programatically generating methods

What is the best way to programatically generate a CompiledMethod from  
a string?

This is what I currently do:

        mthd := Compiler new
                compile:(String streamContents: [:stream |
                        stream
                                nextPutAll: 'bindingsString'; cr;
                                tab; nextPutAll: 'Dynamically generated by bindingsString:'  
asSmalltalkComment; cr;
                                tab; nextPutAll: '^ '; nextPutAll: aString  
surroundedBySingleQuotes; nextPut: $.; cr.
                         ])
                in:self
                classified:'bindings'
                notifying:nil
                ifFail:[:x | ^ x].
        self class addSelector:#bindingsString withMethod: mthd generate.
        self class organization classify: #bindingsString under:#bindings  
suppressIfDefault:false.

This works, actually, though I'm sure the "ifFail" block is totally  
wrong.  However, it doesn't "feel" like the right way.  For one thing,  
I am not sure that method of Compiler is a public API.  For another, I  
have to repeat a lot of information such as the selector and  
classification.

This is in relation to my last message: I am trying to generate  
methods that will return constant data so that it survives  
Monticello.  (I also tried hacking around with MCClassDefinition, but  
that seemed like a lot of work...).  The constant data is only a  
string, so if there is an easier way I'm all ears.


thanks!

Niko

Actually, there is one additional problem with this technique:

        the comment which I insert gets lost.



thanks!
Niko

On Dec 7, 2007, at 11:28 AM, Niko Matsakis wrote:

Hi Niko,

2007/12/7, Niko Matsakis <[hidden email]>:
> What is the best way to programatically generate a CompiledMethod from
> a string?

I'm not sure you want to play with CompiledMethod. What about:

Object  compile: 'method
        ^ 123.' classified: 'accessing'

--
Damien Cassou

That looks easy.  I'll give it a try, thanks!


Niko

On Dec 7, 2007, at 11:39 AM, Damien Cassou wrote:

... that's Computational Fluid Dynamics.  I'm specifically interested
in running a Navier-Stokes simulation using something like the Volume
of Fluid model to deal with free surface movement.  My first goal is
to be able to duplicate something like the following simulation of a
breaking wave:

        http://www.coastal.udel.edu/faculty/jpuleo/RIPPLE/ripple.mov

I want to be able to run similar simulations for waves breaking
against different shaped walls.

Most people use big programs in C or Fortran (!), and half the
trouble is learning how to use them, since there's no prayer of
changing them. However the basic equations can be written in half a
page, and I don't need it to run very fast.

It could be fun but so far it looks like it will be simpler to build
a wave tank :-(.

Thanks for any useful pointers.

        - Dan

I don't know of any Squeak code to do this.  The starting-point paper about this kind of simulation (for computer graphics, anyway) is "Stable Fluids" by Jos Stam (http://www.dgp.toronto.edu/people/stam/reality/Research/pdf/ns.pdf).  I'm not sure if there is a later paper that is more approachable; the others I've seen are more complicated.  Also note that I don't believe that Stam's method can model breaking waves... it doesn't allow for a dynamic boundary between different fluids (eg: air and water).  

Dan Ingalls wrote:

> It could be fun but so far it looks like it will be simpler to build
> a wave tank :-(.

That is very likely.
 
> Thanks for any useful pointers.

No actual pointers, sorry, but I do note most of the complexity of the
available software is in getting the data in and then getting the
results out. This is something that perhaps Croquet could help with? For
the actual equations something like StarSqueak or Kedama might be a good
match, specially if extended for 3D.

-- Jecel

>... that's Computational Fluid Dynamics.  I'm specifically interested in running a Navier-Stokes simulation using something like the Volume of Fluid model to deal with free surface movement.  My first goal is to be able to duplicate something like the following simulation of a breaking wave:
>
> http://www.coastal.udel.edu/faculty/jpuleo/RIPPLE/ripple.mov
>
>I want to be able to run similar simulations for waves breaking against different shaped walls.
>
>Most people use big programs in C or Fortran (!), and half the trouble is learning how to use them, since there's no prayer of changing them. However the basic equations can be written in half a page, and I don't need it to run very fast.
>
>It could be fun but so far it looks like it will be simpler to build a wave tank :-(.

I'm getting more serious about this.  As a point of reference I did build a wave tank...

        http://www.youtube.com/watch?v=joRv6aiST3s

It's fun to build stuff (and I now have a 300-pound structure in my driveway waiting for the next time when the tide is right), but I would still like to be able to do some simulations, even if only in 2D at first.

I have a feeling that if one understands the basic equations, plus the volume-of-fluid method for the surface and other boundaries (as in http://www.math.rug.nl/~veldman/preprints/ECCOMAS2004.pdf), it wouldn't take more than a couple of pages of Squeak, and it might even fit the StarSqueak or Kedama framework as suggested by Jecel.

Sadly I have only personal funds to work with at this point  but, if someone thought they could duplicate, eg, the little ripple.mov above in a few pages of Squeak, I'd pay real money for it, and maybe more for future experiments.  Let me know if this appeals to you and perhaps we can negotiate some sort of agreement.

        - Dan

Dan-

Cool wave tank video.

This existing Java applet is maybe not as fancy as you want eventually, but
check out this link for a ripple tank simulator for what is easily possible
if you are willing to just have a non-particle meshwork ripple model at first:
  http://www.falstad.com/ripple/
"This java applet is a simulation of a ripple tank. It demonstrates waves in
two dimensions, including such wave phenomena as interference, diffraction
(single slit, double slit, etc.), refraction, resonance, phased arrays, and
the Doppler effect. To get started with the applet, just go through the
items in the Setup menu in the upper right. You can also draw on the screen
with the mouse. The predefined setups are just starting points; you can
modify the sources and walls as you desire. Click the 3-D View checkbox to
see a 3-D view. "

The code is there too in Java, but only for two dimensions (shore waves do
some funky circular particle rolling thing I think, if you want to be really
accurate).
  http://en.wikipedia.org/wiki/Ocean_surface_wave
(that links to a simpler wave applet, by the way).

It should not be that hard for someone like you to translate that "ripple"
WaveBox applet to Smalltalk. Only about 1700 lines:
  ~/downloaded_files$ wc WaveBox.java
   1693  5881 48302 WaveBox.java
It's a surprisingly complex applet as far as the GUI, so way more than half
of the code seems related to the GUI and the drawing. From a casual glance
at it, I'm guessing the core part is, as you say, only a few pages of code,
mostly towards the end. There is no license statement in the file though. :-(

>From the start of the source file:
// WaveBox.java (c) 2001 by Paul Falstad, www.falstad.com.
// Rendering algorithm in this applet is based on the description of
// the algorithm used in Atom in a Box by Dean Dauger (www.dauger.com).
// We raytrace through a 3-d dataset, sampling a number of points and
// integrating over them using Simpson's rule.
(So, not much on the wave part, and no other detailed comments in the file).

Anyway, this sort of numerically oriented wave library is exactly the kind
of thing you might want to code in Smalltalk and then translate to Scala to
run on the JVM. It would only go 1/2 to 1/3 of native C or Fortran speeds
probably, but it would be run fast everywhere Squeak/JVM would run. Or
course, having the JVM set up right is a hassle.  Or, of course, you could
write is such that Slang could translate part of it to C. But running that
code everywhere might be harder.

This example makes me think it might be nice to have a Java->Smalltalk
translator as well as a Smalltalk->Java/Scala one as part of a Squeak/JVM,
because there sure is a lot of Java code out there of educational interest.
(Too bad Jim Spohrer's EOE educaitonal Java applet directory site seems not
to be up these days, last I looked.)

--Paul Fernhout

Dan Ingalls wrote:

Dan Ingalls schrieb: During my master thesis, I have experimented a bit with fluid
visualization using Marching Cubes algorithm in squeak.
http://www.gaguls.de/danilo/downloads/PlayfulAnimation_low.pdf   (page 51)

The simulation where particle based, but I get more and more interested
to do fluid simulation using Navier Strokes algorithm. My personal
interest is visualization and the interest in how it works.
I simply did not found the time yet. I can inform you, when I have time
for it. Maybe we find a way to cooperate.

Danilo

On Feb 13, 2008, at 1:50 AM, Danilo wrote:
Neat paper Danilo!  Thanks for the link.

How was the VerletFluids app implemented?  Does it use a VM plugin for  
speed, for either the particle animation or the isosurface  
extraction?  Or is Squeak fast enough?

Thanks,
Josh

Joshua Gargus wrote: The particle simulation and isosurface extraction were pure smalltalk,
no special plugins. It worked only interactive for small simulations.
The screenshot in the paper was from an interactive simulation, that
used 16 particles and a grit for surface extraction of about 15x15x15 cells.

Danilo