Clump size limited to 256 bodies.

[Kneib François]

Hi,

I noticed that there is a limited number of bodies you can add to clump::members : 256. Whatever method you use, if you put 257 members into a clump, the c++ map "members" will have the right number of keys (257), but all corresponding values of Se3r will be set to "nan". If you create a clump of 256 bodies, it will be ok first, but if you add one more body all clump members will be corrupted (see the script attached).

I think the problem doesn't come from the "map" type of C++ or the "dict" type of python, because they always have the right size, even if the clump size exceeds 256. It looks like a variable doesn't have sufficient memory allocated and I really wonder where ...

Does someone have an idea ?

Thanks

[Kneib François (francois-kneib)]

I would like to add one more precision :
Into the script I have sent, all spheres are aligned and if I put more than 256 spheres into the clump it will not work (with NaN). But if you replace (l. 37)

utils.sphere((i,0,0),0.5

by :

utils.sphere((i,rand(),rand()),0.5

then you have one chance out of two not to get the bug.

And if you put seed(x) somewhere at the top of the script (replacing x by whatever int you want), then it will always bugs, or always works depending on the int you choose.

[Kneib François (francois-kneib)]

I noticed that it seems to happen only when the spheres are geometrically tangent or nearly tangent (you can try this by modifying the radius into the script).
I looked at the code and the problem seems to come from the "updateProperties()" method, it seems that "state->ori" is NaN. It comes from the inertia tensor eigen vectors, if some values are NaN (see the comment line 205 : "has NaNs for identity matrix??").
As this is a *very* special case, I change the bug importance.

[Christian Jakob (jakob-ifgt)]

fixed in

https://github.com/yade/trunk/commit/5f72cd29784babc19bfc8ed028185e7937b68af9