can't copy tensorflow/python/pywrap_tensorflow_internal.py

[briancheung]

porter@fattire:~/Projects/tensorflow$ bazel-bin/tensorflow/tools/pip_package/build_pip_package /tmp/tensorflow_pkg
Thu Apr 6 17:10:48 PDT 2017 : === Using tmpdir: /tmp/tmp.kCYPVwxjU5
~/Projects/tensorflow/bazel-bin/tensorflow/tools/pip_package/build_pip_package.runfiles ~/Projects/tensorflow
~/Projects/tensorflow
/tmp/tmp.kCYPVwxjU5 ~/Projects/tensorflow
Thu Apr 6 17:10:49 PDT 2017 : === Building wheel
error: can't copy 'tensorflow/python/pywrap_tensorflow_internal.py': doesn't exist or not a regular file

This happens after building tensorflow with bazel and running the binary build by bazel:

bazel-bin/tensorflow/tools/pip_package/build_pip_package /tmp/tensorflow_pkg

pywrap_tensorflow_internal.py doesn't exist in the source directories:

Tensorflow commit f8dce81
Bazel 0.4.5
CUDA release 8.0, V8.0.26, cuDNN 5105
GPU: NVIDIA Titan 6GB

[CharlesShang]

I met the same problem when installing r1.1.
bazel build --config=opt --config=cuda //tensorflow/tools/pip_package:build_pip_package was completed sucessfully.
Then I run find . -name "pywrap_tensorflow_internal.py" in the tf root dir, got nothing.
There's another relevant issue #8078

[CharlesShang]

A workaround:
copy ./tensorflow/python/pywrap_tensorflow.py to ./bazel-bin/tensorflow/python/pywrap_tensorflow_internal.py.
Then I got a different error /tensorflow/python/gen_resource_variable_ops.py doesn't exist or not a regular file

I think it's likely because some bazel compiling rules may be missed in r1.1.

[lzamparo]

+1, experiencing the same issue

Tensorflow commit d48ef8f
Bazel 0.4.5
CUDA release 8.0, V8.0.26, cuDNN 5105
GPU: NVIDIA GTX 1080

The file in question clearly exists:

[zamparol@server tensorflow]$ ls -lh tensorflow/python/pywrap_tensorflow.py
-rw-r--r-- 1 zamparol cllab 2.5K Apr 25 15:47 tensorflow/python/pywrap_tensorflow.py

[lzamparo]

FWIW, I resolved my issue (sol'n provided towards the end of #1334), I'm not sure what the problem is but I was able to build a wheel by doing the following from the root of the repo:

git submodule update --init
bazel clean --expunge
bazel build -c opt --config=cuda --verbose_failures //tensorflow/tools/pip_package:build_pip_package
bazel-bin/tensorflow/tools/pip_package/build_pip_package ~/utils/tensorflow_pkg

Maybe if one of the bazel build steps fails, you need to do bazel clean --expunge, and try again?

[jart]

Does that solve your issue @CharlesShang?

[CharlesShang]

Unfortunately, It failed again.
The same error mssg was prompted out.
I'm on commit 1ec6ed5.

[rogerxujiang]

I had a similar issue and was able to fix it by removing the VERBS support from the configure. This is what (I guess) happened: bazel build with the VERBS support actually failed, but I didn't notice it and it still created the file of 'bazel-bin/tensorflow/tools/pip_package/build_pip_package'. And this file gives me the error.

[CharlesShang]

@jiangxu87 Could you please give a more detailed howto?

[nishalpereira]

@CharlesShang

Was your error solved?
I have the same error. please help

[CharlesShang]

No.
@nishalpereira, I've installed the pre-built binary instead.

[nishalpereira]

@CharlesShang
Thanks for your reply.
Can you tell me how to install the pre-built binary.

[CharlesShang]

This problem is random.
On another machine (same OS, same cuda version, but different GPU), tf can be built successfully. So I just copy the built file to my machine, install it..

[tensorflowbutler]

It has been 14 days with no activity and this issue has an assignee.Please update the label and/or status accordingly.

[tensorflowbutler]

Nagging Assignee: It has been 14 days with no activity and this issue has an assignee. Please update the label and/or status accordingly.

[jart]

Closing due to inactivity. If it's still an issue let us know and I'll reopen.