Is collision constraint proposed in PBD paper correct?

[apapaxionga]

When I implement the collision detection and response proposed in "Position Based Dynamichttp://matthias-mueller-fischer.ch/publications/posBasedDyn.pdf". I found it only test vertex&triangle penetration for collision. No edge or face collision test. However, although high iteration number may partially hidden this problem, this kind of collision will miss some collision cases such as triangles only intersect on edge or edge and cylinder collision. Is this method correct? Do I make some mistakes in understanding this paper?

[fishboy82]

I have the same question, how to make a robust softbody surface-softbody surface collision algo,and make it suitable for use in the PBD

[Kundelstein]

Actually this is the same thing I wanted to ask.
From my tests (2d) I can conclude that tests: edge vs edge (for triangles) and half-edge vs half-edge (for similar rectangles) are necessary. This was proposed in Real-Time Collision Detection by Christer Ericson. Too bad that in the book there is also brief description.

I would love to hear some opinion about this.

PS. Did anyone of you did the proposed collision constraint for dynamic objects (this triangle thing)?