Generally indeed fine and interesting and straightforward.
One thing I'm missing some explanation of why we need MOLD_ROOT and a second find_path. Quick text with an example directory structure (similar to how you wrong examples of accessing components of MOLD_BIN) could clarify things a lot.

In D13807#366516, @Brecht Van Lommel (brecht) wrote:

I had not heard about this, and confirm the speedup, this is great..

For reference, this configuration works with Clang on Linux (the WITH_LINKER options are GCC only at the moment).

CMAKE_EXE_LINKER_FLAGS:STRING=-fuse-ld=/home/brecht/tools/mold/mold

Good to know, I looked into clang support but -fuse-ld in the C/C++ flags was giving many unused-argument warnings, CMAKE_EXE_LINKER_FLAGS works without warnings (we could check on using this for GCC too - if there are no down-sides).

Updated with Clang support on Linux, similar logic should apply to macOS.

In D13807#366524, @Campbell Barton (campbellbarton) wrote:

Good to know, I looked into clang support but -fuse-ld in the C/C++ flags was giving many unused-argument warnings, CMAKE_EXE_LINKER_FLAGS works without warnings (we could check on using this for GCC too - if there are no down-sides).

Because linker is not used during compilation (;
string(APPEND CMAKE_CXX_FLAGS " -B \"${MOLD_BIN_DIR}\"")
This would recompile the entire codebase even though ideally only link step should be affected. Even if linker flags are needed for static libraries like libbf_blenfont.a, suggest using CMAKE_STATIC_LINKER_FLAGS.

Updated with Clang support on Linux, similar logic should apply to macOS.

https://github.com/rui314/mold/issues/245
mold is pre-alpha on macOS and doesn't get past this -dynamic error.

lld + darwinnew flag would need llvm toolchain

• which might affect compiler detection in different generators
• which is huge

Only viable option for mac is zld to my knowledge. Have been using it for months now and cuts link time in half.
https://github.com/michaeleisel/zld

EDIT the state of mac shouldn't block this patch.. I can propose that in a separate one.

In D13807#366538, @Ankit Meel (ankitm) wrote:

In D13807#366524, @Campbell Barton (campbellbarton) wrote:

Good to know, I looked into clang support but -fuse-ld in the C/C++ flags was giving many unused-argument warnings, CMAKE_EXE_LINKER_FLAGS works without warnings (we could check on using this for GCC too - if there are no down-sides).

Because linker is not used during compilation (;

Right, it's just that it's how it's done with GCC which doesn't raise warnings, perhaps we should be using CMAKE_*_LINKER_FLAGS instead everywhere.

If anyone knows which of the (CMAKE_{EXE/SHARED/STATIC/MODULE}_LINKER_FLAGS) are important to set, it would be good to know. Setting all of them gives errors with mold but I didn't dig into this further.

https://developer.blender.org/diffusion/B/browse/master/CMakeLists.txt;19a62203089ffa3e8866a1989aefef7774744aca$1294-1295
Only EXE I think. MODULE would affect bpy.so
intermediate static archives don't use a linker as per https://cmake.org/cmake/help/latest/command/add_link_options.html so that's out
and we don't build shared libraries so that's also out.