search

jdonnert / WVTICs

SPH initial conditions using Weighted Voronoi Tesselations
MIT License
8 stars 4 forks source link

Code hangups #11

Closed YominCarr closed 7 years ago

YominCarr commented 7 years ago

Sometimes the code suddenly hangs completely. Never happened to me at the same spot and never early in the relaxation process. Seems to happen since merging the particle redistribution scheme, but it seems not to hang in there.

tlrh314 commented 7 years ago

This bug may have already been around before merging the particle redistribution scheme. I never did encouter it when using Toycluster, but I remember running into an infinite loop every now and then after I hacked the physics out of Toycluster with the goal to sample the Magneticum logo. Back then I placed print statements to figure out what was going on and the code seemed to hang in the ngb search if I remember correctly, but my quick'n'dirty solution was to slightly change some parameters that did not encounter the bug because the goal was just to create SPH art ...

jdonnert commented 7 years ago

If this is a subtle thing in the tree or ngbfind, it'll be a pain to debug. I would recommend printing a single letter for every stage of the relaxation process without \n to have some idea in the log where the code is hanging. Then from there we can see further.

YominCarr commented 7 years ago

@Timo:

I am kind of relieved to hear that. Actually this was exactly what I thought, because the neighbour search seems to be the only spot where it can actually get stuck in an infinite loop.

I just never had this issue before and now I get it frequently. So maybe something drives the particles into a state where this happens. I try to investigate it. I started a debugging session without optimisation yesterday which ran through without any issues. But that does not proove anything...

@Julius:

Yep, I will do that. At the moment I still try to catch it in gdb while working on the final parts of the paper.

Am 13.09.2017 um 14:04 schrieb Julius Donnert:

If this is a subtle thing in the tree or ngbfind, it'll be a pain to debug. I would recommend printing a single letter for every stage of the relaxation process without \n to have some idea in the log where the code is hanging. Then from there we can see further.

— You are receiving this because you were assigned. Reply to this email directly, view it on GitHub https://github.com/jdonnert/WVTICs/issues/11#issuecomment-329146646, or mute the thread https://github.com/notifications/unsubscribe-auth/AI7IYHLvy9NELN68Lzh8r1zc_oUz8xjgks5sh8TVgaJpZM4PV_WQ.

YominCarr commented 7 years ago

I actually managed to capture a hangup in the debugger over night and I found a bug in the particle redistribution scheme. I fixed it in testproblems_rebased2 and test it right now before cherry-picking it for master.

This does not mean that there is nothing in the tree, but this should account for most hang ups at the very least.

Am 13.09.2017 um 14:04 schrieb Julius Donnert:

If this is a subtle thing in the tree or ngbfind, it'll be a pain to debug. I would recommend printing a single letter for every stage of the relaxation process without \n to have some idea in the log where the code is hanging. Then from there we can see further.

— You are receiving this because you were assigned. Reply to this email directly, view it on GitHub https://github.com/jdonnert/WVTICs/issues/11#issuecomment-329146646, or mute the thread https://github.com/notifications/unsubscribe-auth/AI7IYHLvy9NELN68Lzh8r1zc_oUz8xjgks5sh8TVgaJpZM4PV_WQ.