Post #4365579

@vitaut@mastodon.social I argue the implementations that try to preserve those as `long double` are the real whacky actors here. The "we're going to give the hardware a few more bits to make implementing transcendentals easier" seems like a reasonable idea