How to handle manylinux1 and py3.4 packages on pypi #411
Description
Since #389 we've begun publishing our nightly images as manylinux2010 which aligns with the TF core packages now produced in this build environment.
Unfortunately, the user is required to have a relatively new version of pip (v19+) in order to install a manylinux2010 package. This means that for many (probably most) users, if they try to pip install tensorflow-addons they will get an out of date version. If TF2 release is only available as a manylinux2010 package this may be less of an issue (ostensibly users will have already updaded pip). But is TF core team aware of the need to inform users to upgrade pip?
One option is to delete all addons manylinux1 packages from PyPi (they're breaking spec anyway).
I wanted to create this issue to discuss if that action makes sense or if we can expect this to be handled when users install TF2?
Our py34 packages will always break with manylinux2010 TF2 so I think the answer there is to delete them.
cc @tensorflow/sig-addons-maintainers @karmel @martinwicke @gunan @yifeif @dynamicwebpaige for visibility and thoughts.