On Wed, 18 Jun 2025, Gábor Németh wrote: > > If normal user code that includes <limits> gets this warning (built with > > an installed compiler), that needs fixing in some way. If not, what is > > different about how GCC uses its own libstdc++? (For example, it's > > important to include system headers with -isystem not -I, does GCC get > > this right when building with the libstdc++ from a previous stage of > > bootstrap?) > > No, normal user code doesn't get the warning, only gcc. On your clue I > checked stage 2 flags though, and it turns out that libstdc++ headers are > normally included with -I, except with -isystem in lean mode. (Indeed, `make > bootstrap-lean` builds fine with my patch 2/2 sans the changes to compile > flags, whereas `make bootstrap` fails without them.) The differentiation > between normal and lean modes were added in [1] as a fix for PR58572 [2], > before which it was always -I since forever [3]. > > I can create a separate patch to use -isystem unconditionally, however there > seems to be a contradiction between what you're stating about its use and what > earlier authors did. Can you advise?
I think it makes no sense for the use of -isystem to be conditional on bootstrap-lean and it should always be used when including what are logically system headers. -- Joseph S. Myers josmy...@redhat.com
