In an offline email exchange with jr, I drifted off into a discussion 
(OK, it was more of a monologue) about how the text{} object has a 
secrete rotation to avoid axis aligned numerical issues. The rotation is 
the equivalent of a user specifying - always - 'rotate 
<0.001,0.001,0.001>' on text{} objects.

It's something that was added to the parse code at some point. It's 
there in v3.7 - probably earlier versions too. The v3.8 source added a 
TODO note about trying to find some better approach.

For the next release of the yuqk fork (R19) I've currently reduced the 
rotation to: 'rotate <0.0001,0.0001,0.0001>' as a longer trial, but I'm 
having second thoughts (*).

After setting up some more formal test code, I'm left unsure of the best 
way to proceed - in no small part because I don't know what really 
caused the issues for which the patch was applied. If anyone remembers 
anything, please pass it along.

I'm starting to think pulling the patch altogether might be a better 
long trial. A user can always add the rotation. In fact, likely a 
cleaner, more minimal rotation about just the z axis) when artefacts 
show up.

Attached is an image and a yuqk scene file where I've set the thickness 
of the text objects at 1e-6; Artifacts show up even at a thickness of 
1.0. The rotation is effectively another issue that can bite or not 
depending on many things. Centering the text{} object about the origin 
lessons the chance the secret rotation will cause a problem.

The left render shows the result for the current secrete rotation of 
<0.001,0.001,0.001>.

In the middle no secrete rotation is applied.

On the right, the current R19 rotation of <0.0001,0.0001,0.0001>. It 
helps, but I think not in a way that really addresses the issue and, who 
knows, maybe it's not large enough for the situation for which the patch 
was added.

Bill P.

(*) - Another thing I played with was to add a counter / inverse 
rotation in SDL to the secrete one in an attempt to get back to no 
rotation. While it sometimes works, there is too much numerical noise in 
rotating and un-rotating (on combining the two opposing matrices). The 
overall results were less stable.

Post a reply to this message

Attachments:
Download 'textsecreterotation.jpg' (22 KB) Download 'textsecreterotation.pov.txt' (2 KB)

Preview of image 'textsecreterotation.jpg'

William F Pokorny <ano### [at] anonymousorg> wrote:
> Centering the text{} object about the origin
> lessons the chance the secret rotation will cause a problem.

I've certainly learned that you're not spelling that correctly.

> The left render shows the result for the current secrete rotation of
> <0.001,0.001,0.001>.

It must be a small glandular thing.


Thanks for uncovering this and pointing it out.
It explains something that I've noticed and ignored, when messing up text via
the coincident surface phenomenon.  There's always been half of the text visible
in this sort of diagonal partition.  Since the text is actually rotated, albeit
by a small amount, this makes perfect sense.

- BW

Post a reply to this message

On 3/13/25 15:44, Bald Eagle wrote:
>> Centering the text{} object about the origin
>> lessons the chance the secret rotation will cause a problem.
> I've certainly learned that you're not spelling that correctly.
> 
>> The left render shows the result for the current secrete rotation of
>> <0.001,0.001,0.001>.
> It must be a small glandular thing.

Oh my... I laughed out loud. +11  :-)

Bill P.

Post a reply to this message