Re: Paul Sephton's article ...

[Michel Hack (1-914-784-7648) <hack@xxxxxxxxxxxxxx>]

I think Nick is way too generous!
> so clueless that the terms 'right' and 'wrong' don't really make sense.

Definitely.

> Most of the individual statements are correct, but the gestalt is twaddle.

I didn't count "individual statements", but several very specific ones are
clearly wrong.  He forgets to count the hidden bit, for example, and seems
ignorant of the variable effective precision when viewed in decimal (from 15
to 17 digits).  The most serious error of course is that he does not seem
to realise that rounding occurs in binary, not in the decimal approximation.
(Actually, later the author does mention 53 bits in the context of "full 15
decimal digits".)

The focus on "consistently appplied at decimal 15" is the root of the
misunderstanding.  The author really ought to find out about DFP!

Btw, I wonder what happened to my glibc bug report #3479.  The glibc
conversions are generally quite good, correctly rounding for arbitrary
decimal precisions -- except for decimal->binary of near-half-exact
numbers with lots of digits.  (That bug was also assigned to Ulrich
Drepper, who is indeed the primary author of the relevant glibc pieces.)

I looked at Septhon's glibc bug report #4943 -- sad story...

I also looked at mine -- it's still marked as "New".  I wonder why
it has not been "accepted"...  Am I supposed to take another step?

Michel.
Sent: 2007-10-03 16:46:09 UTC