PythonObjects were being incorrectly ref-counted. This problem was
pervasive throughout the codebase, leading to an unknown number of memory
leaks and potentially use-after-free.
The issue stems from the fact that Python native methods can either return
"borrowed" references or "owned" references. For the former category, you
*must* incref it prior to decrefing it. And for the latter category, you
should not incref it before decrefing it. This is mostly an issue when a
Python C API method returns a `PyObject` to you, but it can also happen with
a method accepts a `PyObject`. Notably, this happens in `PyList_SetItem`,
which is documented to "steal" the reference that you give it. So if you
pass something to `PyList_SetItem`, you cannot hold onto it unless you
incref it first. But since this is one of only two exceptions in the
entire API, it's confusing and difficult to remember.
Our `PythonObject` class was indiscriminantely increfing every object it
received, which means that if you passed it an owned reference, you now
have a dangling reference since owned references should not be increfed.
We were doing this in quite a few places.
There was also a fair amount of manual increfing and decrefing prevalent
throughout the codebase, which is easy to get wrong.
This patch solves the problem by making any construction of a
`PythonObject` from a `PyObject` take a flag which indicates whether it is
an owned reference or a borrowed reference. There is no way to construct a
`PythonObject` without this flag, and it does not offer a default value,
forcing the user to make an explicit decision every time.
All manual uses of `PyObject` have been cleaned up throughout the codebase
and replaced with `PythonObject` in order to make RAII the predominant
pattern when dealing with native Python objects.
Differential Revision: http://reviews.llvm.org/D13617
Reviewed By: Greg Clayton
llvm-svn: 250195
With this change, liblldb is 95% of the way towards being able
to work under both Python 2.x and Python 3.x. This should
introduce no functional change for Python 2.x, but for Python
3.x there are some important changes. Primarily, these are:
1) PyString doesn't exist in Python 3. Everything is a PyUnicode.
To account for this, PythonString now stores a PyBytes instead
of a PyString. In Python 2, this is equivalent to a PyUnicode,
and in Python 3, we do a conversion from PyUnicode to PyBytes
and store the PyBytes.
2) PyInt doesn't exist in Python 3. Everything is a PyLong. To
account for this, PythonInteger stores a PyLong instead of a
PyInt. In Python 2.x, this requires doing a conversion to
PyLong when creating a PythonInteger from a PyInt. In 3.x,
there is no PyInt anyway, so we can assume everything is a
PyLong.
3) PyFile_FromFile doesn't exist in Python 3. Instead there is a
PyFile_FromFd. This is not addressed in this patch because it
will require quite a large change to plumb fd's all the way
through the system into the ScriptInterpreter. This is the only
remaining piece of the puzzle to get LLDB supporting Python 3.x.
Being able to run the test suite is not addressed in this patch.
After the extension module can compile and you can enter an embedded
3.x interpreter, the test suite will be addressed in a followup.
llvm-svn: 249886
Zachary Turner