Details
-
Bug
-
Status: Closed (View Workflow)
-
Low
-
Resolution: Duplicate
-
6.14/04
-
None
-
/cvmfs/sft.cern.ch/lcg/views/LCG_94python3/x86_64-slc6-gcc62-opt
ROOT 6.14/04
Description
Hi ! I recently noticed some strange behaviours when using some Roofit objects in python. For instance, when a RooWorkspace is set as an attribute of a class, it can trigger a memory error (for instance when importing stuff inside the RooWorkspace)
Here is a minimal non working example:
import ROOT as r |
class Dummy: |
def __init__(self):
|
self.wk = r.RooWorkspace("wk") |
self.wk.factory("my_var[1,0,3]") |
ds = r.RooDataSet( "ds", "ds", self.wk.allVars() ) |
getattr( self.wk, "import" ) (ds) |
self.wk.Print()
|
|
my_dummy = Dummy()
|
The memory error is thrown at the end:
*** glibc detected *** python: free(): invalid pointer: 0x0000000004743a2c *** |
======= Backtrace: =========
|
/lib64/libc.so.6[0x3479275e5e] |
/lib64/libc.so.6[0x3479278cad] |
/cvmfs/sft.cern.ch/lcg/releases/ROOT/6.14.04-38745/x86_64-slc6-gcc62-opt/lib/libRooFitCore.so(_ZN13RooL |
inkedList6DeleteEPKc+0x23)[0x7fd24602d9d3] |
/cvmfs/sft.cern.ch/lcg/releases/ROOT/6.14.04-38745/x86_64-slc6-gcc62-opt/lib/libRooFitCore.so(_ZN12RooW |
orkspaceD2Ev+0x3d)[0x7fd246102fed] |
/cvmfs/sft.cern.ch/lcg/releases/ROOT/6.14.04-38745/x86_64-slc6-gcc62-opt/lib/libRooFitCore.so(_ZN12RooW |
orkspaceD0Ev+0x9)[0x7fd246103269] |
/cvmfs/sft.cern.ch/lcg/views/LCG_94python3/x86_64-slc6-gcc62-opt/lib/libCore.so(_ZN6TClass10DestructorE
|
Pvb+0x31)[0x7fd254991551] |
/cvmfs/sft.cern.ch/lcg/views/LCG_94python3/x86_64-slc6-gcc62-opt/lib/libPyROOT.so(_ZN6PyROOT17op_deallo
|
c_nofreeEPNS_11ObjectProxyE+0x1b7)[0x7fd25835a7a7] |
/cvmfs/sft.cern.ch/lcg/views/LCG_94python3/x86_64-slc6-gcc62-opt/lib/libPyROOT.so(+0x69819)[0x7fd25835a |
819] |
/cvmfs/sft.cern.ch/lcg/views/LCG_94python3/x86_64-slc6-gcc62-opt/lib/libpython3.6m.so.1.0(+0xed4cc)[0x7 |
fd25ecaa4cc]
|
/cvmfs/sft.cern.ch/lcg/views/LCG_94python3/x86_64-slc6-gcc62-opt/lib/libpython3.6m.so.1.0(+0xc70d7)[0x7 |
fd25ec840d7]
|
/cvmfs/sft.cern.ch/lcg/views/LCG_94python3/x86_64-slc6-gcc62-opt/lib/libpython3.6m.so.1.0(+0xed541)[0x7 |
fd25ecaa541]
|
/cvmfs/sft.cern.ch/lcg/views/LCG_94python3/x86_64-slc6-gcc62-opt/lib/libpython3.6m.so.1.0(PyDict_SetIte |
m+0x397)[0x7fd25ec882a7] |
/cvmfs/sft.cern.ch/lcg/views/LCG_94python3/x86_64-slc6-gcc62-opt/lib/libpython3.6m.so.1.0(_PyModule_Cle |
arDict+0x228)[0x7fd25ec97b88] |
/cvmfs/sft.cern.ch/lcg/views/LCG_94python3/x86_64-slc6-gcc62-opt/lib/libpython3.6m.so.1.0(PyImport_Clea |
nup+0x3c4)[0x7fd25ed49174] |
/cvmfs/sft.cern.ch/lcg/views/LCG_94python3/x86_64-slc6-gcc62-opt/lib/libpython3.6m.so.1.0(Py_FinalizeEx |
+0x6b)[0x7fd25ed57fdb] |
/cvmfs/sft.cern.ch/lcg/views/LCG_94python3/x86_64-slc6-gcc62-opt/lib/libpython3.6m.so.1.0(Py_Main+0x6b9 |
)[0x7fd25ed796e9] |
python(main+0x177)[0x400a97] |
/lib64/libc.so.6(__libc_start_main+0x100)[0x347921ed20] |
python[0x400b51] |
======= Memory map: ========
|
00400000-00401000 r-xp 00000000 00:26 4153050 /exp/LHCb/fdesse/ewp-Bd2Kstee-AngAna/venv/angAna/bin/python |
00601000-00602000 r--p 00001000 00:26 4153050 /exp/LHCb/fdesse/ewp-Bd2Kstee-AngAna/venv/angAna/bin/python |
00602000-00603000 rw-p 00002000 00:26 4153050 /exp/LHCb/fdesse/ewp-Bd2Kstee-AngAna/venv/angAna/bin/python |
00f13000-04a3c000 rw-p 00000000 00:00 0 [heap] |
3478e00000-3478e20000 r-xp 00000000 08:02 8300 /lib64/ld-2.12.so |
3479020000-3479021000 r--p 00020000 08:02 8300 /lib64/ld-2.12.so |
3479021000-3479022000 rw-p 00021000 08:02 8300 /lib64/ld-2.12.so |
I also witnessed a similar bug using RooMCStudy in the same fashion.
A temporary workaroud proposed by Chris Burr is to add the line:
atexit.register(self.wk.Delete)
|
inside the definition of the python class to force RooFit to clean up. It would probably be better if RooFit would do it properly in the first place I guess.
cheers,
Fabrice
Attachments
Issue Links
- is duplicated by
-
-
- Closed
-