When using Ninja build system, CMake can specify job pools for number of concurrent compilers
and concurrent linkers. This PR employes the heuristic of max number of compilers as available
physical ram / 1 GiB and a single linker instance to prevent using excessive RAM. Modern
linkers are multithreaded and should not be run concurrently due to risk of resource exhaustion.
clang uses considerably less memory on Linux allowing more concurrent instances than with g++.
Using clang reduced build times on a machine from 63m to 32m.
Brings over the functionality in https://github.com/FreeCAD/FreeCAD-Bundle
to build Linux AppImage, macOS .dmg, and Windows 7z releases.
This version also creates a tagged release for each build, creating an
archive of the weekly builds as well as the ability to easily tie each
build to a git commit. This will make running 'git bisect' easy to
identify sources of regressions.
Python is not provided with debug libraries, so debug builds are not possible.
Building as RelWithDebInfo will build debug information for the binary, while
using the available release libraries.
Modify CMakePresets.json to use -DCMAKE_EXPORT_COMPILE_COMMANDS:BOOL=ON
in all the presets.
The compile_commands.json file can be used by tools like clangd to get
autocompletion and go-to-definition.
This patch also adds compile_commands.json to the .gitignore, as often
you need to symlink the compile_commands.json from the build directory
into the source directory for it to be picked by tools. That is the case
with clangd at least.
Jacob Oursland