Yup, -fuse-ld=mold solves the linker issue, and the sample code runs. Great! S.
On Mon, Oct 27, 2025 at 3:48 PM Tobias Burnus <[email protected]> wrote: > Hi Patrick & Salvatore, hi all, > > Patrick Begou wrote: > > I met this problem many times on a really large application. > > Can you (plural) try the following workaround? -fuse-ld=mold ? [1] > > If it works, please report back. (It seems to handle the reduced testcase.) > > * * * > > For Salvatore's original issue: He provided me with a large testcase, > which took a while to reduce. [2] > > → https://gcc.gnu.org/PR122432 > > Tobias > > [1] 'mold' is an alternative linker; you need a newer version than end > of Feb 2024; distros also package this linker; should be enough to be in > the $PATH > > [2] I used > > https://gcc.gnu.org/wiki/A_guide_to_testcase_reduction#Legacy_reduction_tools, > > including to reduce the number of .o files in the linked-in archive. The > testscript not only build it, but also run it checking for the expected > run-time error. A large library, polymorphic types and a complex > hierarchy didn't speed it up, but at the end it was MPI's use of common > symbols that caused the issue for some reasons. > >
