Unbiasing the Semigroup instance for Map

Change Proposals
23

To follow up with this, here are the changes we would make if we went with the approach above described by Harry:

  • drop the Semigroup and Monoid instance for Data.Map
  • define a newtype in Data.Map called SemigroupMap. Its name makes it more obvious why it exists. Its Semigroup instance would be unbiased. It would have all the type class instances of Data.Map. However, it does not have any of the API of Data.Map (e.g. lookup, etc.). Thus, you would only use it when you need Data.Map to have a Semigroup instance.

This comes with the following advantages

  • Those who do rely on the left-bias nature of the current Map will not experience a “silent” change. Rather, they will be forced to update their code. In situations where they are using this behavior, they will have to coerce the Map k v to Map k (First v) and then wrap it in the SemigroupMap newtype to get the same behavior.
  • Those who want the flexibility of the unbiased Semigroup instance can have it now rather than later.
  • We don’t have two Map types that may confuse beginners or make it harder to import the correct function.
  • In v0.15.0, I believe we would deprecate the SemigroupMap newtype and add the unbiased Semigroup and Monoid instance back to Data.Map with a Warn that says something like “Instance differs from the instance defined in PS releases <= 0.13.x.” In v0.16.0, we would remove the SemigroupMap newtype and remove the Warn constraint on the Semigroup instance.
  • Those who jump two major releases would be notified of the change via the Warn constraint. Those who jump three major releases would not be notified, but that is likely very rare.
4 Likes