bug - Manual_Cluster_ID can go missing with feature as color and Surface Data

[Cryaaa]

This probably is somewhat of an edgecase but I found that if you:

1. use the cluster plotter and select a feature to color the plot,
2. then delete the layers for which the plot was made
3. load Surface data with new features

This causes the "Manual_Cluster_ID" column to go missing from the dropdown and you cannot manually annotate labels again until you close and reopen the plotter -> the Reset button does not fix this. I guess this has something to do with the fact that for surface data we have to hard code the features instead of loading them with the viewer.add_surface() function. I guess there is some kind of event that is being skipped because of this

[jo-mueller]

Hi @Cryaaa woah, good catch! Your hypothesis is probably on point - surface_layer.features is currently only tolerated but not really supported. In other words, viewer.add_surface(surface, features=...) won't complain if it receives features and will put them into the layer.features but it probably also doesn't hook up other functionality such as events.

Maybe we could add a little extra statement in the reset button that inserts the MANUAL_CLUSTER_ID if it's not present?

Edit: Just did a little digging and I think from the napari side it's fine as this is now fully supported. Will check!

Edit2: Maaayyybe the underlying problem here is that the clusters-plotter doesn't clean up behind itself properly in case every layer is removed from the viewer. I can reproduce the problem with any sequence like this:

1. Use clusters-plotter on layer A
2. close layer A (but not the plotter
3. open new layer B --> coloring messed up

[Cryaaa]

Edit2: Maaayyybe the underlying problem here is that the clusters-plotter doesn't clean up behind itself properly in case every layer is removed from the viewer. I can reproduce the problem with any sequence like this:

1. Use clusters-plotter on layer A
2. close layer A (but not the plotter
3. open new layer B --> coloring messed up

Actually for me it works fine if the new layer that is added is a points or labels layer. It just snaps back to what it was if reopening the layer or resets when adding a new layer type. For me it's just the surface layer which messes up.

Maybe we could add a little extra statement in the reset button that inserts the MANUAL_CLUSTER_ID if it's not present?

I think this would definitely be the best approach because it was also my first instinct to use it when realizing the bug so I guess other users might use it similarly

[jo-mueller]

I was able to identify a bunch of related issues. For instance, if you

• add some layers
• do some clustering
• delete the layers
• click reset
• Segfault 😞

#379 should fix it a bit more globally 👍