Retreiving valid PyEnvironment from PyAbstract1

[pascalito]

Hello Caterina,

The environment property of the PyAbstract1 objects causes my Python program to crash with a segmentation fault (and even my Python debugger when inspecting such objects).

Here is a minimalist version that should go wrong on any computer:

import gc

from apronpy.var import PyVar
from apronpy.environment import PyEnvironment
from apronpy.box import PyBoxMPQManager
from apronpy.abstract1 import PyAbstract1

env = PyEnvironment([PyVar('x')])
man = PyBoxMPQManager()
abs_value = PyAbstract1.top(man, env)

inner_env_A = abs_value.environment
inner_env_B = abs_value.environment
inner_env_C = abs_value.environment

del inner_env_A
del inner_env_B
del inner_env_C  # Fails with SIGSEGV or double free or invalid pointer

gc.collect()  # Just to be sure that __del__ was called

Is it a bug ?

Note that this program does not use the APRON ctypes directly but uses only their Python proxies (Py*). In my humble opinion, this should not end with a segmentation fault.

Fix suggestion

I guess that this problem is triggered by the code of the environment property of the PyAbstract1 class. This code builds a PyEnvironment without increment of the APRON environment reference counter but eventually triggers a decrement through the __del__ method of the newly created PyEnvironment.

Adding the increment in the PyEnvironment constructor or in the environment property fixes this issue. I would not swear that these fixes are sound and memory-leak free...

Improvement suggestion

In passing, it would be nice that the Py*1 classes could all offer a way of retrieving the underlying environment as a PyEnvironment object.

[caterinaurban]

Thanks, good catch! I will implement your improvement suggestion as soon as I have a moment. It's delicate to do, to avoid memory leaks or segmentation faults.

[pascalito]

The fix works fine. Thanks!