Segmentation fault on pytest exit when creating a SerializableObjectWithMetadata #1474
Labels
bug
A problem, flaw, or broken functionality.
Comments
ssteinbach
changed the title
|
When OTIO is built in debug mode, we get a little bit more details: |
|
The fact that it crashes it metadata={} is given as an argument kind of suggests me it has to do with the round-trip in https://github.com/AcademySoftwareFoundation/OpenTimelineIO/blob/main/src/py-opentimelineio/opentimelineio-bindings/otio_utils.cpp#L99-L121. This part of the code is pretty painful to debug... |
Alright, with GDB atatched, we get this. The interesting part of that backtrace is: Notice the |
|
Some more information: https://github.com/AcademySoftwareFoundation/OpenTimelineIO/blob/main/src/py-opentimelineio/opentimelineio/core/_core_utils.py#L61 is executed properly, which means that at least Also found pytest-dev/pyfakefs#693 which kind of means that pyfakefs can break stuff. Looking at the pyfakefs code, it seems extremely intrusive. |
|
By doing a small change, we can verify if diff --git a/src/py-opentimelineio/opentimelineio-bindings/otio_serializableObjects.cpp b/src/py-opentimelineio/opentimelineio-bindings/otio_serializableObjects.cpp
index b3338eb..4f37429 100644
--- a/src/py-opentimelineio/opentimelineio-bindings/otio_serializableObjects.cpp
+++ b/src/py-opentimelineio/opentimelineio-bindings/otio_serializableObjects.cpp
@@ -188,7 +188,8 @@ static void define_bases1(py::module m) {
py::class_<SOWithMetadata, SerializableObject,
managing_ptr<SOWithMetadata>>(m, "SerializableObjectWithMetadata", py::dynamic_attr())
.def(py::init([](std::string name, py::object metadata) {
- return new SOWithMetadata(name, py_to_any_dictionary(metadata));
+ auto asd = py_to_any_dictionary(metadata);
+ return new SOWithMetadata(name, asd);
}),
py::arg_v("name"_a = std::string()),
py::arg_v("metadata"_a = py::none()))
diff --git a/src/py-opentimelineio/opentimelineio-bindings/otio_utils.cpp b/src/py-opentimelineio/opentimelineio-bindings/otio_utils.cpp
index c327374..55ed8ea 100644
--- a/src/py-opentimelineio/opentimelineio-bindings/otio_utils.cpp
+++ b/src/py-opentimelineio/opentimelineio-bindings/otio_utils.cpp
@@ -102,7 +102,8 @@ static void py_to_any(py::object const& o, any* result) {
_value_to_any = core.attr("_value_to_any");
}
- result->swap(_value_to_any(o).cast<PyAny*>()->a);
+ py::object tmp = _value_to_any(o);
+ result->swap(tmp.cast<PyAny*>()->a);
}
AnyDictionary py_to_any_dictionary(py::object const& o) {Then in GDB: So |
|
Stumbled over this issue today, and created a respective issue in pyfakefs. def test_create_clip(capsys, fs):
fs.clear_cache()
otio.core.SerializableObjectWithMetadata(
metadata={}
) |
|
Thanks! We will test this workaround to see if it helps. |
|
Alternatively, you could use a fixture without caching instead of import pytest
from pyfakefs.fake_filesystem_unittest import Patcher
@pytest.fixture
def fs_no_cache():
with Patcher(use_cache=False) as patcher:
yield patcher.fs
...
def test_empty_fs(fs_no_cache):
passCaching is a performance feature that is only relevant if you run a larger number of tests with |
|
Ok, I've made a patch release, please check if your tests work with the current version! |
|
Nice, thanks @mrbean-bremen ! |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
A user reported that there is a segmentation fault when running pytest + objects with metadata + the "fs" (fake filesystem) capture system.
Repro:
Output:
(The warning is related to our use of
impand appears to be a red herring)The text was updated successfully, but these errors were encountered: