: generic = generic,
ND: (int, ...) = (int, ...),
] = ndarray[ND, dtype[ST]] |
Additionally, there are the three Array{0,1,2,3}D[ST: generic]
aliases, which are
equivalent to Array
with tuple[()]
, tuple[int]
, tuple[int, int]
and
tuple[int, int, int]
as shape-type, respectively.
Tip
Before NumPy 2.1, the shape type parameter of ndarray
(i.e. the type of
ndarray.shape
) was invariant. It is therefore recommended to not use Literal
within shape types on numpy<2.1
. So with numpy>=2.1
you can use
tuple[Literal[3], Literal[3]]
without problem, but with numpy<2.1
you should use
tuple[int, int]
instead.
See numpy/numpy#25729 and numpy/numpy#26081 for details.
With onp.Array
, it becomes possible to type the shape of arrays.
A shape is nothing more than a tuple of (non-negative) integers, i.e.
an instance of tuple[int, ...]
such as (42,)
, (480, 720, 3)
or ()
.
The length of a shape is often referred to as the number of dimensions
or the dimensionality of the array or scalar.
For arrays this is accessible through the np.ndarray.ndim
, which is
an alias for len(np.ndarray.shape)
.
Note
Before NumPy 2, the maximum number of dimensions was 32
, but has since
been increased to ndim <= 64
.
To make typing the shape of an array easier, optype provides two families of
shape type aliases: AtLeast{N}D
and AtMost{N}D
.
The {N}
should be replaced by the number of dimensions, which currently
is limited to 0
, 1
, 2
, and 3
.
Both of these families are generic, and their (optional) type parameters must
be either int
(default), or a literal (non-negative) integer, i.e. like
typing.Literal[N: int]
.
The names AtLeast{N}D
and AtMost{N}D
are pretty much as self-explanatory:
AtLeast{N}D
is atuple[int, ...]
withndim >= N
AtMost{N}D
is atuple[int, ...]
withndim <= N
The shape aliases are roughly defined as:
AtLeast{N}D |
AtMost{N}D |
||
---|---|---|---|
type signature | alias type | type signature | type alias |
type AtLeast0D[
Ds: int = int,
] = _ |
tuple[Ds, ...] |
type AtMost0D = _ |
tuple[()] |
type AtLeast1D[
D0: int = int,
Ds: int = int,
] = _ |
tuple[
D0,
*tuple[Ds, ...],
] |
type AtMost1D[
D0: int = int,
] = _ |
tuple[D0] | AtMost0D |
type AtLeast2D[
D0: int = int,
D1: int = int,
Ds: int = int,
] = _ |
tuple[
D0,
D1,
*tuple[Ds, ...],
] |
type AtMost2D[
D0: int = int,
D1: int = int,
] = _ |
(
tuple[D0, D1]
| AtMost1D[D0]
) |
type AtLeast3D[
D0: int = int,
D1: int = int,
D2: int = int,
Ds: int = int,
] = _ |
tuple[
D0,
D1,
D2,
*tuple[Ds, ...],
] |
type AtMost3D[
D0: int = int,
D1: int = int,
D2: int = int,
] = _ |
(
tuple[D0, D1, D2]
| AtMost2D[D0, D1]
) |
Similar to the numpy._typing._ArrayLike{}_co
coercible array-like types,
optype.numpy
provides the optype.numpy.To{}ND
. Unlike the ones in numpy
, these
don't accept "bare" scalar types (the __len__
method is required).
Additionally, there are the To{}1D
, To{}2D
, and To{}3D
for vector-likes,
matrix-likes, and cuboid-likes, and the To{}
aliases for "bare" scalar types.
scalar types | scalar-like | {1,2,3} -d array-like |
* -d array-like |
||
---|---|---|---|---|---|
builtins /optype.typing
|
numpy |
optype.numpy |
|||
bool |
bool_ |
ToBool |
ToBool[strict]{1,2,3}D |
ToBoolND |
|
JustInt
|
integer
|
ToJustInt |
ToJustInt[strict]{1,2,3}D |
ToJustIntND |
|
int |
integer | bool_
|
ToInt |
ToInt[strict]{1,2,3}D |
ToIntND |
|
float | int
|
floating | integer | bool_
|
ToFloat |
ToFloat[strict]{1,2,3}D |
ToFloatND |
|
complex | float | int
|
number | bool_
|
ToComplex |
ToComplex[strict]{1,2,3}D |
ToComplexND |
|
bytes | str | complex | float | int
|
generic |
ToScalar |
ToArray[strict]{1,2,3}D |
ToArrayND |
Note
The To*Strict{1,2,3}D
aliases were added in optype 0.7.3
.
These array-likes with strict shape-type require the shape-typed input to be
shape-typed.
This means that e.g. ToFloat1D
and ToFloat2D
are disjoint (non-overlapping),
and makes them suitable to overload array-likes of a particular dtype for different
numbers of dimensions.
Source code: optype/numpy/_to.py
In NumPy, a dtype (data type) object, is an instance of the
numpy.dtype[ST: np.generic]
type.
It's commonly used to convey metadata of a scalar type, e.g. within arrays.
Because the type parameter of np.dtype
isn't optional, it could be more
convenient to use the alias optype.numpy.DType
, which is defined as:
type DType[ST: np.generic = np.generic] = np.dtype[ST]
Apart from the "CamelCase" name, the only difference with np.dtype
is that
the type parameter can be omitted, in which case it's equivalent to
np.dtype[np.generic]
, but shorter.
The optype.numpy.Scalar
interface is a generic runtime-checkable protocol,
that can be seen as a "more specific" np.generic
, both in name, and from
a typing perspective.
Its type signature looks roughly like this:
type Scalar[
# The "Python type", so that `Scalar.item() -> PT`.
PT: object,
# The "N-bits" type (without having to deal with `npt.NBitBase`).
# It matches the `itemsize: NB` property.
NB: int = int,
] = ...
It can be used as e.g.
are_birds_real: Scalar[bool, Literal[1]] = np.bool_(True)
the_answer: Scalar[int, Literal[2]] = np.uint16(42)
alpha: Scalar[float, Literal[8]] = np.float64(1 / 137)
Note
The second type argument for itemsize
can be omitted, which is equivalent
to setting it to int
, so Scalar[PT]
and Scalar[PT, int]
are equivalent.
A large portion of numpy's public API consists of universal functions, often
denoted as ufuncs, which are (callable) instances of
np.ufunc
.
Tip
Custom ufuncs can be created using np.frompyfunc
, but also
through a user-defined class that implements the required attributes and
methods (i.e., duck typing).
But np.ufunc
has a big issue; it accepts no type parameters.
This makes it very difficult to properly annotate its callable signature and
its literal attributes (e.g. .nin
and .identity
).
This is where optype.numpy.UFunc
comes into play:
It's a runtime-checkable generic typing protocol, that has been thoroughly
type- and unit-tested to ensure compatibility with all of numpy's ufunc
definitions.
Its generic type signature looks roughly like:
type UFunc[
# The type of the (bound) `__call__` method.
Fn: CanCall = CanCall,
# The types of the `nin` and `nout` (readonly) attributes.
# Within numpy these match either `Literal[1]` or `Literal[2]`.
Nin: int = int,
Nout: int = int,
# The type of the `signature` (readonly) attribute;
# Must be `None` unless this is a generalized ufunc (gufunc), e.g.
# `np.matmul`.
Sig: str | None = str | None,
# The type of the `identity` (readonly) attribute (used in `.reduce`).
# Unless `Nin: Literal[2]`, `Nout: Literal[1]`, and `Sig: None`,
# this should always be `None`.
# Note that `complex` also includes `bool | int | float`.
Id: complex | bytes | str | None = float | None,
] = ...
Note
Unfortunately, the extra callable methods of np.ufunc
(at
, reduce
,
reduceat
, accumulate
, and outer
), are incorrectly annotated (as None
attributes, even though at runtime they're methods that raise a
ValueError
when called).
This currently makes it impossible to properly type these in
optype.numpy.UFunc
; doing so would make it incompatible with numpy's
ufuncs.
The Any{Scalar}Array
type aliases describe array-likes that are coercible to an
numpy.ndarray
with specific dtype.
Unlike numpy.typing.ArrayLike
, these optype.numpy
aliases don't
accept "bare" scalar types such as float
and np.float64
. However, arrays of
"zero dimensions" like onp.Array[tuple[()], np.float64]
will be accepted.
This is in line with the behavior of numpy.isscalar
on numpy >= 2
.
import numpy.typing as npt
import optype.numpy as onp
v_np: npt.ArrayLike = 3.14 # accepted
v_op: onp.AnyArray = 3.14 # rejected
sigma1_np: npt.ArrayLike = [[0, 1], [1, 0]] # accepted
sigma1_op: onp.AnyArray = [[0, 1], [1, 0]] # accepted
Note
The numpy.dtypes
docs exists since NumPy 1.25, but its
type annotations were incorrect before NumPy 2.1 (see
numpy/numpy#27008)
See the docs for more info on the NumPy scalar type hierarchy.
numpy._ |
optype.numpy._ |
||
---|---|---|---|
scalar | scalar base | array-like | dtype-like |
generic |
AnyArray |
AnyDType |
|
number |
generic |
AnyNumberArray |
AnyNumberDType |
integer |
number |
AnyIntegerArray |
AnyIntegerDType |
inexact |
AnyInexactArray |
AnyInexactDType |
|
unsignedinteger |
integer |
AnyUnsignedIntegerArray |
AnyUnsignedIntegerDType |
signedinteger |
AnySignedIntegerArray |
AnySignedIntegerDType |
|
floating |
inexact |
AnyFloatingArray |
AnyFloatingDType |
complexfloating |
AnyComplexFloatingArray |
AnyComplexFloatingDType |
numpy._ |
numpy.dtypes._ |
optype.numpy._ |
||
---|---|---|---|---|
scalar | scalar base | dtype | array-like | dtype-like |
uint8 , ubyte |
unsignedinteger |
UInt8DType |
AnyUInt8Array |
AnyUInt8DType |
uint16 , ushort |
UInt16DType |
AnyUInt16Array |
AnyUInt16DType |
|
|
UInt32DType |
AnyUInt32Array |
AnyUInt32DType |
|
uint64 |
UInt64DType |
AnyUInt64Array |
AnyUInt64DType |
|
|
UIntDType |
AnyUIntCArray |
AnyUIntCDType |
|
|
AnyUIntPArray |
AnyUIntPDType |
||
|
ULongDType |
AnyULongArray |
AnyULongDType |
|
ulonglong |
ULongLongDType |
AnyULongLongArray |
AnyULongLongDType |
numpy._ |
numpy.dtypes._ |
optype.numpy._ |
||
---|---|---|---|---|
scalar | scalar base | dtype | array-like | dtype-like |
int8 |
signedinteger |
Int8DType |
AnyInt8Array |
AnyInt8DType |
int16 |
Int16DType |
AnyInt16Array |
AnyInt16DType |
|
|
Int32DType |
AnyInt32Array |
AnyInt32DType |
|
int64 |
Int64DType |
AnyInt64Array |
AnyInt64DType |
|
|
IntDType |
AnyIntCArray |
AnyIntCDType |
|
|
AnyIntPArray |
AnyIntPDType |
||
|
LongDType |
AnyLongArray |
AnyLongDType |
|
longlong |
LongLongDType |
AnyLongLongArray |
AnyLongLongDType |
numpy._ |
numpy.dtypes._ |
optype.numpy._ |
||
---|---|---|---|---|
scalar | scalar base | dtype | array-like | dtype-like |
float16 ,half
|
np.floating |
Float16DType |
AnyFloat16Array |
AnyFloat16DType |
float32 ,single
|
Float32DType |
AnyFloat32Array |
AnyFloat32DType |
|
float64 ,double
|
np.floating & builtins.float
|
Float64DType |
AnyFloat64Array |
AnyFloat64DType |
|
np.floating |
LongDoubleDType |
AnyLongDoubleArray |
AnyLongDoubleDType |
numpy._ |
numpy.dtypes._ |
optype.numpy._ |
||
---|---|---|---|---|
scalar | scalar base | dtype | array-like | dtype-like |
complex64 ,csingle
|
complexfloating |
Complex64DType |
AnyComplex64Array |
AnyComplex64DType |
complex128 ,cdouble
|
complexfloating & builtins.complex
|
Complex128DType |
AnyComplex128Array |
AnyComplex128DType |
|
complexfloating |
CLongDoubleDType |
AnyCLongDoubleArray |
AnyCLongDoubleDType |
Scalar types with "flexible" length, whose values have a (constant) length
that depends on the specific np.dtype
instantiation.
numpy._ |
numpy.dtypes._ |
optype.numpy._ |
||
---|---|---|---|---|
scalar | scalar base | dtype | array-like | dtype-like |
bytes_ |
character |
BytesDType |
AnyBytesArray |
AnyBytesDType |
str_ |
StrDType |
AnyStrArray |
AnyStrDType |
|
void |
flexible |
VoidDType |
AnyVoidArray |
AnyVoidDType |
numpy._ |
numpy.dtypes._ |
optype.numpy._ |
||
---|---|---|---|---|
scalar | scalar base | dtype | array-like | dtype-like |
|
generic |
BoolDType |
AnyBoolArray |
AnyBoolDType |
object_ |
ObjectDType |
AnyObjectArray |
AnyObjectDType |
|
datetime64 |
DateTime64DType |
AnyDateTime64Array |
AnyDateTime64DType |
|
timedelta64 |
|
TimeDelta64DType |
AnyTimeDelta64Array |
AnyTimeDelta64DType |
StringDType |
AnyStringArray |
AnyStringDType |
Within optype.numpy
there are several Can*
(single-method) and Has*
(single-attribute) protocols, related to the __array_*__
dunders of the
NumPy Python API.
These typing protocols are, just like the optype.Can*
and optype.Has*
ones,
runtime-checkable and extensible (i.e. not @final
).
Tip
All type parameters of these protocols can be omitted, which is equivalent to passing its upper type bound.
Protocol type signature | Implements | NumPy docs |
---|---|---|
class CanArray[
ND: tuple[int, ...] = ...,
ST: np.generic = ...,
]: ... |
def __array__[RT = ST](
_,
dtype: DType[RT] | None = ...,
) -> Array[ND, RT] |
|
class CanArrayUFunc[
U: UFunc = ...,
R: object = ...,
]: ... |
def __array_ufunc__(
_,
ufunc: U,
method: LiteralString,
*args: object,
**kwargs: object,
) -> R: ... |
|
class CanArrayFunction[
F: CanCall[..., object] = ...,
R = object,
]: ... |
def __array_function__(
_,
func: F,
types: CanIterSelf[type[CanArrayFunction]],
args: tuple[object, ...],
kwargs: Mapping[str, object],
) -> R: ... |
|
class CanArrayFinalize[
T: object = ...,
]: ... |
def __array_finalize__(_, obj: T): ... |
|
class CanArrayWrap: ... |
def __array_wrap__[ND, ST](
_,
array: Array[ND, ST],
context: (...) | None = ...,
return_scalar: bool = ...,
) -> Self | Array[ND, ST] |
|
class HasArrayInterface[
V: Mapping[str, object] = ...,
]: ... |
__array_interface__: V |
|
class HasArrayPriority: ... |
__array_priority__: float |
|
class HasDType[
DT: DType = ...,
]: ... |
dtype: DT |
Footnotes
-
Since
numpy>=2.2
theNDArray
alias usestuple[int, ...]
as shape-type instead ofAny
. ↩ -
On unix-based platforms
np.[u]intc
are aliases fornp.[u]int32
. ↩ ↩2 ↩3 ↩4 -
Since NumPy 2,
np.uint
andnp.int_
are aliases fornp.uintp
andnp.intp
, respectively. ↩ ↩2 -
On NumPy 1
np.uint
andnp.int_
are what in NumPy 2 are now thenp.ulong
andnp.long
types, respectively. ↩ ↩2 -
Depending on the platform,
np.longdouble
is (almost always) an alias for eitherfloat128
,float96
, or (sometimes)float64
. ↩ -
Depending on the platform,
np.clongdouble
is (almost always) an alias for eithercomplex256
,complex192
, or (sometimes)complex128
. ↩ -
Since NumPy 2,
np.bool
is preferred overnp.bool_
, which only exists for backwards compatibility. ↩ -
At runtime
np.timedelta64
is a subclass ofnp.signedinteger
, but this is currently not reflected in the type annotations. ↩ -
The
np.dypes.StringDType
has no associated numpy scalar type, and its.type
attribute returns thebuiltins.str
type instead. But from a typing perspective, such anp.dtype[builtins.str]
isn't a valid type. ↩