I think we can constrain the allowed input to be locally manifold, but where multiple topologically different vertices can have the same position in space, as I suggested in my previous email in this thread.

-Marius

Not yet, but I’ll start categorizing more failing tests and list them under the appropriate github issue.

As a start, it would be interesting if you could handle the 8 test cases listed here:
https://github.com/openscad/openscad/issues/1580

They all cause zero area triangles on export from OpenSCAD today.

By first glance it looks like your algorithm is equivalent to iteratively performing edge flips, except you handle all vertex-edge intersections per split operation, instead of just one.
I’ll take a closer look later.

-Marius

On 03. des. 2016 20:25, Ronaldo wrote:

Well, it is possible under certain assumptions, most importantly when a
model can be interpreted as manifold, using coordinate matching, and a
couple of other techniques.

In 3d printing, most models are (or should be) manifolds, so it is a
reasonable assumption to work from when attempting to 'heal'

But the healing algorithms are not able to read your mind, so to turn
your example into a usable manifold (if that is your goal), you need to
edit it.

Carsten Arnholm

On 03. des. 2016 22:00, Marius Kintel wrote:

Not sure I understood which exact 8 cases that is, but I tried these:

issue1580-back-to-back.stl      result: no warnings
issue1580-back-to-back2.stl      result: no warnings
issue1580-zero-area-triangle.stl result: no warnings
issue945e.stl                    result: no warnings
issue945f.stl                    result: no warnings
issue1803.stl                    result: no warnings
adns2610_dev_circuit_inv_ascii.stl  result: no warnings  (ascii)
issue1876.stl                    result: no warnings

See attached file with more info.

In short, all appears to end with no warning, and using simple
visualisation of the resulting AMF file seems to confirm they are all ok.

issue1876.stl  was an interesting case of having also a duplicated face,
it was handled ok.

Yes it is true all edge splits are done done at once for any given edge.

From the above tests, my scheme seems to work, but I guess there are
more complex  problems

Carsten Arnholm

Cool, looks like you’re a good step ahead of me in terms of success rate.

There are more complex ones, but they’re all based on original bug reports which I haven’t properly organized yet.

-Marius