Some ways to improve the build speed of kvrocks

[jihuayu]

While working on the kvrocks code, I need to repeatedly compile the program in order to debug my code.
I generate the makefile using ./x.py build, and then I use make kvrocks in the build folder to complete the subsequent build.

I noticed that a amount of time is being spent linking the kvrocks file. In typical cases where only one file is being modified, the default configuration takes around 55 seconds to complete.

The build speed is quite slow. I tried some other approaches and found that by disabling the cmake parameters ENABLE_STATIC_LIBSTDCXX and ENABLE_IPO, the build time decreased to 12s, with a linking time of 5s. This has resulted in a significant improvement.

Then, I change the cmake script, replacing the default linker with mold. This resulted in a reduced build time of 8 seconds, with the linking time even less than 1 second. This is truly awesome.

I have make a table below. We can observe and compare the time taken under different circumstances.

Option	Build Time(s)	Link Time(s)
Complier GCC + Default config	55	48
Complier GCC + Disable ENABLE_STATIC_LIBSTDCXX and ENABLE_IPO	12	5
Complier GCC + Disable ENABLE_STATIC_LIBSTDCXX and ENABLE_IPO + Use mold	8	1
Complier Clang + Disable ENABLE_STATIC_LIBSTDCXX and ENABLE_IPO + Use mold	6	1

I believe I can submit a pull request to add support for MOLD's cmake parameters.
Additionally, I suggest creating a documentation guide to help new developers have a more comfortable experience developing kvrocks.

I would like to ask if there are any other best practices that everyone can share?

[git-hulk]

@jihuayu Thanks for your great proposal. I think it's unnecessary to enable the IPO and static link the libc when building with the debug mode. @apache/kvrocks-committers

[PragmaTwice]

If you want to have a better build performance (including incremental build), make sure you are using ninja, clang, and lld (or mold).

You can use lld easily via adding -fuse-ld=lld to cxxflags.

[jihuayu]

Are you saying that we can use the following command directly to use mold?
./x.py build -j14 -DENABLE_STATIC_LIBSTDCXX=OFF -DENABLE_IPO=OFF --compiler=clang -DCMAKE_CXX_FLAGS="-fuse-ld=mold"

I have found some issues when using mold with ninja and clang. I will continue my research in the coming days.