bradb

In the comments above, Qhull reported the maximum error due to roundoff round off error for computing distances is 0.024 This is substantially more than the reported distances of the...

As reported in scipy/scipy#6484, Qhull 2019.1 will fix this problem. It fails on qhull 2015.2 due to qh_checkflipped reporting a flipped facet to qh_initialhull, which incorrectly reverses the facet orientation....

Qhull 2019.1 continues to fail for Bisaloo's data. Even if qh_maxsimplex searches for the best initial simplex ('Qs'), qh_intialhull reports that qh.interior_point is barely more than qh.DISTround (4e-16) to two...

This is a precision issue due to Qhull computing the Delaunay triangulation by projecting the input to a paraboloid. Your points are in a narrow range [50783..., 6959...] that is...

This will be fixed in Qhull 2019.1. See my note for #25 'Qz option can lead...' Thanks Jean-Romain for reporting this problem, and David for not letting it go.

Good catch. Thanks for the corrected entry. The same problem occurs for qhalf documentation. http://www.qhull.org/html/qhalf.htm#outputs The 'o' documentation is correct. http://www.qhull.org/html/qh-opto.htm#o Both problems will be fixed in the next full...

You can use qh_addpoint in llibqhull_r.c to add a point. qh_addpoint requires a facet that is clearly below the point (e.g., qh_findbestfacet in poly2_r.c). This is a slow method to...

qh_addpoint is faster than rebuilding the convex hull. Unless the point is interior to the convex hull, qh_addpoint requires a distance test for each facet. For adding many points, qh_addpoint...

Qhull is designed for error handling. All computations are done in floating point arithmetic which is inherently imprecise. The geometric computations are used to make topological decisions. All kinds of...

fflush(NULL) appears to be well defined. It should flush all open file descriptors. I think the problem is elsewhere. Can you reproduce the /dev/null failure with qhull tracing its progress...