Cmake optimization with warp2

[zdenop]

• Don't use GLOB in CMake files - fixes RFC: Don't use GLOB in CMake files #4442
• Optimization to improve build speed:
  • Precompiled Headers (PCH) - Enabled by default via ENABLE_PRECOMPILED_HEADERS=ON
  • Unity Builds (Jumbo Builds) - Disabled by default via ENABLE_UNITY_BUILD=OFF. Can reduce compilation time by 30-60% but may increase memory usage
  • ccache Integration - Enabled by default via ENABLE_CCACHE=ON
  • Ninja Job Pools - Enabled by default via ENABLE_NINJA_POOL=ON
• fix several CMake macos issues

Usage Examples

cmake --build build  -G Ninja -DCMAKE_BUILD_TYPE=Release -DENABLE_UNITY_BUILD=ON
cmake --build build --config Release --target install

[egorpugin]

What is warp2?

[egorpugin]

It is better to put any code changes outside of this CMake PR.

[zdenop]

macOS CMake CI failed without these non-CMake changes. This is reason why they are in this PR.

[zdenop]

Regarding warp see https://www.warp.dev/

[egorpugin]

What is the purpose of all this?

Someone says GLOBs = bad?
I say GLOBs = good.

What about the rest?
Need of unity/pch/ccache/job pools?

[amitdo]

@zdenop,

Since this PR was merged, the macOS+GCC CI job fails.

Could you fix the issue?

[zdenop]

@amitdo: this PR was merged on Aug 15

GA started to fail after Sep 17:

This issue seems specific to the combination of macOS, CMake, and GCC; other configurations such as macOS with CMake and Clang, or Ubuntu with CMake and GCC, do not show the problem.

I do not have access to a macOS device for debugging and testing this scenario.

Additionally, I noticed that the autotools-macos uses only Clang as the compiler on macOS. Perhaps we should consider skipping GCC builds in the CMake workflow for macOS as well.

[zdenop]

Seems like it if fixed.