Qreal in QVariant

Jupiter

Hi,

in my application i put a qreal in a QVariant.

how ever in very rare situations (not directly reproduce able) converting it back to a qreal fails.
f.e.

@qreal d = 0.0f;
QVariant v = qVariantFromValue<qreal>(d);
qreal d2 = v.value<qreal>();@

when i debug the code d = something like 1.8e-308. so almost 0.0. thats ok.
but when i step further sometimes d2 is something like 1.8e308 thats something completely different. how can i fix the problem? and it happens not every time. only some times.

thanks in advance

giesbert

a floating point value is always fuzzy 0, not exact :-)
The reason is the binary representation of the value.

Jupiter

yes i know that. but both values aren't the same the second isn't even fuzzy 0 since the exponent has an positive sign!

lgeyer

I've done some testing and I cannot reproduce your problem. You might see some jitter due to the copying in and out from QVariant but not such a flip. There is most probably other broken code interfereing or you your hardware is flipping bits.

fluca1978

I've tested the following:

@ qreal qr = 0.0f;
QVariant qv = qVariantFromValue<qreal>( qr );
qreal qr2 = qv.value<qreal>();

qDebug() << "Original value is " << qr;
qDebug() << "QVariant value is " << qr2;
qDebug() << "QVariant value is " << qv.value<qreal>();

@

and all I see is zero either in the output or in the debugger. I'm running 4.7.4 on linux x86_64. Despite the "real" value, having the exponent to change its sign seems really strange....

goetz

According to K&R C book, the notation 0.0f defines a float value of zero. qreal in in Qt is usually "a typedef for double on all platforms except for those using CPUs with ARM architectures"[1]. So it may be that you run into a conversion problem at the first line (although I don't believe that's the case here). Do you have any specific reasons to fix the constant to type float? I would leave out the f entirely.

I just tested this on my Mac here, and the variables always have a value of 0.0. Even if I put 1.8e-308 into d manually, v and d2 it always have this value.

I suspect something else going wrong here.

fn1. http://doc.qt.nokia.com/4.7/qtglobal.html#qreal-typedef

Jupiter

well actually the initializing with 0.0f was just a test, first it was 0 then it was 0.0 then it is 0.0f

i also tested to init it with @memset(&d, 0, sizeof(d));@

i don't see a problem with my other code, since it is working in like 99% of all cases. and only 1% get messed up. i actually works over a signal slots connection, where the signal takes a QVariant where i put the qreal into, and on the slot the QVariant is converted back to a qreal. at that position it has the wrong sign of the exponent sometimes

dangelog

Zero is always representable without any rounding error -- and if the code is actually

@
float f = 0.0f;
@

there's no math involved, therefore there should be no roundoff errors.

I agree with Volker and suspect there's something else going on.

lgeyer

Well, can you provide a small example that reproduces your problem or at least any real world code. The one you've posted works like a charm. Have you tried to debug your application to find out when and where your value is actually altered?

Jupiter

well its the same code called like 1000 times. and 1 time the error occurs, so you can't really debug that unfortunately

fluca1978

[quote author="Jupiter" date="1320931268"]well its the same code called like 1000 times. and 1 time the error occurs, so you can't really debug that unfortunately[/quote]

Could you log the output each time and for each iteration? With, of course, as much information as possible. Could this code be in a method that is called with wrong arguments? Is it possible to have a small example to try?

Jupiter

in my code its the following:

object A creates object B.
B sets its member "member" to 0.0 in the constructor.

now when A calls the getter b->member() 1 line after it sometimes receives the unexpected number.
there is no other class with access to b.

when b sets its member a signal with the variant is emitted. the signal signature is signal(const QVariant& v) so the receivers also can't change it.

lgeyer

If you really want any reasonable help you won't get around providing some real world code.