Prev: workbook saveas: how? (interop)
Next: reference type
From: hswerdfe on 4 Apr 2006 13:18
I have a function to convert Strings to Doubles
double StringToFloat(CString a_test);

some times StringToFloat returns a -1.#IND
for values that can't be converted


This value is supposed to be in a Range.
and sometimes that range is not bounded -1.#INF to 1.#INF

so if I do something Like this, when GetMin() returns -1.#INF and
GetMax() return 1.#INF

double testNum = StringToFloat(strBAdString); //testNum is now -1.#IND

bool tmp = testNum <= m_max;
bool tmp2 = testNum >= GetMin();
bool tmp3 = testNum <= GetMax();
bool tmp4 = testNum == GetMax();
double tmpMax = GetMax();
bool tmp5 = testNum <= tmpMax;

I have a problem that all the tmp bool's are true, except tmp3. tmp3 is
false.

GetMax() is exactly a one line function
that looks like this.
double myClass::GetMax() const
{
return m_max;
}
and m_max is a member function
double m_max;

So my question is Why? Why is tmp3 false?

not only that but shouldn't tmp4 be false. why is it true?

I'm using VC++6 running in Debug mode.

thanks , how
From: Simon Trew on 4 Apr 2006 19:34
"hswerdfe" <hswerdfe(a)example.com> wrote in message
news:e0u986$757$1(a)nrc-news.nrc.ca...
>
> So my question is Why? Why is tmp3 false?

Don't know, but there is something fishy about m_max-- what is it? This
complete program sets b to true:

#include <limits>

double GetMax()
{
return std::numeric_limits<double>::quiet_NaN();
}

int main(int, char**)
{
double max = GetMax();
bool const b = max <= max;
return b;
}

How is m_max initialized? NaNs never compare equal to themselves, but I'm
not sure about <=. If we assume the tautology (a<=b) === !(b>a) then we'd
hope it was always consistent, but I'm not sure that's actually required.


From: hswerdfe on 5 Apr 2006 12:48
Simon Trew wrote:
> "hswerdfe" <hswerdfe(a)example.com> wrote in message
> news:e0u986$757$1(a)nrc-news.nrc.ca...
>
>>So my question is Why? Why is tmp3 false?
>
>
> Don't know, but there is something fishy about m_max-- what is it? This
> complete program sets b to true:
>
> #include <limits>
>
> double GetMax()
> {
> return std::numeric_limits<double>::quiet_NaN();
> }
>
> int main(int, char**)
> {
> double max = GetMax();
> bool const b = max <= max;
> return b;
> }
>
> How is m_max initialized?

if no value is passed into the Object construtor
SetMax(HUGE_VAL);

where Set Max looks like this
BOOL myClass::SetMax(const double a_newVal)
{
m_max = a_newVal;
return true;
}

> NaNs never compare equal to themselves, but I'm
> not sure about <=. If we assume the tautology (a<=b) === !(b>a) then we'd
> hope it was always consistent, but I'm not sure that's actually required.

now I am a little fuzzy I though '(a<=b) == !(b>a)' this had to be true.
I guess math works diffrent in the world of Nans.


as a follow up
I thought the problem might have something to do with the fact that
GetMax was a const function

so I made 2 Test Functions

double myClass::TestConst() const
{
double Zero = 0.0;
double NaN = 0.0/ Zero;

bool tmp = NaN <= m_max;
bool tmp2 = NaN >= GetMin();
bool tmp3 = NaN <= GetMax();
//bool tmp4 = NaN <= GetMaxNotConst();
return 1.0;
}
double myClass::TestNotConst()
{
double Zero = 0.0;
double NaN = 0.0/ Zero;

bool tmp = NaN <= m_max;
bool tmp2 = NaN >= GetMin();
bool tmp3 = NaN <= GetMax();
bool tmp4 = NaN <= GetMaxNotConst();
return 1.0;
}

where GetMaxNotConst() looks like this
double myClass::GetMaxNotConst()
{
return m_max;
}

this had no effect
tmp, tmp2 are true in both of the above functions
and tmp3, tmp4 are both false.


at anyrate I have conluded that it is probably not safe to assume
anything when comparing with any opperator #IND, and #INF.
so I am now using
int _finite( double x );
and
int _fpclass( double x );

to check for this kind of stuff.
and returning the compaison value I want ..:)


It would still be nice to know why?
but as my old DOS based risk game used to say.
* ours is not to reason why ours is but to do or die.
back to work, now.

Thanks, how

>
>






From: Murrgon on 5 Apr 2006 14:28
There is a really good article in "Game Programming Gems 6"
called "Floating-Point Tricks". It should provide you with
some clarity. Here is a small excerpt which may help you:

"For two float values v1 and v2, is it true that one of the
clauses v1 > v2, v1 == v2 or v1 < v2 is true? Again, NO.
All comparisons of any flavor fail if one operand is a NaN."

Murrgon
From: hswerdfe on 5 Apr 2006 17:07

Murrgon wrote:
> There is a really good article in "Game Programming Gems 6"
> called "Floating-Point Tricks". It should provide you with
> some clarity. Here is a small excerpt which may help you:
>
> "For two float values v1 and v2, is it true that one of the
> clauses v1 > v2, v1 == v2 or v1 < v2 is true? Again, NO.
> All comparisons of any flavor fail if one operand is a NaN."

while this may be the C++ standard, (I wouldn't know).
This is not result I experience, with the implementation I am using (MS
VC++ 6.0)

as sometimes they do return true.

> Murrgon

thanks for your help Murrgon.
 |  Next  |  Last
Pages: 1 2
Prev: workbook saveas: how? (interop)
Next: reference type