""" Ref provides basic support for nesting Pattern objects within each other, by adding offset, rotation, scaling, and other such properties to the reference. """ #TODO more top-level documentation from typing import Dict, Optional, Sequence, Mapping, TYPE_CHECKING, Any, TypeVar import copy import numpy from numpy import pi from numpy.typing import NDArray, ArrayLike from .error import PatternError from .utils import is_scalar, annotations_t from .repetition import Repetition from .traits import ( PositionableImpl, RotatableImpl, ScalableImpl, Mirrorable, PivotableImpl, Copyable, RepeatableImpl, AnnotatableImpl, ) if TYPE_CHECKING: from . import Pattern R = TypeVar('R', bound='Ref') class Ref( PositionableImpl, RotatableImpl, ScalableImpl, Mirrorable, PivotableImpl, Copyable, RepeatableImpl, AnnotatableImpl, ): """ `Ref` provides basic support for nesting Pattern objects within each other, by adding offset, rotation, scaling, and associated methods. """ __slots__ = ( '_target', '_mirrored', # inherited '_offset', '_rotation', 'scale', '_repetition', '_annotations', ) _target: Optional[str] """ The name of the `Pattern` being instanced """ _mirrored: NDArray[numpy.bool_] """ Whether to mirror the instance across the x and/or y axes. """ def __init__( self, target: Optional[str], *, offset: ArrayLike = (0.0, 0.0), rotation: float = 0.0, mirrored: Optional[Sequence[bool]] = None, scale: float = 1.0, repetition: Optional[Repetition] = None, annotations: Optional[annotations_t] = None, ) -> None: """ Args: target: Name of the Pattern to reference. offset: (x, y) offset applied to the referenced pattern. Not affected by rotation etc. rotation: Rotation (radians, counterclockwise) relative to the referenced pattern's (0, 0). mirrored: Whether to mirror the referenced pattern across its x and y axes. scale: Scaling factor applied to the pattern's geometry. repetition: `Repetition` object, default `None` """ self.target = target self.offset = offset self.rotation = rotation self.scale = scale if mirrored is None: mirrored = (False, False) self.mirrored = mirrored self.repetition = repetition self.annotations = annotations if annotations is not None else {} def __copy__(self) -> 'Ref': new = Ref( target=self.target, offset=self.offset.copy(), rotation=self.rotation, scale=self.scale, mirrored=self.mirrored.copy(), repetition=copy.deepcopy(self.repetition), annotations=copy.deepcopy(self.annotations), ) return new def __deepcopy__(self, memo: Optional[Dict] = None) -> 'Ref': memo = {} if memo is None else memo new = copy.copy(self) new.repetition = copy.deepcopy(self.repetition, memo) new.annotations = copy.deepcopy(self.annotations, memo) return new # target property @property def target(self) -> Optional[str]: return self._target @target.setter def target(self, val: Optional[str]) -> None: if val is not None and not isinstance(val, str): raise PatternError(f'Provided target {val} is not a str or None!') self._target = val # Mirrored property @property def mirrored(self) -> Any: # TODO mypy#3004 NDArray[numpy.bool_]: return self._mirrored @mirrored.setter def mirrored(self, val: ArrayLike) -> None: if is_scalar(val): raise PatternError('Mirrored must be a 2-element list of booleans') self._mirrored = numpy.array(val, dtype=bool, copy=True) def as_pattern( self, *, pattern: Optional['Pattern'] = None, library: Optional[Mapping[str, 'Pattern']] = None, ) -> 'Pattern': """ Args: pattern: Pattern object to transform library: A str->Pattern mapping, used instead of `pattern`. Must contain `self.target`. Returns: A copy of the referenced Pattern which has been scaled, rotated, etc. according to this `Ref`'s properties. """ if pattern is None: if library is None: raise PatternError('as_pattern() must be given a pattern or library.') assert self.target is not None pattern = library[self.target] pattern = pattern.deepcopy() if self.scale != 1: pattern.scale_by(self.scale) if numpy.any(self.mirrored): pattern.mirror2d(self.mirrored) if self.rotation % (2 * pi) != 0: pattern.rotate_around((0.0, 0.0), self.rotation) if numpy.any(self.offset): pattern.translate_elements(self.offset) if self.repetition is not None: combined = type(pattern)() for dd in self.repetition.displacements: temp_pat = pattern.deepcopy() temp_pat.translate_elements(dd) combined.append(temp_pat) pattern = combined return pattern def rotate(self: R, rotation: float) -> R: self.rotation += rotation if self.repetition is not None: self.repetition.rotate(rotation) return self def mirror(self: R, axis: int) -> R: self.mirrored[axis] = not self.mirrored[axis] self.rotation *= -1 if self.repetition is not None: self.repetition.mirror(axis) return self def get_bounds( self, *, pattern: Optional['Pattern'] = None, library: Optional[Mapping[str, 'Pattern']] = None, ) -> Optional[NDArray[numpy.float64]]: """ Return a `numpy.ndarray` containing `[[x_min, y_min], [x_max, y_max]]`, corresponding to the extent of the `Ref` in each dimension. Returns `None` if the contained `Pattern` is empty. Args: library: Name-to-Pattern mapping for resul Returns: `[[x_min, y_min], [x_max, y_max]]` or `None` """ if pattern is None and library is None: raise PatternError('as_pattern() must be given a pattern or library.') if pattern is None and self.target is None: return None if library is not None and self.target not in library: raise PatternError(f'get_bounds() called on dangling reference to "{self.target}"') return self.as_pattern(pattern=pattern, library=library).get_bounds() def __repr__(self) -> str: name = f'"{self.target}"' if self.target is not None else None rotation = f' r{self.rotation*180/pi:g}' if self.rotation != 0 else '' scale = f' d{self.scale:g}' if self.scale != 1 else '' mirrored = ' m{:d}{:d}'.format(*self.mirrored) if self.mirrored.any() else '' return f''