Line 79 reads cmake_parse_arguments(_PCH "FORCEINCLUDE" "SOURCE_CXX:SOURCE_C" "" ${ARGN}) Is it a typo that "SOURCE_CXX:SOURCE_C" should be "SOURCE_CXX;SOURCE_C"? Currently it is a colon ":" but it seems that it should be a semicolon “;”

The mechanism for the fix is from https://cmake.org/pipermail/cmake-developers/2012-March/015380.html.

There's also a fix for an issue I ran into where the Windows precompiled header files didn't work out of the box on Linux.

Main CMakeLists.txt

add_library (${PROJECT_NAME} STATIC ${HEADERS} ${SOURCES})
add_precompiled_header (${PROJECT_NAME} ${ALLOCATORS_INCLUDE_DIR}/pch.h 
${ALLOCATORS_SOURCES_DIR}/pch.c)
target_include_directories (${PROJECT_NAME} PRIVATE ${ALLOCATORS_INCLUDE_DIR})

add_subdirectory (tests)

CMakeLists.txt for tests:

foreach (__TEST__ ${TESTS})
    add_executable (${__TEST__} ${ALLOCATORS_TESTS_DIR}/${__TEST__}.c)
endforeach ()

When I generate the project for VS2017, I will get precompiled headers for tests(other projects). I don't need use precompiled header in the tests. Is there a solution to this problem?

Now different .pch files are used for debug/release/etc in visual studio projects.

Before this change switching from debug/release in visual studio would result in lots of pch related errors until you do a rebuild.

If you rename PrecompiledHeader.cmake to FindPrecompiledHeader.cmake, then you can use FIND_PACKAGE( PrecompiledHeader REQUIRED ) rather than include...

GCC pre-compiled header found in different folder (Idea)

I got gcc pre-compiled headers to work with a regular makefile and tried your code and played with it. The whole forced include is necessary, if one uses #include "header.h. If one uses #include <header.h> instead, the pre-compiled header file can reside in the build folder, if that directory is included as the first one. Even though the documentation says that header.h and header.h.gch must be in the same directory - this is actually not true. I let cmake create a <target>_inc directory, where I let Ninja put the pre-compiled header. The FORCEINCLUDE is no longer necessary.

For gcc in release mode, -DNDEBUG flag is missing when pch is generated.

As the pch is not generated with the same flags, it is not used during the compilation of objects. For example, the following warning appears: precompiled.h.gch/.c++: not used because `NO_INLINE' not defined [-Winvalid-pch]

Could we have the same flags for the pch generation and the compilation?

There are projects which have include files outside the current source directory. For example if the project has an include and an src folder, and the current CMake file is in the src folder.

In these cases, the pch file has to be referenced as ../include/pch.h.

While this works with the FORCEINCLUDE option (it doesn't without that), it also creates files outside the _pch_binary_dir.

I think the best approach to resolve this would be the addition of another optional parameter, PCH_PATH, such as:

add_precompiled_header( target ../include/foo/bar/pch.h PCH_PATH foo/bar/pch.h )

Most likely this parameter should be required if the header path starts with "..", or if it an absolute path.

This would solve both this issue, and the other I reported.

At least for GCC, without the FORCEINCLUDE option, the directory where the pch file is generated is added as an include path.

get_filename_component(_name ${_input} NAME)
// ...
set(_pchfile "${_pch_binary_dir}/${_input}")
// ...
list(APPEND _pch_compile_flags -include "${_pchfile}")

This means that if the pch file is in a subdirectory (e.g. "foo/pch.h"), and that directory itself isn't in the include path, the compiler won't be able to find the gch file, and the original header will be used instead.