Since employing ccache on our CI server, we find that the bottleneck in terms of build time is now our static analysis pass, that uses clang-tidy, among other tools. Does anyone know of a way to accelerate clang-tidy in a similar way to how ccache does so with a regular compiler?
Thank you for the question and everybody for the answers.
After struggling to make ejfitzgerald/clang-tidy-cache working w/o any decent documentation or help I've given a try matus-chochlik/ctcache. It's indeed much easier in installation and configuration.
Here's a snippet we use to automatically use it with cmake when it's installed
find_program (CLANG_TIDY_CACHE_PATH NAMES "clang-tidy-cache")
if (CLANG_TIDY_CACHE_PATH)
find_program (_CLANG_TIDY_PATH NAMES "clang-tidy" "clang-tidy-15" "clang-tidy-14" "clang-tidy-13" "clang-tidy-12")
# Why do we use ';' here?
# It's a cmake black magic: https://cmake.org/cmake/help/latest/prop_tgt/LANG_CLANG_TIDY.html#prop_tgt:%3CLANG%3E_CLANG_TIDY
# The CLANG_TIDY_PATH is passed to CMAKE_CXX_CLANG_TIDY, which follows CXX_CLANG_TIDY syntax.
set (CLANG_TIDY_PATH "${CLANG_TIDY_CACHE_PATH};${_CLANG_TIDY_PATH}" CACHE STRING "A combined command to run clang-tidy with caching wrapper")
else ()
find_program (CLANG_TIDY_PATH NAMES "clang-tidy" "clang-tidy-15" "clang-tidy-14" "clang-tidy-13" "clang-tidy-12")
endif ()
set (CMAKE_CXX_CLANG_TIDY "${CLANG_TIDY_PATH}")
With it, if clang-tidy-cache is in the path, it will be used together with any found clang-tidy.
During the build, we define CTCACHE_DIR and preserve them between builds.
In our CI it caused reducing the linting time from 330 minutes to 22, so it's almost 15x faster!