Difference between Qt 5.6.x and 5.7.x ; floating precision

[Desch]

Hi everyone,
I try to understand why this :

float b = 11.25;
qDebug() << QString::number(b, 'f', 1);

gives :

• In Qt 5.6.x: 11.2
• In Qt 5.7.x (and above ) : 11.3

For testing I'm using Visual Studio 2013.

It might be this reason, but not sure :
Added the ability to convert a floating point to its shortest, exact string form, without having to pre-calculate how many digits that is; QVariant uses this

In order to evaluate the impact of this change on my project, is it possible to go back to 11.2 (using a specific flag, a function or linking to other libraries) ?

Thanks a lot for your help and merry Christmas to all of you.

David

[VRonin]

Hi, would you want 11.26 to still be 11.2 or go 11.3? basically what is the threshold for 11.2 to become 11.3?

[Desch]

11.26 has to be 11.3. It's Ok.
There is no specific threshold. if you round the value 11.25 with one decimal with Qt 5.6, it was 11.2. With Qt 5.7 it is 11.3. So simply why this difference between both Qt versions?
Now if you know the reason why, do you know a way to round 11.25 to 11.2 with Qt 5.7 and above? (probably the answer is to add an epsilon...)

Thanks again...
David

[Christian Ehrlicher]

Qt5.7 and above is using libdouble-conversion which could lead to a difference here. And since 11.25f can be converted to 11.2499999d there was maybe an error in the old implementation.
This is also discussed here: https://bugreports.qt.io/browse/QTBUG-44039

[kshegunov]

@Christian-Ehrlicher said in Difference between Qt 5.6.x and 5.7.x ; floating precision:

And since 11.25f can be converted to 11.2499999d

11.25 is representable exactly in IEEE754. If it was rounded down then it's definitely a bug.

[Desch]

@kshegunov said in Difference between Qt 5.6.x and 5.7.x ; floating precision:
Thanks a lot to everyone.