#14195: Generalize makeStableName# -------------------------------------+------------------------------------- Reporter: andrewthad | Owner: (none) Type: feature request | Status: patch Priority: low | Milestone: Component: Compiler | Version: 8.2.1 Resolution: | Keywords: Operating System: Unknown/Multiple | Architecture: | Unknown/Multiple Type of failure: None/Unknown | Test Case: Blocked By: | Blocking: Related Tickets: | Differential Rev(s): Phab:D3928 Wiki Page: | -------------------------------------+------------------------------------- Description changed by bgamari: Old description:

The function `makeStableName#` has the following type:

{{{ makeStableName# :: a -> State# RealWorld -> (#State# RealWorld, StableName# a#) }}}

I believe that it could safely be changed to:

{{{ makeStableName# :: a -> State# s -> (#State# s, StableName# a#) }}}

Currently, I have some code that looks like this:

{{{ {-# LANGUAGE MagicHash #-} {-# LANGUAGE UnboxedTuples #-} module Example ( Stable , stabilize , destabilize ) where

import GHC.Prim import GHC.ST

data Stable s a = Stable (StableName# a) a

stabilize :: a -> ST s (Stable s a) stabilize a = ST $ \ s -> case makeStableName# a (unsafeCoerce# s) of (# s', sn #) -> (# unsafeCoerce# s', Stable sn a #)

destabilize :: Stable s a -> a destabilize (Stable _ a) = a

instance Eq a => Eq (Stable s a) where Stable sn1 a1 == Stable sn2 a2 = case eqStableName# sn1 sn2 of 0# -> a1 == a2 _ -> True }}}

I want to be working in `ST`, not in `IO`, and I believe that this use of `unsafeCoerce#` is safe. However, it would be nice to not have to use it at all.

New description: The function `makeStableName#` has the following type: {{{#!hs makeStableName# :: a -> State# RealWorld -> (#State# RealWorld, StableName# a#) }}} I believe that it could safely be changed to: {{{#!hs makeStableName# :: a -> State# s -> (#State# s, StableName# a#) }}} Currently, I have some code that looks like this: {{{ {-# LANGUAGE MagicHash #-} {-# LANGUAGE UnboxedTuples #-} module Example ( Stable , stabilize , destabilize ) where import GHC.Prim import GHC.ST data Stable s a = Stable (StableName# a) a stabilize :: a -> ST s (Stable s a) stabilize a = ST $ \ s -> case makeStableName# a (unsafeCoerce# s) of (# s', sn #) -> (# unsafeCoerce# s', Stable sn a #) destabilize :: Stable s a -> a destabilize (Stable _ a) = a instance Eq a => Eq (Stable s a) where Stable sn1 a1 == Stable sn2 a2 = case eqStableName# sn1 sn2 of 0# -> a1 == a2 _ -> True }}} I want to be working in `ST`, not in `IO`, and I believe that this use of `unsafeCoerce#` is safe. However, it would be nice to not have to use it at all. -- -- Ticket URL: http://ghc.haskell.org/trac/ghc/ticket/14195#comment:2 GHC http://www.haskell.org/ghc/ The Glasgow Haskell Compiler

#14195: Generalize makeStableName# -------------------------------------+------------------------------------- Reporter: andrewthad | Owner: (none) Type: feature request | Status: patch Priority: low | Milestone: Component: Compiler | Version: 8.2.1 Resolution: | Keywords: Operating System: Unknown/Multiple | Architecture: | Unknown/Multiple Type of failure: None/Unknown | Test Case: Blocked By: | Blocking: Related Tickets: | Differential Rev(s): Phab:D3928 Wiki Page: | -------------------------------------+------------------------------------- Comment (by andrewthad): You're right. To actually make this be safe, we would have to redefine `StableName#` as well: {{{#!hs data StableName# s a makeStableName# :: a -> State# s -> (#State# s, StableName# s a#) }}} And now you can't use `runST` create a `StableName` at the top level like you did before. But, now this becomes a more annoying change to make. I'm going to go ahead and close this because, as you've pointed out, it's not as simple to do as I thought, and because I've realized that I don't actually need to use `StableName` to accomplish the thing I was originally planning on doing. -- Ticket URL: http://ghc.haskell.org/trac/ghc/ticket/14195#comment:4 GHC http://www.haskell.org/ghc/ The Glasgow Haskell Compiler