Re: number->string radix patch

[Kevin Ryde]

Richard Todd <address@hidden> writes:
>
> Those second
> two answers are only different in the least significant numbers, which
> looks like different amounts of rounding error to me.

Neither input nor output code keeps a tight rein on accuracy as far as
I can tell, it loses on decimal I think.

But what's not good is that there's some rounding in hex output, for
instance

        (number->string #i#x1FFFFFFFFFFFF 16)
        => "2.0eC"

where I expected "1.FFFFFFFFFFFFeC" or whatever.  Non-power-of-2 bases
are going to round, but I'd have thought an important point for power
of 2 bases would be that they don't round.

I haven't tried to look where it goes wrong, but this is on my i386
debian, built with CFLAGS=-g (ie. no optimizations).  I added some
printfs into idbl2str and saw the value from the input is correct,
ie. contains lots of 1 bits, the rounding to 2.0 is apparently in the
output.