After having failed following the googled tutorial in HOpenGL programming, I thought I’d write down the steps I actually can get to work in a tutorial-like fashion. It may be a good idea to read this in parallell to the tutorial linked, since Panitz actually brings a lot of good explanations, even though his syntax isn’t up to speed with the latest HOpenGL at all points.
First of all, we’ll want to load the OpenGL libraries, throw up a window, and generally get to grips with what needs to be done to get a program running at all.
main = do
(progname, _) <- getArgsAndInitialize
createWindow "Hello World"
mainLoop
This code throws up a window, with a given title. Nothing more happens, though. This is the skeleton that we’ll be building on to. Save it to HelloWorld.hs and compile it by running
ghc -package GLUT HelloWorld.hs -o HelloWorld
However, as a skeleton, it is profoundly worthless. It doesn’t even redraw the window, so we should definitely make sure to have a function that takes care of that in there somewhere. Telling the OpenGL-system what to do is done by using state variables, and these, in turn are handled by the datatype Data.IORef. So we modify our code to the following
main = do
(progname, _) <- getArgsAndInitialize
createWindow "Hello World"
displayCallback $= clear [ ColorBuffer ]
mainLoop
This sets the global state variable carrying the callback function responsible for drawing the window to be the function that clears the color state. Save this to the HelloWorld.hs, recompile, and rerun. This program no longer carries the original pixels along, but rather clears everything out.
The displayCallback is a globally defined IORef, which can be accessed through a host of functions defined in Data.IORef. However, deep within the OpenGL code, there are a couple of definition providing the interface functions $= and get to fascilitate interactions with these. Thus we can do things like
So, we have a window, we have a display callback that clears the canvas. Don’t we want more out of it? Sure we do. So let’s draw some things.
myPoints :: [(GLfloat,GLfloat,GLfloat)]
myPoints = map (\k -> (sin(2*pi*k/12),cos(2*pi*k/12),0.0)) [1..12]
main = do
(progname, _) <- getArgsAndInitialize
createWindow "Hello World"
displayCallback $= display
mainLoop
display = do
clear [ColorBuffer]
renderPrimitive Points $ mapM_ (\(x, y, z)->vertex$Vertex3 x y z) myPoints
flush
Now, the important thing to notice in this codeextract is that last line. It starts a rendering definition, gives the type to be rendered, and then a sequence of function calls, each of which adds a vertex to the rendering canvas. The statement is basically equivalent to something along the lines of
for appropriate triples of coordinate values at the appropriate places. This results in the rendition here:

We can replace Points with other primitives, leading to the rendering of:

Each three coordinates following each other define a triangle. The last n mod 3 coordinates are ignored.
Keyword Triangles

Triangles are drawn according to a “moving window” of size three, so the two last coordinates in the previous triangle become the two first in the next triangle.
Keyword TriangleStrip

Triangle fans have the first given coordinate as a basepoint, and takes each pair of subsequent coordinates to define a triangle together with the first coordinate.
Keyword TriangleFan

Each pair of coordinates define a line.
Keyword Lines

With line loops, each further coordinate defines a line together with the last coordinate used. Once all coordinates are used up, an additional line is drawn back to the beginning.
Keyword LineLoop

Line strips are like line loops, only without the last link added.
Keyword LineStrip

For the Quads keyword, each four coordinates given define a quadrangle.
Keyword Quads

And a Quadstrip works as the trianglestrip, only the window is 4 coordinates wide and steps 2 steps each time, so each new pair of coordinates attaches a new quadrangle to the last edge of the last quadrangle.
Keyword QuadStrip

A Polygon is a filled line loop. Simple as that!
Keyword Polygon
There are more things we can do on our canvas than just spreading out coordinates. Within the command list constructed after a renderPrimitive, we can give several different commands that control what things are supposed to look like, so for instance we could use the following
in order to produce these four coloured squares

where each color command sets the color for the next item drawn, and the vertex commands give vertices for the four squares.
We have already seen one callback in action: displayCallBack. The Callbacks are state variables of the HOpenGL system, and are called in order to handle various things that may happen to the place the drawing canvas lives. For a first exercise, go resize the latest window you’ve used. Go on, do it now.
I bet it looked ugly, didn’t it?
This is because we have no code handling what to do if the window should suddenly change. Handling this is done in a callback, residing in the IORef reshapeCallback. Similarily, repainting is done in displayCallback, keyboard and mouse input is in keyboardMouseCallback, and so on. We can refer to the HOpenGL documentation for window callbacks and for global callbacks. Window callbacks are things like display, keyboard and mouse, and reshape. Global callbacks deal with timing issues (for those snazzy animations) and the menu interface systems.
In order for a callback to possibly not be defined, most are typed within the Maybe monad, so by setting the state variable to Nothing, a callback can be disabled. Thus, setting callbacks is done using the keyword Just. We’ll add a callback for reshaping the window to our neat code, changing the main function to
reshape s@(Size w h) = do
viewport $= (Position 0 0, s)
postRedisplay Nothing
Here, the code for the reshape function resizes the viewport so that our drawing area contains the entire new window. After setting the new viewport, it also tells the windowing system that something has happened to the window, and that therefore, the display function should be called.
So, in conclusion, so far we can display a window, post basic callbacks to get the windowhandling to run smoothly, and draw in our window. Next installment of the tutorial will bring you 3d drawing, keyboard and mouse interactions, the incredible power of matrices and the ability to rotate 3d objects for your leisure. Possibly, we’ll even look into animations.
14 People had this to say...
Thank you!
Keep more comming
Awesome. Like Leif said, keep it going.
I’ve been waiting for someone to fill the OpenGL tutorial void in haskell.
Also, I happen to think it’s important work.
OpenGL is everywhere and is an excellent example of a low level popular C library for haskell to interface with. The more use it, the better the OpenGL bindings get, the better haskell’s interaction with normal libraries get.
One point worth noting is that the added call to postRedisplay I used in the reshape callback is completely redundant. This wasn’t clear to me when writing, however, since I would do most of my testing over an ssh-tunneled X session. Thus updating was slow in general, and in particular resulted in redisplays not occuring as fast as they should.
A locally run test showed me that the behaviour asked actually does occur with reshape defined as
- September 15th, 2006 at 12:19
(x, y, z)->vertex$Vertex3 x y zdoes not compile for me. Maybe the intention was
\(x, y, z)->vertex$Vertex3 x y zer, in general your backslashes are disappearing, as in
map (k -> (sin(2*pi*k/12),cos(2*pi*k/12),0.0)) [1..12]Also, the keyword is Quads not Quad.
Thanks for the corrections. I think the case with the disappearing backslashes is to a certain extent the blogpost being “hung over” from problems I had when editing - where all of a sudden ALL “special” characters ended up escaped with \ all over the place; and at some point I just removed all backslashes (without thinking about the consequences).
I’ll incorporate your points in an edit now. Thanks.
[...] As we left off the last installment, we were just about capable to open up a window, and draw some basic things in it by giving coordinate lists to the command renderPrimitive. The programs we built suffered under a couple of very infringing and ugly restraints when we wrote them - for one, they weren’t really very modularized. The code would have been much clearer had we farmed out important subtasks on other modules. For another, we never even considered the fact that some manipulations would not necessarily be good to do on the entire picture. [...]
With all due respect, the code above is pretty ugly. Is there any real advantage to using Haskell for OpenGL programming?
First off, there are definitive ways to write the code prettier than I did - this tutorial is written almost as much for my own learning as it is for teaching purposes.
Secondly, I wouldn’t be able to pinpoint a good knock-out argument why OpenGL in Haskell would be a better idea than, say, OpenGL in C or OpenGL with Python bindings et.c. other than at most being able to manhandle lists and what-not with the Haskell built-in tools. You also will not get the most funky and modern stuff at the current state of HOpenGL, so I’d explicitly recommend against it for, say, games programming in general.
I remember OpenGL in C as being unwieldy and annoying when I coded it, and the way it’s encapsulated in Haskell to be neat in comparison, but this may be taken care of in other language bindings just as well.
Thanks for the tutorial
[...] [1] http://blog.mikael.johanssons.org/archive/2006/09/opengl-programming-in-haskell-a-tutorial-part-1/ [...]
Very cute!
I should take some time to play with this.
[...] best OpenGL tutorial for Haskell that I’ve found is this one from Michi’s blog, using GLUT to interface with X. For this tutorial we are going to use the Gtk GLDrawingArea [...]
Thanks a lot for this tutorial
This is exactly how a tutorial should be — begin by showing how to draw some frigging points rather than with 15 pages of boilerplate, side remarks and whatnot. I can usually figure out much of a library from the docs once I got the ball rolling like this, but getting there is a pain from a reference manual.
Minor point: As of now (GLUT-2.1.1.2) the resizing works just fine without the reshape callback.
Want your say?