type (++) = (<>)

[Edward Kmett]

Yes, I attempted to address why this was left unmolested in my last couple
of paragraphs.

For comparison, we have not generalized map to have the same signature as
fmap because we do not have to. fmap already exists,
having the less general signature directly indicates why it isn't possible
for you to use

instance Functor Foo where
  map = ...

and a large portion of the community would like it to remain ungeneralized
(often for teaching purposes).

Given that we already have a polymorphic combinator in Prelude there is no
need to make their teaching tool more awkward to teach with. As it is,
teachers already have to get into typeclasses before they can talk about
foldr, and this already causes a fair bit of consternation.

If we had everything to over again and no installed base? I'd probably say
map should have been the member of Functor, like it is in most other
languages that have anything even remotely resembling type classes at this
point. The pain of transition doesn't warrant it, however.

Similarly, (<>) is winding up in Prelude as part of the Semigroup/Monoid
proposal. (++) was explicitly un-generalized in Haskell 98 for teaching and
error purposes, and we have (<>) present and moving into Prelude, so there
isn't a real reason to go and generalize the second name and there is at
least a weak justification for leaving the status quo in place.

Having two operators for the same thing means another side-condition to
learn, another quirk to learn, etc. You still need to memorize which one
you are allowed to define in Semigroup. The folks who write that same
fiddly kind of pretty printer code do derive some benefit from keeping
those operators separate.

Finally, The status quo at least lets us dream the dream that we could
remove (++) entirely from the language some day. (Even if we're unlikely to
act on this impulse due to the fixity concerns raised above.)

-Edward

On Tue, Jul 4, 2017 at 3:31 AM, Vassil Ognyanov Keremidchiev <
varosi at gmail.com> wrote:

> "plus" symbol is natural denoting joining two pieces in almost all
> languages. That is why I would like not to get rid of it, but to be a bit
> more general. So in new code that will use the new prelude to be able to
> use it on broader range of containers (monoids/semigroups).
>
> I really mean:
> (++) :: Semigroup s => s -> s -> s
>
> About the fixity, it could retain it's original fixity.
>
> 2017-07-04 3:24 GMT+03:00 Edward Kmett <ekmett at gmail.com>:
>
>> Note: I realize nobody is directly saying that we should use (++) instead
>> of (<>) in this conversation just yet, but I want to clear a few things up.
>>
>> One of the early options when the operator (<>) was coined was to try to
>> say we should just generalize the type of (++) instead to make it mappend.
>> (Note: it originally was mplus, in a Haskell version long long ago, so it
>> keeps getting repurposed!) Unfortunately, this plan ran afoul of the fact
>> that the primary libraries using the (<>) notation at the time (pretty
>> printing libraries) also mixed it heavily with (++), exploiting the
>> different fixities involved. (Finding a decent fixity for (<>) was a huge
>> chunk of the conversation at the time.)
>>
>> There is a deliberate fixity difference between (++) and (<>), a good
>> chunk of code out there mixes them that deals with pretty printing that
>> would break pretty horribly if we just outright removed (++), and trying to
>> do a visual search and replace for (++) with (<>) in light of them having
>> different fixities is a very error prone process, so we aren't currently
>> planning on deprecating the existing (++) operator any time soon. At least,
>> nobody has put a proposal to the core libraries committee to that end.
>>
>> Since the call was made to make (<>) become the new operator, we
>> ultimately decided to leave (++) untouched, even though it could be
>> generalized to match (<>), for much the same reason that map still exists,
>> despite there being a more general fmap: Ultimately, there isn't a
>> reasonable migration plan to make (++) or map become the way you define the
>> instance involved, and at least this way the name duplication can be
>> leveraged by the folks who want a less polymorphic combinator.
>>
>> Would the world be a little tidier without map or (++) hanging about?
>> Sure. But the hate mail levels in my inbox would skyrocket commensurately.
>> ;)
>>
>> -Edward
>>
>> On Mon, Jul 3, 2017 at 5:01 PM, Erik de Castro Lopo <mle+hs at mega-nerd.com
>> > wrote:
>>
>>> Vassil Ognyanov Keremidchiev wrote:
>>>
>>> > What do you think of making (++) the same as (<>) so we could use ++ as
>>> > concatenation of any monoid, not just lists in Haskell 2020?
>>> > This will be more intuitive for beginners, too.
>>>
>>> Two symbolic operators that are synonymous seems a bit of a waste. I
>>> would
>>> much rather see (++) be deprecated in favour of (<>). At work we have a
>>> custom prelude which already does this.
>>>
>>> Erik
>>> --
>>> ----------------------------------------------------------------------
>>> Erik de Castro Lopo
>>> http://www.mega-nerd.com/
>>> _______________________________________________
>>> Haskell-prime mailing list
>>> Haskell-prime at haskell.org
>>> http://mail.haskell.org/cgi-bin/mailman/listinfo/haskell-prime
>>>
>>
>>
>> _______________________________________________
>> Haskell-prime mailing list
>> Haskell-prime at haskell.org
>> http://mail.haskell.org/cgi-bin/mailman/listinfo/haskell-prime
>>
>>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mail.haskell.org/pipermail/haskell-prime/attachments/20170704/7c67a638/attachment.html>