Proposal: Add conspicuously missing Functor instances for tuples
matthewtpickering at gmail.com
Tue Jan 19 21:35:41 UTC 2016
To directly answer the initial proposal, there seems to be little harm in
functor instances for n-tuples. Functor instances are usually ok as one
must always provide two arguments to fmap which agree with each other.
There is little chance of using the wrong instance by accident.
However, I am strongly against adding foldable and traversable instances
for n-tuples. I thank Ryan for articulating why he finds them useful but
due to the pervasive use of tuples to return multiple values from functions
I find even the 2-tuple instance distasteful .
It is very easy to introduce bugs when refactoring with the existence of
these instances. If altering a function which returns a useful foldable
such as lists to return a pair then if you ever call length on the returned
value then you have introduced a difficult to track down bug. Unlike calls
to fmap, there is no sanity check to make sure you are using the right
If there are other cases where such instances are useful then I would
reconsider my position but I find the arguments for (a canonical instance
and for consistency) unconvincing. It should not be the case that we
propagate bad behaviour for the sake of consistency. I even think Ryan's
use case would be better served by a separate datatype for that purpose.
On 19 Jan 2016 21:04, <amindfv at gmail.com> wrote:
> If the only uses we can imagine for these instances are
> - Cases where it's not a great design decision
> - Cases where it's used accidentally and results in a silent runtime
> failure instead of a compile-time error
> then I'm a strong -1
> El 19 ene 2016, a las 15:54, Alois Cochard <alois.cochard at gmail.com>
> I don't have any use case like that. I'm in favor of this proposal for
> consistency sake.
> That last part of my comment about usefulness/discouraging usage was about
> using Functor instance on tuple in general, no matter the arity.
> On 19 January 2016 at 21:32, <amindfv at gmail.com> wrote:
>> El 19 ene 2016, a las 10:24, Alois Cochard <alois.cochard at gmail.com>
>> Agree for consistency, I can also see those instances as being useful in
>> some specific context, even if I agree with Andreas that in general they
>> should be discouraged (especially for newcomers).
>> Can you give us an example where using e.g. the Functor instance for a
>> 5-tuple would be the correct/best design decision?
>> On 19 January 2016 at 09:20, Herbert Valerio Riedel <hvriedel at gmail.com>
>>> On 2016-01-18 at 21:10:07 +0100, David Feuer wrote:
>>> > For some reason I really can't imagine, it seems the only tuple type
>>> > with a Functor instance is (,) a. I was astonished to find that
>>> > fmap (+1) (1,2,3)
>>> > doesn't work. Since this is *useful*, and there is *only one way to do
>>> > it*, I propose we add the following:
>>> > instance Functor ((,,) a b) where
>>> > fmap f (a,b,c) = (a,b,f c)
>>> > instance Functor ((,,,) a b c) where
>>> > fmap f (a,b,c,d) = (a,b,c,f d)
>>> > etc.
>>> As stated elsewhere in this thread already, there is the issue about
>>> consistency. Here's a relevant section from the Haskell 2010 report:
>>> > 6.1.4 Tuples
>>> > ...
>>> > However, every Haskell implementation must support tuples up to size
>>> > 15, together with the instances for Eq, Ord, Bounded, Read, and Show.
>>> IMO, we either have no `Functor` instances for tuples at all, or we have
>>> them for all tuples up to size 15. The current situations of having them
>>> defined only for 2-tuples is inconsistent.
>>> Libraries mailing list
>>> Libraries at haskell.org
>> Libraries mailing list
>> Libraries at haskell.org
> Libraries mailing list
> Libraries at haskell.org
-------------- next part --------------
An HTML attachment was scrubbed...
More information about the Libraries