HistStack in Pad a crash in rootpy, but not pyroot

[wlav]

Hi,

see here:
http://root.cern.ch/phpBB3/viewtopic.php?f=14&t=17597

could you have a look and let me know whether it is just a side effect of that earlier NameError, or if anything needs changing on the PyROOT side?

Thanks,
Wim

[pwaller]

Interesting. Thanks for reporting this upstream. I'm not using root or rootpy anymore, so unfortunately my motivation to fix this is low. I'm the offender having written the code though so I feel obliged to help a little.

The problem is coming from my nasty hack to make importing faster 😢. I arranged it like this because at the time on network filesystems it meant a significant speedup (up to tens of seconds in extreme circumstances) for ROOT imports (and subsequent dynamic library loads) which were then not actually used (for example because you wrote --help or some non-root thing was happening).

Looking at the users code it appears correct, It's not clear that it should fail. I don't have a ROOT installation handy to test, @ndawe can you confirm if you're able to reproduce?

It is interesting that module is not defined, which is used inside that lambda. Looking back I think it's supposed to be a way to access the self.module from ModuleFacade in ROOT.py (because its locals() is ROOT.__dict__). It looks to me like our log error handler was called, which interacted in a bad way with the quick import mechanism.

It would be really useful to see if the same code is executed with NO_ROOTPY_HANDLER set if we get a message.

I'm not in a position (lack of time, working software installation or access to an appropriate box) to look into trying to reproduce this myself now. It seems to me likely that the crash in Py_Finalize is due to our hackery and not ROOT's fault.

P.S, apologies @wlav for shamelessly using undocumented internal interfaces such that they found their way into this public library. It might be that rootpy should be fixed to avoid this in places.

[ndawe]

For easy reference, here is the offending code:

import rootpy
from rootpy.plotting import *

h = Hist(10,0.,10.,name='htest')
h.Fill(2.)

hs = HistStack()
hs.Add(h)

c = Canvas()
p = Pad(0.,0.5,1.,1.)
p.Draw()

p.cd()
hs.Draw()

The reported code works fine on my end, however, this is actually not the first time I've seen this NameError in quickroot. There seems to be a race condition if the moment the logger is first initialized corresponds to a crash on the ROOT side. I'll try to reproduce this with something else.

On the ROOT side, the signal handler was called at #5 0x000000010869bb91 in THashList::RecursiveRemove () after the pad was destructed.

[ndawe]

I can't tell from the post on the forum what ROOT version was used but there could potentially be an issue on the ROOT side that also happens to trigger this race condition in rootpy.

[ndawe]

I will follow up with the user on the forum.

[pwaller]

Closing because it's a dead end, we never heard back from the user. Please reopen if you encounter this.