Install failing from the master: Failed building wheel

[jcwchen]

🐛 Bug

To Reproduce

Steps to reproduce the behavior:

1.pip install -q git+https://github.com/pytorch/vision.git

Expected behavior

Build successfully without any error message

Environment

• PyTorch / torchvision Version (e.g., 1.0 / 0.4.0): master
• OS (e.g., Linux): ubuntu16.04
• How you installed PyTorch / torchvision (conda, pip, source): pip
• Build command you used (if compiling from source): pip install -q git+https://github.com/pytorch/vision.git
• Python version: 3.6
• CUDA/cuDNN version:
• GPU models and configuration:
• Any other relevant information: clang7

Additional context

It occurs a week ago and keeps failing. The error message is as follows:

+ pip install -q git+https://github.com/pytorch/vision.git
  Failed building wheel for torchvision
Command "/tmp/venv/bin/python3.6 -u -c "import setuptools, tokenize;__file__='/tmp/pip-req-build-7wrzlist/setup.py';f=getattr(tokenize, 'open', open)(__file__);code=f.read().replace('\r\n', '\n');f.close();exec(compile(code, __file__, 'exec'))" install --record /tmp/pip-record-bbtn3535/install-record.txt --single-version-externally-managed --compile --install-headers /tmp/venv/include/site/python3.6/torchvision" failed with error code 1 in /tmp/pip-req-build-7wrzlist/

I suppose it's a bug from the master branch? If so, when can it be fixed?
If not, how can I fix this?
Thanks.

[pmeier]

Hi @jcwchen, you left out the traceback which contains the information you need:

Traceback (most recent call last):
      File "<string>", line 1, in <module>
      File "/tmp/pip-req-build-vx375of5/setup.py", line 14, in <module>
        import torch
    ModuleNotFoundError: No module named 'torch'

torchvision depends on torch which has to available before you run the torchvision setup. The master of torchvision requires the master of torch (see compatibility table in the README). You can either install it from source (pip install https://github.com/pytorch/pytorch.git) or go with the pre-built nightly version instead.

[jcwchen]

Hi @pmeier, Thank you for the fast reply.

I did install torch first, but I still encountered this error. The same build code worked a week ago.
Besides, if I pip install from an older version of torchvision (7/6) like:
pip install -q git+https://github.com/pytorch/vision.git@86b6c3e22e9d7d8b0fa25d08704e6a31a364973b
It can work, too. That's why I suppose it might be a bug from master branch.

BTW, where did you get that traceback information? (Knowing this could help debug this)
Thank you.

[pmeier]

I did install torch first, but I still encountered this error.

Huh, that is weird. I did a fresh install with

pip install numpy
pip install --pre torch -f https://download.pytorch.org/whl/nightly/cu102/torch_nightly.html
pip install git+https://github.com/pytorch/vision.git

(remember to adapt the link to your CUDA version) and this is working fine. Could you try this and report your results?

Could you also run

wget https://raw.githubusercontent.com/pytorch/pytorch/master/torch/utils/collect_env.py
python collect_env.py

and post the output? From what you have provided so far the only significant difference between our systems is that I used gcc and not clang. @fmassa Anything comes to mind that could have caused this?

BTW, where did you get that traceback information? (Knowing this could help debug this)

That is simply my output of pip install git+https://github.com/pytorch/vision.git when torch is not installed.

[jcwchen]

The output from collect_env.py is here:

2020-07-17 19:16:54 (86.0 MB/s) - 'collect_env.py' saved [13234/13234]

+ python collect_env.py
Collecting environment information...
PyTorch version: 1.7.0a0+7eb71b4
Is debug build: No
CUDA used to build PyTorch: None

OS: Ubuntu 16.04.6 LTS
GCC version: Could not collect
CMake version: version 3.5.1

Python version: 3.6
Is CUDA available: No
CUDA runtime version: No CUDA
GPU models and configuration: No CUDA
Nvidia driver version: No CUDA
cuDNN version: No CUDA

Versions of relevant libraries:
[pip3] numpy==1.19.0
[pip3] torch==1.7.0a0+7eb71b4
[conda] Could not collect

Originally, I pip install pytorch from their master branch.
I tried your build code, but it still encountered the same error.

I don't have CUDA in my machine and I saw there are some changes in recent PRs of pytorch/vision related to CUDA, will it be an issue?

[pmeier]

Just to recap:

• You've installed torch from source and it is working as expected
• With this setup you can install 86b6c3e22e9d7d8b0fa25d08704e6a31a364973b without errors
• Installing from the current master fails

Did I miss anything?

---

#2388 might be the offender (Cc @andfoy). It added the following to the README:

libpng and libjpeg must be available at compilation time in order to be available. Make sure that it is available on the standard library locations, otherwise, add the include and library paths in the environment variables TORCHVISION_INCLUDE and TORCHVISION_LIBRARY, respectively.

@jcwchen Could you check if libpng and libjpeg are available?

apt list --installed | grep -E "lib(png|jpeg)"

[andfoy]

Hi @jcwchen, could you please try to compile torchvision using pip install -vvv git+https://github.com/pytorch/vision.git@86b6c3e22e9d7d8b0fa25d08704e6a31a364973b or python setup.py develop and post the logs here?

[jcwchen]

Just to recap:

• You've installed torch from source and it is working as expected
• With this setup you can install 86b6c3e22e9d7d8b0fa25d08704e6a31a364973b without errors
• Installing from the current master fails

Did I miss anything?

#2388 might be the offender (Cc @andfoy). It added the following to the README:

libpng and libjpeg must be available at compilation time in order to be available. Make sure that it is available on the standard library locations, otherwise, add the include and library paths in the environment variables TORCHVISION_INCLUDE and TORCHVISION_LIBRARY, respectively.

@jcwchen Could you check if libpng and libjpeg are available?

apt list --installed | grep -E "lib(png|jpeg)"

The recap is correct. I will check whether libpng and libjpeg are available later. (The machine is a CI)

[jcwchen]

Hi @jcwchen, could you please try to compile torchvision using pip install -vvv git+https://github.com/pytorch/vision.git@86b6c3e22e9d7d8b0fa25d08704e6a31a364973b or python setup.py develop and post the logs here?

pip install -vvv git+https://github.com/pytorch/vision.git@86b6c3e22e9d7d8b0fa25d08704e6a31a364973b works without any error message.
I am trying python setup.py develop now.