Re: [updated] HP-33S failures Message #2 Posted by Karl Schneider on 12 Dec 2006, 6:48 p.m., in response to message #1 by Antonio Maschio (Italy)
Hi, Antonio --
A previous bug list for the HP-33S was posted more than two years ago:
http://www.hpmuseum.org/cgi-sys/cgiwrap/hpmuseum/archv014.cgi?read=62596#62596
Three of the five listed there are the ones mentioned in a recent thread, and were fixed by KinHPo sometime in 2005.
The "Radians" bug linked in the thread is an initialization error. I don't know if that was fixed.
The "Solve" bug (finding only one root) I discovered subsequently was caused by implementing logic directly from the HP-17B/27S solver. I proposed more-sophisticated logic in the same thread:
http://www.hpmuseum.org/cgi-sys/cgiwrap/hpmuseum/archv014.cgi?read=62665#62665
You have also mentioned the estimated error of integration that differed from those of the HP-32SII. I found that the differences resulted from an implementation of the HP48G method of estimation, replacing that of the HP-15C/34C/32S/32SII, although using the same syntax.
You also mentioned the "Rmdr" bug that had been discussed in 2004, Here's a link to that thread:
http://www.hpmuseum.org/cgi-sys/cgiwrap/hpmuseum/archv014.cgi?read=53359#53359
The quick summary is that "Rmdr" functions like "MOD" on the HP-41 and HP-42. "INT/" gives results that are always consistent with those of "Rmdr". However, when the divisor and dividend are of opposite sign, the results from "INT/" are not consistent with integer division on the HP-16C or with regular division on the HP-33S using an integer-only base (HEX, OCT, BIN). For example, in octal mode, try the following:
------ HP-33S ----- -- HP-16C --
divide INT/ Rmdr divide RMD
-31 ENTER 4 -6 -7 3 -6 -1
31 ENTER -4 -6 -7 -3 -6 1
(IMPORTANT: Set the word size to 36 bits and 2's complement on the HP-16C for complete consistency with the HP-33S.)
Finally, there's the trigonometric bug Gerson Barbosa found, which has many manifestations, as have been discussed recently.
So, that's eight operational "flaws" that I recall being discussed. A summary:
- Three (Polar, HMS, Comb/Perm) are bugs that have been fixed.
- "Radians" is an intialization error that may have been fixed.
- "Solve" is not a bug per se, but could be addressed by algorithmic refinements.
- "Integ estimated error" is not a bug, but should be addressed by improved documentation.
- "Rmdr" is actually "MOD"; "INT/" is consistent with MOD ("Rmdr"), but not always with proper integer division.
- "Trigonometrics" is indeed a bug -- fairly minor in most cases, but unacceptable nonetheless.
Consolidating this into a single discussion with links will take some work. Maybe after Christmas...
Regards,
-- KS
Edited: 14 Dec 2006, 1:26 a.m.
|