[BUILD IMPROVEMENT] Deprecated installing external projects using cmake "ExaternalProject_Add"

[yiakwy-enterprise-roborock]

Argument: this is a not a good practice and the operation might face many practical problems

1. We don't to have to clone the whole tensorflow project, only the released one. The URL specified in cmake/TensorflowBase.cmake is incorrect! (git repo ssh address with a shallow depth). Or we can just use URL key word to fetch the released tar.gz files.

2. native cmake does not support curl with https. And there might be other factors like git cache. proxy server causing permanent hung In command line

3. User can provide the released source code of tensorflow without downloading it.

A temporal work around

Solution 1

change "cmake/TensorflowBase.cmake", comment the lines :

ExternalProject_Add(
tensorflow_base
# GIT_REPOSITORY https://github.com/tensorflow/tensorflow.git
# GIT_TAG "${TENSORFLOW_TAG}" 
# TMP_DIR "/tmp"
# STAMP_DIR "tensorflow-stamp"
# DOWNLOAD_DIR "tensorflow"
SOURCE_DIR "${TENSORFLOW_SOURCE}"
BUILD_IN_SOURCE 1
UPDATE_COMMAND ""
CONFIGURE_COMMAND ""
BUILD_COMMAND "${CMAKE_CURRENT_BINARY_DIR}/build_tensorflow.sh"
INSTALL_COMMAND "${CMAKE_CURRENT_SOURCE_DIR}/cmake/copy_links.sh" bazel-bin
)

Solution 2

Or equivalently, copy the generated "build/build_tensorflow.sh" to the source folder of downloaded tensorflow.

Conclusion

Drop the ExternalProject_Add and using shell commands, execute_process to do the relevant jobs.

[FloopCZ]

Hi, yes, the current way of downloading the whole repository is very slow. #214 should do the trick.

1. Not sure about what is wrong with GIT_URL though, it seems just fine to me?
2. Fortunately, it is not the case for the supported systems (Ubuntu 18.04+ and Arch Linux).
3. I would rather not let the user do this unless she knows what is she doing and in that case it is not that hard to edit the CMake files.