Re: floating point war casualty list

[Stephen L Moshier <steve at moshier dot net>]

>  Changes in this area are really delicate, and
> could use more careful study before being implemented. Keeping
> floating-point working correctly on all targets is really tricky, since
> very few people know what the floating-point characteristics are on
> all targets.

A compiler company once paid me to track down and replicate
a decimal - to - binary conversion mistake in the Microsoft compiler.
I did so, but after the system was running nicely, generating
deliberately wrong answers, I went to the manager and asked why
in the world would they do this, why didn't they simply get
Microsoft to fix the bug.  Answer, "Microsoft doesn't _want_
to fix the bug!"

I will eventually remember the bug-compatiblity reasons why the DEC
format is done the way it is, but there is no telling how long it will
take to find out why some of those other format conversions are done
the way they are.  It is just not a wise move to disturb this stuff,
because it will take a very long time for the dust to settle down.
The most generous explanation offered for the recent changes is that
somebody was on a cosmetic cleanup campaign.  It would be far better
to let it be, leave it ugly.