Leak testing #490
Leak testing #490
Conversation
ccummingsNV
commented
Sep 5, 2025
ccummingsNV
commented
- If live object tracking is enabled, uses it to detect tests that leak objects
- Adds a marker that says a test is known to leak, and what it leaks
skallweitNV
left a comment
skallweitNV
left a comment
There was a problem hiding this comment.
Looks mostly fine.
I don't like the new report_live_objects API. I also think we don't need the to_string function on the LiveObjectInfo struct as this really just serves the purpose of populating a python dictionary.
Then I would really welcome if we can move all the leak tracking code into a new leak_tracking.py file or similar, just leaving the hooks in plugin to call into the other module. This will make it much easier to separate it out again if need be.
| #if SGL_ENABLE_OBJECT_TRACKING | ||
| /// Reports current set of live objects. | ||
| static void report_live_objects(); | ||
| static std::vector<LiveObjectInfo> report_live_objects(bool log_to_tty = true); |
There was a problem hiding this comment.
I don't like this API too much, mostly because we have similar calls in slang-rhi and D3D12. Let's leave report_live_objects as is (printing to stdout).
Can we add something like get_live_object_infos() to return the list?
|
|
||
| [tool.pytest.ini_options] | ||
| pythonpath = ["."] | ||
| markers = [ |
There was a problem hiding this comment.
Can we move this definition to the slangpy pytest plugin somehow?
| std::string LiveObjectInfo::to_string() | ||
| { | ||
| return fmt::format( | ||
| "address={}, self_py={}, ref_count={}, class_name={}", |
There was a problem hiding this comment.
these to_string functions usually return something including the class name LiveObjectInfo(...)
| # Types to ignore when checking for leaked objects | ||
| # - The reflection types are created and cached per device when buffers are loaded, so are hard | ||
| # to identify as actual leaks. | ||
| # - CoopVec is created on demand within the device when the coopvec api is used, and so will appear |
There was a problem hiding this comment.
| # - CoopVec is created on demand within the device when the coopvec api is used, and so will appear | |
| # - The reflection types are created and cached per device when buffers are loaded, so are hard | |
| # to identify as actual leaks. | |
| # - CoopVec is created on demand within the device when the coopvec API is used, and so will appear | |
| # as a leak for cached devices. | |
| IGNORE_LIVE_OBJECT_TYPES = ["NativeSlangType", "TypeLayoutReflection", "TypeReflection", "CoopVec"] | |
| raise RuntimeError("Unsupported platform") | ||
|
|
||
| # If live object tracking is supported, enable leak tracking | ||
| LEAK_TRACKING_ENABLED = hasattr(spy.Object, "report_live_objects") |
There was a problem hiding this comment.
Should we move all the leak checking specifics into a new leak_tracking.py module?
|
@ccummingsNV Is this still being worked on, or should it be moved to draft? |
