[Haskell-cafe] Help with TH trick

Michael Sloan mgsloan at gmail.com
Fri Nov 4 02:45:36 UTC 2022


And the point of the "instantiators" is to support generating instances
where you also require some value-level definition.  This can now be done
via deriving via, except that deriving via will reveal implementation
details to users which may be undesirable.  Very roughly (I did not look at
how instantiators actually work in the package):

data BoolLike = Trueish | Falseish

$($(derive [d|
  instance EqViaBijection Bool BoolLike where
    to True = Trueish
    to False = Falseish
    from Trueish = True
    from Falseish = False
]|))

and the generated code would be an `Eq` instance, and the bijection used
would not be visible to outside users (whereas it is if you use deriving
via for this)

On Thu, Nov 3, 2022 at 8:37 PM Michael Sloan <mgsloan at gmail.com> wrote:

> Ah, it allows the mechanism to be open, so if you create a new typeclass
> you can also declare how to use TH to derive instances.  Roughly (if TH
> allowed splices in more spots in quotes):
>
> class Tangy a where
>   -- ...
>
> instance Deriver (Tangy a) where
>   runDeriver Proxy cxt ty = [d| instance $(cxt) => Tangy $(ty) where {-
> ... -} |]
>
> Now in some other module I can do:
>
> $($(derive [d| instance Deriving (Tangy X) |]))
>
> The first splice will generate the code that dispatches to the instance,
> and the next splice will generate the actual output.  This approach can
> also allow you to use more complex dispatch to the deriver if you wish.
> For example, maybe you have a Serializable class that comes with some
> format type that is openly extensible.  You could define `instance
> Deriver (Serializable Json a)` / `instance Deriver (Serializable Yaml a)`
> etc and `derive` will dispatch to them when interpreting the outer splice.
>
> On Thu, Nov 3, 2022 at 8:23 PM David Feuer <david.feuer at gmail.com> wrote:
>
>> Why the double splice? What about this makes you generate code that when
>> run will generate code?
>>
>> On Thu, Nov 3, 2022, 10:10 PM Michael Sloan <mgsloan at gmail.com> wrote:
>>
>>> Hmm, the documentation does list 5 bullets of why this is a nice way of
>>> doing things.  But yes, considering this didn't catch on no doubt either
>>> the docs are poor quality or the thing isn't really something people want.
>>> FWIW I wrote this many years ago, before deriving via.  I just figured you
>>> might be interested as it is very similar to your query.
>>>
>>> The examples could be better.  One thing to note is that you can list a
>>> bunch of instances:
>>>
>>> $($(derive [d|
>>>         instance Deriving (Storable X)
>>>         instance Deriving (Eq X)
>>>         instance Deriving (Show X)
>>>         |]))
>>>
>>>
>>> (purely an example, afaik it wouldn't make sense to use TH to generate
>>> some canonical Eq and Show instances)
>>>
>>> Whereas with typical `$(mkStorable [t| Storable X|] $)` style you need
>>> multiple splices and you need to know the names of these TH functions.
>>>
>>> In retrospect this would probably be better without the `Deriving`
>>> wrapper, I think I needed that to disambiguate from the instantiators,
>>> which indeed aren't properly documented at all (I don't recall how they
>>> work, and don't feel like digging into it much further)
>>>
>>> On Tue, Oct 25, 2022 at 11:21 PM David Feuer <david.feuer at gmail.com>
>>> wrote:
>>>
>>>> I must admit that looks rather mysterious to me. The documentation I
>>>> saw doesn't make the benefit of the nested splice terribly obvious.
>>>>
>>>> On Tue, Oct 25, 2022, 10:53 PM Michael Sloan <mgsloan at gmail.com> wrote:
>>>>
>>>>> You might also be interested in
>>>>> https://hackage.haskell.org/package/th-utilities-0.2.5.0/docs/TH-Derive.html -
>>>>> though afaik it didn't catch on.
>>>>>
>>>>> The cleverness there is to use nesting splicing, where the inner one
>>>>> generates code involving something like "runDeriver (Proxy @ Storable X)
>>>>> ...", where runDeriver is a method of a typeclass.  The instances of this
>>>>> class implement the code generation that is ultimately used for the output.
>>>>>
>>>>> $($(derive [d|
>>>>>         instance Deriving (Storable X)
>>>>>         |]))
>>>>>
>>>>>
>>>>> On Sat, Oct 22, 2022, 13:48 David Feuer <david.feuer at gmail.com> wrote:
>>>>>
>>>>>> Okay, I found it: recursion-schemes.
>>>>>>
>>>>>> On Sat, Oct 22, 2022, 3:05 PM David Feuer <david.feuer at gmail.com>
>>>>>> wrote:
>>>>>>
>>>>>>> That doesn't work out so well when the class has a bunch of methods.
>>>>>>> I definitely saw the nice way on Hackage ... somewhere.
>>>>>>>
>>>>>>> On Sat, Oct 22, 2022, 2:49 PM Brandon Allbery <allbery.b at gmail.com>
>>>>>>> wrote:
>>>>>>>
>>>>>>>> I'm not aware of anything specifically like that, but the `makeā€¦`
>>>>>>>> functions in
>>>>>>>> https://hackage.haskell.org/package/deriving-compat-0.6.1/docs/Data-Deriving.html
>>>>>>>> may allow you to do something similar?
>>>>>>>>
>>>>>>>> On Sat, Oct 22, 2022 at 2:32 PM David Feuer <david.feuer at gmail.com>
>>>>>>>> wrote:
>>>>>>>> >
>>>>>>>> > I remember seeing a package that offered a function used
>>>>>>>> something like this:
>>>>>>>> >
>>>>>>>> > $(deriveThingy [d| instance Foo a => Thingy a |])
>>>>>>>> >
>>>>>>>> > to allow the user to specify the constraint(s) for a generated
>>>>>>>> instance. I'd love to borrow the code for that, but I can't remember what
>>>>>>>> package it was in, what class it derived, etc. Can anyone help?
>>>>>>>> > _______________________________________________
>>>>>>>> > Haskell-Cafe mailing list
>>>>>>>> > To (un)subscribe, modify options or view archives go to:
>>>>>>>> > http://mail.haskell.org/cgi-bin/mailman/listinfo/haskell-cafe
>>>>>>>> > Only members subscribed via the mailman list are allowed to post.
>>>>>>>>
>>>>>>>>
>>>>>>>>
>>>>>>>> --
>>>>>>>> brandon s allbery kf8nh
>>>>>>>> allbery.b at gmail.com
>>>>>>>>
>>>>>>> _______________________________________________
>>>>>> Haskell-Cafe mailing list
>>>>>> To (un)subscribe, modify options or view archives go to:
>>>>>> http://mail.haskell.org/cgi-bin/mailman/listinfo/haskell-cafe
>>>>>> Only members subscribed via the mailman list are allowed to post.
>>>>>
>>>>>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mail.haskell.org/pipermail/haskell-cafe/attachments/20221103/97b04b78/attachment.html>


More information about the Haskell-Cafe mailing list