Celestia and the Eigen library

[duds26]

C++ also has a vector and an array class.

What a pretty it isn't usable like the Eigen lib is.
(Or otherwise there would be no point in the Eigen lib.)

Meh, it doesn't matter if it works.

[cartrite]

I'm wondering if this is right?
The first change was basically a copy of what Pat did to the GTK gui for stars.
The second left "getAstrocentricPosition" as is (before I changed it to getPosition) by this modification.

Code: Select all

Point3d bodyPos = ptFromEigen(body->getAstrocentricPosition(appSim->getTime()));

This also gets the kde gui to work on Suse11.1 using gcc 4.3.2.

patched.txt.zip

I'm thinking that this should work as it did before Eigen upgrade?
cartrite

[t00fri]

open Suse 11.1 ,Athlon64 3200+ 3gb ram, GeForce9800gt video card with 1gb ram

Hey Steve,

what do I see here? Looks like a pretty mighty system? Much faster than mine anyway...

Incidentally, how do you like SuSE 11.1 meanwhile? I am asking since my lab has just ordered a new machine for my office (Core2Duo 2.6 GHz + 4GB ram + same graphics card as you have).

Fridger

[cartrite]

Actually, it's the same old system with some new parts. I replaced the graphics card and hard drive. Then I installed Suse 11.1 on the new hard drive. I still want to get a new system board. I want to go with an intel board with a socket 1366 and a i7 chip. Don't got the money for that yet though. They are supposed to support 24 gb of ram. I want one of them so I can start working on a Mars texture and also finish the cloud texture I started last year.

The graphics card is being bottlenecked by my old system board. The card has ddr3 ram and the system board just has a ddr bus. The sata drive I got is the newer 3gbit/sec but the system board only has 1.5gbit/sec sata ports. So I got a jumper in the hard drive to limit it's speed. Can't wait till I can afford the new board.

I like Suse 11.1. It's been a lot easier then the last time I had it installed. I guess there have been a lot of updates to work out the bugs.

Not sure about the Core2Duo 2.6 GHz. As of now, I had only run AMD.
cartrite

[t00fri]

...
you might be able to set up regression tests before making the changes as follows;
1. Create a celURL at the scene you want to test.
2. Capture a PNG image of the scene.
The celURL together with the corresponding image forms the baseline test.

After making any code changes, go back to the celURL, re-capture a new image, then use something like Imagemagick to compare the 2 images. (You can script operations such as subtracting one image from the other in Imagemagick I think.)
You could set up a number of these regression tests and repeat them all after any code changes.

...
CC

CC,

FYI, in the context of testing the many Eigenized rotations involved in my galaxy code, I used an image based method that should be pretty sensitive wrto bugs in matrix operations... and it works well.

For the pre-Eigen SVN code, I stored the "triangle shaped" configuration of 3
galaxies M 65, M 66 and NGC 3628 under a FoV of 1 deg 0' 15.5" as bookmark and
made a fullscreen screenshot.
------------------
I noted in this context that under Windows, the FoV is NOT stored in the
bookmarks, which I think is a bug!
-------------------

Then I upgraded to today's Eigenized SVN rev 4789, recalled the bookmark under
precisely the same FoV and made another screenshot. Everything fullscreen.

Next, I loaded both screenshots into GIMP as aligned layers on top of each other and quickly moved the transparency between 100% and 0% and back again several times.

This way smallest modifications in the various orientations based on Matrix
rotations can be easily spotted.

In fact, I noted NO movements whatsoever over the whole images! This sounds comforting. This sort of test should be done with more galaxies, like the ones in the Coma cluster or notably with the famous Stephans quintet (NGC 7320 etc).

Fridger

[cartrite]

In rev 4792, I noticed what may be a bug in this Eigen code. It is present in the GTK, KDE, and the QT4 interfaces built with Linux. Many stars are not selectable. The svn version 4751, which is about where 1.6.0 was built from, all the stars are selectable.

An example can be seen with this cel url.

Code: Select all

cel://Follow/Sol:Earth/2009-07-21T23:09:04.30223?x=ABjM5UqhbP///////////w&y=GrWX9x8s&z=aD+An6mkI////////////w&ow=0.189758&ox=-0.00360401&oy=0.981795&oz=-0.00754101&select=PI%20Sgr&fov=29.1981&ts=1&ltd=0&p=0&rf=6027&lm=2056&tsrc=0&ver=3


If you check out this screen shot, a small irregular box surrounds one of the stars that can't be selected. There are many.

test.jpg

This cel url

Code: Select all

cel://Follow/PI%20Sgr/2009-07-21T23:11:08.64523?x=AABUfOrDVE3S/////////w&y=AEBa2OnPQOf7/////////w&z=AAD4ceWabzxj/////////w&ow=0.141371&ox=-0.00470799&oy=0.989866&oz=-0.0125264&select=PI%20Sgr&fov=29.1981&ts=1&ltd=0&p=0&rf=6027&lm=2056&tsrc=0&ver=3

shows another star in another place.
See image. Of the 2 stars in the box, the one on the right is selectable and the one on the left is not.

test2.jpg

Both of these examples are selectable with svn r4751. In fact every star that I clicked on was selected with r4751.
note. The GTK version is from an older svn version that was done after the Eigen updates to star code.
Actually, I just rebuilt the svn GTK version at r 4792 and got the same non selectable stars.
cartrite