> On 5 Aug 2025, at 14:31, Simon Richter <[email protected]> wrote: > > On 8/5/25 20:29, Matthias Klose wrote: > >> I'm proposing to do this "transition" inline, doing just no-change binNMUs >> in most cases. long double in the ABI is not that common. > > Isn't that effectively just a repeat of the long double transition for > 64->128 bit? That did affect quite a number of packages IIRC. > > (Also, we can use the opportunity to drop the "ldbl" suffix from > libgmpxx4ldbl and libnewmat10ldbl)
Maybe it does not affect this case, but I was also thinking, long double can be surprisingly common due to projects over-focused on 32-bit x86 and never being fixed properly. For example, boost::math which I had no success in getting fixed to not use long double where it not only makes no sense but causes quite some harm: https://github.com/boostorg/math/pull/1220 (I believe that affects PowerPC just as much as Arm, but I did not investigate much)
