[Haskell-cafe] Constructors on left and right hand sides of equation
Erik Hesselink
hesselink at gmail.com
Thu Sep 18 07:44:17 UTC 2014
On Thu, Sep 18, 2014 at 4:08 AM, Richard A. O'Keefe <ok at cs.otago.ac.nz> wrote:
>
> On 18/09/2014, at 2:03 AM, PATRICK BROWNE wrote:
>
>> Dear list
>> In Eq 1 is (NewGlass i) on the LHS distinct from (NewGlass i) on the RHS?
>> In Eq2 are the occurrences of the data constructor Fill on the LHS and RHS distinct?
>>
>> data Glass = NewGlass Int | Fill Glass Int deriving Show
>> drink :: Glass -> Int -> Glass
>> drink (NewGlass i) j = (NewGlass i) -- Eq 1
>> drink (Fill m i) j
>> | full (Fill m i) = Fill (NewGlass (size m)) ((size m) - j) -- Eq 2
>> | (i >= j) = Fill m (i-j)
>> | otherwise = drink m (j-1)
>
> The question question is "what do you MEAN by 'are they distinct'"?
>
> They are two distinct references to a single binding occurrence
> of an identifier. In Eq1, the first occurrence is in a pattern,
> and the second is in an expression.
>
> I wonder if you were interested in the question whether
> the second occurrence of NewGlass i would allocate new storage
> or whether it would use whatever the first occurrence matched?
> In that case, I believe the answer is "it's up to the compiler".
> You can *ask* for the matched data to be shared:
>
> drink g@(NewGlass _) _ = g
> drink g@(Fill m i ) j
> | full g = Fill (NewGlass (size m)) (size m - j)
> | i >= j = Fill m (i - j)
> | True = drink m (j - 1)
>
> Someone else has already pointed out
>
> data Proxy a = Proxy
>
> -- foo :: Proxy a -> Proxy b
> foo Proxy = Proxy -- Eq3
>
> where the two occurrences of Proxy in Eq3 can have different types.
> But note that
>
> -- bar :: Proxy a -> Proxy a
> bar p at Proxy = p -- Eq4
>
> The two occurrences of the variable p in Eq4 must have
> the same type (and the same value).
Since we're talking about allocating new storage: even though 'Eq3'
has a second occurrence of 'Proxy' on the right hand side, I believe
in GHC's implementation these are shared and there is only one Proxy
at runtime. Of course this is implementation dependent.
Regards,
Erik
More information about the Haskell-Cafe
mailing list