code review needed: StringToFloat

mpark

The FloatString object in obex does not have a string-to-float conversion, so I wrote my own and added it to my private copy of FloatString. I'm wondering about updating the official FloatString but thought I should let you guys review the code first. So here it is. Let me know if it's useful, correct (within its limitations), etc.

PUB StringToFloat(strptr) : flt | significand, ssign, places, exp, esign 
{{
  Converts string to floating-point number
  entry:
      strptr = pointer to z-string

  exit:
      flt = floating-point number


  Assumes the following floating-point syntax: [noparse][[/noparse]-] [noparse][[/noparse]0-9]* [noparse][[/noparse] . [noparse][[/noparse]0-9]* ] [noparse][[/noparse] e|E [noparse][[/noparse]-|+] [noparse][[/noparse]0-9]* ]
                                               ┌── ┌───── ┌─────────── ┌───────────────────
                                               │   │      │            │     ┌──── ┌─────
    Optional negative sign ────────────────────┘   │      │            │     │     │
    Digits ────────────────────────────────────────┘      │            │     │     │
    Optional decimal point followed by digits ────────────┘            │     │     │
    Optional exponent ─────────────────────────────────────────────────┘     │     │
      optional exponent sign ────────────────────────────────────────────────┘     │
      exponent digits ─────────────────────────────────────────────────────────────┘

  Examples of recognized floating-point numbers:
  "123", "-123", "123.456", "123.456e+09"
  
  Conversion stops as soon as an invalid character is encountered. No error-checking.
    
  Based on Ariba's StrToFloat in http://forums.parallax.com/forums/default.aspx?f=25&m=280607
}}
  significand~
  ssign~
  exp~
  esign~
  places~
  repeat
    case byte[noparse][[/noparse]strptr]
      "-":
        ssign~~
      ".":
        places := 1
      "0".."9":
        significand := significand * 10 + byte[noparse][[/noparse]strptr] - "0"
        if places
          ++places                    'count decimal places
      "e", "E":
        ++strptr ' skip over the e or E
        repeat
          case byte[noparse][[/noparse]strptr]
            "+":
              ' ignore
            "-":
              esign~~
            "0".."9":
              exp := exp * 10 + byte[noparse][[/noparse]strptr] - "0"
            other:
              quit
          ++strptr
        quit              
      other:
        quit
    ++strptr
    
  if ssign
    -significand
  flt := f.FFloat(significand)

  ifnot esign  ' tenf table is in decreasing order, so the sign of exp is reversed
    -exp
    
  if places
    exp += places - 1
        
  flt := f.FMul(flt, tenf[noparse][[/noparse]exp])              'adjust flt's decimal point

cessnapilot

mpark: your code is correct and elegant. I have made a small application to check it (see attached file). If you allow me, I will upgrade a GPS_Float object in OBEX with your·StringToFloat ·procedure , since some NMEA sentences may contain wider range of floats than the original (very simple) conversion routine expects.
Thanks

mpark

Thanks for your comments. You're free to use it any way you like. I'll look into adding it to the official FloatString.

I'm not familiar with GPS_Float, but if it uses FloatString wouldn't it be better just to use the updated FloatString?

mpark

Tried to update FloatString on obex and got an error. Will Parallax system administrators investigate immediately, as the error message says? We shall see.

Edit: I don't know if anyone investigated, but I tried again and this time it worked. FloatString 1.2 is good to go.

Post Edited (mpark) : 3/7/2009 8:48:55 AM GMT