dumblob
commented
3 weeks ago Just a bit crazy brainstroming here.
What if we saw the whole problem even more holistically than just focusing on the one operation right after the one which creates this "ref. faces problem"?
What if we somehow "split" the part into two (conceptually something like booleans but kind of "generalized") right at that point and try to "visually" preserve the "concept" (intent) the user has done. The AI hype all around can give you a sense of what the "visual concept" (intent, essence) might mean. Note, I would not want to use any kind of AI (even if it would be just its purest form - i.e. fuzzy logic) but it can help us develop a holistic answer to the "reference faces problem".
E.g. let us have a cube 3x3x3 [op1]. On one face we will draw a circle and subtract its 360deg rotation from the cube [op2]. Then we will select the half-cylindric face and somewhere close to its centroid we will create a touching sketch with a triangle [op3] which we will extrude outwards from the cube [op4].
Then we will change op2 to "draw and rotate a triange forming a cone which we will add to the cube" [op2b].
My proposal would be then to separate visually the whole initial creation right at the place where the change started. That means we take the status after op2 and then subtract it from state at op4 resulting in a temporary shape [tmp1]. This difference [dif1] will then be the guidance for the intent. Then we will just take op1, apply op2b and then try to find a way how to apply op3, etc. to mimic tmp1 (i.e. the original "look & feel"). In this case we can take a look at tmp1 center of gravity, tmp1 vector along its longest dimension, tmp1 "rotation" relative to the shape after op1, we can try to translate tmp1 a little bit for validation of different options (sounds like another gradient descent algo...), IDK, just try to fit it in somehow :smiley:.
Yeah, it is on the verge of dreaming. But why not to dream at this stage :rofl:
This is a discussion about how to solve the "referenced faces problem", both for 2D and 3D faces. 2D faces are enclosed, "colored" areas of a sketch that one can extrude, 3D faces are the flat surfaces that 3D objects are comprised of (a 3D cube has 6 faces).
The problem is simple: How do you refer to a selected face, e.g. during an extrusion?
Just ID
This is the simplest solution and what we've done till now, but it has a huge deal breaker: we're calculating faces and they're not static.
This simplest way to break this is approach is to create a sketch with 2 rectangles that share a line, do an extrusion on the second rectangle and then edit the sketch and remove the shared line. Now there's only 1 face and the face ID 2 does not exist
Centroid anchor
A very interesting idea @MattFerraro introduced with the document is to use anchors to refer to faces.
Every time a face is generated its centroid is also calculated and thus during extrusion the face centroid is referenced, not the face or its ID
After a sketch edit the face with centroid closest to the saved centroid is selected.
This approach though is also problematic:
Some possible fixes for the second problem is:
About the failing problem I can't think of a good enough solution. Maybe we can store the amount of "independant" (they share no features) faces and if that drops we fail? Do we just throw a warning when more than 1 extrusion with different centroids end up referring to the same face?
I think the correct approach is to introduce some kind of hashing to the faces that is able to produce a "hash distance" or "not found". the face with the closest hash distance (if any) is selected.
But what do we include the hash and how do we hash it so that a "distance" can be found?