core.py

# Licensed under a 3-clause BSD style license - see LICENSE.rst

from collections import defaultdict
from functools import partial

import numpy as np
from matplotlib import rcParams
from matplotlib.artist import Artist
from matplotlib.axes import Axes, subplot_class_factory
from matplotlib.transforms import Affine2D, Bbox, Transform

from astropy.coordinates import BaseCoordinateFrame, SkyCoord
from astropy.utils import minversion
from astropy.utils.compat.optional_deps import HAS_PIL
from astropy.wcs import WCS
from astropy.wcs.wcsapi import BaseHighLevelWCS, BaseLowLevelWCS

from .coordinates_map import CoordinatesMap
from .frame import RectangularFrame, RectangularFrame1D
from .transforms import CoordinateTransform
from .utils import get_coord_meta, transform_contour_set_inplace
from .wcsapi import IDENTITY, transform_coord_meta_from_wcs

__all__ = ["WCSAxes", "WCSAxesSubplot"]

VISUAL_PROPERTIES = ["facecolor", "edgecolor", "linewidth", "alpha", "linestyle"]


class _WCSAxesArtist(Artist):
    """This is a dummy artist to enforce the correct z-order of axis ticks,
    tick labels, and gridlines.

    FIXME: This is a bit of a hack. ``Axes.draw`` sorts the artists by zorder
    and then renders them in sequence. For normal Matplotlib axes, the ticks,
    tick labels, and gridlines are included in this list of artists and hence
    are automatically drawn in the correct order. However, ``WCSAxes`` disables
    the native ticks, labels, and gridlines. Instead, ``WCSAxes.draw`` renders
    ersatz ticks, labels, and gridlines by explicitly calling the functions
    ``CoordinateHelper._draw_ticks``, ``CoordinateHelper._draw_grid``, etc.
    This hack would not be necessary if ``WCSAxes`` drew ticks, tick labels,
    and gridlines in the standary way.
    """

    def draw(self, renderer):
        self.axes.draw_wcsaxes(renderer)


class WCSAxes(Axes):
    """
    The main axes class that can be used to show world coordinates from a WCS.

    Parameters
    ----------
    fig : `~matplotlib.figure.Figure`
        The figure to add the axes to
    *args
        ``*args`` can be a single ``(left, bottom, width, height)``
        rectangle or a single `matplotlib.transforms.Bbox`.  This specifies
        the rectangle (in figure coordinates) where the Axes is positioned.
        ``*args`` can also consist of three numbers or a single three-digit
        number; in the latter case, the digits are considered as
        independent numbers.  The numbers are interpreted as ``(nrows,
        ncols, index)``: ``(nrows, ncols)`` specifies the size of an array
        of subplots, and ``index`` is the 1-based index of the subplot
        being created.  Finally, ``*args`` can also directly be a
        `matplotlib.gridspec.SubplotSpec` instance.
    wcs : :class:`~astropy.wcs.WCS`, optional
        The WCS for the data. If this is specified, ``transform`` cannot be
        specified.
    transform : `~matplotlib.transforms.Transform`, optional
        The transform for the data. If this is specified, ``wcs`` cannot be
        specified.
    coord_meta : dict, optional
        A dictionary providing additional metadata when ``transform`` is
        specified. This should include the keys ``type``, ``wrap``, and
        ``unit``. Each of these should be a list with as many items as the
        dimension of the WCS. The ``type`` entries should be one of
        ``longitude``, ``latitude``, or ``scalar``, the ``wrap`` entries should
        give, for the longitude, the angle at which the coordinate wraps (and
        `None` otherwise), and the ``unit`` should give the unit of the
        coordinates as :class:`~astropy.units.Unit` instances. This can
        optionally also include a ``format_unit`` entry giving the units to use
        for the tick labels (if not specified, this defaults to ``unit``).
    transData : `~matplotlib.transforms.Transform`, optional
        Can be used to override the default data -> pixel mapping.
    slices : tuple, optional
        For WCS transformations with more than two dimensions, we need to
        choose which dimensions are being shown in the 2D image. The slice
        should contain one ``x`` entry, one ``y`` entry, and the rest of the
        values should be integers indicating the slice through the data. The
        order of the items in the slice should be the same as the order of the
        dimensions in the :class:`~astropy.wcs.WCS`, and the opposite of the
        order of the dimensions in Numpy. For example, ``(50, 'x', 'y')`` means
        that the first WCS dimension (last Numpy dimension) will be sliced at
        an index of 50, the second WCS and Numpy dimension will be shown on the
        x axis, and the final WCS dimension (first Numpy dimension) will be
        shown on the y-axis (and therefore the data will be plotted using
        ``data[:, :, 50].transpose()``)
    frame_class : type, optional
        The class for the frame, which should be a subclass of
        :class:`~astropy.visualization.wcsaxes.frame.BaseFrame`. The default is to use a
        :class:`~astropy.visualization.wcsaxes.frame.RectangularFrame`

    Attributes
    ----------
    coords : :class:`~astropy.visualization.wcsaxes.CoordinatesMap`
        Container for coordinate information.
    """

    def __init__(
        self,
        fig,
        *args,
        wcs=None,
        transform=None,
        coord_meta=None,
        transData=None,
        slices=None,
        frame_class=None,
        **kwargs,
    ):
        """ """

        super().__init__(fig, *args, **kwargs)
        self._bboxes = []

        if frame_class is not None:
            self.frame_class = frame_class
        elif wcs is not None and (
            wcs.pixel_n_dim == 1 or (slices is not None and "y" not in slices)
        ):
            self.frame_class = RectangularFrame1D
        else:
            self.frame_class = RectangularFrame

        if transData is not None:
            # User wants to override the transform for the final
            # data->pixel mapping
            self.transData = transData

        self.reset_wcs(
            wcs=wcs, slices=slices, transform=transform, coord_meta=coord_meta
        )
        self._hide_parent_artists()
        self.format_coord = self._display_world_coords
        self._display_coords_index = 0
        fig.canvas.mpl_connect("key_press_event", self._set_cursor_prefs)
        self.patch = self.coords.frame.patch
        self._wcsaxesartist = _WCSAxesArtist()
        self.add_artist(self._wcsaxesartist)
        self._drawn = False

    def _display_world_coords(self, x, y):
        if not self._drawn:
            return ""

        if self._display_coords_index == -1:
            return f"{x} {y} (pixel)"

        pixel = np.array([x, y])

        coords = self._all_coords[self._display_coords_index]

        world = coords._transform.transform(np.array([pixel]))[0]

        coord_strings = []
        for idx, coord in enumerate(coords):
            if coord.coord_index is not None:
                coord_strings.append(
                    coord.format_coord(world[coord.coord_index], format="ascii")
                )

        coord_string = " ".join(coord_strings)

        if self._display_coords_index == 0:
            system = "world"
        else:
            system = f"world, overlay {self._display_coords_index}"

        coord_string = f"{coord_string} ({system})"

        return coord_string

    def _set_cursor_prefs(self, event, **kwargs):
        if event.key == "w":
            self._display_coords_index += 1
            if self._display_coords_index + 1 > len(self._all_coords):
                self._display_coords_index = -1

    def _hide_parent_artists(self):
        # Turn off spines and current axes
        for s in self.spines.values():
            s.set_visible(False)

        self.xaxis.set_visible(False)
        if self.frame_class is not RectangularFrame1D:
            self.yaxis.set_visible(False)

    # We now overload ``imshow`` because we need to make sure that origin is
    # set to ``lower`` for all images, which means that we need to flip RGB
    # images.
    def imshow(self, X, *args, **kwargs):
        """
        Wrapper to Matplotlib's :meth:`~matplotlib.axes.Axes.imshow`.

        If an RGB image is passed as a PIL object, it will be flipped
        vertically and ``origin`` will be set to ``lower``, since WCS
        transformations - like FITS files - assume that the origin is the lower
        left pixel of the image (whereas RGB images have the origin in the top
        left).

        All arguments are passed to :meth:`~matplotlib.axes.Axes.imshow`.
        """
        origin = kwargs.pop("origin", "lower")

        # plt.imshow passes origin as None, which we should default to lower.
        if origin is None:
            origin = "lower"
        elif origin == "upper":
            raise ValueError("Cannot use images with origin='upper' in WCSAxes.")

        if HAS_PIL:
            from PIL.Image import Image

            if minversion("PIL", "9.1"):
                from PIL.Image import Transpose

                FLIP_TOP_BOTTOM = Transpose.FLIP_TOP_BOTTOM
            else:
                from PIL.Image import FLIP_TOP_BOTTOM

            if isinstance(X, Image) or hasattr(X, "getpixel"):
                X = X.transpose(FLIP_TOP_BOTTOM)

        return super().imshow(X, *args, origin=origin, **kwargs)

    def contour(self, *args, **kwargs):
        """
        Plot contours.

        This is a custom implementation of :meth:`~matplotlib.axes.Axes.contour`
        which applies the transform (if specified) to all contours in one go for
        performance rather than to each contour line individually. All
        positional and keyword arguments are the same as for
        :meth:`~matplotlib.axes.Axes.contour`.
        """
        # In Matplotlib, when calling contour() with a transform, each
        # individual path in the contour map is transformed separately. However,
        # this is much too slow for us since each call to the transforms results
        # in an Astropy coordinate transformation, which has a non-negligible
        # overhead - therefore a better approach is to override contour(), call
        # the Matplotlib one with no transform, then apply the transform in one
        # go to all the segments that make up the contour map.

        transform = kwargs.pop("transform", None)

        cset = super().contour(*args, **kwargs)

        if transform is not None:
            # The transform passed to self.contour will normally include
            # a transData component at the end, but we can remove that since
            # we are already working in data space.
            transform = transform - self.transData
            transform_contour_set_inplace(cset, transform)

        return cset

    def contourf(self, *args, **kwargs):
        """
        Plot filled contours.

        This is a custom implementation of :meth:`~matplotlib.axes.Axes.contourf`
        which applies the transform (if specified) to all contours in one go for
        performance rather than to each contour line individually. All
        positional and keyword arguments are the same as for
        :meth:`~matplotlib.axes.Axes.contourf`.
        """
        # See notes for contour above.

        transform = kwargs.pop("transform", None)

        cset = super().contourf(*args, **kwargs)

        if transform is not None:
            # The transform passed to self.contour will normally include
            # a transData component at the end, but we can remove that since
            # we are already working in data space.
            transform = transform - self.transData
            transform_contour_set_inplace(cset, transform)

        return cset

    def _transform_plot_args(self, *args, **kwargs):
        """
        Apply transformations to arguments to ``plot_coord`` and
        ``scatter_coord``.
        """
        if isinstance(args[0], (SkyCoord, BaseCoordinateFrame)):
            # Extract the frame from the first argument.
            frame0 = args[0]
            if isinstance(frame0, SkyCoord):
                frame0 = frame0.frame

            native_frame = self._transform_pixel2world.frame_out
            # Transform to the native frame of the plot
            frame0 = frame0.transform_to(native_frame)

            plot_data = []
            for coord in self.coords:
                if coord.coord_type == "longitude":
                    plot_data.append(frame0.spherical.lon.to_value(coord.coord_unit))
                elif coord.coord_type == "latitude":
                    plot_data.append(frame0.spherical.lat.to_value(coord.coord_unit))
                else:
                    raise NotImplementedError(
                        "Coordinates cannot be plotted with this "
                        "method because the WCS does not represent longitude/latitude."
                    )

            if "transform" in kwargs.keys():
                raise TypeError(
                    "The 'transform' keyword argument is not allowed,"
                    " as it is automatically determined by the input coordinate frame."
                )

            transform = self.get_transform(native_frame)
            kwargs.update({"transform": transform})

            args = tuple(plot_data) + args[1:]

        return args, kwargs

    def plot_coord(self, *args, **kwargs):
        """
        Plot `~astropy.coordinates.SkyCoord` or
        `~astropy.coordinates.BaseCoordinateFrame` objects onto the axes.

        The first argument to
        :meth:`~astropy.visualization.wcsaxes.WCSAxes.plot_coord` should be a
        coordinate, which will then be converted to the first two parameters to
        `matplotlib.axes.Axes.plot`. All other arguments are the same as
        `matplotlib.axes.Axes.plot`. If not specified a ``transform`` keyword
        argument will be created based on the coordinate.

        Parameters
        ----------
        coordinate : `~astropy.coordinates.SkyCoord` or `~astropy.coordinates.BaseCoordinateFrame`
            The coordinate object to plot on the axes. This is converted to the
            first two arguments to `matplotlib.axes.Axes.plot`.

        See Also
        --------
        matplotlib.axes.Axes.plot :
            This method is called from this function with all arguments passed to it.

        """
        args, kwargs = self._transform_plot_args(*args, **kwargs)

        return super().plot(*args, **kwargs)

    def text_coord(self, *args, **kwargs):
        """
        Print a text string using `~astropy.coordinates.SkyCoord` or
        `~astropy.coordinates.BaseCoordinateFrame` objects onto the axes.

        The first argument to
        :meth:`~astropy.visualization.wcsaxes.WCSAxes.text_coord` should be a
        coordinate, which will then be converted to the first two parameters to
        `matplotlib.axes.Axes.text`. All other arguments are the same as
        `matplotlib.axes.Axes.text`. If not specified a ``transform`` keyword
        argument will be created based on the coordinate.

        Parameters
        ----------
        coordinate : `~astropy.coordinates.SkyCoord` or `~astropy.coordinates.BaseCoordinateFrame`
            The coordinate object to plot on the axes. This is converted to the
            first two arguments to `matplotlib.axes.Axes.text`.

        See Also
        --------
        matplotlib.axes.Axes.text :
            This method is called from this function with all arguments passed to it.

        """
        args, kwargs = self._transform_plot_args(*args, **kwargs)

        return super().text(*args, **kwargs)

    def scatter_coord(self, *args, **kwargs):
        """
        Scatter `~astropy.coordinates.SkyCoord` or
        `~astropy.coordinates.BaseCoordinateFrame` objects onto the axes.

        The first argument to
        :meth:`~astropy.visualization.wcsaxes.WCSAxes.scatter_coord` should be a
        coordinate, which will then be converted to the first two parameters to
        `matplotlib.axes.Axes.scatter`. All other arguments are the same as
        `matplotlib.axes.Axes.scatter`. If not specified a ``transform``
        keyword argument will be created based on the coordinate.

        Parameters
        ----------
        coordinate : `~astropy.coordinates.SkyCoord` or `~astropy.coordinates.BaseCoordinateFrame`
            The coordinate object to scatter on the axes. This is converted to
            the first two arguments to `matplotlib.axes.Axes.scatter`.

        See Also
        --------
        matplotlib.axes.Axes.scatter : This method is called from this function with all arguments passed to it.
        """
        args, kwargs = self._transform_plot_args(*args, **kwargs)

        return super().scatter(*args, **kwargs)

    def reset_wcs(self, wcs=None, slices=None, transform=None, coord_meta=None):
        """
        Reset the current Axes, to use a new WCS object.
        """
        # Here determine all the coordinate axes that should be shown.
        if wcs is None and transform is None:
            self.wcs = IDENTITY

        else:
            # We now force call 'set', which ensures the WCS object is
            # consistent, which will only be important if the WCS has been set
            # by hand. For example if the user sets a celestial WCS by hand and
            # forgets to set the units, WCS.wcs.set() will do this.
            if wcs is not None:
                # Check if the WCS object is an instance of `astropy.wcs.WCS`
                # This check is necessary as only `astropy.wcs.WCS` supports
                # wcs.set() method
                if isinstance(wcs, WCS):
                    wcs.wcs.set()

                if isinstance(wcs, BaseHighLevelWCS):
                    wcs = wcs.low_level_wcs

            self.wcs = wcs

        # If we are making a new WCS, we need to preserve the path object since
        # it may already be used by objects that have been plotted, and we need
        # to continue updating it. CoordinatesMap will create a new frame
        # instance, but we can tell that instance to keep using the old path.
        if hasattr(self, "coords"):
            previous_frame = {
                "path": self.coords.frame._path,
                "color": self.coords.frame.get_color(),
                "linewidth": self.coords.frame.get_linewidth(),
            }
        else:
            previous_frame = {"path": None}

        if self.wcs is not None:
            transform, coord_meta = transform_coord_meta_from_wcs(
                self.wcs, self.frame_class, slices=slices
            )

        self.coords = CoordinatesMap(
            self,
            transform=transform,
            coord_meta=coord_meta,
            frame_class=self.frame_class,
            previous_frame_path=previous_frame["path"],
        )

        self._transform_pixel2world = transform

        if previous_frame["path"] is not None:
            self.coords.frame.set_color(previous_frame["color"])
            self.coords.frame.set_linewidth(previous_frame["linewidth"])

        self._all_coords = [self.coords]

        # Common default settings for Rectangular Frame
        for ind, pos in enumerate(
            coord_meta.get("default_axislabel_position", ["b", "l"])
        ):
            self.coords[ind].set_axislabel_position(pos)

        for ind, pos in enumerate(
            coord_meta.get("default_ticklabel_position", ["b", "l"])
        ):
            self.coords[ind].set_ticklabel_position(pos)

        for ind, pos in enumerate(
            coord_meta.get("default_ticks_position", ["bltr", "bltr"])
        ):
            self.coords[ind].set_ticks_position(pos)

        if rcParams["axes.grid"]:
            self.grid()

    def draw_wcsaxes(self, renderer):
        if not self.axison:
            return
        # Here need to find out range of all coordinates, and update range for
        # each coordinate axis. For now, just assume it covers the whole sky.

        self._bboxes = []
        # This generates a structure like [coords][axis] = [...]
        ticklabels_bbox = defaultdict(partial(defaultdict, list))

        visible_ticks = []

        for coords in self._all_coords:
            # Draw grids
            coords.frame.update()
            for coord in coords:
                coord._draw_grid(renderer)

        for coords in self._all_coords:
            # Draw tick labels
            for coord in coords:
                coord._draw_ticks(renderer, self._bboxes)

                visible_ticks.extend(coord.ticklabels.get_visible_axes())
                # Save ticklabel bboxes
                ticklabels_bbox[coord] = coord.ticklabels._axis_bboxes
                self._bboxes += coord.ticklabels._all_bboxes

        for coords in self._all_coords:
            # Draw axis labels
            for coord in coords:
                coord._draw_axislabels(
                    renderer,
                    bboxes=self._bboxes,
                    ticklabels_bbox=ticklabels_bbox,
                    visible_ticks=visible_ticks,
                )

        self.coords.frame.draw(renderer)

    def draw(self, renderer):
        """Draw the axes."""
        # Before we do any drawing, we need to remove any existing grid lines
        # drawn with contours, otherwise if we try and remove the contours
        # part way through drawing, we end up with the issue mentioned in
        # https://github.com/astropy/astropy/issues/12446
        for coords in self._all_coords:
            for coord in coords:
                coord._clear_grid_contour()

        # In Axes.draw, the following code can result in the xlim and ylim
        # values changing, so we need to force call this here to make sure that
        # the limits are correct before we update the patch.
        locator = self.get_axes_locator()
        if locator:
            pos = locator(self, renderer)
            self.apply_aspect(pos)
        else:
            self.apply_aspect()

        if self._axisbelow is True:
            self._wcsaxesartist.set_zorder(0.5)
        elif self._axisbelow is False:
            self._wcsaxesartist.set_zorder(2.5)
        else:
            # 'line': above patches, below lines
            self._wcsaxesartist.set_zorder(1.5)

        # We need to make sure that that frame path is up to date
        self.coords.frame._update_patch_path()

        super().draw(renderer)

        self._drawn = True

    # Matplotlib internally sometimes calls set_xlabel(label=...).
    def set_xlabel(self, xlabel=None, labelpad=1, loc=None, **kwargs):
        """Set x-label."""
        if xlabel is None:
            xlabel = kwargs.pop("label", None)
            if xlabel is None:
                raise TypeError(
                    "set_xlabel() missing 1 required positional argument: 'xlabel'"
                )
        for coord in self.coords:
            if (
                "b" in coord.axislabels.get_visible_axes()
                or "h" in coord.axislabels.get_visible_axes()
            ):
                coord.set_axislabel(xlabel, minpad=labelpad, **kwargs)
                break

    def set_ylabel(self, ylabel=None, labelpad=1, loc=None, **kwargs):
        """Set y-label."""
        if ylabel is None:
            ylabel = kwargs.pop("label", None)
            if ylabel is None:
                raise TypeError(
                    "set_ylabel() missing 1 required positional argument: 'ylabel'"
                )

        if self.frame_class is RectangularFrame1D:
            return super().set_ylabel(ylabel, labelpad=labelpad, **kwargs)

        for coord in self.coords:
            if (
                "l" in coord.axislabels.get_visible_axes()
                or "c" in coord.axislabels.get_visible_axes()
            ):
                coord.set_axislabel(ylabel, minpad=labelpad, **kwargs)
                break

    def get_xlabel(self):
        for coord in self.coords:
            if (
                "b" in coord.axislabels.get_visible_axes()
                or "h" in coord.axislabels.get_visible_axes()
            ):
                return coord.get_axislabel()

    def get_ylabel(self):
        if self.frame_class is RectangularFrame1D:
            return super().get_ylabel()

        for coord in self.coords:
            if (
                "l" in coord.axislabels.get_visible_axes()
                or "c" in coord.axislabels.get_visible_axes()
            ):
                return coord.get_axislabel()

    def get_coords_overlay(self, frame, coord_meta=None):
        """Get coordinates overlay on given frame.

        Parameters
        ----------
        frame : str, `~astropy.coordinates.BaseCoordinateFrame`
            Frame to get overlay for. If a string must correspond to
            one of the coordinate frames registered in the astropy
            frame transform graph.

        coord_meta : dict
            Metadata for the coordinates overlay.

        Returns
        -------
        overlay : `~astropy.visualization.wcsaxes.CoordinatesMap`
            Coordinates overlay.
        """
        # Here we can't use get_transform because that deals with
        # pixel-to-pixel transformations when passing a WCS object.
        if isinstance(frame, WCS):
            transform, coord_meta = transform_coord_meta_from_wcs(
                frame, self.frame_class
            )
        else:
            transform = self._get_transform_no_transdata(frame)

        if coord_meta is None:
            coord_meta = get_coord_meta(frame)

        coords = CoordinatesMap(
            self,
            transform=transform,
            coord_meta=coord_meta,
            frame_class=self.frame_class,
        )

        self._all_coords.append(coords)

        # Common settings for overlay
        coords[0].set_axislabel_position("t")
        coords[1].set_axislabel_position("r")
        coords[0].set_ticklabel_position("t")
        coords[1].set_ticklabel_position("r")

        self.overlay_coords = coords

        return coords

    def get_transform(self, frame):
        """
        Return a transform from the specified frame to display coordinates.

        This does not include the transData transformation

        Parameters
        ----------
        frame : :class:`~astropy.wcs.WCS` or :class:`~matplotlib.transforms.Transform` or str
            The ``frame`` parameter can have several possible types:
                * :class:`~astropy.wcs.WCS` instance: assumed to be a
                  transformation from pixel to world coordinates, where the
                  world coordinates are the same as those in the WCS
                  transformation used for this ``WCSAxes`` instance. This is
                  used for example to show contours, since this involves
                  plotting an array in pixel coordinates that are not the
                  final data coordinate and have to be transformed to the
                  common world coordinate system first.
                * :class:`~matplotlib.transforms.Transform` instance: it is
                  assumed to be a transform to the world coordinates that are
                  part of the WCS used to instantiate this ``WCSAxes``
                  instance.
                * ``'pixel'`` or ``'world'``: return a transformation that
                  allows users to plot in pixel/data coordinates (essentially
                  an identity transform) and ``world`` (the default
                  world-to-pixel transformation used to instantiate the
                  ``WCSAxes`` instance).
                * ``'fk5'`` or ``'galactic'``: return a transformation from
                  the specified frame to the pixel/data coordinates.
                * :class:`~astropy.coordinates.BaseCoordinateFrame` instance.
        """
        return self._get_transform_no_transdata(frame).inverted() + self.transData

    def _get_transform_no_transdata(self, frame):
        """
        Return a transform from data to the specified frame.
        """
        if isinstance(frame, (BaseLowLevelWCS, BaseHighLevelWCS)):
            if isinstance(frame, BaseHighLevelWCS):
                frame = frame.low_level_wcs

            transform, coord_meta = transform_coord_meta_from_wcs(
                frame, self.frame_class
            )
            transform_world2pixel = transform.inverted()

            if self._transform_pixel2world.frame_out == transform_world2pixel.frame_in:
                return self._transform_pixel2world + transform_world2pixel

            else:
                return (
                    self._transform_pixel2world
                    + CoordinateTransform(
                        self._transform_pixel2world.frame_out,
                        transform_world2pixel.frame_in,
                    )
                    + transform_world2pixel
                )

        elif isinstance(frame, str) and frame == "pixel":
            return Affine2D()

        elif isinstance(frame, Transform):
            return self._transform_pixel2world + frame

        else:
            if isinstance(frame, str) and frame == "world":
                return self._transform_pixel2world

            else:
                coordinate_transform = CoordinateTransform(
                    self._transform_pixel2world.frame_out, frame
                )

                if coordinate_transform.same_frames:
                    return self._transform_pixel2world
                else:
                    return self._transform_pixel2world + coordinate_transform

    def get_tightbbox(self, renderer, *args, **kwargs):
        # FIXME: we should determine what to do with the extra arguments here.
        # Note that the expected signature of this method is different in
        # Matplotlib 3.x compared to 2.x, but we only support 3.x now.

        if not self.get_visible():
            return

        # Do a draw to populate the self._bboxes list
        self.draw_wcsaxes(renderer)
        bb = [b for b in self._bboxes if b and (b.width != 0 or b.height != 0)]
        bb.append(super().get_tightbbox(renderer, *args, **kwargs))

        if bb:
            _bbox = Bbox.union(bb)
            return _bbox
        else:
            return self.get_window_extent(renderer)

    def grid(self, b=None, axis="both", *, which="major", **kwargs):
        """
        Plot gridlines for both coordinates.

        Standard matplotlib appearance options (color, alpha, etc.) can be
        passed as keyword arguments. This behaves like `matplotlib.axes.Axes`
        except that if no arguments are specified, the grid is shown rather
        than toggled.

        Parameters
        ----------
        b : bool
            Whether to show the gridlines.
        axis : 'both', 'x', 'y'
            Which axis to turn the gridlines on/off for.
        which : str
            Currently only ``'major'`` is supported.
        """
        if not hasattr(self, "coords"):
            return

        if which != "major":
            raise NotImplementedError(
                "Plotting the grid for the minor ticks is not supported."
            )

        if axis == "both":
            self.coords.grid(draw_grid=b, **kwargs)
        elif axis == "x":
            self.coords[0].grid(draw_grid=b, **kwargs)
        elif axis == "y":
            self.coords[1].grid(draw_grid=b, **kwargs)
        else:
            raise ValueError("axis should be one of x/y/both")

    def tick_params(self, axis="both", **kwargs):
        """
        Method to set the tick and tick label parameters in the same way as the
        :meth:`~matplotlib.axes.Axes.tick_params` method in Matplotlib.

        This is provided for convenience, but the recommended API is to use
        :meth:`~astropy.visualization.wcsaxes.CoordinateHelper.set_ticks`,
        :meth:`~astropy.visualization.wcsaxes.CoordinateHelper.set_ticklabel`,
        :meth:`~astropy.visualization.wcsaxes.CoordinateHelper.set_ticks_position`,
        :meth:`~astropy.visualization.wcsaxes.CoordinateHelper.set_ticklabel_position`,
        and :meth:`~astropy.visualization.wcsaxes.CoordinateHelper.grid`.

        Parameters
        ----------
        axis : int or str, optional
            Which axis to apply the parameters to. This defaults to 'both'
            but this can also be set to an `int` or `str` that refers to the
            axis to apply it to, following the valid values that can index
            ``ax.coords``. Note that ``'x'`` and ``'y``' are also accepted in
            the case of rectangular axes.
        which : {'both', 'major', 'minor'}, optional
            Which ticks to apply the settings to. By default, setting are
            applied to both major and minor ticks. Note that if ``'minor'`` is
            specified, only the length of the ticks can be set currently.
        direction : {'in', 'out'}, optional
            Puts ticks inside the axes, or outside the axes.
        length : float, optional
            Tick length in points.
        width : float, optional
            Tick width in points.
        color : color, optional
            Tick color (accepts any valid Matplotlib color)
        pad : float, optional
            Distance in points between tick and label.
        labelsize : float or str, optional
            Tick label font size in points or as a string (e.g., 'large').
        labelcolor : color, optional
            Tick label color (accepts any valid Matplotlib color)
        colors : color, optional
            Changes the tick color and the label color to the same value
             (accepts any valid Matplotlib color).
        bottom, top, left, right : bool, optional
            Where to draw the ticks. Note that this can only be given if a
            specific coordinate is specified via the ``axis`` argument, and it
            will not work correctly if the frame is not rectangular.
        labelbottom, labeltop, labelleft, labelright : bool, optional
            Where to draw the tick labels. Note that this can only be given if a
            specific coordinate is specified via the ``axis`` argument, and it
            will not work correctly if the frame is not rectangular.
        grid_color : color, optional
            The color of the grid lines (accepts any valid Matplotlib color).
        grid_alpha : float, optional
            Transparency of grid lines: 0 (transparent) to 1 (opaque).
        grid_linewidth : float, optional
            Width of grid lines in points.
        grid_linestyle : str, optional
            The style of the grid lines (accepts any valid Matplotlib line
            style).
        """
        if not hasattr(self, "coords"):
            # Axes haven't been fully initialized yet, so just ignore, as
            # Axes.__init__ calls this method
            return

        if axis == "both":
            for pos in ("bottom", "left", "top", "right"):
                if pos in kwargs:
                    raise ValueError(f"Cannot specify {pos}= when axis='both'")
                if "label" + pos in kwargs:
                    raise ValueError(f"Cannot specify label{pos}= when axis='both'")

            for coord in self.coords:
                coord.tick_params(**kwargs)

        elif axis in self.coords:
            self.coords[axis].tick_params(**kwargs)

        elif axis in ("x", "y") and self.frame_class is RectangularFrame:
            spine = "b" if axis == "x" else "l"

            for coord in self.coords:
                if spine in coord.axislabels.get_visible_axes():
                    coord.tick_params(**kwargs)


# In the following, we put the generated subplot class in a temporary class and
# we then inherit it - if we don't do this, the generated class appears to
# belong in matplotlib, not in WCSAxes, from the API's point of view.


class WCSAxesSubplot(subplot_class_factory(WCSAxes)):
    """
    A subclass class for WCSAxes.
    """

    pass