License issue with 1.6.0.rc1

[roywei]

Following are the license issue found by mentors during 1.6.0.rc1 vote @ general.

https://lists.apache.org/thread.html/rb83ff64bdac464df2f0cf2fe8fb4c6b9d3b8fa62b645763dc606045f%40%3Cgeneral.incubator.apache.org%3E

I'm compiling this list of issues so the community can help fix them.

• NOTICE incorrectly states "Copyright 2017 and onwards”. This has been pointed out in a number of releases. Is someone going to fix it?
• I can see several files now contain "Copyright 2019 Intel Corporation” and " Copyright (c) 2019 by Contributors” it looks like some 3rd party code have been updated but the LICENSE file has not.
• license for this files [2] is missing and these [9] and these [10]
• files copyright The MathJax Consortium e.g [3]
• file copyright Martin Hensel [4]
• file copyright Bytedance Inc [5] of unknown license
• file copyright NumPy Developers [6] this file look to incorrectly have an ASF header on it
• file copyright Xiao Liu + others [7] of unknown license which also incorrectly looks to have an ASF header on it
• file copyright Manuel Carrasco Moñin [8] which again seems to incorrectly have two headers

• 1. Pybind11 license issue from onnx tensorrt #15547
• 2. ./docs/python_docs/themes/mx-theme/mxtheme/static/material-design-icons-3.0.1/iconfont/MaterialIcons-Regular.*
• 3. ./3rdparty/mkldnn/doc/assets/mathjax/jax/output/CommonHTML/fonts/TeX/SansSerif-Regular.js
• 4. ./3rdparty/mkldnn/doc/assets/mathjax/extensions/TeX/mhchem3/mhchem.js
• 5. ./3rdparty/ps-lite/src/ibverbs_van.h
• 6. ./src/operator/numpy/np_einsum_path_op-inl.h
• 7 ./example/image-classification/predict-cpp/image-classification-predict.cc
• 8. ./tests/nightly/sh2ju.sh
• 9. ./docs/python_docs/themes/mx-theme/mxtheme/static/font/Roboto/Roboto-Regular.*
• 10. ./docs/python_docs/themes/mx-theme/mxtheme/static/webfonts/fa-solid-900.*
• 11. Revisit white list for rat license check.

How to fix them

Here is the guideline from Apache: https://www.apache.org/legal/src-headers.html
Note the copy right and license header requirements are different for 3rdparty files and mxnet source files.

More references on license issues:
https://www.apache.org/legal/resolved.html

[roywei]

@PatricZhao could you help with the "Copyright 2019 Intel Corporation” and 3 and 4 under mkldnn? Thanks

[roywei]

@hzfan @reminisce @haojin2 Looks like we need to remove the ASF header if this file is directly copied from numpy, and add it to whitelist of the header check.
https://github.com/apache/incubator-mxnet/blob/master/src/operator/numpy/np_einsum_op-inl.h

[roywei]

@aaronmarkham any idea how to fix the font files in #2? Can we move it to s3 and download it when building the docs?

[aaronmarkham]

@aaronmarkham any idea how to fix the font files in #2? Can we move it to s3 and download it when building the docs?

I'd research the fonts' licenses, add the appropriate license references, and leave them where they are. Moving to s3 would just add to site latency, and we don't want that.

[leezu]

@hzfan @reminisce @haojin2 Looks like we need to remove the ASF header if this file is directly copied from numpy, and add it to whitelist of the header check.
https://github.com/apache/incubator-mxnet/blob/master/src/operator/numpy/np_einsum_op-inl.h

@roywei The file includes modifications, such as

https://github.com/apache/incubator-mxnet/blob/bd67723da96e6d36e72c9a42535a4fe68f234a71/src/operator/numpy/np_einsum_op-inl.h#L1088

It seems valid to re-license the file under Apache license given that the original license header is included. (Above link suggests this approach is not recommended though)

If we choose not to relicense files with minor modifications, https://github.com/apache/incubator-mxnet/blob/master/cmake/Modules/FindCUDAToolkit.cmake also needs to be updated to remove the ASF header.
The file contains the following minor modifications

https://github.com/apache/incubator-mxnet/blob/bd67723da96e6d36e72c9a42535a4fe68f234a71/cmake/Modules/FindCUDAToolkit.cmake#L713-L715

[roywei]

@hzfan @reminisce @haojin2 Looks like we need to remove the ASF header if this file is directly copied from numpy, and add it to whitelist of the header check.
https://github.com/apache/incubator-mxnet/blob/master/src/operator/numpy/np_einsum_op-inl.h

@roywei The file includes modifications, such as

https://github.com/apache/incubator-mxnet/blob/bd67723da96e6d36e72c9a42535a4fe68f234a71/src/operator/numpy/np_einsum_op-inl.h#L1088

It seems valid to re-license the file under Apache license given that the original license header is included. (Above link suggests this approach is not recommended though)

If we choose not to relicense files with minor modifications, https://github.com/apache/incubator-mxnet/blob/master/cmake/Modules/FindCUDAToolkit.cmake also needs to be updated to remove the ASF header.
The file contains the following minor modifications

https://github.com/apache/incubator-mxnet/blob/bd67723da96e6d36e72c9a42535a4fe68f234a71/cmake/Modules/FindCUDAToolkit.cmake#L713-L715

I'll leave the authors to decide whether to add Apache license if it's modified by MXNet contributors. Either way, we also need to acknowledge the original License and CopyRight in the LICENSE file.

[leezu]

@roywei thanks. When removing the ASF header from incubator-mxnet/cmake/Modules/FindCUDAToolkit.cmake, how to pass the RAT license check? There is no .rat-excludes files in the mxnet repository. How to exclude the file?

[roywei]

@roywei thanks. When removing the ASF header from incubator-mxnet/cmake/Modules/FindCUDAToolkit.cmake, how to pass the RAT license check? There is no .rat-excludes files in the mxnet repository. How to exclude the file?

here it is: https://github.com/apache/incubator-mxnet/blob/master/tests/nightly/apache_rat_license_check/rat-excludes

It's used by CI step here: https://github.com/apache/incubator-mxnet/blob/master/ci/docker/runtime_functions.sh#L1457

More reference and history: https://cwiki.apache.org/confluence/display/MXNET/MXNet+Source+Licenses

[TaoLv]

@roywei Regarding [3], [4], do you think we need list the copyrights/license in MXNet top LICENSE file or ask MKL-DNN itself to change? I noticed that in the latest version of MKL-DNN, it already added MathJax copyrights/license to its LICENSE file. See https://github.com/intel/mkl-dnn/blob/master/LICENSE#L352

[roywei]

@roywei Regarding [3], [4], do you think we need list the copyrights/license in MXNet top LICENSE file or ask MKL-DNN itself to change? I noticed that in the latest version of MKL-DNN, it already added MathJax copyrights/license to its LICENSE file. See https://github.com/intel/mkl-dnn/blob/master/LICENSE#L352

Hi @TaoLv , yes, the existing MathJax is for the docs file, we need to also include mkldnn part(add additional file path that use the same license). Basically for all newly added source files, if there is a different license or copyright statement (different than the Apache contributor copyright) we need to state in the top level license file clearly: which file used what license and what’s the copyright. So we also need to list all files or file paths that has intel copyright