junrushao commented on PR #228:
URL: https://github.com/apache/tvm-ffi/pull/228#issuecomment-3591066100

   gpt-5.1-codex-max says:
   
   > I dug through the new commit `fa8e0e5` and it only adds the header 
`include/tvm/ffi/extra/stl.h`, a new C++ test source 
`tests/python/cpp_src/test_stl.cc`, the Python driver 
`tests/python/test_stl.py`, and a single friend line in 
`include/tvm/ffi/container/map.h` to allow the STL type traits to call 
`MapObj::CreateFromRange`. No core runtime or shutdown logic was touched.
   >
   > I tried to reproduce locally by running `pytest tests/python/test_stl.py`, 
but the first build of that test module didn’t finish within the timeout, so I 
couldn’t observe a crash. There’s nothing in the diff that runs at interpreter 
teardown; the only new runtime path is the `test_stl` module exercising the new 
STL type traits.
   >
   > Given that, the segfault at pytest exit is likely either (a) pre‑existing 
in the core runtime, or (b) triggered by `test_stl` exercising a latent bug, 
but not caused by structural changes in this commit. To pin it down quickly, 
I’d suggest:
   >
   > 1. rerun the failing command with `PYTHONFAULTHANDLER=1` (or under `gdb 
--args pytest -vvs tests/python`) to grab a C stack trace at the crash;
   > 2. rerun the same pytest command on the parent commit (`7cd2e50`) to see 
if it still segfaults;
   > 3. if you want a fast signal, run `pytest -vvs tests/python --deselect 
tests/python/test_stl.py` once; if the segfault still appears, it’s not tied to 
the new STL code.
   >
   > With a stack trace we can investigate the actual crashing thread/function 
and decide next fixes.


-- 
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.

To unsubscribe, e-mail: [email protected]

For queries about this service, please contact Infrastructure at:
[email protected]


---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]

Reply via email to