Re: Building a different GCC version in SDK vs GCC used to build the rootfs/image

Khem Raj

On 6/10/21 7:57 AM, greghwang@... wrote:
The backstory is I'm supporting a legacy system built with the Jethro branch and gcc 5.x.  I'm wondering instead of upgrading the underlying yocto layers to a more recent branch like Thud which supports gcc 8.2, is there a way to leave the image built with gcc 5, but build a new cross compiler toolchain use gcc 8.2.  The application team I'm supporting really wants to use C++17 features, and is unable to with gcc 5.x as their crosscompiler.  I would try to put the 2.28 glibc and 8.2 gcc-runtime libraries on the image and rely on forward compatibility of the libs to ensure compatibility with gcc 5.  Otherwise, the image would still be using gcc-5 as the compiler for all the recipes.
I did some experiments on Thud poky and was able to get a different version of gcc in the toolchain by modifying PREFERRED_VERSION_gcc_corss-canadian-${TRANSLATED_TARGET_ARCH}, however I quickly ran into problem trying compile hello-world with g++.  It looked like the compiler was trying to find its header files in
sysroots/aarch64-oe-linux/usr/include/c++/<gcc version>
but now I have a mismatched gcc version from what is on the sysroot and the cross compiler can't find it's headers.
I'm wondering if forcing a different version of gcc in the SDK is even feasible, or maybe I should be looking at external toolchains or some other solution to support my application team's need for an upgraded version of gcc.
I think this will be hard to support unless your applications are self contained and will bring its own runtime as well in that case you can build a SDK from newer releases which matches compiler you need and let them use it, if you want to mix and match then it will require a bit of extra work e.g. check if runtime from newer compiler will still work with older libs/binaries or find a way to package two versions seprately and ensure that existing packages on image can keep using older libs

So usually upgrading to relevant yocto release would be cleanest way to move forward and would be most reliable of all options but if thats not an option then
Perhaps upgrading compiler across the board might be better approach

Join to automatically receive all group messages.