[jira] [Commented] (ARROW-14526) installing pyarrow breaks the conda environment

["Weston Pace (Jira)" <ji...@apache.org>]

    [ https://issues.apache.org/jira/browse/ARROW-14526?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17436966#comment-17436966 ] 

Weston Pace commented on ARROW-14526:
-------------------------------------

If you have gdb installed can you figure out a bit more information for us?  It should be something like:

{noformat}
gdb --args python -c "import pandas"
...
(gdb) run
...
Program received signal SIGSEGV, Segmentation fault.
...
(gdb) disassemble
...
=> 0x00007ffff7d616a7 <+71>:	???
...
{noformat}

Alternatively, if you can capture and attach a core file that would show the same.  A common cause is an attempt to run some kind of extension (e.g. AVX2) that isn't supported by the processor.  Generally Arrow should detect this and not make such calls but I don't know if we have an Opteron in our CI infrastructure.

> installing pyarrow breaks the conda environment
> -----------------------------------------------
>
>                 Key: ARROW-14526
>                 URL: https://issues.apache.org/jira/browse/ARROW-14526
>             Project: Apache Arrow
>          Issue Type: Bug
>          Components: Python
>         Environment: Conda version: conda 4.10.3
> The issue occurs on RedHat Linux Server running on AMD Opteron Processor 4180. (Works with intel processor)
>            Reporter: Naman
>            Priority: Blocker
>              Labels: pyarrow
>
> I am working with pyarrow for parquet and feather IO. My environment has the following requirements python=3.8, pandas, geopandas, pyarrow, matplotlib, seaborn, jupyter
> You can reproduce the env using
> {code:java}
> $ conda create -n testpyarrow python=3.8 pandas geopandas matplotlib seaborn jupyter notebook pyarrow 
> $ conda activate testpyarrow{code}
> _you can skip pyarrow in the above command and then install it separately from conda-forge as instructed in the pyarrow github (*doesn't make a difference*) page using_
> {code:java}
> $ conda install -c conda-forge pyarrow{code}
> After the env is created and activated, opening a python terminal and try importing to import anything
> {code:java}
> $ python
> >>>import pandas{code}
> yields
> {code:java}
> Illegal instruction (core dumped){code}
> If you uninstall pyarrow, all the imports start working. So I have zeroed down that pyarrow is breaking the imports.
> As it is core dumped I can't trace the issue further. Illegal instruction is raised irrespective of the library I try to import  (i.e. seaborn, matplotlib, geopandas, pandas, etc)



--
This message was sent by Atlassian Jira
(v8.3.4#803005)