Save/save_precision problem ?

[Joao Cardoso]

Hi,

There seems to exist a problem with the save command with the -ascii option,
the save_precision option and (eventually) printf.

I noticed the problem when I could't reproduce Octave anwsers with an external
program feed from octave.

It seems that if the save_precision is too high (the default of 17 decimal
places) the saved variables are unrounded (yes...).

It does not seem to be a problem of internal rounding because of binary
inability to exactly represent decimal numbers.

Look bellow:

octave
Octave, version 1.1.1.
Copyright (C) 1992, 1993, 1994, 1995 John W. Eaton.
This is free software with ABSOLUTELY NO WARRANTY.
For details, type `warranty'.

octave:1> load hemophil.asc
octave:2> hemophil
hemophil =

  -0.00560  -0.16570   0.00000
  -0.16980  -0.15850   0.00000
  -0.34690  -0.18790   0.00000
  -0.08940   0.00640   0.00000

...stuff deleted...

octave:3> save -ascii popo hemophil
octave:4> system("more popo")
# name: hemophil
# type: matrix
# rows: 75
# columns: 3
 -0.0055999999999999999 -0.16569999999999999 0
 -0.16980000000000001 -0.1585 0
 -0.34689999999999999 -0.18790000000000001 0
 -0.089399999999999993 0.0064000000000000003 0
 -0.16789999999999999 0.071300000000000002 0
 -0.083599999999999994 0.0106 0

octave:5> save_precision=5
save_precision = 5
octave:6> save -ascii popo hemophil
octave:7> system("more popo")
# name: hemophil
# type: matrix
# rows: 75
# columns: 3
 -0.0056 -0.1657 0
 -0.1698 -0.1585 0
 -0.3469 -0.1879 0
 -0.0894 0.0064 0
 -0.1679 0.0713 0
 -0.0836 0.0106 0

everyting from save_precision=5 to save_precision=15 goes OK.

save_precision = 16
octave:33> save -ascii popo hemophil
octave:34> system("more popo")
# name: hemophil
# type: matrix
# rows: 75
# columns: 3
 -0.0056 -0.1657 0
 -0.1698 -0.1585 0
 -0.3469 -0.1879 0
 -0.08939999999999999 0.0064 0
 -0.1679 0.0713 0
 -0.08359999999999999 0.0106 0
 -0.1979 -0.0005 0

octave:35> save_precision=17
save_precision = 17
octave:36> save -ascii popo hemophil
octave:37> system("more popo")mophil
# name: hemophil
# type: matrix
# rows: 75
# columns: 3
 -0.0055999999999999999 -0.16569999999999999 0
 -0.16980000000000001 -0.1585 0
 -0.34689999999999999 -0.18790000000000001 0
 -0.089399999999999993 0.0064000000000000003 0
 -0.16789999999999999 0.071300000000000002 0
 -0.083599999999999994 0.0106 0

If it where an internal representation problem, this should be in error,

octave:39> format long
octave:40> hemophil
hemophil =

  -0.005600000000000  -0.165700000000000   0.000000000000000
  -0.169800000000000  -0.158500000000000   0.000000000000000
  -0.346900000000000  -0.187900000000000   0.000000000000000
  -0.089400000000000   0.006400000000000   0.000000000000000

octave:43> hemophil (4,1)
ans = -0.0894000000000000
octave:44> hemophil (4,1) - -0.0894000000000000
ans = 0

Well, is a new bug or is just binary inability to represent some decimal
numbers? If it is shouldn't be better to use a default save_precision
of 15 or lower?

No, I don't have a Pentium :-(

Thanks,

Joao