Re: Program Computes incorrectly under VPC.

user_at_domain.invalid
Date: 02/18/05 

Date: Fri, 18 Feb 2005 07:17:38 -0500

John McGhie [MVP - Word and Word Macintosh] wrote:

> Hi Rob:
>
> On 17/2/05 23:13, in article eZZqYoOFFHA.3376@TK2MSFTNGP12.phx.gbl,
> "user@domain.invalid" <user@domain.invalid> wrote:
>
>
>>>That's why both offer refunds on the product if it doesn't do what you
>>>need it to do.
>>>
>>>I don't see how, "try this and if doesn't work, we'll take it back" is
>>>negligent.
>>
>>The problem is, a scientist or engineer etc who is using this program is
>>unaware that his results may be wrong. As I previously stated, it gave
>>me a capacity 5 lbs for a particular stud instead of 2000lbs. Granted,
>>this was a mistake on the conservative side, but if it wasn't I could
>>potentially have a structural failure and lives (like your family) could
>>be lost.
>>
>>This IS a serious issue - just like the Pentium issue was. It is not
>>simply that the program is giving warning messages or crashing.
>>
>>Again, Microsoft REALLY needs to make it clear to current and future
>>owners of this software that it should not be used for floating point math.
>
>
> Please... Take a deep breath and consider the illogic of your statement :-)
>
> Microsoft Excel is the dominant spread*** in the world. It is used by
> literally millions of users, many of whom perform very mission-critical
> functions with it.
>
> Its calculations are inaccurate.
>
> Almost NONE of its users know that. The information is right there in the
> Help file. The limits are clearly stated. Yet almost none of the Excel
> users in the world today know that the thing makes rounding errors. Almost
> NONE of the spreadsheets in the world today (billions of them!) have EVER
> been audited to detect such errors.
>
> One of my jobs during the Y2K lead-up was working with the team that checked
> the corporation's spreadsheets for date errors. We were all horrified: not
> only at the number of date errors, but at the number of spreadsheets, the
> serious usage they were being put to, and the fact that absolutely NONE of
> them had ever been audited. Excel's errors paled into insignificance
> compared to the logic errors in those spreadsheets that had never been
> detected.
>
> You can't just put a warning label on something and assume that people will
> understand it.
>
> If it were true that the PowerPC could not be used for floating point
> calculations, you wouldn't be reading this. Your PowerPC and mine are doing
> millions of floating point calculations per second just to draw the displays
> we are using.
>
> You can't say for certain that "All scientific programs will give errors if
> they are used in VPC." It's not true. Most of them won't. Only a tiny
> percentage of algorithms will fail in VPC. Probably fewer will be
> inaccurate in VPC than are inaccurate on either the Mac or a real PC.
> Getting this error in modern software is very rare. But it's still
> possible.
>
> The warning label, to be accurate, would have to be along the lines of "VPC
> emulates a Pentium, however the floating point processor in the Apple can,
> under rare circumstances, produce rounding errors when used for recursive
> operations with a high number of significant digits." If you had read a
> warning like that, you "might" have understood its impact. But I doubt it:
> you have no idea how the software author has coded his algorithm either.
>
> You see, we have no definitive statement yet from the software author that
> this "is" a rounding error. He doesn't know either (or he's not saying...).
> We have yet to eliminate the real possibility that this is simply a "bug",
> which just happens to have become apparent in YOUR calculation, which you
> just happen to have performed on a PowerPC. Chances are quite high that if
> you knew exactly what was happening, you could produce a very similar error
> with this software on a real Pentium.
>
> I think you really might conclude that your position is unsupportable.
>
> It may not have occurred to you, but the reason software companies are
> always less than forthcoming about the limitations of their products -- is
> because of the atmosphere of toxic shame that has been produced by the USA
> legal system. The world would have much better software, and far fewer
> disappointments, if Americans did not automatically start blaming one
> another and heading off to lawyers and courts whenever anything went wrong.
>
> It seems to me that you want to "force" Microsoft to admit liability for,
> and accept blame for, something they didn't do, have no control over, that
> cannot be fixed. That attitude has produced a legal climate in the USA
> where companies dare not admit that "anything" is wrong, because the moment
> they do, some lawyer will try to put them out of business.
>
> If human beings are expected to operate in an environment where they can
> never make a mistake, they will never actually make "anything". This has
> already bought us the Y2K problem and the dot-Bomb stock market bubble. Is
> that enough? Do we have to keep doing this?
>
> Or could we all suddenly say "--it happens. I need to carry a roll of
> toilet paper with me, and WHEN it happens, have a quick check to make sure
> that the a---hole involved is not ME!"
>
> Nah... Maybe not... I have a lawyer friend and she needs the work :-)
>
> Cheers
>
John,

Thanks for all of your information. It has been very enlightening. What
is curious to me is that this is not an extremely complicated algortithm
and it does not iterate a significant amount of times. It just seems
unlikley to me that ANY processor couldn't run it correctly. I could
actually run thru all of its iterative calculations in a half a day on a
calculator. Is there any possibility it is another issue? If in fact
this algorithm won't run correctly natively on a PPC then I
wholeheartedly apologize for all of my ramblings.

-Regards,

Rob

Loading