#2467: orphan instance warnings are badly behaved
---------------------------------+------------------------------------------
Reporter: duncan | Owner:
Type: bug | Status: new
Priority: normal | Milestone: 6.12 branch
Component: Compiler | Version: 6.8.2
Severity: normal | Resolution:
Keywords: | Difficulty: Unknown
Testcase: | Os: Unknown/Multiple
Architecture: Unknown/Multiple |
---------------------------------+------------------------------------------
Comment (by Syzygies):
Reflecting further, I noticed that these manual sections go together
nicely:
> 5.1.2. Command line options in source files
> 5.6.12. Orphan modules and instance declarations
In particular, the phrase stands out
> in some circumstances, the OPTIONS_GHC pragma is the Right Thing.
Moving ''-fno-warn-orphans'' from the command line to a particular orphan
file is idiomatic GHC for my situation. I already have a module containing
helper functions such as
{{{
foldrC ∷ (Foldable d, Foldable e, Foldable f) ⇒
(c a → d (e (f a))) → (a → b → b) → b → c a → b
foldrC gi h y = foldr (flip $ foldr (flip $ foldr h)) y . gi
}}}
for defining deep Foldable instances that go through a newtype and three
layers of structure.
{{{
instance Foldable ((,) a) where foldr f z (_,y) = f y z
}}}
can arise as one of those layers, so this is a natural place to put such
orphans.
Instance declarations are just as hard on the programmer as on GHC itself.
For example, ''Functor ((,) a)'' is documented in Control.Monad, but
importing Control.Monad doesn't expose this instance. ''Functor ((,) a)''
is not documented in Data.Foldable, but importing Data.Foldable does
expose this instance. This left me perplexed as to why I could define deep
Functor instances that relied on ''Functor ((,) a)'', but I couldn't write
''foldr (+) 2 (1,2)'' in ghci. I was left believing that this was some
compiler magic I didn't understand, until I started using ''ghc --show-
iface'', and spelunking the library source code.
If I could have one wish related to instance declarations, it wouldn't
have to do with these orphan warnings, but rather with better control over
duplicate instance declarations. For example, I can't import Data.Foldable
and at the same time try to redefine ''instance Monad ((->) a)''; there is
no ''hiding'' clause that affects instances. Perhaps this is a moot point,
as I have yet to find a standard library instance that has a reasonable
alternative definition. One can imagine Haskell' providing missing
instances anytime it can prove that there is only one possible definition.
--
Ticket URL: <http://hackage.haskell.org/trac/ghc/ticket/2467#comment:9>
GHC <http://www.haskell.org/ghc/>
The Glasgow Haskell Compiler_______________________________________________
Glasgow-haskell-bugs mailing list
Glasgow-haskell-bugs@haskell.org
http://www.haskell.org/mailman/listinfo/glasgow-haskell-bugs
