Re: Ariane 5 Failure from 1996

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

On Jul 10, 4:04 pm, Martin <martin.do...@btopenworld.com> wrote:

> My understanding of the Ariane pretty slim but if Wikipedia is
> accurate (http://en.wikipedia.org/wiki/Ariane_5) then it seems likely
> that the same error would have occurred, as my reading is that it was
> the deliberate removal of Ada checks that led to the 32-bit Float
> being assigned to a 16-bit value, i.e. that part was C-in-Ada-syntax.

In the Ariane 4 flight profile the float couldn't have overflowed the
16-bit target (I think the physical value concerned was the ground
velocity) so they decided they could save cycles by not checking for
potential overflow before assigning and/or not setting up an exception
handler (these protections were retained elsewhere).

In Ariane 5 the flight profile was such that overflow could, and did,
occur, resulting in an exception (presumably a constraint error).. I
don't exactly what is said to have happened after that, there was
mention of a subsystem that shouldn't have been running after launch
feeding diagnostic data in place of real data to a higher subsystem
which shut itself down ...

So, in C, the overflowed data might well not have resulted in the
exception which doomed the rocket. As someone below has said, it's
quite possible that something else would have gone wrong!