Addons: 0.8.2, 0.8.1 version is not compatible with Mac OS version < 10.15

[nbowon]

Describe the feature and the current behavior/state.
I'm using Mac OS 10.14.6 version and I cannot install 0.8.2 or 0.8.1 version with pip
Seems 0.8.2 is only compatible with Mac OS version 10.15.x
Could you make 0.8.2 version of tensorflow-addons compatible with Mac OS 10.14.x?

Relevant information

• Are you willing to contribute it (yes/no): no
• Are you willing to maintain it going forward? (yes/no): yes
• Is there a relevant academic paper? (if so, where): no
• Is there already an implementation in another framework? (if so, where): no
• Was it part of tf.contrib? (if so, where): I'm not sure

Which API type would this fall under (layer, metric, optimizer, etc.)

Who will benefit with this feature?
anyone who are using Mac OS 10.13, 10.14 version

Any other info.

[seanpmorgan]

Thanks for the issue! So the reason for this is when we switched builds to GitHub actions we didn't notice that the MacOS environment was running 10.15. Previously we had built for 10.13.

Simple fix just need to copy TF-IO's strategy:
https://github.com/tensorflow/io/blob/master/.github/workflows/build.yml#L81-L82

[seanpmorgan]

@nbowon Could you please try to install tfa-nightly and see if it works. The pip package is still labeled as 10.15 but I'm not sure if pip prevented you from installing it or if the package crashed on older versions.

Re-opening this until it's confirmed to be fixed. Once it is we will create a 0.8.3 patch release.

[nbowon]

I can install tfa-nightly,
when I try
"pip install tfa-nightly"
"tfa-nightly-0.8.0.dev20200125" is installed.
I still cannot install 0.8.2 version
I'm not sure how can I resolve this..
I guess that there is no 10.13 or 10.14 labeled package for 0.8.2 or above version in pypi

[seanpmorgan]

As a short term option you can build Addons from source:
https://github.com/tensorflow/addons#installing-from-source

We'll continue to work to get the packages fixed though. Please follow along with this issue.

@yongtang just FYI it appears that TF-IO packages are being built against 10.15 as well since moving to GitHub actions:
https://pypi.org/project/tensorflow-io-nightly/#files

[yongtang]

@seanpmorgan Thanks! I think the issue happens in setup tools. As python setup.py... creates the package with macosx_10_15 shown up in the filename of whl created. May need to look into setuptools package to see if there is a ways to pass the expected name instead. Otherwise the build may have to happen on macOS 10.13.

[yongtang]

Looks like there is a --plat-name in setup tools that could be used to specify a different platform:

Options for 'build' command:
....
  --plat-name (-p)   platform name to build for, if supported (default: macosx
                     -10.15-x86_64)
  --compiler (-c)    specify the compiler type
....

[yongtang]

@seanpmorgan I created a PR #797 for tensorflow/io. Think it might work on add-on as well.

Still need to find out a way to sanity check to make sure the generated binary works on 10.13.

[nbowon]

@nbowon seanpmorgan Thanks. I'm trying to use it by building from source.
I hope it will be resolve soon.

[yongtang]

@seanpmorgan adding the following line will resolve the issue on addon:

--plat-name macosx_10_13_x86_64

[seanpmorgan]

@nbowon When you get a chance could you try pip installing tfa-nightly and seeing how it works now. Packages look to support 10.13 now:
https://pypi.org/project/tfa-nightly/#files

[nbowon]

@seanpmorgan I can install tea-nightly-0.9.0.dev20200222 version and it is labeled with machos_10_13
looks like it worked.
Thank you