Too much memory used at creation/removal of interactions ? (JCFpm)

[Jérôme Duriez]

Hi,

The problem is described here : https://answers.launchpad.net/yade/+question/242890

I'm attaching the script describing this simple compression, with the .spheres file (in a 2d comment) that is loaded to "ymport" the sample. In this script/in the on-line yade version, the number of interactions that were requested to erase, despite a geometrical overlap, is not accessible.

Let me know if I forgot something to let run the simulation.

Jérôme

[Jérôme Duriez (jduriez)]

The .sphere file

[Bruno Chareyre (bruno-chareyre)]

Thanks for reporting. Could you check if the following replacement fix the problem?

scene->interactions->requestErase(contact->getId1(),contact->getId2());
====>
scene->interactions->requestErase(contact);

I don't know if I failed some search/replace or if it has been commited afterward, but the (id1,id2) form should not be used after [1] (two years ago).

Even if it fixes your problem, please don't close the bug. It should be simply impossible to write leaking code. The mechanism of queuing pending erase has been removed, but this nasty function filling the queue are still there... my mistake.

Besides, you can't really erase interactions when they overlap. They will effectively be created then reset at each step (but without the memory leak). It would be better to not erase in terms of performance. Plus, I don't understand how you can decide to ignore them if you erase them: history is lost, in the next iteration they look new.

Bruno

https://github.com/yade/trunk/commit/15182aca8473be86c74b139731dbb88b0007f759

[Jérôme Duriez (jduriez)]

It solves the memory problem !

The JCF model was surely only on Luc's PC when you did this commit, and then the use of requestErase(id1,id2) was re-introduced when JCF code was commited after.

I'll commit the change in JCFpm soon, but why not erasing this (id1,id2) version from InteractionContainer ? I noticed these different versions, but could not deduce directly that there was right and wrong ones...

For the question of erasing or not such interactions in JCFpm, I agree with you. Probably this will change after some discussion with other users. (but I don't think there might be another problem than performance : such interactions will have Fn=Ft=0 at their re-birth)

[Jérôme Duriez (jduriez)]

https://github.com/yade/trunk/commit/131a602b0fafd2bd67cc03354f3f8e81a2268374

[Bruno Chareyre (bruno-chareyre)]

"please don't close the bug" -> reopening.

[Jérôme Duriez (jduriez)]

Hi Bruno,

Is it this change https://github.com/yade/trunk/commit/a9d7963ff501c4ac873564094e4a47f5bf8a3b5f that you had in mind before closing the bug ?

[Bruno Chareyre (bruno-chareyre)]

Absolutely.
Closing :)