pcolormesh: hatch-parameter not appearing

[oschoudhury]

There seems to be a problem with pcolormesh and the hatch-parameter. The hatch-symbol does not seem to appear, if you execute the basic example

x = np.arange(10)
y = np.arange(10)
z = np.arange(100).reshape((10, 10))
plt.pcolormesh(x, y, z, hatch = '*')
plt.show()

I only see the the image but not the *-symbol.

[efiring]

I don't think hatching was ever implemented in collections.QuadMesh, and I don't think it makes sense to do so. The problem here is that our automatic docstring generation doesn't have a mechanism for filtering out kwargs that don't apply. The hatch kwarg was inherited from collections.Collection, but it makes sense only for a subset of Collection subclasses.

[oschoudhury]

I disagree that it does not make sense to use the hatch kwarg here. Let me show it with a basic example (the example can be better, but it does indicate where it might be useful):

x = np.arange(10)
y = np.arange(10)
z1 = np.arange(100).reshape((10, 10))
z2 = 100 + np.arange(100).reshape((10, 10))
mask1, mask2 = np.random.randint(0, 2, size=(2, 10, 10))
masked1 = np.ma.masked_array(z1, mask1)
masked2 = np.ma.masked_array(z2, mask2)
plt.pcolormesh(x, y, masked1, vmin = 0, vmax = 200, alpha = 0.3, hatch = '/')
plt.pcolormesh(x, y, masked2, vmin = 0, vmax = 200, alpha = 0.3, hatch = '\\')

It is hard from the colors alone to demonstrate where the regions overlap, but the addition of the hatch-kwarg would make it a lot easier.

[pelson]

FWIW @oschoudhury - plt.pcolor does allow you to do this (of course, pcolormesh would be more efficient).

Simply adding the line gc.set_hatch(self.get_hatch()) to the draw method of quadmesh enables hatching of the whole mesh, but does not mask the hatch as it does colours if there is missing data.

[petehuang]

Devs, what is the general consensus on implementing hatching in QuadMesh?

[efiring]

I don't know if there is a consensus, but I will continue to vote "no" until someone provides a nice, clean PR that adds hatch support with little extra code. QuadMesh is a strange and complicated thing, so making it do more is not to be undertaken lightly. It's strength is in making relatively dense image-like plots, but unlike imshow, it can handle non-Cartesian grids. Neither pcolormesh nor imshow implements "hatch", although imshow chokes on it rather than silently ignoring it.

1. For situations where hatching is needed, use pcolor instead of pcolormesh. Yes, it's slow, but this is problematic only if the grid is dense. Do the applications where hatching is helpful include such dense grids?

2. The problem of silently ignoring an invalid kwarg should be addressed.

Side comment: pcolormesh and QuadMesh badly need some API and relatively high-level refactoring. I would like to do it, but like so many things, it hasn't risen high enough in priority that I have been able to get to it.

[ozanmert]

Here is another workaround for this missing functionality:

https://stackoverflow.com/a/69059584/2952639

[github-actions]

This issue has been marked "inactive" because it has been 365 days since the last comment. If this issue is still present in recent Matplotlib releases, or the feature request is still wanted, please leave a comment and this label will be removed. If there are no updates in another 30 days, this issue will be automatically closed, but you are free to re-open or create a new issue if needed. We value issue reports, and this procedure is meant to help us resurface and prioritize issues that have not been addressed yet, not make them disappear. Thanks for your help!