#14196: Replace ArrayArray# with either UnliftedArray# or Array# -------------------------------------+------------------------------------- Reporter: andrewthad | Owner: (none) Type: feature request | Status: new Priority: normal | Milestone: Component: Compiler | Version: 8.2.1 Resolution: | Keywords: | LevityPolymorphism Operating System: Unknown/Multiple | Architecture: | Unknown/Multiple Type of failure: None/Unknown | Test Case: Blocked By: | Blocking: Related Tickets: | Differential Rev(s): Wiki Page: | -------------------------------------+------------------------------------- Comment (by simonpj): OK now I get it. I had to look at [https://downloads.haskell.org/~ghc/7.4.2/docs/html/libraries/ghc- prim-0.2.0.0/GHC-Prim.html#v:indexArrayArrayArray-35- the documentation for ArrayArray#]. If I may say it like this: The current primitive data type `ArrayArray#` is a heap-allocated array of pointers to unlifted objects. Any kind of unlifted objects would be fine, provide they are represented by a pointer. Specifically, we can put both `Array#` and `ByteArray#` (and I suppose another `ArrayArray#`) inside an `ArrayArray#`. But doing so is jolly awkwerd because `ArrayArray#` is not a parameterised type. Why isn't it parameterised? Becuase previously we had no way to quantify over unlifted types. But now we do. So we can make `ArrayArray#` into a paremterised type, namely your new {{{ data UnliftedArray# (a :: TYPE 'UnliftedRep) }}} Your idea here is that `UniftedArray#` is an array of pointers to ''unlifted'' values. Cool, I like it. Levity polymorphism is great. -- Ticket URL: http://ghc.haskell.org/trac/ghc/ticket/14196#comment:3 GHC http://www.haskell.org/ghc/ The Glasgow Haskell Compiler