Re: VarRound implementation

[Fabian Cenedese]

> Without a test how can you know if this implementation produces the correct values?  What about variant type conversions?  It seems silly to send a patch without testing it...

Now I'm confused. I'm adding more types to VarRound and I'm also implementing
the test_VarRound function (programmer's pride :)

For this I tested more cases on Windows and got strange results (in my view).

VT_R4: 7.850 -> VarRound(1) -> 7.800
VT_R4: 7.851 -> VarRound(1) -> 7.900

Is this the binary representation (in)accuracy?
How can I test if the rounding worked ok when the numbers may differ in
the last (invisible) bits?

And how should I round a currency? It's valid on Windows but I couldn't
interpret the results. Is this some kind of fixed comma type?

bye  Fabi

VarRound implementation

[Dmitry Timoshkov]

"Fabian Cenedese"  wrote:
> For this I tested more cases on Windows and got strange results (in my view).
>
> VT_R4: 7.850 -> VarRound(1) -> 7.800
> VT_R4: 7.851 -> VarRound(1) -> 7.900
>
> Is this the binary representation (in)accuracy?
> How can I test if the rounding worked ok when the numbers may differ in
> the last (invisible) bits?
>
> And how should I round a currency? It's valid on Windows but I couldn't
> interpret the results. Is this some kind of fixed comma type?

dlls/oleaut32/tests/vartype.c around line 4514 has the following comment:

/* MSDN states that rounding of R4/R8 is dependent on the underlying
     * bit pattern of the number and so is architecture dependent. In this
     * case Wine returns .2 (which is more correct) and Native returns .3
     */

So, I'd suggest just surround your test by todo_wine and add a comment
explaining the situation.