Feature Requestion: filled step plot

[jseabold]

This was asked a long time ago, and I just ran into a similar need. It would be nice to be able to easily make a filled step plot. The solutions I've come up with are suboptimal.

http://www.mail-archive.com/matplotlib-users@lists.sourceforge.net/msg13072.html

[lebek]

This sounds like a bug to me, as Eric says in the mailing list:

The Line2D object should not be modifying the
coordinates it is given. At the very least, its get_path() should
return the actual path it is drawing; at present it does not.

[lebek]

Also see #643, especially the fix proposed by @ycopin.

[Tillsten]

I tried to solve this issue, but there some small refactoring in Lines2D is necessary:
self._path is set by recache(). The path drawn for step-lines is only calculated when drawing.
The easiest solution seems to be to move the calculation from "real" path to recache.
Because i am not sure about side effects, i need some input from more experienced
developers? @efiring @mdboom @dmcdougall

Also while studying the code, i found the following lines:

    transformed_path = self._get_transformed_path()
    path, affine = transformed_path.get_transformed_path_and_affine()
    path = affine.transform_path(path)

Are the last two lines not redundant? (But i don't know matplotlib interals good enough to be sure.)

[Tillsten]

A easier solution would be to just add a get_drawn_path method, which returns "real" path (and if necessary calculates it). The step-calculation function would be shared with the draw methods.

[lebek]

We fixed this as part of a group assignment on my CS course. We did a bit of refactoring in Line2D and made get_ydata and get_xdata return the actual drawn data (when orig=False). Filled step plots are then as easy as:

lines = plt.step(x, y)
plt.fill_between(lines[0].get_xdata(orig=False), 0, lines[0].get_ydata(orig=False))
plt.show()

To make things more convenient we added a fill kwarg to plt.plot and plt.step:

plt.step(x, y, fill='red')
plt.show()

Not sure if either change would be desirable upstream but I'll try to put together a PR.

[mdboom]

@Tillsten: Those lines are not redundant. The first performs the non-affine part of the transform, the second performs the affine part.

[mdboom]

@Tillsten: I'm not sure what the side effects you're suggesting would be -- it probably depends on what ways users are relying on the way it works now -- but maybe as a start we could do that refactor and run it against the test suite...

[parrenin]

An additionnal request is to have an equivalent of 'fill_between' for step.

[tacaswell]

Closing as duplicate of #643