Update shapes

[kevinyamauchi]

This PR updates the shapes SVG writer. In particular, it adds support for colored shapes and updates the handling of opacity.

I accidentally started the PR here: https://github.com/sofroniewn/napari-svg/pull/1

Closes #6, #10

Testing

To test this, I checked out the PR in a fresh environment, pip installed locally and then installed napari from this PR.

[kevinyamauchi]

I am bringing this test error over from the other PR:

@sofroniewn , when you ran tests before did you get errors from vispy for the image and labels layers?

FAILED napari_svg/_tests/test_write_layer.py::test_write_image_no_extension[image] - KeyError: 'colormap name gray not found'

.venv/lib/python3.7/site-packages/vispy/color/colormap.py:1120: KeyError

I think it might be these lines: https://github.com/sofroniewn/napari-svg/blob/834f7d104065618ae8a71ed78c399379b4404078/napari_svg/layer_to_xml.py#L67-L70

I think it should be grays:

https://github.com/vispy/vispy/blob/bb4606dc6a41832732474a037bbb92628646815d/vispy/color/colormap.py#L1049-L1059

It looks like 'gray' is a matplotlib colormap that is supported by vispy and 'grays' is a vispy-native colormap. It looks like vispy tries to import matplotlib and if it can't, it skips trying to return a matplotlib colormap. Do you think that pytest is unable to import matplotlib for some reason?

From the same environment, I can both import matplotlib from the python terminal and use the svg writer plugin to make a working SVG. Any thoughts @tlambert03 ?

[sofroniewn]

It looks like 'gray' is a matplotlib colormap that is supported by vispy and 'grays' is a vispy-native colormap. It looks like vispy tries to import matplotlib and if it can't, it skips trying to return a matplotlib colormap. Do you think that pytest is unable to import matplotlib for some reason?

Hmm strange, not sure what's going on there. Seems like going with the vispy default is safest for now

[kevinyamauchi]

We can switch it to grays in the test. The default colormap in the Image layer is gray, so we would also have to either pass a colormap argument in with the layer metadata in the tests or change the default in the Image layer. To be honest, I am a little surprised this issue hasn't come up because neither napari nor vispy explicitly depends on matplotlib. I guess matplotlib is just everywhere 😛

I will do a bit more testing of the tests to see if I can definitely pin down the root cause.

On a somewhat-related note, I just realized that calling the non-data parameters of the layer metadata for the plugins is a little confusing given that all of the layers also have a metadata parameter.

[sofroniewn]

We can switch it to grays in the test. The default colormap in the Image layer is gray, so we would also have to either pass a colormap argument in with the layer metadata in the tests or change the default in the Image layer.

Maybe we should switch to grays everywhere, including napari

On a somewhat-related note, I just realized that calling the non-data parameters of the layer metadata for the plugins is a little confusing given that all of the layers also have a metadata parameter.

Indeed, we tend to use meta for the full layer metadata, and then metadata for the dict on each layer. I don't love it, but it's where we are at right now ....

[kevinyamauchi]

Okay, I think the shapes layer writer is working with colors and I have addressed the empty shapes layer and grayscale colormap issue. An SVG generated with the plugin with a layer with colored shapes and an empty shapes layer:

Colored shapes issue (#6 )

• The changes in https://github.com/napari/napari/pull/1248 expose the Nx4 color arrays for the edge and face colors. we now use those to color the shapes in the SVG
• Opacity is now set by the layer opacity

Empty shapes layer issue (#10 )

• made the extreema calculation return np.nan for empty layers
• the full extrema calculation is down with np.nanmax() and np.nanmin(), so the elements with nan extreema do not affect the calculation (i.e., the extreema are calculated ignoring the emtpy layers).
• if the final XML has np.nan extreema (i.e., all layers were empty), it outputs an empty SVG with height and width 1 (it seems that a height of 0 is invalid 😛)

In terms of behavior this means that:

• Writing SVGs with all empty layers ouputs an empty SVG with a height and width of 1.
• Writing SVGs with an empty shapes layer, but other layers with things (including other shapes layers) writs an SVG with the extreema calculated ignoring all empty layers with all of the elements from the non-empty layers.

Grayscale colormap issue (#13):

• As discussed in #13, I convert the name of the grayscale color map for Image/Labels layers from gray (the matplotlib name) to grays the vispy name.
• There are likely other colormaps from matplotlib that are vendored in napari that will cause the SVG writer to choke if the user hasn't separately installed matplotlib.

[kevinyamauchi]

@sofroniewn, okay sounds good. I will remove those comments. I am going to bed soon, so I will leave it up in case anybody has any comments and then merge tomorrow.

[kevinyamauchi]

@sofroniewn , it looks like I don't have write access to this repo, so I think you will have to merge it.

[sofroniewn]

Just gave you write access. I will merge now. Should I also make the new release 0.1.3 release or are we waiting on anything else?

[kevinyamauchi]

I just tested this with https://github.com/napari/napari/pull/1248. All of the napari tests pass and the writer is able to handle empty shapes layers, so I think it is good to go. Merging now!

@sofroniewn , it would be great if you could make the 0.1.3 release!