[
https://issues.apache.org/jira/browse/ARROW-1960?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16312713#comment-16312713
]
Robert Nishihara commented on ARROW-1960:
-----------------------------------------
I actually haven't seen this problem, but I'm sure I will...
Note that importing tensorflow is slow... just timed it on my laptop and it
took 1.27s, so doing this pre-emptively has some drawbacks.
> [Python] Pre-emptively import TensorFlow if it is available
> -----------------------------------------------------------
>
> Key: ARROW-1960
> URL: https://issues.apache.org/jira/browse/ARROW-1960
> Project: Apache Arrow
> Issue Type: Bug
> Components: Python
> Reporter: Wes McKinney
> Fix For: 0.9.0
>
>
> To work around some ABI incompatibility issues with libstdc++ (TF is using a
> newer compiler), we should consider importing tensorflow first if it is
> available before loading Arrow's shared libraries built with the manylinux1
> toolchain. See discussion in
> https://github.com/apache/arrow/issues/1450
> One question is whether TensorFlow's symbols will clash with Arrow's in some
> way that might impact its functioning.
--
This message was sent by Atlassian JIRA
(v6.4.14#64029)
