On Fri, 12 Sept 2025 at 12:51, Cassio Neri wrote: > > The current implementation of generate_canonical yields NAN for float16_t and > fails to compile for float128_t on platforms where long double is IBM-128.
Hi Cassio, The standard doesn't require anything in <random> to work with extended floating-point types, only with float, double and long double. If making 16-bit FP types work is doable without too much effort, that seems fine, but supporting std::float128_t for IBM double-double targets is not a good use of anybody's time. Nathan is in the middle of rewriting generate_canonical, the latest iteration of the patch is at https://gcc.gnu.org/pipermail/gcc-patches/2025-September/695230.html
