-
Notifications
You must be signed in to change notification settings - Fork 10
Description
Since benchmarks probably would really follow the data type authors approach "make counterproductive things unrepresentable" (like not providing features where data types are completely unsuited for according features, for example [] not having built-in snoc) & if data type does not provides an interface - benchmarks really should not construct a composite operation to do what was not provided by authors, because then questions to the implementation can arise & benchmark advertises & benchmarks an operation that is absent in the interface.
From the participation of particular data types in particular benchmark groups - the data type feature sets can be inferred (do they provide indexing, sorting, filtering, filtering by index) - that table in itself can be autogenerated & it would at once give a map of data type abilities. So keeping the benchmarks would draw a feature table, & having a feature table would help people to from the get-go see what data types apply to their stage/case/usage without doing a mental superposition of "have I accounted for everything".