Is NoRepType right for immutable vectors?

[andy-morris]

The following GHCI session illustrates what I'm talking about:

λ» :set -XOverloadedLists 
λ» :m Data.Set Data.Vector Data.Generics.Uniplate.Data

λ» getInts xs = universeBi xs :: [Int]

λ» getInts ([1, 2, 3] :: [Int])
[1,2,3]

λ» getInts ([1, 2, 3] :: Set Int)
[1,2,3]

λ» getInts ([1, 2, 3] :: Vector Int)
[]

Since all vectors declare themselves to be NoRepTypes, the uniplate library (and probably others) immediately stop traversing on encountering a vector. This seems to me like it's probably not the intended behaviour for immutable vectors. (For mutable ones it's a different matter, of course.)

So, unless I'm mistaken, I'd like to suggest that immutable vector types follow the lead of Set and friends, and have Data instances along the lines of the following (but presumably defined once and for all in ...Generic like now):

instance Data a => Data (Vector a) where
    gfoldl f z m   = z fromList `f` toList m
    toConstr _     = fromListConstr -- was: error "toConstr"
    gunfold k z c  = case constrIndex c of -- was: error "gunfold"
        1 -> k (z fromList)
        _ -> error "gunfold"
    dataTypeOf _   = vectorDataType -- was: mkNoRepType "Data.Vector.Vector"
    dataCast1 f    = gcast1 f

fromListConstr :: Constr
fromListConstr = mkConstr vectorDataType "fromList" [] Prefix

vectorDataType :: DataType
vectorDataType = mkDataType "Data.Vector.Vector" [fromListConstr]

[RyanGlScott]

Indeed, the current Data instances seem strangely inconsistent, since we implement gfoldl f z m = z fromList f toList m but don't implement other methods in a similar fashion.

Your proposed implementation sounds sensible to me—care to make a pull request?

[andy-morris]

You probably saw, but if not, I've put it in #205.