C++ API Build instructions result in an error

[seemethere]

🐛 Bug

To Reproduce

Following the current instructions in the README to build the C++ API result in an error.

Repro:

mkdir build
cd build
# Add -DWITH_CUDA=on support for the CUDA if needed
cmake ..
make
make install

Resulting Error:

CMake Warning at CMakeLists.txt:13 (find_package):
  By not providing "FindPython3.cmake" in CMAKE_MODULE_PATH this project has
  asked CMake to find a package configuration file provided by "Python3", but
  CMake did not find one.

  Could not find a package configuration file provided by "Python3" with any
  of the following names:

    Python3Config.cmake
    python3-config.cmake

  Add the installation prefix of "Python3" to CMAKE_PREFIX_PATH or set
  "Python3_DIR" to a directory containing one of the above files.  If
  "Python3" provides a separate development package or SDK, be sure it has
  been installed.


CMake Error at CMakeLists.txt:14 (find_package):
  By not providing "FindTorch.cmake" in CMAKE_MODULE_PATH this project has
  asked CMake to find a package configuration file provided by "Torch", but
  CMake did not find one.

  Could not find a package configuration file provided by "Torch" with any of
  the following names:

    TorchConfig.cmake
    torch-config.cmake

  Add the installation prefix of "Torch" to CMAKE_PREFIX_PATH or set
  "Torch_DIR" to a directory containing one of the above files.  If "Torch"
  provides a separate development package or SDK, be sure it has been
  installed.


-- Configuring incomplete, errors occurred!

Expected behavior

Build completes successfully

Environment

Please copy and paste the output from our
environment collection script
(or fill out the checklist below manually).

You can get the script and run it with:

wget https://raw.githubusercontent.com/pytorch/pytorch/master/torch/utils/collect_env.py
# For security purposes, please check the contents of collect_env.py before running it.
python collect_env.py

PyTorch version: 1.5.0
Is debug build: No
CUDA used to build PyTorch: 10.2

OS: Ubuntu 18.04.3 LTS
GCC version: (Ubuntu 7.4.0-1ubuntu1~18.04.1) 7.4.0
CMake version: version 3.10.2

Python version: 3.8
Is CUDA available: No
CUDA runtime version: Could not collect
GPU models and configuration:
GPU 0: Quadro GP100
GPU 1: Quadro GP100

Nvidia driver version: 418.116.00
cuDNN version: Could not collect

Versions of relevant libraries:
[pip] numpy==1.18.1
[pip] torch==1.5.0
[conda] blas                      1.0                         mkl
[conda] mkl                       2020.0                      166
[conda] mkl-service               2.3.0            py38he904b0f_0
[conda] mkl_fft                   1.0.15           py38ha843d7b_0
[conda] mkl_random                1.1.0            py38h962f231_0
[conda] pytorch                   1.5.0           py3.8_cuda10.2.89_cudnn7.6.5_0    pytorch-test

Additional context

cc @chauhang

[jinuhwang]

Check your cmake version.
What you are missing is FindPython3.cmake module.
Upgrading cmake to the latest version fixed my issue.

[fmassa]

@seemethere does updating CMake fixes it for you?

[Edwardmark]

@fmassa what if I do not want to updating CMake?

[fmassa]

@Edwardmark good question, and I don't know the answer. @bmanga would you know a solution for this?

[bmanga]

It seems that the first cmake release to have the module is cmake 3.12, which is effectively ahead of what ships in Ubuntu 18.04 LTS (3.10).
The question is whether it whether it is worth supporting 3.10, especially given how easy it is to just upgrade cmake (it has prebuilt binaries on the website, and there are pip and conda packages available). But I also understand that in some circumstances it's difficult to upgrade.

@Edwardmark there is the possibility of building and using cmake without actually installing it, but I'm not sure that addresses your concern either.

@fmassa I can look into backporting the FindPython3 module if you think it's worth it.