BUG: pandas 1.5.2 segfault with pypy 3.9-7.3.11

[nicolaerosia]

Pandas version checks

• I have checked that this issue has not already been reported.

• I have confirmed this bug exists on the latest version of pandas.

• I have confirmed this bug exists on the [main branch] (https://pandas.pydata.org/docs/dev/getting_started/install.html#installing-the-development-version-of-pandas) of pandas.

Reproducible Example

docker run -it --rm \
--entrypoint /bin/bash \
docker.io/pypy:3.9-7.3.11

pypy3.9 -m venv venv-pandas-152-pypy39
source venv-pandas-152-pypy39/bin/activate
pip install pandas==1.5.2
python -c 'import pandas as pd; pd.Timedelta("5min")'

Issue Description

Segmentation fault when using pypy 3.9-7.3.11 with pandas 1.5.2

Expected Behavior

no segfault

Installed Versions

python -c 'import pandas as pd; pd.show_versions()'

INSTALLED VERSIONS
------------------
commit           : 8dab54d6573f7186ff0c3b6364d5e4dd635ff3e7
python           : 3.9.16.final.0
python-bits      : 64
OS               : Linux
OS-release       : 6.1.6-200.fc37.x86_64
Version          : #1 SMP PREEMPT_DYNAMIC Sat Jan 14 16:55:06 UTC 2023
machine          : x86_64
processor        : 
byteorder        : little
LC_ALL           : None
LANG             : C.UTF-8
LOCALE           : en_US.UTF-8

pandas           : 1.5.2
numpy            : 1.24.1
pytz             : 2022.7.1
dateutil         : 2.8.2
setuptools       : 58.1.0
pip              : 22.3.1
Cython           : None
pytest           : None
hypothesis       : None
sphinx           : None
blosc            : None
feather          : None
xlsxwriter       : None
lxml.etree       : None
html5lib         : None
pymysql          : None
psycopg2         : None
jinja2           : None
IPython          : None
pandas_datareader: None
bs4              : None
bottleneck       : None
brotli           : None
fastparquet      : None
fsspec           : None
gcsfs            : None
matplotlib       : None
numba            : None
numexpr          : None
odfpy            : None
openpyxl         : None
pandas_gbq       : None
pyarrow          : None
pyreadstat       : None
pyxlsb           : None
s3fs             : None
scipy            : None
snappy           : None
sqlalchemy       : 1.4.46
tables           : None
tabulate         : None
xarray           : None
xlrd             : None
xlwt             : None
zstandard        : 0.19.0
tzdata           : None

[lithomas1]

@nicolaerosia

Thanks for reporting this. I'll try to take a look at this over the weekend.

Do note, though, that PyPy is not one of our officially supported platforms.

I set up the infrastructure to test pandas on PyPy a long while back, but as of right now it basically only tests that compiling pandas on PyPy works (the tests are allowed to fail without failing our CI checks).

[nicolaerosia]

Thank you, I’m aware it is best effort. FYI 1.4.4 works.

[jorisvandenbossche]

(the tests are allowed to fail without failing our CI checks)

It seems that those tests have been segfaulting as well for a long time. It already fails on pytest initialization (reading of conftest?) or collection:

Run ci/run_tests.sh
PYTHONHASHSEED=2937057230
xvfb-run pytest -r fEs -n auto --dist=loadfile --max-worker-restart 0 pandas -m "not slow and not network and not single_cpu"
Segmentation fault (core dumped)

The failure is visible in the CI logs as far back as the logs are still available, and based on the time that the build takes (that is still visible for older builds), it seems to certainly go back to 1.5.0rc0.
Using those timings, that traces it back to #47641 (that's the first PR where the timing drops from ~40min to ~15min).

I can also reproduce this locally, and the gdb backtrace confirms the relationship with Timedelta:

Thread 1 "python" received signal SIGSEGV, Segmentation fault.
0x00007fffb2971aec in __pyx_getprop_6pandas_5_libs_6tslibs_10timedeltas_10_Timedelta_days () from /home/joris/scipy/pandas/pandas/_libs/tslibs/timedeltas.pypy39-pp73-x86_64-linux-gnu.so
(gdb) bt
#0  0x00007fffb2971aec in __pyx_getprop_6pandas_5_libs_6tslibs_10timedeltas_10_Timedelta_days () from /home/joris/scipy/pandas/pandas/_libs/tslibs/timedeltas.pypy39-pp73-x86_64-linux-gnu.so
#11 0x00007fffb296e2f5 in __pyx_tp_new_6pandas_5_libs_6tslibs_10timedeltas__Timedelta () from /home/joris/scipy/pandas/pandas/_libs/tslibs/timedeltas.pypy39-pp73-x86_64-linux-gnu.so
#20 0x00007fffb298aa7d in __pyx_f_6pandas_5_libs_6tslibs_10timedeltas__timedelta_from_value_and_reso () from /home/joris/scipy/pandas/pandas/_libs/tslibs/timedeltas.pypy39-pp73-x86_64-linux-gnu.so
#21 0x00007fffb298df6c in __pyx_pw_6pandas_5_libs_6tslibs_10timedeltas_10_Timedelta_25_from_value_and_reso () from /home/joris/scipy/pandas/pandas/_libs/tslibs/timedeltas.pypy39-pp73-x86_64-linux-gnu.so
#29 0x00007fffb29af130 in __pyx_pw_6pandas_5_libs_6tslibs_10timedeltas_9Timedelta_1__new__ () from /home/joris/scipy/pandas/pandas/_libs/tslibs/timedeltas.pypy39-pp73-x86_64-linux-gnu.so
#41 0x00007fffb2a488bb in __pyx_pw_6pandas_5_libs_6tslibs_10timestamps_10_Timestamp_9__add__ () from /home/joris/scipy/pandas/pandas/_libs/tslibs/timestamps.pypy39-pp73-x86_64-linux-gnu.so
#50 0x00007fffb2abe869 in __pyx_pw_6pandas_5_libs_6tslibs_7offsets_11BusinessDay_5_apply () from /home/joris/scipy/pandas/pandas/_libs/tslibs/offsets.pypy39-pp73-x86_64-linux-gnu.so
#55 0x00007fffb2af0fbd in __pyx_pw_6pandas_5_libs_6tslibs_7offsets_11apply_wraps_1wrapper () from /home/joris/scipy/pandas/pandas/_libs/tslibs/offsets.pypy39-pp73-x86_64-linux-gnu.so
#65 0x00007fffb2ac5166 in __pyx_pw_6pandas_5_libs_6tslibs_7offsets_10BaseOffset_11__add__ () from /home/joris/scipy/pandas/pandas/_libs/tslibs/offsets.pypy39-pp73-x86_64-linux-gnu.so
#72 0x00007fffb2aa010c in __pyx_pw_6pandas_5_libs_6tslibs_7offsets_10BaseOffset_13__radd__ () from /home/joris/scipy/pandas/pandas/_libs/tslibs/offsets.pypy39-pp73-x86_64-linux-gnu.so
#81 0x00007fffb2ac12ee in __pyx_pw_6pandas_5_libs_6tslibs_7offsets_10BaseOffset_41rollforward () from /home/joris/scipy/pandas/pandas/_libs/tslibs/offsets.pypy39-pp73-x86_64-linux-gnu.so

cc @jbrockmendel

[jorisvandenbossche]

And I can also reproduce this with a simple Timedelta call:

>>>> import pandas as pd
>>>> pd.Timedelta(1, unit="D")
Segmentation fault (core dumped)

[lithomas1]

Sorry for the silence here, was traveling over the past day, and thanks for the assistance in debugging Joris.

Something is going wrong in calling the function _ensure_components()

__pyx_t_1 = ((struct __pyx_vtabstruct_6pandas_5_libs_6tslibs_10timedeltas__Timedelta *)__pyx_v_self->__pyx_vtab)->_ensure_components(__pyx_v_self); if (unlikely(!__pyx_t_1)) __PYX_ERR(0, 1034, __pyx_L1_error)

is the code used to call _ensure_components.

However, (struct __pyx_vtabstruct_6pandas_5_libs_6tslibs_10timedeltas__Timedelta *)__pyx_v_self->__pyx_vtab is NULL, which results in a segfault.

[lithomas1]

Maybe the problem is since days is defined by datetime.timedelta here.
https://github.com/cython/cython/blob/69b6107922efe895f83441b4b9d4dd64ff3d46b6/Cython/Includes/cpython/datetime.pxd#L176-L179

Best guess is that cython is getting confused between the cdef'ed days and the Python property days.

Renaming days to days1, seconds to seconds1, and microseconds to microseconds1 fixed it for me.