I've forgotten what I intended in the STG paper, but GHC's Core language certainly allows case on a function; all it does is to force the function to head normal form. Simon | -----Original Message----- | From: ghc-devs [mailto:ghc-devs-bounces@haskell.org] On Behalf Of William | Knop | Sent: 18 June 2014 23:54 | To: ghc-devs@haskell.org | Subject: Re: Case expressions in STG | | Hi Tom, | | SPJ is surely more qualified to answer than I, but I'll take a stab. | | In general, it is computationally infeasible to compare functions. | Granted, in your example, the function isn't being compared-- and | therefore the case expr is extraneous. I don't think there exists a | feasible, uncontrived example, so I imagine that's why STG forbids it | (though I don't know where it's specified). | | Cheers, | Will | | | > On Jun 18, 2014, at 9:43 AM, Tom Ellis wrote: | > | > I am reading SPJ's seminal work "Implementing lazy functional languages | on | > stock hardware: the Spinless Tagless G-machine" (1992). The paper is | > available here | > | > http://citeseer.ist.psu.edu/viewdoc/summary?doi=10.1.1.53.3729 | > | > On p62 we see "The expression scrutinised by a case expression must | > eventually evaluate either to a primitive value or a constructor | > application.". | > | > This doesn't make sense to me. Isn't the following a valid STG | program? | > What am I missing? Where is it specified that the scrutinee of a case | > expression cannot be of a function type? | > | > {x} \n {z} -> let f = \{x} \n {y} -> g x y | > in case f of f' -> f' z | > | > Thanks, | > | > Tom | > _______________________________________________ | > ghc-devs mailing list | > ghc-devs@haskell.org | > http://www.haskell.org/mailman/listinfo/ghc-devs | _______________________________________________ | ghc-devs mailing list | ghc-devs@haskell.org | http://www.haskell.org/mailman/listinfo/ghc-devs

On Wed, Jun 18, 2014 at 07:40:26PM -0400, William Knop wrote:

f = \x -> x g = \x -> (x,1) h = \x -> fst (g x) i = \x -> case f of f -> True _ -> False

i f => True i h => ?

If g isn't inlined into h and fst optimized out, wouldn't the head normal form of f and h be different, and the comparison fail even though it shouldn't? Or should it? I've taken function equality to be "f and g are equal iff f x == g x, for all x."

You mean 'i = \x -> case x of' ... Anyway, I'm not suggesting using case to *compare* functions, simply force their thunk before proceeding with a single default alternative. Tom

| i = \x -> case f of | f -> True | _ -> False This case expression always succeeds with the first pattern. The second is entirely overlapped. Variable patterns always match. Simon | -----Original Message----- | From: William Knop [mailto:william.knop.nospam@gmail.com] | Sent: 19 June 2014 00:40 | To: Simon Peyton Jones | Cc: ghc-devs@haskell.org; tom-lists-ghc-devs@jaguarpaw.co.uk | Subject: Re: Case expressions in STG | | Whoops, looks like my phone dropped Tom from the CC (fixed). | | Simon, what about the following: | | f = \x -> x | g = \x -> (x,1) | h = \x -> fst (g x) | i = \x -> case f of | f -> True | _ -> False | | i f => True | i h => ? | | If g isn't inlined into h and fst optimized out, wouldn't the head | normal form of f and h be different, and the comparison fail even though | it shouldn't? Or should it? I've taken function equality to be "f and g | are equal iff f x == g x, for all x." | | Will | | | > On Jun 18, 2014, at 7:04 PM, Simon Peyton Jones | wrote: | > | > I've forgotten what I intended in the STG paper, but GHC's Core | language certainly allows case on a function; all it does is to force | the function to head normal form. | > | > Simon | > | > | -----Original Message----- | > | From: ghc-devs [mailto:ghc-devs-bounces@haskell.org] On Behalf Of | > | William Knop | > | Sent: 18 June 2014 23:54 | > | To: ghc-devs@haskell.org | > | Subject: Re: Case expressions in STG | > | | > | Hi Tom, | > | | > | SPJ is surely more qualified to answer than I, but I'll take a stab. | > | | > | In general, it is computationally infeasible to compare functions. | > | Granted, in your example, the function isn't being compared-- and | > | therefore the case expr is extraneous. I don't think there exists a | > | feasible, uncontrived example, so I imagine that's why STG forbids | > | it (though I don't know where it's specified). | > | | > | Cheers, | > | Will | > | | > | | > | > On Jun 18, 2014, at 9:43 AM, Tom Ellis | devs@jaguarpaw.co.uk> wrote: | > | > | > | > I am reading SPJ's seminal work "Implementing lazy functional | > | > languages | > | on | > | > stock hardware: the Spinless Tagless G-machine" (1992). The paper | > | > is available here | > | > | > | > http://citeseer.ist.psu.edu/viewdoc/summary?doi=10.1.1.53.3729 | > | > | > | > On p62 we see "The expression scrutinised by a case expression | > | > must eventually evaluate either to a primitive value or a | > | > constructor application.". | > | > | > | > This doesn't make sense to me. Isn't the following a valid STG | > | program? | > | > What am I missing? Where is it specified that the scrutinee of a | > | > case expression cannot be of a function type? | > | > | > | > {x} \n {z} -> let f = \{x} \n {y} -> g x y | > | > in case f of f' -> f' z | > | > | > | > Thanks, | > | > | > | > Tom | > | > _______________________________________________ | > | > ghc-devs mailing list | > | > ghc-devs@haskell.org | > | > http://www.haskell.org/mailman/listinfo/ghc-devs | > | _______________________________________________ | > | ghc-devs mailing list | > | ghc-devs@haskell.org | > | http://www.haskell.org/mailman/listinfo/ghc-devs