Sure, but that's okay, because I never need to expose the type to anyone. I'm just nervous because I haven't seen documentation saying those values are type-specific. On Feb 11, 2018 10:37 PM, "Edward Kmett" wrote:

Keep in mind a newtype of Foo will have a different TypeRep, but will compare as equal under reallyUnsafePtrEquality#.

-Edward

On Sun, Feb 11, 2018 at 5:14 AM, David Feuer wrote:

...

Can I use reallyUnsafePtrEquality# reliably to identify whether a value is a nullary constructor of a particular type? For example, if I have

data Foo = Foo

Can I write

isFoo :: a -> Bool isFoo !a = isTrue# (reallyUnsafePtrEquality# a Foo)

instead of

isFoo :: forall a. Typeable a => a -> Bool isFoo a | Just Refl <- eqTypeRep (typeRep @a) (typeRep @Foo) , Foo <- a = True | otherwise = False

The reason I'm asking is because this would let me (potentially) raiseIO# a nullary constructor and then catch# it and see if it was what I was looking for rather than having to open a SomeException to get to an Exception dictionary, open that to get a TypeRep, and then peer inside that to check a Fingerprint. That is, I'd get lighter-weight exceptions that only carry the information I actually need.

Thanks, David

_______________________________________________ Haskell-Cafe mailing list To (un)subscribe, modify options or view archives go to: http://mail.haskell.org/cgi-bin/mailman/listinfo/haskell-cafe Only members subscribed via the mailman list are allowed to post.

That's a really interesting idea, thanks for linking it. On Tue, Feb 20, 2018 at 9:43 PM, Joachim Breitner wrote:

Hi,

Am Dienstag, den 20.02.2018, 17:56 -0500 schrieb Gregory Collins:

...

You mean like this? https://github.com/gregorycollins/hashtables/ blob/master/src/Data/HashTable/Internal/UnsafeTricks.hs#L72

My experience is that this works, except when doing coverage, where the compiler adds instrumentation code that breaks the technique. I #ifdef'd my way around the problem, swapping out a slower impl when I was doing code coverage.

it seems that in this case, https://ghc.haskell.org/trac/ghc/ticket/14826 would very much apply, wouldn’t it?

Cheers, Joachim

-- Joachim Breitner mail@joachim-breitner.de http://www.joachim-breitner.de/

_______________________________________________ Haskell-Cafe mailing list To (un)subscribe, modify options or view archives go to: http://mail.haskell.org/cgi-bin/mailman/listinfo/haskell-cafe Only members subscribed via the mailman list are allowed to post.

-- Gregory Collins