On Mon, Sep 21, 2026 at 8:50 PM Andrea Pinski <[email protected]> wrote: > > On Mon, Sep 21, 2026 at 5:46 PM Matt Turner <[email protected]> wrote: > > > > On Mon, Sep 21, 2026 at 8:30 PM Andrea Pinski > > <[email protected]> wrote: > > > > > > On Mon, Sep 21, 2026 at 5:12 PM Matt Turner <[email protected]> wrote: > > > > > > > > Several blocks of longlong.h do not build with clang, because clang > > > > rejects casts on asm output operands. This affects libgcrypt, which > > > > imports longlong.h, and wide-int.cc in GCC itself. > > > > > > Again NO. Go talk with libgcrypt to pick up the GMP version of > > > longlong.h. > > > > I don't see where you said NO before. > > > > I also think you must have missed my email about who is the upstream. > > Here's a link to it: > > https://gcc.gnu.org/pipermail/gcc-patches/2026-September/732164.html > > Again libgcrypt is wrong when it comes to longlong.h. Though you > could say there are 2 different versions of longlong.h now. The GMP > longlong.h is the portable upstream version. While the one in GCC is > the GCC specific version.
I still don't think you read my email? The GMP longlong.h cannot be imported into libgcrypt without a licensing change. You don't even have to care about libgcrypt for this series. PATCH v3 4/4 should be enough to justify the core change. I don't understand why you're responding so reflexively.
