Description

samsja

opened on Sep 6, 2023

Issue body actions

Context

Tensorflow now supports proto 4 https://github.com/tensorflow/tensorflow/blob/master/requirements_lock_3_10.txt#L314.

We should bump tensorflow version in the pyproject.toml to latest version that support proto4

Activity

1. samsja

   added this to DocArray backlogon Sep 6, 2023
2. samsja

   converted this from a draft issue on Sep 6, 2023
3. JohannesMessner

   changed the title [-]Bump the lates tensorflow version[/-] [+]chore: bump to latest tensorflow version[/+] on Sep 6, 2023
4. JohannesMessner commented on Sep 6, 2023

   

   JohannesMessner

   on Sep 6, 2023

   Member

   More actions

   I don't think we have a tf version specified at all right now, we just install a certain version in the test CI.
   If we now specify this new version that would imply dropping support for all older versions, right?
5. samsja commented on Sep 6, 2023

   

   samsja

   on Sep 6, 2023

   MemberAuthor

   More actions

   If we now specify this new version that would imply dropping support for all older versions, right?

   yes

   I don't think we have a tf version specified at all right now
   We don't have a tensorflow version specified because proto version was caped on tensorflow.

   This would allow us to have proper dependency management for tensorflow
6. JoanFM commented on Sep 6, 2023

   

   JoanFM

   on Sep 6, 2023

   Member

   More actions

   But here u are proposing only to cap the TF version to a minimum version? If we do not have any reason not to support the previous versions, what would be the benefit?
7. samsja commented on Sep 6, 2023

   

   samsja

   on Sep 6, 2023

   MemberAuthor

   More actions

   But here u are proposing only to cap the TF version to a minimum version? If we do not have any reason not to support the previous versions, what would be the benefit?

   benefit would be that tensorflow could be registered as a dependency in docarray.

   As of today no package manager know that docarray can have tensorflow has a dependecy
8. JohannesMessner commented on Sep 6, 2023

   

   JohannesMessner

   on Sep 6, 2023

   Member

   More actions

   But here u are proposing only to cap the TF version to a minimum version? If we do not have any reason not to support the previous versions, what would be the benefit?

   benefit would be that tensorflow could be registered as a dependency in docarray.

   As of today no package manager know that docarray can have tensorflow has a dependecy

   That seems more like a benefit to our dev workflow rather than a user benefit.
   Yes, it would be nice for them to do pip install docarray[tf], but in return making it incompatible with every but the very newest tf version doesn't seem like a good tradeoff to me.
9. samsja commented on Sep 6, 2023

   

   samsja

   on Sep 6, 2023

   MemberAuthor

   More actions

   well not only our dev workflow but the dev workflow of our user ...

   But yeah I see the argument, lets keep it like this then, but in 6 month probably we should reconsider

   Reacted by Johannes Messner

Issue actions