briand | 16 Jun 22:03 2013

opengl type confusion

This,

wireframe :: Double -> Double -> Double -> IO ()
wireframe wx wy wz = do 
  -- yz plane
  renderPrimitive LineLoop $ do
                       vertex $ Vertex3 0.0 0.0 0.0
                       vertex $ Vertex3 0.0 wy 0.0
                       vertex $ Vertex3 0.0 wy wz
                       vertex $ Vertex3 0.0 0.0 wz

produces this:

    No instance for (VertexComponent Double)
      arising from a use of `vertex'
    Possible fix:
      add an instance declaration for (VertexComponent Double)
    In the expression: vertex
    In a stmt of a 'do' block: vertex $ Vertex3 0.0 wy 0.0
    In the second argument of `($)', namely
      `do { vertex $ Vertex3 0.0 0.0 0.0;
            vertex $ Vertex3 0.0 wy 0.0;
            vertex $ Vertex3 0.0 wy wz;
            vertex $ Vertex3 0.0 0.0 wz }'

and thusly this :-(

Changing the declaration to GLdouble -> GLdouble -> GLdouble -> IO() and using
(0.0::GLdouble) fixes it, and I'm not clear on why it's not automagic.  There are many times I see the
compiler doing type conversion an numerican arguments although sometimes the occasional
(Continue reading)

Brandon Allbery | 16 Jun 22:15 2013
Picon

Re: opengl type confusion

On Sun, Jun 16, 2013 at 4:03 PM, <briand <at> aracnet.com> wrote:
Changing the declaration to GLdouble -> GLdouble -> GLdouble -> IO() and using
(0.0::GLdouble) fixes it, and I'm not clear on why it's not automagic.  There are many times I see the

Haskell never "automagic"s types in that context; if it expects GLdouble, it expects GLdouble. Pretending it's Double will not work. It "would" in the specific case that GLdouble were actually a type synonym for Double; however, for performance reasons it is not. Haskell Double is not directly usable from the C-based API used by OpenGL, so GLdouble is a type synonym for CDouble which is.

compiler doing type conversion an numerican arguments although sometimes the occasional fracSomethingIntegralorOther is required.

I presume the reason the type specification for numeric literals is because there is no defaulting (and probably can't be without introducing other strange type issues) for GLdouble.

In any case, the very fact that you refer to "automagic" and "type conversion" indicates that you don't really have an understanding of how Haskell's numeric types work; this will lead you into not only this kind of confusion, but worse problems later. In particular, you're going to get into dreadful messes where you expect Haskell to transparently deal with strange combinations of numeric types as if Haskell were (almost-typeless) Perl or something, and you'll have real trouble getting that code to work until you sit down and figure out how strong typing and Haskell's numeric typeclasses interact.

--
brandon s allbery kf8nh                               sine nomine associates
allbery.b <at> gmail.com                                  ballbery <at> sinenomine.net
unix, openafs, kerberos, infrastructure, xmonad        http://sinenomine.net
_______________________________________________
Haskell-Cafe mailing list
Haskell-Cafe <at> haskell.org
http://www.haskell.org/mailman/listinfo/haskell-cafe
briand | 16 Jun 22:42 2013

Re: opengl type confusion

On Sun, 16 Jun 2013 16:15:25 -0400
Brandon Allbery <allbery.b <at> gmail.com> wrote:

> On Sun, Jun 16, 2013 at 4:03 PM, <briand <at> aracnet.com> wrote:
> 
> > Changing the declaration to GLdouble -> GLdouble -> GLdouble -> IO() and
> > using
> > (0.0::GLdouble) fixes it, and I'm not clear on why it's not automagic.
> >  There are many times I see the
> 
> 
> Haskell never "automagic"s types in that context; if it expects GLdouble,
> it expects GLdouble. Pretending it's Double will not work. It "would" in
> the specific case that GLdouble were actually a type synonym for Double;
> however, for performance reasons it is not. Haskell Double is not directly
> usable from the C-based API used by OpenGL, so GLdouble is a type synonym
> for CDouble which is.
> 
> compiler doing type conversion an numerican arguments although sometimes
> > the occasional fracSomethingIntegralorOther is required.
> >
> 
> I presume the reason the type specification for numeric literals is because
> there is no defaulting (and probably can't be without introducing other
> strange type issues) for GLdouble.
> 

What I was thinking about, using a very poor choice of words, was this :

*Main> let a = 1
*Main> :t a
a :: Integer
*Main> let a = 1::Double
*Main> a
1.0
*Main> :t a
a :: Double
*Main> 

so normally 1 would be interpreted as an int, but if I declare 'a' a Double then it gets "promoted" to a Double
without me having to call a conversion routine explicitly.

That seems automagic to me.

(0.0::GLdouble) works to make the compiler happy.  So it appears to be taking care of the conversion automagically.

So maybe a better question, I hope, is:

How can I simply declare 0.0 to be (0.0::GLdouble) and have the functional call work.  Doesn't a conversion
have to be happening, i.e. shouldn't I really have to do (realToFrac 0.0) ?

Brian
L Corbijn | 16 Jun 23:10 2013
Picon

Re: opengl type confusion




On Sun, Jun 16, 2013 at 10:42 PM, <briand <at> aracnet.com> wrote:
On Sun, 16 Jun 2013 16:15:25 -0400
Brandon Allbery <allbery.b <at> gmail.com> wrote:

> On Sun, Jun 16, 2013 at 4:03 PM, <briand <at> aracnet.com> wrote:
>
> > Changing the declaration to GLdouble -> GLdouble -> GLdouble -> IO() and
> > using
> > (0.0::GLdouble) fixes it, and I'm not clear on why it's not automagic.
> >  There are many times I see the
>
>
> Haskell never "automagic"s types in that context; if it expects GLdouble,
> it expects GLdouble. Pretending it's Double will not work. It "would" in
> the specific case that GLdouble were actually a type synonym for Double;
> however, for performance reasons it is not. Haskell Double is not directly
> usable from the C-based API used by OpenGL, so GLdouble is a type synonym
> for CDouble which is.
>
> compiler doing type conversion an numerican arguments although sometimes
> > the occasional fracSomethingIntegralorOther is required.
> >
>
> I presume the reason the type specification for numeric literals is because
> there is no defaulting (and probably can't be without introducing other
> strange type issues) for GLdouble.
>

What I was thinking about, using a very poor choice of words, was this :


*Main> let a = 1
*Main> :t a
a :: Integer
*Main> let a = 1::Double
*Main> a
1.0
*Main> :t a
a :: Double
*Main>

so normally 1 would be interpreted as an int, but if I declare 'a' a Double then it gets "promoted" to a Double without me having to call a conversion routine explicitly.

That seems automagic to me.

(0.0::GLdouble) works to make the compiler happy.  So it appears to be taking care of the conversion automagically.

So maybe a better question, I hope, is:

How can I simply declare 0.0 to be (0.0::GLdouble) and have the functional call work.  Doesn't a conversion have to be happening, i.e. shouldn't I really have to do (realToFrac 0.0) ?

Brian


_______________________________________________
Haskell-Cafe mailing list
Haskell-Cafe <at> haskell.org
http://www.haskell.org/mailman/listinfo/haskell-cafe

_______________________________________________
Haskell-Cafe mailing list
Haskell-Cafe <at> haskell.org
http://www.haskell.org/mailman/listinfo/haskell-cafe
Brandon Allbery | 16 Jun 23:58 2013
Picon

Re: opengl type confusion

On Sun, Jun 16, 2013 at 4:42 PM, <briand <at> aracnet.com> wrote:
On Sun, 16 Jun 2013 16:15:25 -0400
Brandon Allbery <allbery.b <at> gmail.com> wrote:
> On Sun, Jun 16, 2013 at 4:03 PM, <briand <at> aracnet.com> wrote:
> > Changing the declaration to GLdouble -> GLdouble -> GLdouble -> IO() and
> > using
> > (0.0::GLdouble) fixes it, and I'm not clear on why it's not automagic.
> >  There are many times I see the
>
> I presume the reason the type specification for numeric literals is because
> there is no defaulting (and probably can't be without introducing other
> strange type issues) for GLdouble.

What I was thinking about, using a very poor choice of words, was this :

*Main> let a = 1
*Main> :t a
a :: Integer
*Main> let a = 1::Double
*Main> a
1.0
*Main> :t a
a :: Double
*Main>

so normally 1 would be interpreted as an int, but if I declare 'a' a Double then it gets "promoted" to a Double without me having to call a conversion routine explicitly.

That seems automagic to me.

No magic involved, although some automation is. Take a look at the `default` keyword in the Haskell Report (this is the "defaulting" I mentioned earlier).


The "default `default`" is `default (Integer, Double)` which means that it will try to resolve a numeric literal as type Integer, and if it gets a type error it will try again with type Double.

You should use this same mechanism to make numeric literals work with OpenGL code: neither Integer nor Double will produce a valid type for the expression, but at the same time the compiler cannot infer a type because there are two possibilities (GLfloat and GLdouble). You could therefore add a declaration `default (Integer, Double, GLdouble)` so that it will try GLdouble to resolve numeric literals when neither Integer nor Double will work.

> How can I simply declare 0.0 to be (0.0::GLdouble) and have the functional call work.  Doesn't a conversion have to be happening, i.e. shouldn't I really have to do (realToFrac 0.0) ?

The first part I just answered. As to the second, a conversion *is* happening, implicitly as defined by the language; the question being, to what type. A numeric literal has type (Num a => a), implemented by inserting a call to `fromIntegral` for literals without decimal points and `fromRational` for others. But the compiler can't always work out what `a` is in (Num a => a) without some help (the aforementioned `default` declaration).

--
brandon s allbery kf8nh                               sine nomine associates
allbery.b <at> gmail.com                                  ballbery <at> sinenomine.net
unix, openafs, kerberos, infrastructure, xmonad        http://sinenomine.net
_______________________________________________
Haskell-Cafe mailing list
Haskell-Cafe <at> haskell.org
http://www.haskell.org/mailman/listinfo/haskell-cafe
L Corbijn | 17 Jun 00:22 2013
Picon

Re: opengl type confusion

I seem to making a mess of it, first accidentally posting an empty message and then forgetting to reply to the list. Thirdly I forgot to mention that my message only describes the 'GHCi magic'.

Lars

P.S. Conclusion, I shouldn't write complicated email this late on the evening.

---------- Forwarded message ----------
From: L Corbijn <aspergesoepje <at> gmail.com>
Date: Mon, Jun 17, 2013 at 12:07 AM
Subject: Re: [Haskell-cafe] opengl type confusion
To: briand <at> aracnet.com


On Sun, Jun 16, 2013 at 11:10 PM, L Corbijn <aspergesoepje <at> gmail.com> wrote:



On Sun, Jun 16, 2013 at 10:42 PM, <briand <at> aracnet.com> wrote:
On Sun, 16 Jun 2013 16:15:25 -0400
Brandon Allbery <allbery.b <at> gmail.com> wrote:

> On Sun, Jun 16, 2013 at 4:03 PM, <briand <at> aracnet.com> wrote:
>
> > Changing the declaration to GLdouble -> GLdouble -> GLdouble -> IO() and
> > using
> > (0.0::GLdouble) fixes it, and I'm not clear on why it's not automagic.
> >  There are many times I see the
>
>
> Haskell never "automagic"s types in that context; if it expects GLdouble,
> it expects GLdouble. Pretending it's Double will not work. It "would" in
> the specific case that GLdouble were actually a type synonym for Double;
> however, for performance reasons it is not. Haskell Double is not directly
> usable from the C-based API used by OpenGL, so GLdouble is a type synonym
> for CDouble which is.
>
> compiler doing type conversion an numerican arguments although sometimes
> > the occasional fracSomethingIntegralorOther is required.
> >
>
> I presume the reason the type specification for numeric literals is because
> there is no defaulting (and probably can't be without introducing other
> strange type issues) for GLdouble.
>

What I was thinking about, using a very poor choice of words, was this :


*Main> let a = 1
*Main> :t a
a :: Integer
*Main> let a = 1::Double
*Main> a
1.0
*Main> :t a
a :: Double
*Main>

so normally 1 would be interpreted as an int, but if I declare 'a' a Double then it gets "promoted" to a Double without me having to call a conversion routine explicitly.

That seems automagic to me.

(0.0::GLdouble) works to make the compiler happy.  So it appears to be taking care of the conversion automagically.

So maybe a better question, I hope, is:

How can I simply declare 0.0 to be (0.0::GLdouble) and have the functional call work.  Doesn't a conversion have to be happening, i.e. shouldn't I really have to do (realToFrac 0.0) ?

Brian


_______________________________________________
Haskell-Cafe mailing list
Haskell-Cafe <at> haskell.org
http://www.haskell.org/mailman/listinfo/haskell-cafe


Oops sorry for the empty reply, I accidentally hit the sent button.

What you are seeing is the defaulting (see http://www.haskell.org/onlinereport/haskell2010/haskellch4.html#x10-790004.3.4). Which roughly speaking means that if you need a specific instance of a number first try Integer then Double and as a last resort fail.

Prelude> :t 1
1 :: Num a => a
Prelude> :t 1.0
1.0 :: Fractional a => a

So normally a number can be just any instance of the Num class, and any number with a decimal can be any Fractional instance. And now with let bindings


The need for defaulting is caused by the monomorphism restriction (http://www.haskell.org/haskellwiki/Monomorphism_restriction), which states that let binds should be monomorphic, or roughly speaking it should contain no type variables (unless of course you provide a type signature).

Prelude> let b = 1
Prelude> :t b
b :: Integer

Prelude> let c = 1.0
Prelude> :t c
c :: Double

So here you see the result of the combination. The monomorphism restriction doesn't allow 'Num a => a' as type for 'b'. So the defaulting kicks in and finds that its first guess 'Integer' fits. Therefore 'b'  gets type Integer. Though for 'c' the guess 'Integer' fails as it isn't a Fractional. Its second guess, Double, is a fractional so 'c' gets type Double.

You can see that the monomorphism restriction is to blame by disabling it

Prelude> :set -XNoMonomorphismRestriction
Prelude> let b = 1
Prelude> :t b
b :: Num a => a

But you shouldn't normally need to do this, as you can provide a specific type signature.

(in a fresh GHCi session)
Prelude> let {b :: Num a => a; b = 1}
Prelude> :t b
b :: Num a => a



On Sun, Jun 16, 2013 at 10:42 PM, <briand <at> aracnet.com> wrote:
On Sun, 16 Jun 2013 16:15:25 -0400
Brandon Allbery <allbery.b <at> gmail.com> wrote:

> On Sun, Jun 16, 2013 at 4:03 PM, <briand <at> aracnet.com> wrote:
>
> > Changing the declaration to GLdouble -> GLdouble -> GLdouble -> IO() and
> > using
> > (0.0::GLdouble) fixes it, and I'm not clear on why it's not automagic.
> >  There are many times I see the
>
>
> Haskell never "automagic"s types in that context; if it expects GLdouble,
> it expects GLdouble. Pretending it's Double will not work. It "would" in
> the specific case that GLdouble were actually a type synonym for Double;
> however, for performance reasons it is not. Haskell Double is not directly
> usable from the C-based API used by OpenGL, so GLdouble is a type synonym
> for CDouble which is.
>
> compiler doing type conversion an numerican arguments although sometimes
> > the occasional fracSomethingIntegralorOther is required.
> >
>
> I presume the reason the type specification for numeric literals is because
> there is no defaulting (and probably can't be without introducing other
> strange type issues) for GLdouble.
>

What I was thinking about, using a very poor choice of words, was this :


*Main> let a = 1
*Main> :t a
a :: Integer
*Main> let a = 1::Double
*Main> a
1.0
*Main> :t a
a :: Double
*Main>

so normally 1 would be interpreted as an int, but if I declare 'a' a Double then it gets "promoted" to a Double without me having to call a conversion routine explicitly.

That seems automagic to me.

(0.0::GLdouble) works to make the compiler happy.  So it appears to be taking care of the conversion automagically.

So maybe a better question, I hope, is:

How can I simply declare 0.0 to be (0.0::GLdouble) and have the functional call work.  Doesn't a conversion have to be happening, i.e. shouldn't I really have to do (realToFrac 0.0) ?

Brian


_______________________________________________
Haskell-Cafe mailing list
Haskell-Cafe <at> haskell.org
http://www.haskell.org/mailman/listinfo/haskell-cafe

_______________________________________________
Haskell-Cafe mailing list
Haskell-Cafe <at> haskell.org
http://www.haskell.org/mailman/listinfo/haskell-cafe
Tom Ellis | 16 Jun 23:19 2013
Picon

Re: opengl type confusion

On Sun, Jun 16, 2013 at 01:03:48PM -0700, briand <at> aracnet.com wrote:
> wireframe :: Double -> Double -> Double -> IO ()
> wireframe wx wy wz = do 
>   -- yz plane
>   renderPrimitive LineLoop $ do
>                        vertex $ Vertex3 0.0 0.0 0.0
>                        vertex $ Vertex3 0.0 wy 0.0
>                        vertex $ Vertex3 0.0 wy wz
>                        vertex $ Vertex3 0.0 0.0 wz
[...]
> 
>     No instance for (VertexComponent Double)
>       arising from a use of `vertex'
[...]
> 
> Changing the declaration to GLdouble -> GLdouble -> GLdouble -> IO() and using
> (0.0::GLdouble) fixes it

Vertex3 takes three arguments, all of which must be of the same instance of
VertexComponent.  Specifying GLdoubles in the signature of wireframe
specifies the types in the last three calls to Vertex3, but (0.0 ::
GLdouble) is still requried on the first to fix the type there.  How else
could the compiler know that you mean 0.0 to be a GLdouble and not a
GLfloat?

Tom
briand | 17 Jun 02:22 2013

Re: opengl type confusion

On Sun, 16 Jun 2013 22:19:22 +0100
Tom Ellis <tom-lists-haskell-cafe-2013 <at> jaguarpaw.co.uk> wrote:

> On Sun, Jun 16, 2013 at 01:03:48PM -0700, briand <at> aracnet.com wrote:
> > wireframe :: Double -> Double -> Double -> IO ()
> > wireframe wx wy wz = do 
> >   -- yz plane
> >   renderPrimitive LineLoop $ do
> >                        vertex $ Vertex3 0.0 0.0 0.0
> >                        vertex $ Vertex3 0.0 wy 0.0
> >                        vertex $ Vertex3 0.0 wy wz
> >                        vertex $ Vertex3 0.0 0.0 wz
> [...]
> > 
> >     No instance for (VertexComponent Double)
> >       arising from a use of `vertex'
> [...]
> > 
> > Changing the declaration to GLdouble -> GLdouble -> GLdouble -> IO() and using
> > (0.0::GLdouble) fixes it
> 
> Vertex3 takes three arguments, all of which must be of the same instance of
> VertexComponent.  Specifying GLdoubles in the signature of wireframe
> specifies the types in the last three calls to Vertex3, but (0.0 ::
> GLdouble) is still requried on the first to fix the type there.  How else
> could the compiler know that you mean 0.0 to be a GLdouble and not a
> GLfloat?
> 
> Tom
> 

it's curious that 

(0.0::GLdouble) 0.0 0.0 

is good enough and that 

(0.0::GLdouble) (0.0::GLdouble) (0.0::GLdouble)

is not required.  I suspect that's because, as you point out, they all have to be the same argument and ghc is
being smart and saying if the first arg _must_ be GLdouble (because I'm explicitly forcing the type), then
the rest must be too.

Meanwhile 4.3.4 about the default is quite interesting. Didn't know about that :-)

Thanks very much for the responses !

Brian
Tom Ellis | 17 Jun 09:05 2013
Picon

Re: opengl type confusion

On Sun, Jun 16, 2013 at 05:22:59PM -0700, briand <at> aracnet.com wrote:
> > Vertex3 takes three arguments, all of which must be of the same instance of
> > VertexComponent.  Specifying GLdoubles in the signature of wireframe
> > specifies the types in the last three calls to Vertex3, but (0.0 ::
> > GLdouble) is still requried on the first to fix the type there.  How else
> > could the compiler know that you mean 0.0 to be a GLdouble and not a
> > GLfloat?
> 
> it's curious that 
> 
> (0.0::GLdouble) 0.0 0.0 
> 
> is good enough and that 
> 
> (0.0::GLdouble) (0.0::GLdouble) (0.0::GLdouble)
> 
> is not required.  I suspect that's because, as you point out, they all
> have to be the same argument and ghc is being smart and saying if the
> first arg _must_ be GLdouble (because I'm explicitly forcing the type),
> then the rest must be too.

That is exactly the reason.

Tom

Gmane