On Fri, Dec 28, 2012 at 11:42 AM, Herbert Valerio Riedel <hvr@gnu.org> wrote:

Hi,

Michael Snoyman <michael@snoyman.com> writes:
[...]

>> Because they're properly represented as semigroups too, just like Min and
>> Max (i.e. they don't behave sensibly "out of the box" on empty lists). The
>> semigroups package already provides the proper types and instances. And
>> just like Min and Max, the semigroups package lets you "lift" its First and
>> Last into Monoids with the Option type.

> Just because such a lifting is possible doesn't mean that it will be
> intuitive or obvious to new users. (I wouldn't know how to make this
> switch, for example.)

...this seems to call for better documentation (including
examples) of the Data.Semigroup.Option Monoid instance then.

Do you have an example where it might not be obvious how to use

Option (Semigroup.First a)

instead of

Monoid.First a

?

cheers,
hvr

I'm simply approaching this from the standpoint of a new user. Monoid is enough of a hurdle to try and learn and understand properly. Now that the user has learnt about Monoid, he/she goes to Data.Monoid because the Hoogle documentation leads there, and currently finds the First and Last datatypes. Case closed, objective achieved.

If we remove First and Last, then the user now needs to:

1. Realize that semigroups exist

2. Realize that they are related to Monoid

3. Find the documentation for them

4. Understand that Option can be composed with First to create a Monoid

Perhaps adding a bunch of documentation to Data.Monoid could solve this, but I simply don't see that the current First and Last datatypes are problematic enough that they should actually be removed.

Michael