What to do with 0 or negative values?

[vwwv]

Even though on the docs it says it should work only with positive values, sometimes it is easy missed introducing a hard to find bug (at least It happened to me :/ ). As an example, this can be unintuitive fromListOccur [(x,1),(x,-1)] == fromListOccur [(x,0)] /= fromListOccur [] . I think it might be a good idea either:

• Detect when a zero or negative has been used and throw an error "negative or zero occurrences not supported".
• Extend the implementation such it handle zero and negative values. (Internally only storing elements with an occurrence different than 0 ).

I think the second option is more interesting cause it eliminates the problem and makes the library useful for new use-cases; but it would slightly change the current semantics of some functions.

[jml]

One option might be to change the type to Natural? It wouldn't improve compile-time safety much but at least the exception would be better, and the intent more clearly communicated.

Wouldn't catch 0 either.

[jwaldmann]

This seems to be related to #10 (currently open?)

fromListOccur should check that multiplicities are positive (or natural - and then ignore zeroes) (it should not silently drop negative multiplicities).
I cannot imagine a situation where that would hurt performance - it iterates over a list anyway, and this check is just one extra step per element.

If anyone really needs the non-checking version then (they should provide a benchmark and)
it could be renamed to ...unchecked or be moved to Data.MultiSet.Unchecked.fromListOccur.

On the other hand, the current (risky) API is consistent with, e.g.,
Data.Set.fromAscList https://hackage.haskell.org/package/containers-0.6.0.1/docs/Data-Set.html#v:fromAscList which is also non-checking (leading to fromAscList [2,0] == fromAscList [0,2] happily returning False). I could not find any previous discussion about this.