Continuing the drive to TF2

[h-vetinari]

This is an attempt (likely a bit of a hail-mary) to continue #104 by @jschueller, which has grown some conflicts with master. I fixed them as best as I could.

If someone wants to collaborate on this PR, I'll happily grant them push-access to my fork of this feedstock.

[conda-forge-linter]

Hi! This is the friendly automated conda-forge-linting service.

I just wanted to let you know that I linted all conda-recipes in your PR (recipe) and found it was in an excellent condition.

[xhochy]

@conda-forge/tensorflow @h-vetinari @njzjz This is ready for a round of review. I uploaded Linux builds to my channel https://anaconda.org/uwe.korn/repo and will upload osx-* builds later today when they finish. In accordance with CFEP-03, here are the build logs:

• Linux Python 3.6
• Linux Python 3.7
• Linux Python 3.8

[h-vetinari]

@xhochy
Looks very good to me, thanks for taking on the lion's share of the work here! I have some cosmetic improvements that I'd like push (no semantic changes), is that fine for you? I could also open a PR on my fork against this PR, if you want to review first.

[jschueller]

no py39 build ?

[xhochy]

no py39 build ?

I would put that in a follow-up PR as this probably requires a bit more work. Upstream also doesn't support 3.9 as some of the dependencies don't seem to support it.

[xhochy]

@xhochy
Looks very good to me, thanks for taking on the lion's share of the work here! I have some cosmetic improvements that I'd like push (no semantic changes), is that fine for you? I could also open a PR on my fork against this PR, if you want to review first.

Feel free to push, I can review the diff here.

[h-vetinari]

I like having this meta-information in patches; taken from the commits with which you added the "header-less" patches.

[h-vetinari]

Combined both osx-arm64 patches, because there didn't seem to be a strong reason for them to be separate; message taken from the file-name of the patch

[xhochy]

@h-vetinari Thanks, change look fine ✅

[xhochy]

osx-arm64 builds are now also available with the corresponding log: https://gist.github.com/d5076de61632b25d5106219d63643066

[izahn]

no py39 build ?

I would put that in a follow-up PR as this probably requires a bit more work. Upstream also doesn't support 3.9 as some of the dependencies don't seem to support it.

The pkgs/main channel has 3.9 variants for tensorflow 2.4.1, and I don't see any 3.9 specific work-arounds in https://github.com/AnacondaRecipes/tensorflow_recipes, so this might actually "just work". OTOH tensorflow/tensorflow#44485 is still open...

[xhochy]

@conda-forge/core In accordance with CFEP-03, I have built the packages locally for commit ee305da and uploaded them to my channel: https://anaconda.org/uwe.korn/repo The channel solely contains the packages for this build, i.e. everything should be copied over. Please someone from core: upload them and merge the PR afterwards.

Build logs:

• osx_64_python3.6.____cpython
• osx_64_python3.7.____cpython
• osx_64_python3.8.____cpython
• osx_arm64_
• linux_64_python3.6.____cpython
• linux_64_python3.7.____cpython
• linux_64_python3.8.____cpython

For reference: It takes ~48h to build the osx-64 packages in serial. This is really heavy, the build is roughly 1/3 of the time with lief/binary rewrite(?) being the majority.

[isuruf]

I had a look and it looks like the correct CONDA_BUILD_SYSROOT was used.

Thanks for the PR. Since you have access to do upload and merge, please go ahead and upload to conda-forge/main. If you want me to do it, let me know.

[xhochy]

Thanks! Copied and merged. 🎆

[h-vetinari]

Awesome, thanks so much @xhochy! 🥳