map_top_n udf proposal

Description: This function takes a map as input and returns the top n keys based on their values as a map. This feature is used frequently in a lot of scenarios so having a udf would help write simpler queries in short time.

map_top_n(map(K, V), integer n) -> map(K, V)

SELECT map_top_n(MAP(ARRAY[1, 2, 3, 4], ARRAY[120, 930, 200, 301]), 2); -- {2 -> 930, 4 -> 301}
SELECT map_top_n(MAP(ARRAY[1, 2, 3, 4], ARRAY['abc', 'abcd', 'efg', 'aaa']), 2); -- {2 -> 'abcd', 3 -> 'efg'}
SELECT map_top_n(null, 2); -- null
SELECT map_top_n(MAP(ARRAY[1, 2, 3], ARRAY[200, 100, 100]), 2); -- {1 -> 200, 2 -> 100}
SELECT map_top_n(MAP(ARRAY[1, 2], ARRAY[200, 100]), 4); -- {1 -> 200, 2 -> 100}

[rongrong]

map_filter_n is a very strange concept. The lambda is not really defining the filter, but a sorting order. So what you really want is map_first_n(map, (k1, v1, k2, v2) -> ..., n), which is basically massaging all parameters in map_from_entries(slice(array_sort(map_entries(map), v -> ...), 1, n)) into one function. I’m not sure how appealing is that. Once we support SQL expression functions (#9613), this can be added as a SQL expression function instead. A more generic need for map with lambda function is probably a map_reduce similar to the array reduce function. It allows you to keep a state while looping through entries. Though in this case the state would be quite complicated. It’s unlikely to have an implementation with lambda that’s more readable or performant than converting to array, sort, slice then convert back.

[avk49]

Yes that definitely works ! But just curious as to why not have a udf similar to map_filter exclusively for handling maps.