rhettinger
commented
5 months ago I did a quick check for recent changes to Objects/unicodeobject.c::replace() but the code has been stable. The last edit was 17b4733f2ff Serhiy Storchaka 2020-04-01. That said, a lot of other things have changed in CPython (the memory allocator, method dispatch, and the next of include files).
It would help a great deal if you could bisect the problem to a particular edit or construct a minimal reproducer that excludes third-party extensions.
The-Compiler
markshannon
vstinner
Crash report
What happened?
I'm trying to run the qutebrowser testsuite with Python 3.13, and am running into an issue where a test reproducibly fails (usually by crashing the interpreter), but only when I run the entire testsuite (not when run in isolation, or even just the tests in the same subfolder).
Given those circumstances, it seems tricky to get to a minimal example. I thought I'd open this issue in the hope of distilling things down further, and arriving at such an example. In the meantime, the best reproduction steps I can come up with are:
A few tests will fail with a
--with-pydebugbuild due to timeouts, those can be ignored. After a while (~13 minutes with--with-pydebugunder gdb on my system), one of the tests intests/unit/mainwindow/test_messageview.pywill fail, usually due to a failing assertion becausePyUnicode_KINDdid return an invalid value.Failing Python code:
stacktrace:
, globals=, locals=Python Exception : There is no member named ready.
) at Python/ceval.c:598
#67 0x00007211a3c51950 in builtin_exec_impl (module=, closure=, locals=Python Exception : There is no member named ready.
, globals=Python Exception : There is no member named ready.
, source= ) at Python/bltinmodule.c:1145 #68 builtin_exec (module=, args=, nargs=, kwnames=) at Python/clinic/bltinmodule.c.h:556
#69 0x00007211a395a4ea in cfunction_vectorcall_FASTCALL_KEYWORDS (func=, args=0x7211a414a180, nargsf=, kwnames=0x0) at Objects/methodobject.c:441
#70 0x00007211a39fcce9 in _PyObject_VectorcallTstate (kwnames=0x0, nargsf=9223372036854775810, args=0x7211a414a180, callable=, tstate=0x7211a3ecb1f0 <_PyRuntime+294032>)
at ./Include/internal/pycore_call.h:168
#71 PyObject_Vectorcall (callable=, args=0x7211a414a180, nargsf=9223372036854775810, kwnames=0x0) at Objects/call.c:327
#72 0x00007211a3889b59 in _PyEval_EvalFrameDefault (tstate=0x7211a3ecb1f0 <_PyRuntime+294032>, frame=0x7211a414a0d8, throwflag=6) at Python/generated_cases.c.h:813
#73 0x00007211a3c36c04 in PyObject_Call (kwargs=0x0, args=Python Exception : There is no member named ready.
, callable=) at Objects/call.c:373
#74 pymain_run_module (modname=, set_argv0=set_argv0@entry=1) at Modules/main.c:297
#75 0x00007211a3c498d2 in pymain_run_python (exitcode=0x7ffecd272a68) at Modules/main.c:633
#76 Py_RunMain () at Modules/main.c:718
#77 0x00007211a3f72cd0 in ??? () at /usr/lib/libc.so.6
#78 0x00007211a3f72d8a in __libc_start_main () at /usr/lib/libc.so.6
#79 0x00005e3a1e856055 in _start ()
```
Sometimes I've also seen a
MemoryErroron the line callingstr.replace, or an ominous:with the following Python stack (note that jinja is involved, which might or might not be a trigger?):
which leads me to the conclusion that there must be some sort of memory corruption going on there.
On one run, I've also see a GC-related crash, which I'm not sure is related:
Python stack:
C stack:
, globals=, locals=Python Exception : There is no member named ready.
)
at Python/ceval.c:598
#48 0x00007f311c651950 in builtin_exec_impl (module=, closure=, locals=Python Exception : There is no member named ready.
, globals=Python Exception : There is no member named ready.
, source= ) at Python/bltinmodule.c:1145 #49 builtin_exec (module=, args=, nargs=, kwnames=)
at Python/clinic/bltinmodule.c.h:556
#50 0x00007f311c35a4ea in cfunction_vectorcall_FASTCALL_KEYWORDS
(func=, args=0x7f311cb33180, nargsf=, kwnames=0x0) at Objects/methodobject.c:441
#51 0x00007f311c3fcce9 in _PyObject_VectorcallTstate
(kwnames=0x0, nargsf=9223372036854775810, args=0x7f311cb33180, callable=, tstate=0x7f311c8cb1f0 <_PyRuntime+294032>) at ./Include/internal/pycore_call.h:168
#52 PyObject_Vectorcall
(callable=, args=0x7f311cb33180, nargsf=9223372036854775810, kwnames=0x0) at Objects/call.c:327
#53 0x00007f311c289b59 in _PyEval_EvalFrameDefault
(tstate=0x7f311c8cb1f0 <_PyRuntime+294032>, frame=0x7f311cb330d8, throwflag=6) at Python/generated_cases.c.h:813
#54 0x00007f311c636c04 in PyObject_Call (kwargs=0x0, args=Python Exception : There is no member named ready.
, callable=)
at Objects/call.c:373
#55 pymain_run_module (modname=, set_argv0=set_argv0@entry=1) at Modules/main.c:297
#56 0x00007f311c6498d2 in pymain_run_python (exitcode=0x7ffc5a236bd8) at Modules/main.c:633
#57 Py_RunMain () at Modules/main.c:718
#58 0x00007f311c043cd0 in ??? () at /usr/lib/libc.so.6
#59 0x00007f311c043d8a in __libc_start_main () at /usr/lib/libc.so.6
#60 0x000064c810b08055 in _start ()
```
I'm lost here on how to best debug this further. My best bet would be to try and at least get an example that I can run more quickly, and then try and bisect CPython in order to find the offending change. If there are any other guesses or approaches to debug what could be going on here, I'd be happy to dig in further.
CPython versions tested on:
CPython main branch
Operating systems tested on:
Linux
Output from running 'python -VV' on the command line:
Python 3.13.0b1 (main, May 23 2024, 09:21:12) [GCC 13.2.1 20240417]
Linked PRs