`tracemalloc` aborts when run from threads in no-gil

Crash report

What happened?

It's possible to get two different aborts from failed assertions, only with PYTHON_GIL=0, using the code below:

from threading import Thread
import _tracemalloc

alive = [
    Thread(target=_tracemalloc.start, args=(
        (memoryview(bytearray(b"abc\xe9\xff")),))),
    Thread(target=_tracemalloc.start, args=()),
    Thread(target=_tracemalloc.get_traceback_limit, args=()),
    Thread(target=_tracemalloc.start, args=()),
    Thread(target=_tracemalloc.stop, args=()),
    Thread(target=_tracemalloc.reset_peak, args=()),
    Thread(target=_tracemalloc.get_tracemalloc_memory, args=())
]

for obj in alive:
    print('START', obj)
    try:
        obj.start()
    except Exception:
        pass

The errors are:

python: Python/tracemalloc.c:445: tracemalloc_remove_trace: Assertion `tracemalloc_config.tracing' failed.
Aborted

And:

python: Python/tracemalloc.c:469: tracemalloc_add_trace: Assertion `tracemalloc_config.tracing' failed.
Aborted

Found with fusil by @vstinner.

CPython versions tested on:

CPython main branch

Operating systems tested on:

Linux

Output from running 'python -VV' on the command line:

Python 3.14.0a1+ experimental free-threading build (heads/main:d467d9246c, Nov 1 2024, 09:05:56) [GCC 11.4.0]

Start by reproducing the supplied threaded example with an experimental free-threading build and PYTHON_GIL=0. Inspect Python/tracemalloc.c around tracemalloc_remove_trace and tracemalloc_add_trace, then verify that the example no longer aborts when concurrent tracemalloc operations are used.