[GHC] #13328: Foldable, Functor, and Traversable deriving handle phantom types badly
GHC
ghc-devs at haskell.org
Mon Mar 6 15:32:21 UTC 2017
#13328: Foldable, Functor, and Traversable deriving handle phantom types badly
-------------------------------------+-------------------------------------
Reporter: dfeuer | Owner: (none)
Type: bug | Status: new
Priority: normal | Milestone: 8.4.1
Component: Compiler | Version: 8.1
Resolution: | Keywords: deriving-perf
Operating System: Unknown/Multiple | Architecture:
Type of failure: Runtime | Unknown/Multiple
performance bug | Test Case:
Blocked By: | Blocking:
Related Tickets: | Differential Rev(s):
Wiki Page: |
-------------------------------------+-------------------------------------
Comment (by RyanGlScott):
Two more thoughts to consider.
One: how should this interact with your feature request in #13117? That
is, you had previously requested that if you wrote this:
{{{#!hs
data V a deriving Functor
}}}
then GHC should generate this:
{{{#!hs
instance Functor V where
fmap _ x = case x of {}
}}}
But `V`'s type parameter is phantom here, so you could just as well
implement it like this (as Eric Mertens originally pointed out in
https://mail.haskell.org/pipermail/libraries/2017-January/027603.html):
{{{#!hs
instance Functor V where
fmap _ = coerce
}}}
Which choice should we make here?
Two (a follow-up question to comment:1, after reading Eric's comment about
role annotations in
https://mail.haskell.org/pipermail/libraries/2017-January/027603.html):
consider this code:
{{{#!hs
data Foo a = Foo
deriving instance Foldable Foo
}}}
In this scenario, the type parameter to `Foo` is phantom, so the generated
`Foldable` instance is:
{{{#!hs
instance Foldable Foo where
foldMap _ _ = mempty
}}}
Now what happens if you add a role annotation to `Foo`?
{{{#!hs
type role Foo nominal
data Foo a = Foo
deriving instance Foldable Foo
}}}
Now, since `Foo`'s type parameter is no longer phantom, the generated
`Foldable` instance will be:
{{{#!hs
instance Foldable Foo where
foldMap _ Foo = mempty
}}}
That is to say, the choice of role can affect what the strictness of
`foldMap` will be. Is this desirable?
--
Ticket URL: <http://ghc.haskell.org/trac/ghc/ticket/13328#comment:4>
GHC <http://www.haskell.org/ghc/>
The Glasgow Haskell Compiler
More information about the ghc-tickets
mailing list