blog

One common oversight in Haskell compilers is failing to intern identifiers using Ints and failing to prefer IntMaps and IntSets. The PureScript compiler, for instance, uses Maps as of writing.

I always do the following:

newtype Unique = Unique { unUnique :: Int }
data Name = Name { name   :: T.Text
                 , unique :: !Unique
                 ...
                 }
instance Eq (Name a) where
    (==) (Name _ u) (Name _ u') = u == u'
instance Ord (Name a) where
    compare (Name _ u) (Name _ u') = compare u u'

Note that making the name field lazy makes things faster.

Using Uniques makes equality and comparisons faster; moreover, one can use Names as keys in IntMaps and IntSets, which have more efficient implementations. I came across the dramatic performance advantage in my Kempe compiler; liveness analysis is drastically faster with IntSets compared to Sets.