missingmask of all layers in rasterstack

[tiemvanderdeure]

Changes the default behaviour of missingmask when applied on a RasterStack or RasterSeries, so the mask covers all layers.

It's probably possible to reduce the memory use of this operation by using missingmask!, but I can't really see how to guarantee that that works when not all dimensions are shared.

The dispatch on RasterSeries might run into errors if the dimensions are not shared between the layers, so maybe we don't want it?

[rafaqz]

Yeah it will have to be using mask!.

broadcast_dims! can probably handle this pretty easily. First make the mask array matching the total size of the stack.
then map over layers(stack) and use broadcast_dims! on each layer to update the values.

Should only need one allocation. Will work even if stack layer dims are in weird orders, missing, etc.

Bad part is missing values in small layers propagate to every other dimension, but I guess we want that.

[tiemvanderdeure]

I definitely think we want missing values to propagate across dimensions. I can't come up with another consistent way to do this.

[tiemvanderdeure]

Alright, tried my best. But it might be you can come up with some wizardry to make this even cleaner?

[codecov-commenter]

Codecov Report

Attention: 1 lines in your changes are missing coverage. Please review.

Comparison is base (18dcf5d) 82.09% compared to head (9e96152) 82.17%.

Files	Patch %	Lines
src/methods/mask.jl	93.75%	1 Missing ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##             main     #599      +/-   ##
==========================================
+ Coverage   82.09%   82.17%   +0.08%     
==========================================
  Files          61       61              
  Lines        4228     4242      +14     
==========================================
+ Hits         3471     3486      +15     
+ Misses        757      756       -1     

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

[rafaqz]

Is there a test for mixed size stacks?

[tiemvanderdeure]

I added some.

Right now missingmask will fail on a RasterSeries if some layers have dimensions that the first layer does not have. For a RasterStack it just calls dims, but we cannot do this for a RasterSeries.

What do you think the most logical behaviour would be in that case?

[rafaqz]

RasterSeries is for fixed dimensions so thats fine.

[tiemvanderdeure]

In that case I think we're good to go.