[Dri-devel] Re: The Z-buffer (was)Re: [Bug #100500] Yet another Performer model (office.unc) strange artifacts

[Brian P. <br...@pr...>]

Brynn Rogers wrote:
> 
> no...@so... wrote:
> 
> >
> > Date: 01/16/00 10:01
> > By: brianp
> >
> > Comment:
> > The problems with office.unc is Z fighting due to poor positioning of the near and far clipping planes.  In perfly, click on "FOV panel", then drag the "Near Clip" slider to the right to about 100.0 or until the problem goes away.
> >
> > The DRI_ALWAYS_INDIRECT problem as fixed, as reported in a
> > different bug.
> 
> Brian,   I built from CVS today (1-16) and I still see problems with this that are not a clip plane thing.
> I run 'perfly office.unc', clicked FOV, and sliding the NEAR CLIP does not help show the shelves.
> I can run the slider out till the room disappears, and no shelves.
> HOWEVER, if I take the FAR CLIP slider and run it forward just a bit (one 'notch') the back wall of the office
> disappears, and the shelves appear.   It is some kind of z-buffer thing, thinking the shelves are behind the wall (but spinning the model around you can't see anything behind the wall.  Only when I moved the far clip in did I see it.
> And yes, now DRI_ALWAY_INDIRECT works fine [well, Sloooow] and this problem appears the same with the voodoo3 on or off [except for being sloooooower]
> 
> Even the perfly  "IRIS PERFORMER" splash screen has got some problems for me now.  If I had to describe it it looks like the 3D text is inside out or something.
> 
> As a sanity check, this morning (before my Vikings ugly performance) I renamed my xc and got an entire new checkout of xc (using cvs ... update -A xc).   I do a 'make World >& world.log' ; check the log for errors [none]; and su root; make install.   I
> usually check to see if the tdfx.o driver changed, and build and install it by hand if it did.
> The result of this looks exactly the same as what I was getting last night.
> 
> There are z-buffer problems like this all over the models now (you see the cows feet through the body, ...) that weren't there before darylls big code merge.
> 
> After I saw an interaction between perfly and Enlightenment, I today tried running X with no window manager, and started perfly from a telnet session (with DISPLAY set right of course).   This didn't run.  just now I retested (after killing fvwm) to give
> the full error message and it worked fine.
> 
> To see the Z-buffer thing I am talking about easily, do 'perfly cow.obj' and no matter what angle you view it from, the tail is always drawn in front of any other part of the cow.   I almost said 'tail is always drawn first' but it would actually be drawn
> last, right?


Ah, I was testing office.unc on my dispatch-0-0-1 branch.  When I
switched
back to the main branch I found the same thing.  If you run "perfly -n 5
office.unc" do you see messages about the depth buffer size being 0?
It appears that performer is not getting a depth buffer so depth testing
isn't working.

After asking Daryll about this yesterday (he doesn't see it) I
figured it was something funny in my local build.  Looks like I'm
not the only one.

I may not have time to track this down for a day or two.

Anyone else seeing this new problem?

-Brian