You can not select more than 25 topics Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.
nvim_config/typings/matplotlib/transforms.pyi

605 lines
13 KiB

"""
This type stub file was generated by pyright.
"""
import numpy as np
from .path import Path
from .patches import Patch
from .figure import Figure
from numpy.typing import ArrayLike
from collections.abc import Iterable, Sequence
from typing import Literal
DEBUG: bool
class TransformNode:
INVALID_NON_AFFINE: int
INVALID_AFFINE: int
INVALID: int
is_bbox: bool
@property
def is_affine(self) -> bool:
...
pass_through: bool
def __init__(self, shorthand_name: str | None = ...) -> None:
...
def __copy__(self) -> TransformNode:
...
def invalidate(self) -> None:
...
def set_children(self, *children: TransformNode) -> None:
...
def frozen(self) -> TransformNode:
...
class BboxBase(TransformNode):
is_bbox: bool
is_affine: bool
def frozen(self) -> Bbox:
...
def __array__(self, *args, **kwargs):
...
@property
def x0(self) -> float:
...
@property
def y0(self) -> float:
...
@property
def x1(self) -> float:
...
@property
def y1(self) -> float:
...
@property
def p0(self) -> tuple[float, float]:
...
@property
def p1(self) -> tuple[float, float]:
...
@property
def xmin(self) -> float:
...
@property
def ymin(self) -> float:
...
@property
def xmax(self) -> float:
...
@property
def ymax(self) -> float:
...
@property
def min(self) -> tuple[float, float]:
...
@property
def max(self) -> tuple[float, float]:
...
@property
def intervalx(self) -> tuple[float, float]:
...
@property
def intervaly(self) -> tuple[float, float]:
...
@property
def width(self) -> float:
...
@property
def height(self) -> float:
...
@property
def size(self) -> tuple[float, float]:
...
@property
def bounds(self) -> tuple[float, float, float, float]:
...
@property
def extents(self) -> tuple[float, float, float, float]:
...
def get_points(self) -> np.ndarray:
...
def containsx(self, x: float) -> bool:
...
def containsy(self, y: float) -> bool:
...
def contains(self, x: float, y: float) -> bool:
...
def overlaps(self, other: BboxBase) -> bool:
...
def fully_containsx(self, x: float) -> bool:
...
def fully_containsy(self, y: float) -> bool:
...
def fully_contains(self, x: float, y: float) -> bool:
...
def fully_overlaps(self, other: BboxBase) -> bool:
...
def transformed(self, transform: Transform) -> Bbox:
...
coefs: dict[str, tuple[float, float]]
def anchored(self, c: tuple[float, float] | str, container: BboxBase | None = ...) -> Bbox:
...
def shrunk(self, mx: float, my: float) -> Bbox:
...
def shrunk_to_aspect(self, box_aspect: float, container: BboxBase | None = ..., fig_aspect: float = ...) -> Bbox:
...
def splitx(self, *args: float) -> list[Bbox]:
...
def splity(self, *args: float) -> list[Bbox]:
...
def count_contains(self, vertices: ArrayLike) -> int:
...
def count_overlaps(self, bboxes: Iterable[BboxBase]) -> int:
...
def expanded(self, sw: float, sh: float) -> Bbox:
...
def padded(self, w_pad: float, h_pad: float | None = ...) -> Bbox:
...
def translated(self, tx: float, ty: float) -> Bbox:
...
def corners(self) -> np.ndarray:
...
def rotated(self, radians: float) -> Bbox:
...
@staticmethod
def union(bboxes: Sequence[BboxBase]) -> Bbox:
...
@staticmethod
def intersection(bbox1: BboxBase, bbox2: BboxBase) -> Bbox | None:
...
class Bbox(BboxBase):
def __init__(self, points: ArrayLike, **kwargs) -> None:
...
@staticmethod
def unit() -> Bbox:
...
@staticmethod
def null() -> Bbox:
...
@staticmethod
def from_bounds(x0: float, y0: float, width: float, height: float) -> Bbox:
...
@staticmethod
def from_extents(*args: float, minpos: float | None = ...) -> Bbox:
...
def __format__(self, fmt: str) -> str:
...
def ignore(self, value: bool) -> None:
...
def update_from_path(self, path: Path, ignore: bool | None = ..., updatex: bool = ..., updatey: bool = ...) -> None:
...
def update_from_data_x(self, x: ArrayLike, ignore: bool | None = ...) -> None:
...
def update_from_data_y(self, y: ArrayLike, ignore: bool | None = ...) -> None:
...
def update_from_data_xy(self, xy: ArrayLike, ignore: bool | None = ..., updatex: bool = ..., updatey: bool = ...) -> None:
...
@property
def minpos(self) -> float:
...
@property
def minposx(self) -> float:
...
@property
def minposy(self) -> float:
...
def get_points(self) -> np.ndarray:
...
def set_points(self, points: ArrayLike) -> None:
...
def set(self, other: Bbox) -> None:
...
def mutated(self) -> bool:
...
def mutatedx(self) -> bool:
...
def mutatedy(self) -> bool:
...
class TransformedBbox(BboxBase):
def __init__(self, bbox: Bbox, transform: Transform, **kwargs) -> None:
...
def get_points(self) -> np.ndarray:
...
class LockableBbox(BboxBase):
def __init__(self, bbox: BboxBase, x0: float | None = ..., y0: float | None = ..., x1: float | None = ..., y1: float | None = ..., **kwargs) -> None:
...
@property
def locked_x0(self) -> float | None:
...
@locked_x0.setter
def locked_x0(self, x0: float | None) -> None:
...
@property
def locked_y0(self) -> float | None:
...
@locked_y0.setter
def locked_y0(self, y0: float | None) -> None:
...
@property
def locked_x1(self) -> float | None:
...
@locked_x1.setter
def locked_x1(self, x1: float | None) -> None:
...
@property
def locked_y1(self) -> float | None:
...
@locked_y1.setter
def locked_y1(self, y1: float | None) -> None:
...
class Transform(TransformNode):
input_dims: int | None
output_dims: int | None
is_separable: bool
@property
def has_inverse(self) -> bool:
...
def __add__(self, other: Transform) -> Transform:
...
@property
def depth(self) -> int:
...
def contains_branch(self, other: Transform) -> bool:
...
def contains_branch_seperately(self, other_transform: Transform) -> Sequence[bool]:
...
def __sub__(self, other: Transform) -> Transform:
...
def __array__(self, *args, **kwargs) -> np.ndarray:
...
def transform(self, values: ArrayLike) -> np.ndarray:
...
def transform_affine(self, values: ArrayLike) -> np.ndarray:
...
def transform_non_affine(self, values: ArrayLike) -> ArrayLike:
...
def transform_bbox(self, bbox: BboxBase) -> Bbox:
...
def get_affine(self) -> Transform:
...
def get_matrix(self) -> np.ndarray:
...
def transform_point(self, point: ArrayLike) -> np.ndarray:
...
def transform_path(self, path: Path) -> Path:
...
def transform_path_affine(self, path: Path) -> Path:
...
def transform_path_non_affine(self, path: Path) -> Path:
...
def transform_angles(self, angles: ArrayLike, pts: ArrayLike, radians: bool = ..., pushoff: float = ...) -> np.ndarray:
...
def inverted(self) -> Transform:
...
class TransformWrapper(Transform):
pass_through: bool
def __init__(self, child: Transform) -> None:
...
def __eq__(self, other: object) -> bool:
...
def frozen(self) -> Transform:
...
def set(self, child: Transform) -> None:
...
class AffineBase(Transform):
is_affine: Literal[True]
def __init__(self, *args, **kwargs) -> None:
...
def __eq__(self, other: object) -> bool:
...
class Affine2DBase(AffineBase):
input_dims: Literal[2]
output_dims: Literal[2]
def frozen(self) -> Affine2D:
...
@property
def is_separable(self):
...
def to_values(self) -> tuple[float, float, float, float, float, float]:
...
class Affine2D(Affine2DBase):
def __init__(self, matrix: ArrayLike | None = ..., **kwargs) -> None:
...
@staticmethod
def from_values(a: float, b: float, c: float, d: float, e: float, f: float) -> Affine2D:
...
def set_matrix(self, mtx: ArrayLike) -> None:
...
def clear(self) -> Affine2D:
...
def rotate(self, theta: float) -> Affine2D:
...
def rotate_deg(self, degrees: float) -> Affine2D:
...
def rotate_around(self, x: float, y: float, theta: float) -> Affine2D:
...
def rotate_deg_around(self, x: float, y: float, degrees: float) -> Affine2D:
...
def translate(self, tx: float, ty: float) -> Affine2D:
...
def scale(self, sx: float, sy: float | None = ...) -> Affine2D:
...
def skew(self, xShear: float, yShear: float) -> Affine2D:
...
def skew_deg(self, xShear: float, yShear: float) -> Affine2D:
...
class IdentityTransform(Affine2DBase):
...
class _BlendedMixin:
def __eq__(self, other: object) -> bool:
...
def contains_branch_seperately(self, transform: Transform) -> Sequence[bool]:
...
class BlendedGenericTransform(_BlendedMixin, Transform):
input_dims: Literal[2]
output_dims: Literal[2]
is_separable: bool
pass_through: bool
def __init__(self, x_transform: Transform, y_transform: Transform, **kwargs) -> None:
...
@property
def depth(self) -> int:
...
def contains_branch(self, other: Transform) -> Literal[False]:
...
@property
def is_affine(self) -> bool:
...
@property
def has_inverse(self) -> bool:
...
class BlendedAffine2D(_BlendedMixin, Affine2DBase):
def __init__(self, x_transform: Transform, y_transform: Transform, **kwargs) -> None:
...
def blended_transform_factory(x_transform: Transform, y_transform: Transform) -> BlendedGenericTransform | BlendedAffine2D:
...
class CompositeGenericTransform(Transform):
pass_through: bool
input_dims: int | None
output_dims: int | None
def __init__(self, a: Transform, b: Transform, **kwargs) -> None:
...
class CompositeAffine2D(Affine2DBase):
def __init__(self, a: Affine2DBase, b: Affine2DBase, **kwargs) -> None:
...
@property
def depth(self) -> int:
...
def composite_transform_factory(a: Transform, b: Transform) -> Transform:
...
class BboxTransform(Affine2DBase):
def __init__(self, boxin: BboxBase, boxout: BboxBase, **kwargs) -> None:
...
class BboxTransformTo(Affine2DBase):
def __init__(self, boxout: BboxBase, **kwargs) -> None:
...
class BboxTransformToMaxOnly(BboxTransformTo):
...
class BboxTransformFrom(Affine2DBase):
def __init__(self, boxin: BboxBase, **kwargs) -> None:
...
class ScaledTranslation(Affine2DBase):
def __init__(self, xt: float, yt: float, scale_trans: Affine2DBase, **kwargs) -> None:
...
class AffineDeltaTransform(Affine2DBase):
def __init__(self, transform: Affine2DBase, **kwargs) -> None:
...
class TransformedPath(TransformNode):
def __init__(self, path: Path, transform: Transform) -> None:
...
def get_transformed_points_and_affine(self) -> tuple[Path, Transform]:
...
def get_transformed_path_and_affine(self) -> tuple[Path, Transform]:
...
def get_fully_transformed_path(self) -> Path:
...
def get_affine(self) -> Transform:
...
class TransformedPatchPath(TransformedPath):
def __init__(self, patch: Patch) -> None:
...
def nonsingular(vmin: float, vmax: float, expander: float = ..., tiny: float = ..., increasing: bool = ...) -> tuple[float, float]:
...
def interval_contains(interval: tuple[float, float], val: float) -> bool:
...
def interval_contains_open(interval: tuple[float, float], val: float) -> bool:
...
def offset_copy(trans: Transform, fig: Figure | None = ..., x: float = ..., y: float = ..., units: Literal["inches", "points", "dots"] = ...) -> Transform:
...