[RESOLVED] Double appears to have different values in VB6 and C++

[wavering]

I have a variable, dblScoreFD, which is declared in my Visual Basic 6 project as a double and then passed on to my C++ DLL and then back again

I have had strange things happen and after a lot of digging I discover that in C++ it has the value
1030.7899314588

whereas when it immediately comes back to VB6 it has the value
1030.78993145882

The screen shots are below.

Any help appreciated
Bob
.

[dilettante]

There are a ton of factors that can combine to make high precision comparisons fail.

Everything from differences in how a "double" is internally implemented, to implicit intermediate operations you aren't aware of, to differing compiler optimizations chosen, etc.

If you want the level of precision you seem to be after scaled integers may be your best bet (VB6's Currency type or Variant Decimal subtype come to mind). Floating point arithmetic has limited precision by its nature and approximation errors accumulate.

[techgnome]

more than likely it's a display issue ... if the values were truly different, you would see the larger number in C++ and the truncated one upon returning to VB6... since VB6 reports the longer number... that is what it is getting from the C++ code... which tells me that the tooltip is simply truncating the display.

-tg

[dilettante]

That String compare is bizarre anyway. CStr() is locale-aware, so some users where decimal point is not "." would never see the VB code test as equal.

[wavering]

That String compare is bizarre anyway. CStr() is locale-aware, so some users where decimal point is not "." would never see the VB code test as equal.

Agreed! That was just a test. The nub of the problem is that both sides of the equation are revealed to have the same value by tooltip value and both are of the same type but VB6 is saying the values are NOT the same. Here is the process

1. Find a solution and note the Score in C++
2. In Vb6 see if this score is ALREADY present in the Pool, if not put it in
3. Repeat the process

I am finding duplicate entries and the reason is that it finds a solution and puts it into the Pool. It then finds the SAME solution again but when it does the comparison it refuses to accept that it is the same value but it DOES accept they are different if I convert them both to a string (like I say, just a test)

So, it seems to me that passing the double to the DLL and back to VB6 seems to have subtle and unexpected effects at the limits of precision
Thanks for comments
Bob

Edit: See Screen Shots below - how did it get past the first test?
.
.

[szlamany]

There is marshaling involved in moving the arguments between VB and C - I am doing a lot of that with VB.Net and C++ right now myself. How do you pass the parameters??

I can't imagine the score-precision is too important - you are doing floating point math here so you have to accept rounding and mantisa display issues.

Can you lose that digit on both sides so that you "force" the values to be the same - and thus eliminate the dups like that?

[Milk]

Converting a double to string often loses accuracy so identical strings do not indicate identical doubles.

Passing the double to and from the DLL will not alter the double.

You could use your own level of accuracy for equality comparisons...

Const TINY As Double = 0.000000001

If Abs(PageStatsFD.Score(lngRank) - dblScoreFD) < TINY Then...