primitive-unlifted-1.0.0.0 fails Core Lint on GHC 9.0 and later #24
Comments
|
@treeowl has written most of the current incantation of At the least, it's probably doing a hackage revision to clarify that this is not expected to work with GHC 9.0. @treeowl What do you think? |
|
I'll have to take a look, but yeah, this should all be obsolete soon. Do
you know if the primop updates will happen in that version?
…On Tue, Mar 9, 2021, 2:01 PM Andrew Martin ***@***.***> wrote:
@treeowl <https://github.com/treeowl> has written most of the current
incantation of primitive-unlifted, so he would be more likely to know the
answer. I'm not too concerned about this though. Since boxed rep recently
made it into GHC head and unlifted data types looks like it's about to
land, the whole primitive-unlifted library feels much less useful to me
than it once has.
At the least, it's probably doing a hackage revision to clarify that this
is not expected to work with GHC 9.0. @treeowl
<https://github.com/treeowl> What do you think?
—
You are receiving this because you were mentioned.
Reply to this email directly, view it on GitHub
<#24 (comment)>,
or unsubscribe
<https://github.com/notifications/unsubscribe-auth/AAOOF7I5HDMZBAGRWO5KBOLTCZPBBANCNFSM4Y4HRSSA>
.
|
The |
|
I sure hope that comes soon, because without it, the whole `BoxedRep` thing
is purely a complication with no improvement. Indeed, I believe it's a
complication (rather than an improvement) compared to the original proposal
of using separate primitive types for unlifted arrays, etc., unless and
until someone finds a way to make anything other than primops levity
polymorphic in this sense.
…On Tue, Mar 9, 2021 at 4:12 PM Ryan Scott ***@***.***> wrote:
Do you know if the primop updates will happen in that version?
The BoxedRep commit
<https://gitlab.haskell.org/ghc/ghc/-/commit/3e082f8ff5ea2f42c5e6430094683b26b5818fb8>,
as far as I can see, has only updated RuntimeRep itself, not only
primops. I'm not aware of any work done towards updating primops to use
BoxedRep yet, which is being tracked in GHC#18080
<https://gitlab.haskell.org/ghc/ghc/-/issues/18080>.
—
You are receiving this because you were mentioned.
Reply to this email directly, view it on GitHub
<#24 (comment)>,
or unsubscribe
<https://github.com/notifications/unsubscribe-auth/AAOOF7LHOHRZJBDSJ7WZUITTCZ6LBANCNFSM4Y4HRSSA>
.
|
|
I'm working on that at https://gitlab.haskell.org/ghc/ghc/-/merge_requests/5218 It doesn't seem like it's going to be difficult. Once I figure out what I messed up with |
If you compile
primitive-unlifted-1.0.0.0using the-dcore-lintflag with GHC 9.0 or later, Core Lint will fail on thetryTakeUnliftedMVar#function inData.Primitive.Unlifted.MVar.Primops. Here is a minimized example:{-# LANGUAGE DataKinds #-} {-# LANGUAGE KindSignatures #-} {-# LANGUAGE MagicHash #-} {-# LANGUAGE UnboxedSums #-} {-# LANGUAGE UnboxedTuples #-} {-# LANGUAGE UnliftedNewtypes #-} {-# OPTIONS_GHC -dcore-lint #-} module Bug where import GHC.Exts import GHC.Types newtype UnliftedMVar# s (a :: TYPE UnliftedRep) = UnliftedMVar# (MVar# s Any) tryTakeUnliftedMVar# :: UnliftedMVar# s a -> State# s -> (# State# s, (# (##) | a #) #) {-# NOINLINE tryTakeUnliftedMVar# #-} tryTakeUnliftedMVar# (UnliftedMVar# mv) s = case unsafeCoerce# (tryTakeMVar# mv s) of (# s', 0#, _ #) -> (# s', (#(##)| #)#) (# s', _, a #) -> (# s', (#|a #) #)It's unclear to me if this is a GHC bug or an unsound use of
unsafeCoerce#(in particular, I'm not sure if the coercion fromAny @(*) :: *toa_ax5 :: TYPE 'UnliftedRepis sound), so I thought I'd raise an issue here first.The text was updated successfully, but these errors were encountered: