Issue 200 fix intersection

[danielcu888]

Fix for failing intersection issue #200 which generates an invalid Solid containing two disjoint parts when it should yield a single two part MultiSoild.

In summary, when recovering the vertices pertaining to each Polyhedra component of the desired intersection volume, we now process each 3-cell deemed relevant to the intersection result individually rather than collectively, since they may pertain to disjoint volumes. These volumes are now combined downstream as separate primitives, preserving the disjoint nature of the components of the result if they are indeed disjoint.

Also added a standalone test to demonstrate the fix.

[mhugo]

Many thanks for your proposition.

However, I am very hesitant to touch files in the "CGAL_patches" directory.
This directory is probably a mistake by itself: if I remember correctly it has been added as a quick and dirty solution to comply with CGAL >= 4.13, because we did not manage to maintain compatibility with newer versions of CGAL.

However I don't know what to do with a patch to the patch. Ideally it should be proposed upstream to CGAL ...

@sloriot About the fix to corefinement, what do you think ?

[sloriot]

Indeed you should not hack that code because it is correct :)
If you need to have exactly one connected component per solid, then you have to split that solid (it will save you the trouble of redoing that when you update the coref code).
Don't you have a split function somewhere in SFCGAL (it does remind me something)?

CGAL 5.1 will officially introduce the split function but for earlier releases you can use Face_filtered_graph, connected_components() and copy_face_graph(), like in this example that is missing the copy_face_graph() call that is done in this one.

[danielcu888]

@mhugo, @sloriot thanks for the feedback, much appreciated :)

Firstly, if not done already, a ticket should be raised to remove the CGAL patches if they are undesirable.

Secondly, a disclaimer should be added to any files that the contained code should not be modified, lest anyone else wastes time attempting to do so.

Lastly, I submitted a fix for this issue a while back to use the Nef polyhedra library to post process the results from corefinement into disjoint components, but is was rejected on the basis that that library is non-performant and that corefinement was the recommended way to address this problem, of which this was an attempt, so sorry for the confusion on that. There is evidence that this fix also has other undesirable consequences so I will not re-submit it.

I will attempt to re-address this issue using the recommendations by @sloriot and close this PR.

[danielcu888]

I did salvage the commit associated with fixing the WKT bench test which failed for locally on Mac OS X, and submitted that here: #221