OMCSession.sendExpression uses up all available memory when pyplot is imported

[christopher.schoelzel@…]

I wanted to create a testing script for my modelica models using OMPython when I discovered something very strange: The code example below behaves different depending on whether the pyplot module is imported or not. If the module is imported, the parser call in OMCSession.sendExpression will use up all available memory (at least up to 8 GB). If the module is not imported or if OMCSession.execute is used instead of OMCSession.sendExpression, the memory consumption stays below 100 MB.

# OS: Windows 7
# OMPython version: 2.0.7
# OpenModelica version: 1.9.4
# python version: 2.7.11
# output of 'pip freeze':
#   cycler==0.10.0
#   matplotlib==1.5.1
#   numpy==1.10.4
#   OMPython==2.0.7
#   pyparsing==2.1.0
#   python-dateutil==2.5.0
#   pytz==2015.7
#   six==1.10.0

import OMPython
# comment out the following line and the code below will work for both 'execute' and 'sendExpression'
import matplotlib.pyplot as plt

s = OMPython.OMCSession()
print s.sendExpression("loadModel(Modelica)")
print s.sendExpression("simulate(Modelica.Electrical.Machines.Examples.DCMachines.DCPM_Start,stopTime=100,numberOfIntervals=100000)")
# execute: normal behavior
data = s.execute("readSimulationResult(currentSimulationResult,{time, ground.p.v, ground.p.i})")
# sendExpression: consumes all available RAM
# data = s.sendExpression("readSimulationResult(currentSimulationResult,{time, ground.p.v, ground.p.i})")
del s

[Christoph <buchner@…>]

possibly related: ​https://github.com/OpenModelica/OMPython/issues/3 ?

[Martin Sjölund]

I guess this is to be expected. The simulation result is huge and is sent as a text-string that a parser needs to read.

[Martin Sjölund]

Also, ​https://github.com/OpenModelica/OMPython/issues/3 is unrelated. They are supposed to give different results.

[Martin Sjölund]

It's weird in the sense that this only occurs if matplotlib and pyparsing are used together. I don't have time to look into any specifics, but a reduced test shows that matplotlib severly increases the memory consumption of pyparsing.
Note that matplotlib also uses pyparsing, so it is possible pyparsing has some internal state that is not thread-safe and thus not advisable to use together with OMPython.

[Christoph <buchner@…>]

Does this also occur if you import mpl _after_ OMPython? If not, it could be some import order thing maybe?

[Christoph <buchner@…>]

import mpl _after_ OMPython?

sorry, the other way round of course. :-P

[Martin Sjölund]

Yes, the order does not matter. And I checked the threading module as well (seems to be only 1 thread). I also checked with Python3 :) It would be good if PyParsing could expose a lock or something in order to make it possible for two modules to use it... Perhaps it does...

I'll attach a small example+file that can be tested on pyparsing.

[Martin Sjölund]

Typed parsing grammar

[Martin Sjölund]

File causing memory issues for pyparsing