HP 35S - long live the RPN!

[Dirk E.]

Two days ago I found an emulator not asking for any key, registration or payment on an HP support site, yesterday this page was gone; only a few Q&A left explaining things findable in the manual (or on MoHPC) with slightly more effort. Just three days ago I had received my 'spare' 35s (you never know), that obviously someone had returned to the store (it has the known bugs, otherwise it's ok, no different rev.). My hope is still to become a happy WP 31s and WP 34s owner, maybe even a WP 43s?

Like others here I find myself using the 35s more and more despite it's size and shortcomings of which most annoying to me is that some RPN function positions moved against the 15C e.g. (RCL STO Rdn x<>y no more around ENTER, but a "()" primary function key that does nothing in RPN etc.). Many functions require 3 keystrokes to be reached, that's similiar to "the other" non-graphing programmable scientific calculator fx-5800P. But I doubt that I'll afford to 'teach' the Casio RPN; it's possible, but you can save the programs only on paper or textfile via PC keyboard. Another point is that the hinge for the front cover will likely break soon (as it did on my first unit), so it's not to take-everywhere.

At this place I'd like to vote for more RPN use - many posts refer to 35s EQN or ALG use, but - the bugs aside - in programs they're much slower, and if you take the time to note down at least the important things for a program, there's far less trouble later and an RPN keystroke program doesn't have to look more complicated. To me these endless round bracket chains are more confusing.
Tony Thimet published his version of a performance test with an explicit program listing for HP 35S. The test considered 34 "substantial operations"; if it finishes in less than 5 s, an outer loop should give the measuring more accuracy. Thus the "performance index" calculates

((outer loops) x 34) / time (s).

(It's not exactly operations per second, but it allows for comparison.) Here's my test on HP 35S in RPN and in ALG modes. I broke the single line expressions to show better how the code compares:

Code:

HP 35s

RPN                           ALG
37.05 s  perf.idx 9.2         51.05 s  perf.idx 6.7

B001 LBL B                    A001 LBL A
B002 10       STO A           A002 10 >A
B004 10       STO B           A003 10 >B
B006 1        +               A004 (B + 1
B008 4.567E-4 -                       - 4.567E-4
B010 70       +                       + 70
B012 69       -                       - 69)
B014 7        x                       x 7
B016 11       /                       / 11
B018 RCL B                    A005 B
B019 1        -                      - 1
B021 STO B                           >B
B022 x!=0?    GTO B006        A006 x!=0?     GTO A004
B024 Rv                       A008 SQRT(
B025 LOG                                SQRT(
B026 SIN                                SIN(
B027 vX                                 LOG(
B028 vX                                     REGy))))
B029 RCL A                    A009 A
B030 1        -                      - 1
B032 STO A                           >A
B033 x!=0?    GTO B004        A010 x!=0?     GTO A003
B035 RTN                      A012 RTN

In ALG mode it's almost as slow as the fx-4500PA (slightly slower successor to the fx-4500P from 1989) which I personally like for it's compactness and it's approach to save keystrokes (no mandatory brackets with single-operand functions and avoiding some multiplication signs e.g.)

Code:

fx-4500PA

55.09 s  perf.idx 6.1

Prog PERFM
LINE1   A=10 :
        LBL 0 : B=10 :
        LBL 1 : X=(B+1
                    -4.567E-4
                    +70
                    -69)
                    x7
                    /11 :
                 B=
                   B
                    -1 :
            B>0 => Goto 1 [fi]
            X=v _
               v
                sin
                    log
                        X :
            A=
              A
               -1 :
         A>0 => Goto 0 [fi]
         X

(This is all on one line, broken just for comparison to the above. The "v" stands for square root. The "[fi]" - end-if - is a special character, outlined triangle towards right, delimiting expressions as ":" or CRLF would.)

The competition is more than 3 times faster, it has also commands for structured programming, but these are on the third page of the Program-Function menu (6 keystrokes):

Code:

fx-5800P

10.09 s (10x) perf.idx 33.7               8.42 s (10x) perf.idx 40.4
30.04 s (30x) perf.idx 34.0               24.72 s (30x) perf.idx 41.3

Prog PERFM                               Prog PERFS
 30 ->A :                                 For 30 ->A To 1 Step -1 :
 LBL 0 : 10 ->B :                         For 10 ->B To 1 Step -1 :
 LBL 1 : (B+1                               (B+1
           -4.567E-4                          -4.567E-4
           +70                                +70
           -69)                               -69)
           x7                                 x7
           /11 ->X :                          /11 ->X :
          B                               Next:
           -1
              ->B :
     B>0 => Goto 1 :
          v _(                             v _(
           v  (                             v  (
            sin (                            sin (
             log (                            log (
                  X)))) ->X :                      X)))) ->X :
     A                                    Next:
      -1
         ->A :
  A>0 => Goto 0 :
  X                                       X

(Also the "v" means square root here, and I entered the programs as "1 line" using ":" as expression/command separator only.) Maybe it's not fair to use the advanced loop commands in such a comparison? Should I do another test with DSE (which the 4500P lacks) on the 35s?

- So far about the ALG mode - I'd say: Use RPN :-)

Then there are the limited trigonometric function calculations near n x pi/4. The forum archive has a beautiful demonstration listing in message 27 from KS in the thread "The trigonometric bug is spreading !?".

Somewhere in this or a related thread I read that other calculators give even shorter responses in this range - these are from my collection:

sin 3.1415926 (rad)

free42 (1.5.5 dec) 5.35897932384626...e-8
TI-66: 5.358979323846E-08
fx-4500PA: 5.36E-08
HP 15C LE: 5.359E-08
fx-5800P: 5.35898E-8
HP 35S: 5.358979E-8

If I'd need for sure higher accuracy in this field (but I'm not aware of any such need for myself), I could type in and use Takayuki Hosoda's alternate trig. functions or look for other workarounds like e.g. sine for angles from 1e-2 to 1e-4 deg (below 1e-4 deg the ->RAD does it):

1e-4
->RAD
ENTER
ENTER
3
y^x
6
/
-
result: 1.74532925199E-6

- It contines endlessly: I like the nicely readable display except for some too small annunciators and the exponent out of sight in "all" mode, I like the programming except for missing indirect branches and just single-letter labels, I love the RPN use except for some keys moved around, I like the good readable key labels except for there's no distinction between menus and just functions ... where was binary burried?

Two years after I bought it and more than seven years now after HP released the 35s I have little hope that the industry will issue any new nice and pocketable *RPN* calculator or even fix the boldest issues with this one. Comparing the 300s+ with other manufactuerers' I see almost no difference left. Let's make the best out of what we have and support the efforts around the WP family. (Am I nostalgic?)

Long live RPN!
Cheers, Dirk
_______________
Sanyo CZ1210, TI-66, fx-4500P, HP 15C LE, fx-5800P, HP 35s ...?