Re: About the F-22 software bug

[sjw <simon.j.wright@mac.com>]

On Feb 5, 11:39 pm, Phil Clayton <phil.clay...@lineone.net> wrote:

> Interesting to see that 'cyclic fixed point types' are under
> discussion:
>
> http://www.ada-auth.org/cgi-bin/cvsweb.cgi/ai05s/ai05-0175-1.txt?rev=1.2
>
> http://www.ada-auth.org/ai-files/minutes/min-0911.html#AI175

Interesting that those discussions are in terms of cyclic *fixed
point* types.

I agree that at the boundary you need to handle cyclic fixed point.
Back in the day, the Ferranti name for these was 'standard angle
format'; the F1600 series was a 24-bit machine, and the most
significant bit typically represented -180 degrees. Overflow was
represented by an optional check, so wrap-round was as easy as pie.

But why (on a machine with any sort of floating-point capability)
would you want to preserve this anywhere other than at the boundary?

On my present system, angles are stored in radians. Handling wrap-
round (and sometimes we need to deal with a range -2pi .. 2pi) was not
properly dignified with library support from the start, but we've
achieved a generic which lets you normalise an angle to -pi <= a < pi
or to 0 <= b < 2pi or to -2pi <= c < 2pi with minimal pain. We don't
need to worry about latitude!