On Wed, 17 Jan 2024 at 00:31, Jason Merrill wrote: > > On 1/10/24 04:22, Ken Matsui wrote: > > +/* Return true if T is an integral type. With __STRICT_ANSI__, __int128 > > and > > + unsigned __int128 are not integral types. */ > > This really needs a rationale, since they are actually integer types. I > know __int128 is considered an extension rather than an extended integer > type under the standard, but is there a writeup we can point to for why? > > And even if we don't want to subject it to all the standard requirements > of an extended integer type, why not still say it's an integral type? > flag_iso is only supposed to disable features that could conflict with > obscure but standard-conforming code, and since __int128 is in the > reserved namespace, I'd think it should be safe to support (to the > degree that we do) regardless of flag_iso.
The reason for __int128 not being an integral type is because the standard says that intmax_t must be the largest standard or extended integer type, and intmax_t is fixed by ABI to be a 64-bit type. As a result, GCC has historically said that 128-bit integer types are not part of the "standard or extended integer type"classification, in Joseph's words they're sui generis types. But C2x and C++23 changed this, and now we can just do the obvious, simple thing and say that 128-bit integer types are integer types. This changed with https://cplusplus.github.io/LWG/issue3828 for C++. So we can remove the dependency on __STRICT_ISO__ for 128-bit integer types, and implementing std::is_integral with a built-in seems like the perfect time to do that. But that seems like stage 1 material, as we need to go through the library and see what needs to change.