World: r3wp

:-) yep...

Crashing on Win7 after printing

... (several lines omitted ..)
1362:polygon  : 1362
--------------------------------------------------
POLYGON PRIMITIVE
...
-->Select_Float_Array()
-->HT_Select()
(crash)

For those interested in a quick overview of the changes in the hostkit, 
here's a diff between CGR-A109 and RT-A109:
https://gist.github.com/9d6d0330b88d6d5419b1

I was out of town, back now... Christian, yes it seems there are 
still one or two compile things to figure out, I'm getting random 
crashes too.

oddly, in debug mode its 100% stable.

Andreas thanks for that.... 


that is exactly why I spent a day re-organizing my code base so that 
the host and the CGR system are now almost separate.  

its now very easy to see what little changes are required for CGRs 
to work.  


though a few tweaks in the host would make CGRs a bit faster and 
simple to use.

one note is that a few of the changes are in fact just added documentation 
in the host, which is scarce at best.


I think it would be nice that as people work on and study the host 
files, more documentation be added, if only to improve its understanding.


many of the view files use short variable names which is ok, but 
the sense of what and most especially *why* things are being done 
is currently very
obscure and I can see it being hard to maintain in the long run.


when you look at all the box region math, sometimes being done more 
than once on the same values, its very hard to positively identify 
what is being calculated and why it even needs to be done at that 
point in the various functions.


when I look at the code, there are many very similar ops being done, 
and it might be that some of that should be reshaped into more re-usable 
code.  maybe if only as multi-line macros, to keep it fast.

christian, if you can send me the full trace of the app it might 
help me track the problem...  either mail it to me or just dump it 
to me privately in altme.


I think its a problem with how VC is optimisizing things to be parralelized 
by SSD type instructions.


on my system its not crashing exactly at the same place but within 
the same source file.  seeing a pattern might help me corner it.

Maxim, the demo crashes for me right when it attempts to open a window..I 
tried it also with the torus.r3d dataset so there must be some other 
problem.

I've had that too, but changing the optimizing on Release mode, I 
have just about the same issue as Christian.


in debug mode I've *never* had a crash yet. even on very small or 
huge datasets.


so AFAICT its a dreary compiler issue, and the worst of it is that 
I can't debug it, cause it *only* crashes when I'm not in Debug.

at this point I've cornered it to the host-tools array code which 
is basically just a loop over the host-code.  so its possible that 
the way the core dll was compiled is part of the problem.


I'll try to compile the A107 version in Release mode.. maybe it was 
built using different switches.

Crashes for me too, under XP x64.

on window open or after a ->> / <-- trace line?

On window open.

Maxim, all demos run fine now.

thanks.  Andreas told me he was reaching 70fps using the simple tree 
demo running wine on linux!

closer to 2-5 fps here, but probably due to software OpenGL

(strange that it runs now, when I was warned before that OpenGL wouldn't 
work)

the heavy tree is quite hard... the torus should be real-time even 
using software mode.

also when first showing the window, the view area is only in about 
the upper left quarter of the window. a resize solves this.

I get between 5-10 fps on the torus. I suppose virtualbox doesn't 
do this stuff very well.

wrt resize, yes that is on purpose.  it allows me to test events 
within and without the opengl.

its possible wine allows some of the rendering to find its way to 
the gfx card.

ok

ok, next step: modeling dialect :-)

and yes it is strange that it works now and didn't work before!  
I didn't change any of OpenGL code, though I had a serious memory 
corruption issue, so its possible that on your system, that translated 
to some bogus system call.

it has been behaving strangely just yesterday (openGL issues) and 
today (network flakey), so I'm not even sure it will keep working. 
:-)

what is the .r3d format like?

hehe.  if you look at the demo files, you will see how to specify 
a polygon primitive... its dead simple... a list of vertex... and 
a list of faces (which are indices to the vertices)

look into the torus... its a rebol loadable data file.

the same as the commented out primitive in the tests.

It would be wonderful with an actual dialect:


view [my-torus: torus scale .2 rotate 45 material [specular: blah]]

yep that is how it will be... its already close to a dialect.  I 
just need to add more attributes.

btw notice that I included your last release of the r3_gui.r3 in 
the cgr-apps to make sure I stay forward complient... though you 
may want to do a diff on the version included and yours in order 
to patch crashing events when they originate from a gob which isn't 
a face.

ok, but I suspect many changes and additions are coming. I'm not 
sure if it will be a problem for you.

I'll also add a few default primitives like sphere, cube, teapot, 
etc.

how do I find fps for the Torus demo?

for me it won't change anyting as long as you can retrofit the event 
stuff I am "fixing" as I encounter them..

you count the number of refreshes in a second... printed in the console 
 ;-)

so, it simply renders into a GOB? can you make a demo where it's 
integrated between some styles?

pekr, on windows it shoudn't go beyond 60fps because of timing issues.

henrik yes.  the openGL CGR use the "container" rendering mode which 
allocates a windows control from scratch and adds it to the window.


then when you call show, it gets notified of the change in size and 
automatically updates the control and gl viewport & camera

notice that I do not call show on it in animation.  I call refresh, 
which bypasses view and makes it *much* faster.

henrik, I'll wait for the next release of r3-gui.  I wasn't able 
to launch the style browser.. it gave some error.

cool, I'd like to see a demo with styles, if possible. that will 
be a good indicator of real-world viability of the extension.

Maxim, I'll build a new r3-gui.r for you.

ok, I'll try it out tomorrow.. its 5AM here... got to go get some 
sleep :-)

if you can do one thing to help me.... make a style which doesn't 
allocate any of the gob type data.  (color, text, draw etc) so that 
when you print the gob its basically a none gob.

ok, cool

I use the none type gob as an indicator that it *might* be a cgr, 
this allows me to quickly check any face without it being to slow 
to retrieve the CGR internal data.