>> This seems like an incomplette binfo comming form obj-C++ FE. Martin, do you have some idea? > > Likely PR45580 which I just checked in a patch for. Yes, that fixed the regression. :-) Thanks a lot