In general, merging different filters may be complex - especially so in the upcoming version where attributes are associated to a list of possible values for a given key (e.g. well equal to B03 or B04 or B05).

Example: what does it mean to "merge" filters well=[B03, B04, B05] and well=[B03, B06]? What is the expected outcome?
We can clearly find the most reasonable answer, and implement it.

This would be even more complex if we have both include and exclude (which we actually propose to not pursue - ref #2154).

Note that this definition is strictly necessary for the interactive definition of job filters, as in fractal-analytics-platform/fractal-web#682 (comment). In this modal we have to create the "patch" filters which would produce the expected list of images when combined with existing filters.

With @mfranzon, we suggest to take a more clear direction - and state that attribute filters are never merged, but only replaced by highest-priority values.

Example:
old_attribute_filter = {well=[B03, B04, B05], anotherkey: [value1, value2]}
new_attribute_filter = {well=[B03, B06]}
resulting_filter = {well=[B03, B06], anotherkey: [value1, value2]}

Note: none of this applies to type filters, because there is no ambiguity in merging key=True with a higher-priority key=False. It's already a replacement. This is connected to our proposal to also split attributes and types into different db/API field - ref #2153.