👍 modulo comment warning.

attn@pelson Any comments on the changes to

Could you please add an entry tohttps://github.com/matplotlib/matplotlib/tree/master/doc/users/whats_new ?

@leeonadoh@ryanbelt Can you add an entry in whats_new? This helps to make sure that your new feature is discover-able and highlighted when we release the next feature version.

Sorry for not giving an update. There's currently a better solution in the works, which will generalize this feature to all types of plots. We'll add the whats_new entry when we have that ready.

Ah, great to hear!

Please feel free to share early and often to get feed back.

The updated solution is able to give all types of plots a uniform size when displayed on the legend.

It works by creating a subclass ofLegend calledUniformLegend that specifies its own list of default handlers. These default handlers will ensure an uniform size.

Currently, the code to use UniformLegend has not been written. The idea however, is similar to the old solution where we introduce a new parameterset_uni_size toaxes.legend() andaxes.figlegend(). Whenset_uni_size is specified, we check to ensure that no conflicting parameters are specified (custom handlers, marker scale, etc), and return an instance ofUniformLegend instead of justLegend.

TODO: Unit tests.
Would image testing be required for this? I think checking the marker size would suffice.

If possible I would generally prefer tests checking marker sizes over image comparison tests.

This seems like a lot of work for this feature, and I wonder if there is a neater approach. It seems that the key part of this is toset_markersize on each of the handlers of interest.

Rather than implementing a very specific keyword, I'd be looking at simplifying the ability to change the legend class which is used fromplt.legend()/ax.legend() athttps://github.com/matplotlib/matplotlib/blob/master/lib/matplotlib/axes/_axes.py#L509.

With the ability to switch the legend class, I'd then be looking for the best place to override so that I can get the behaviour I wanted. It isn't entirely clear where the handler responsibility lies within the Legend class, but I'd probably plump forLegend._init_legend_box purely for the fact that the handles are created in therehttps://github.com/matplotlib/matplotlib/blob/master/lib/matplotlib/legend.py#L717

It is worth noting, that there is a bit of a smell in the legend code at the moment - some of the artists from the handles are not referenced within legend itself, as there appears to have been an assumption of 1-to-1 mapping of handler to artists.https://github.com/matplotlib/matplotlib/blob/master/lib/matplotlib/legend_handler.py#L205-L208

So with all of this, I'd be looking at (untested):

from matplotlib.legend import Legendfrom matplotlib.legend_handler as legend_handleruniform_size_cases = {legend_handler.HandlerLine2D: lambda handler: class UniformSizeLegend(Legend):    def __init__(self, *args, **kwargs):        super(UniformSizeLegend, self).__init__(*args, **kwargs)        self.uniform_size = kwargs.get('uniform_size', 30)    def _init_legend_box(self, *args, **kwargs):        super(UniformSizeLegend, self)._init_legend_box(*args, **kwargs)        for artist in self.legendHandles:            if hasattr(artist, '_legmarker'):                artist._legmarker.set_markersize(self.uniform_size)            elif hasattr(artist, 'set_markersize'):                artist.set_markersize(self.uniform_size)

And I'd call it with:

plt.legend(uniform_size=20, legend_class=UniformSizeLegend)

Just an alternative thought to the problem.

HTH

Moved to 'proprosed next point release' meaning this is not a blocker for the next release, but can go in if it is ready.

This needs a re-base and@pelson 's comments on a possible re-think have not been addressed.

Closing as abandoned. Seems like it is also in#11127

#11127 is a bit different, it would create many legend entries for a single scatterplot. The idea here was to unify the size of different scatter plot's handles. Many options are shown inthe stackoverflow question "setting a fixed size for points in legend". The easiest is just:

legend = plt.legend()for legobj in legend.legendHandles:    legobj.set_sizes([64])

My faviourite, which I just added to that thread as well would be to use the update function of the Handler.

def update(handle, orig):    handle.update_from(orig)    handle.set_sizes([64])plt.legend(handler_map={PathCollection : HandlerPathCollection(update_func=update)})