error: Body's boundary metting its opposite boundary

[Sergei Dorofeenko]

When I try to add polydisp spheres into periodic domain I get a error:

FATAL /home/sega/work/yade/trunk/pkg/common/InsertionSortCollider.cpp:343 insertionSortPeri: Inversion of body's #29 boundary with its other boundary, 0.5 meets 0.375
FATAL /home/sega/work/yade/trunk/core/ThreadRunner.cpp:31 run: Exception occured:
/home/sega/work/yade/trunk/pkg/common/InsertionSortCollider.cpp: Body's boundary metting its opposite boundary.

With monodisp spheres no error occurs.

I attach a test script.

[Václav Šmilauer (eudoxos)]

This looks like a genuine bug. My nightmare of a bug in the collider comes true :-| I will try to isolate minimal acse and fix it, but it can take a while. I could disable the exception for now, but I fear that it could just hide some other problems which are not noticed.

[Václav Šmilauer (eudoxos)]

Ok, I run in the same problem with updated regular-sphere-pack.py where there is facet perpendicular to z, where for some reason the maximum is reported to be smaller than the minimum, even though the numbers are the same (to the machine precision). The problem is apparently in the operator< and operator> defined on bounds, which handles same-id situation specially -- with equal coordinates and ids, the minimum is always smaller. I don't know how it can happen that the minimum is above the maximum (hence triggering the inversion). The next commit will contain the fix, which esentially avoids calling handleBoundInversion if the ids are identical.

[Václav Šmilauer (eudoxos)]

Fixed in the subdomains branch (it is not sure when it will be merged back to trunk, though -- I will use Fix Commited instead of Fix Released for now).

[Anton Gladky (gladky-anton)]

Vaclav, is it possible to apply this fix to the current trunk?

[Václav Šmilauer (eudoxos)]

Merged to trunk.