StrToFloat bug report

1
2
Page 1 of 2

Avra

Schreiberling

Gender:
Location: Belgrade, Serbia
Age: 53
Homepage: rs.linkedin.com/in…
Posts: 653
Registered: 07 / 2002

Subject:

StrToFloat bug report

· Posted: 30.03.2015 - 15:59 · #1

Conversion from string to float seams to error in exponent calculation. Problem exists with and without simulator.

Code

var
  f: float;
  s: string[50];
...
    s := '      12345678.9'; // x1
    f := StrToFloat(s);
    s := s + ' = ' + FloatToStr(f) + '   ';
    WriteLn(LCDout, s);      // '12345678.9 = 1.234568E7' (good)

    s := '    1234567890.0'; // x100
    f := StrToFloat(s);
    s := s + ' = ' + FloatToStr(f) + '   ';
    WriteLn(LCDout, s);      // '1234567890.0 = 1.234568E8' (<<< bad, E8 should be E9 !!!)

    s := '12345678900000.0'; // x1.000.000
    f := StrToFloat(s);
    s := s + ' = ' + FloatToStr(f) + '   ';
    WriteLn(LCDout, s);      // '12345678900000.0 = 1.272353E12' ((<<< bad, E12 should be E13 !!!)

Administrator

Gender:
Location: Germany
Age: 24
Homepage: e-lab.de
Posts: 5558
Registered: 03 / 2002

Subject:

Re: StrToFloat bug report

· Posted: 07.04.2015 - 14:29 · #2

Hello Avra,

I tried to find the bug. But the driver is completely written in ASM and
I could not locate the problem. I'm somewhat baffled. Sorry.

rolf

Harald_K

Subject:

Re: StrToFloat bug report

· Posted: 07.04.2015 - 18:27 · #3

just tested with my 3.98 compiler build ....

same problem .... and the first value is converted to 12345678 without exponent
(all only in simulator)

just a remark: the mantissa of the third value in avra's post looks a little bit mixed-up - maybe a problem with frame size or so?

Administrator

Gender:
Location: Germany
Age: 24
Homepage: e-lab.de
Posts: 5558
Registered: 03 / 2002

Subject:

Re: StrToFloat bug report

· Posted: 07.04.2015 - 19:07 · #4

Hello Harald,
sorry. I really don't know what is going on there.
It is many, many years ago and all written in ASM. :aerger:
rolf

Harald_K

Subject:

Re: StrToFloat bug report

· Posted: 07.04.2015 - 22:52 · #5

so ... little bit of ASM-practice helps sometimes ...

this problem occurs only if you use large numbers (more than 8 digits) in non-scientific notation.
lets say 123456.7e13 works good, 12345670000000 doesnt work.

i fear, the same type of error will occure if you convert very small numbers in non-scientific, e.g. 1e-15 will work, 0.000000000000001 might not

i dont know if this is a real "error", because
transferring large strings on a microcontroller with restricted size of RAM is no good idea - stack and frame will catch more space than available, especially when you use some recursive calls.

@rolf:
maybe, one additional inc _accahi at the end of the Str2Flt4-code directly before the str2flt6 entrypoint will fix the problem .... but do some checkout with different sizes of stringnumbers and decimalpoint positions ... i'm still analyzing

Avra

Schreiberling

Gender:
Location: Belgrade, Serbia
Age: 53
Homepage: rs.linkedin.com/in…
Posts: 653
Registered: 07 / 2002

Subject:

Re: StrToFloat bug report

· Posted: 09.04.2015 - 20:16 · #6

Quote by rh

I tried to find the bug. But the driver is completely written in ASM and
I could not locate the problem. I'm somewhat baffled. Sorry.

I will reimplement StrToFloat when I catch time...

Administrator

Gender:
Location: Germany
Age: 24
Homepage: e-lab.de
Posts: 5558
Registered: 03 / 2002

Subject:

Re: StrToFloat bug report

· Posted: 10.04.2015 - 14:13 · #7

Hello Avra,

a Pascal replacement of the StrToFloat function doesn't help.
The bug must be removed in our code generator (ASM) in the
AVRco. User Harald_K already tries to find the bug.

rolf

Administrator

Gender:
Location: Germany
Age: 24
Homepage: e-lab.de
Posts: 5558
Registered: 03 / 2002

Subject:

Re: StrToFloat bug report

· Posted: 11.04.2015 - 20:16 · #8

@Avra,
StrToFloat bug is fixed now.

rolf

1
2
Page 1 of 2

StrToFloat bug report

Registered users in this topic