Panu Lahtinen

Seems that the comments by @sfinkens and me in #1666 are finally needed: parse everything starting with `{` to a dict.

It seems that setting `generate=True` makes the composite to work. This also means a significant performance hit when producing multiple composites sharing channels (instead of single resampling per channel, all...

Just a thought, but could the first use-case be implemented by combining the existing `Background` and `Masking` compositors?

This doesn't seem to help: `lcl = glbl.resample("euro4", resampler="gradient_search", reduce_data=False, generate=True)`

Neither does `lcl.generate_possible_composites(True)` before saving, which outputs `The following datasets were not created and may require resampling to be generated: DataID(name='geo_color')`.

For the `glbl` Scene: ```python ipdb> for k in glbl.keys(): print(k) DataID(name='ir_105', wavelength=WavelengthRange(min=9.8, central=10.5, max=11.2, unit='µm'), resolution=1000, calibration=, modifiers=()) DataID(name='ir_38', wavelength=WavelengthRange(min=3.4, central=3.8, max=4.2, unit='µm'), resolution=1000, calibration=, modifiers=()) DataID(name='vis_04', wavelength=WavelengthRange(min=0.384, central=0.444,...

I adjusted the title and added the short summary what triggers the behaviour to the description.

I tested by pre-resampling the background image to FCI projection and `euro4`, neither of which helped and a second resampling step is still required when `generate=False` is used.

Above I forgot to also resample the land-sea mask. If both background image and LSM are resampled to the final image projection, the composite generation works with a single `resample()`...

@yukaribbba yes, with #2696 I still need to do douple resampling when using `generate=False`. I think the problem is somewhere in the decision tree in a case where the `IncompatibleAreas`...