use FetchContent instead of submodules

[jmoralez]

Proposes using FetchContent instead of submodules to make it easier to install the library, i.e. avoid issues like #6297

[jmoralez]

@jameslamb do you support this? I'm not sure if we could do this

LightGBM/build-cran-package.sh

Lines 140 to 156 in b27d81e

	# Remove 'region', 'endregion', and 'warning' pragmas.
	# This won't change the correctness of the code. CRAN does
	# not allow you to use compiler flag '-Wno-unknown-pragmas' or
	# pragmas that suppress warnings.
	echo "Removing unknown pragmas in headers"
	for file in $(find . -name '*.h' -o -name '*.hpp' -o -name '*.cpp'); do
	sed \
	-i.bak \
	-e 's/^.*#pragma clang diagnostic.*$//' \
	-e 's/^.*#pragma diag_suppress.*$//' \
	-e 's/^.*#pragma GCC diagnostic.*$//' \
	-e 's/^.*#pragma region.*$//' \
	-e 's/^.*#pragma endregion.*$//' \
	-e 's/^.*#pragma warning.*$//' \
	"${file}"
	done
	find . -name '*.h.bak' -o -name '*.hpp.bak' -o -name '*.cpp.bak' -exec rm {} \;

with FetchContent.

[jameslamb]

@jameslamb do you support this? I'm not sure if we could do this

I'm -1 on this.

Things I'm concerned about:

• I think this would make life harder (relative to using git submodules) for repackagers who build in no-internet connection environments, like the one described in Python bindings fail to build without internet connection #5979
  • since the cloning of these sources happens only at CMake configuration time
  • although I guess they could still generate the CMake source directory outside, then transfer it into the no-internet environment 🤔
• to your point, this change would require additional changes for the not-driven-by-CMake build of the R package for CRAN
  • FetchContent() downloads files to the CMake build directory at configuration time, and there is no CMake build directory involved in running sh build-cran-package.sh
  • we could work around that by using CMake to drive the build of the CRAN-style package, similar to what XGBoost does:
    • https://github.com/dmlc/xgboost/blob/bc516198dccd9fb345ac318c56ee7bb48bba5aae/CMakeLists.txt#L298-L300
    • https://github.com/dmlc/xgboost/blob/bc516198dccd9fb345ac318c56ee7bb48bba5aae/R-package/CMakeLists.txt

These are totally solvable problems, but I don't think they're worth the effort just for the benefit "build works without you needing to init git submodules".

What do you think about creating a script in cmake/ like this:

function(assert_submodules_initialized)
  if(NOT EXISTS "${CMAKE_SOURCE_DIR}/external_libs/fmt")
    message(
       FATAL_ERROR
       "Missing required source directory: 'external_libs/fmt'. Run 'git submodule update --init --recursive' from the root of the repo and re-run cmake"
     )
  endif()
  # ... and then another for each of the other submodules ...
endfunction()

And then sourcing it and calling that function early in CMakeLists.txt?

assert_submodules_initialized()

I think that'd generate a clearer configure-time error than "fast_double_parser.h: no such file or directory", hopefully clear enough to allow people to fix their builds without opening issues here.