https://gcc.gnu.org/bugzilla/show_bug.cgi?id=71446
--- Comment #15 from Jakub Jelinek <jakub at gcc dot gnu.org> --- So, I'm afraid I need to give up on the build_aggr_conv+desig11.C part of the above patch, I really don't understand what C++ says that should be done. I've tried to at least test the remaining part of the patch as incremental improvement, but it regresses g++.dg/opt/pr65554.C. The problem is that various spots in the compiler, e.g. reshape_init, add sometimes ce->index. So, I'll try to use a flag on the CONSTRUCTOR to note if there are any user designators (as opposed to something added by the compiler at some point).
