19:46:37jcowananyway, I have two new concerns: I think 8-bit and indeed 16-bit subtypes of string are indeed important, since a huge fraction of all text is within the Latin-1 repertoire and almost all of it is in the Plane 0 repertoire
19:46:48jcowanI am not talking about UTF-8 or UTF-16 encoding here at all
19:48:55jcowansecond, I find the idea of 27-bit single-floats on 32-bit platforms to be a bad one, for two reasons: (a) rounding from a 32-bit computed result to a 27-bit stored one will have to be performed carefully and without hardware assist;
19:49:19jcowan(b) I think it will be very surprising that arithmetic on arguments obtained from specialized float32 arrays (which are mentioned in th array chapter) will be done only to 27-bit precision.