Failure to load backend modules (macosx on non-framework builds and
gtk3 when running headless) now raises ImportError (instead of
RuntimeError and TypeError, respectively).
Third-party backends that integrate with an interactive framework are now
encouraged to define the required_interactive_framework global value to one
of the following values: "qt5", "qt4", "gtk3", "wx", "tk", or "macosx". This
information will be used to determine whether it is possible to switch from a
backend to another (specifically, whether they use the same interactive
framework).
matplotlib.axes.Axes.get_tightbbox adds a new kwarg bbox_extra_artists
to manually specify the list of artists on the axes to include in the
tight bounding box calculation.
Layout tools like Figure.tight_layout, constrained_layout,
and fig.savefig('fname.png',bbox_inches="tight") use
matplotlib.axes.Axes.get_tightbbox to determine the bounds of each axes on
a figure and adjust spacing between axes.
In Matplotlib 2.2 get_tightbbox started to include legends made on the
axes, but still excluded some other artists, like text that may overspill an
axes. This has been expanded to include all artists.
This new default may be overridden in either of three ways:
Make the artist to be excluded a child of the figure, not the axes. E.g.,
call fig.legend() instead of ax.legend() (perhaps using
get_legend_handles_labels to gather handles and
labels from the parent axes).
If the artist is a child of the axes, set the artist property
artist.set_in_layout(False).
Manually specify a list of artists in the new kwarg bbox_extra_artists.
Text.set_text with string argument None sets string to empty¶
Text.set_text when passed a string value of None would set the
string to "None", so subsequent calls to Text.get_text would return
the ambiguous "None" string.
This change sets text objects passed None to have empty strings, so that
Text.get_text returns an empty string.
Axes3D.get_xlim, get_ylim and get_zlim now return a tuple¶
They previously returned an array. Returning a tuple is consistent with the
behavior for 2D axes.
font_manager.list_fonts now follows the platform's casefolding semantics¶
i.e., it behaves case-insensitively on Windows only.
bar / barh no longer accepts left / bottom as first named argument¶
These arguments were renamed in 2.0 to x / y following the change of the
default alignment from edge to center.
Different exception types for undocumented options¶
Passing style='comma' to ticklabel_format()
was never supported. It now raises ValueError like all other
unsupported styles, rather than NotImplementedError.
Passing the undocumented xmin or xmax arguments to
set_xlim() would silently override the left
and right arguments. set_ylim() and the
3D equivalents (e.g. set_zlim3d) had a
corresponding problem.
A TypeError will be raised if they would override the earlier
limit arguments. In 3.0 these were kwargs were deprecated, but in 3.1
the deprecation was undone.
Axes.margins and Axes3D.margins
no longer accept arbitrary keywords. TypeError will therefore be raised
if unknown kwargs are passed; previously they would be silently ignored.
If too many positional arguments are passed, TypeError will be raised
instead of ValueError, for consistency with other call-signature violations.
Axes3D.margins now raises TypeError instead of emitting a deprecation
warning if only two positional arguments are passed. To supply only x and
y margins, use keyword arguments.
PEP 3102 describes keyword-only arguments, which allow Matplotlib
to provide explicit call signatures - where we previously used
*args,**kwargs and kwargs.pop, we can now expose named
arguments. In some places, unknown kwargs were previously ignored but
now raise TypeError because **kwargs has been removed.
matplotlib.axes.Axes.stem() no longer accepts unknown keywords,
and raises TypeError instead of emitting a deprecation.
matplotlib.axes.Axes.stem() now raises TypeError when passed
unhandled positional arguments. If two or more arguments are passed
(ie X, Y, [linefmt], ...) and Y cannot be cast to an array, an error
will be raised instead of treating X as Y and Y as linefmt.
Cleanup decorators and test classes no longer destroy warnings filter on exit¶
The decorators and classes in matplotlib.testing.decorators no longer
destroy the warnings filter on exit. Instead, they restore the warnings
filter that existed before the test started using warnings.catch_warnings.
Non-interactive FigureManager classes are now aliases of FigureManagerBase¶
The FigureManagerPdf, FigureManagerPS, and FigureManagerSVG classes,
which were previously empty subclasses of FigureManagerBase (i.e., not
adding or overriding any attribute or method), are now direct aliases for
FigureManagerBase.
When called with preview=False, image.thumbnail previously returned an
figure whose canvas class was set according to the output file extension. It
now returns a figure whose canvas class is the base FigureCanvasBase (and
relies on FigureCanvasBase.print_figure) to handle the canvas switching
properly).
As a side effect of this change, image.thumbnail now also supports .ps, .eps,
and .svgz output.
FuncAnimation now draws artists according to their zorder when blitting¶
FuncAnimation now draws artists returned by the user-
function according to their zorder when using blitting,
instead of using the order in which they are being passed.
However, note that only zorder of passed artists will be
respected, as they are drawn on top of any existing artists
(see #11369).
Selection of contour levels is now the same for contour and
contourf; previously, for contour, levels outside the data range were
deleted. (Exception: if no contour levels are found within the
data range, the levels attribute is replaced with a list holding
only the minimum of the data range.)
When contour is called with levels specified as a target number rather
than a list, and the 'extend' kwarg is used, the levels are now chosen
such that some data typically will fall in the extended range.
When contour is called with a LogNorm or a LogLocator, it will now
select colors using the geometric mean rather than the arithmetic mean
of the contour levels.
A bug was fixed where the last row and column of data in
streamplot were being dropped.
Changed default AutoDateLocator kwarg interval_multiples to True¶
The default value of the tick locator for dates, dates.AutoDateLocator
kwarg interval_multiples was set to False which leads to not-nice
looking automatic ticks in many instances. The much nicer
interval_multiples=True is the new default. See below to get the
old behavior back:
Axes.get_position used to return the original position unless a
draw had been triggered or Axes.apply_aspect had been called, even
if the kwarg original was set to False. Now Axes.apply_aspect
is called so ax.get_position() will return the new modified position.
To get the old behavior use ax.get_position(original=True).
The ticks for colorbar now adjust for the size of the colorbar¶
Colorbar ticks now adjust for the size of the colorbar if the
colorbar is made from a mappable that is not a contour or
doesn't have a BoundaryNorm, or boundaries are not specified.
If boundaries, etc are specified, the colorbar maintains the
original behavior.
When using hexbin and plotting with a logarithmic color scale, the colorbar
ticks are now correctly log scaled. Previously the tick values were linear
scaled log(number of counts).
PGF backend now explicitly makes black text black¶
Previous behavior with the pgf backend was for text specified as black to
actually be the default color of whatever was rendering the pgf file (which was
of course usually black). The new behavior is that black text is black,
regardless of the default color. However, this means that there is no way to
fall back on the default color of the renderer.
The rc modifier functions rcdefaults,
rc_file_defaults and rc_file
now ignore rcParams in the matplotlib.style.core.STYLE_BLACKLIST set. In
particular, this prevents the backend and interactive rcParams from
being incorrectly modified by these functions.
In particular, this implies that CallbackRegistry.callbacks[signal] is now
a mapping of callback ids to weakref.WeakMethods (i.e., they need to be first called
with no arguments to retrieve the method itself).
The rcParam now defaults to True and is deprecated (i.e., in future versions
of Matplotlib, unicode input will always be supported).
Moreover, the underlying implementation now uses \usepackage[utf8]{inputenc}
instead of \usepackage{ucs}\usepackage[utf8x]{inputenc}.
Return type of ArtistInspector.get_aliases changed¶
ArtistInspector.get_aliases previously returned the set of aliases as
{fullname:{alias1:None,alias2:None,...}}. The dict-to-None mapping
was used to simulate a set in earlier versions of Python. It has now been
replaced by a set, i.e. {fullname:{alias1,alias2,...}}.
This value is also stored in ArtistInspector.aliasd, which has likewise
changed.
Since dateutil and pytz both provide time zones, and
matplotlib already depends on dateutil, matplotlib will now use
dateutil time zones internally and drop the redundant dependency
on pytz. While dateutil time zones are preferred (and
currently recommended in the Python documentation), the explicit use
of pytz zones is still supported.
backend_qt5.FigureCanvasQT.keyAutoRepeat (directly check
event.guiEvent.isAutoRepeat() in the event handler to decide whether to
handle autorepeated key presses).
image._ImageBase.iterpnames, use the interpolation_names property
instead. (this affects classes that inherit from _ImageBase including
FigureImage, BboxImage, and AxesImage)
FigureCanvasQT.keyAutoRepeat (directly check
event.guiEvent.isAutoRepeat() in the event handler to decide whether to
handle autorepeated key presses)
FigureCanvasWx.macros
_ImageBase.iterpnames, use the interpolation_names property instead.
(this affects classes that inherit from _ImageBase including
FigureImage, BboxImage, and AxesImage)
(Legend.draggable may be reintroduced as a property in future releases)
textpath.TextToPath.tex_font_map
matplotlib.cbook.deprecation.mplDeprecation will be removed
in future versions. It is just an alias for
matplotlib.cbook.deprecation.MatplotlibDeprecationWarning. Please
use matplotlib.cbook.MatplotlibDeprecationWarning directly if necessary.
The matplotlib.cbook.Bunch class has been deprecated. Instead, use
types.SimpleNamespace from the standard library which provides the same
functionality.
Axes.mouseover_set is now a frozenset, and deprecated. Directly
manipulate the artist's .mouseover attribute to change their mouseover
status.
The following keyword arguments are deprecated:
passing verts to Axes.scatter (use marker instead)
passing obj_type to cbook.deprecated
The following call signatures are deprecated:
passing a wx.EvtHandler as first argument to backend_wx.TimerWx
The LocatableAxes classes in toolkits have been deprecated. The base Axes
classes provide the same functionality to all subclasses, thus these mixins are
no longer necessary. Related functions have also been deprecated. Specifically:
mpl_toolkits.axes_grid1.axes_divider.LocatableAxesBase: no specific
replacement; use any other Axes-derived class directly instead.
mpl_toolkits.axes_grid1.axes_divider.locatable_axes_factory: no specific
replacement; use any other Axes-derived class directly instead.
Setting or unsetting hold (deprecated in version 2.0) has now
been completely removed. Matplotlib now always behaves as if hold=True.
To clear an axes you can manually use cla(),
or to clear an entire figure use clf().
The matplotlib documentation used the proprietary sphinx directives
..htmlonly::, and ..latexonly::. These have been replaced with the
standard sphinx directives ..only::html and ..only::latex. This
change will not affect any users. Only downstream package maintainers, who
have used the proprietary directives in their docs, will have to switch to the
sphinx directives.
The symlink from lib/mpl_examples to ../examples has been removed.
This is not installed as an importable package and should not affect
end users, however this may require down-stream packagers to adjust.
The content is still available top-level examples directory.