Qt 6.9.0 GraphicScene scrollbars give wrong value ( nan or inf )

zenon

Hello,

The problem occurs with Qt 6.9.0 on Mac Intel and silicon but not with Linux.

I joined a QGraphicsView with a QGraphicsScene . I initialize
g_scene..setSceneRect( 0,0,0,0 );
and then plot a lot of points ( in ex 300 ) :

QGraphicsTextItem * pt
pt->setPos( x, y )
gscene.addItem( pt );

If many plots in a normal range are inserted the scrollbars is stucked because the coord of de scene give qreal that are inf nan.

to test I do this :

QRectF R = g_scene.sceneRect();
and I obseved that the values in R were not a number or infinite. No real values.

It happens when many graphiics are inserted . And these wrong real values do not appear with Qt 6.8.0 on Mac and do not appear with Qt 6.9.0 on Linux

When a certain amount of items are inserted the coordinates of the graphic turns to give wrong real values lile not a number or infinite.

I suspect a bug.

ChrisW67

g_scene..setSceneRect( 0,0,0,0 );

(preserved your typo)

You have told the scene that it has zero width and height (and set the view's default scrolling area also). I do not find it surprising that the view has issues scrolling a zero-size area .

Perhaps you meant to set a null QRectF: g_scene.setSceneRect(QRectF());

zenon

No it's not a problem setting to zero gives extended size . That does not make any change as I've just tested. The wrong real values appears affter inserting smalll text items.

If i posted it's because this misbehaviour appears only with Qt 6.9.0 on Mac not on Linux . I did not test on Windiows. Qt 6.8.0 was right on Mac.

I discovered the bug with

int is_nan(  qreal  a )
{
  return ( boost::math::isinf(a) ||  boost::math::isnan(a) ) ; 
}

Some more updated information if I compile a debug version the bug does not appear but with some optimization

QMAKE_CXXFLAGS += -std=c++17 -O3 -ffast-math -flto -funroll-loops -march=native -DNDEBUG

goes wrong

QMAKE_CXXFLAGS += -std=c++17 -Og

goes well

Regards

zenon

last minute . I found the right way.
In fact it is the -ffast-math compiler option which beheave differently on Mac and Linux with Qt 6.9.0

Qt 6.9.0 does not support any more very fast compiler optiion
on Mac Mx

CFLAGS        = -pipe -Ofast    $(EXPORT_ARCH_ARGS) -isysroot /Library/Developer/CommandLineTools/SDKs/MacOSX.sdk -mmacosx-version-min=15.0 -Wall -Wextra $(DEFINES)
CXXFLAGS      = -pipe -stdlib=libc++ -std=c++17 -Ofast -funroll-loops -flto -march=native -DNDEBUG -Ofast -std=gnu++1z $(EXPORT_ARCH_ARGS) -isysroot /Library/Developer/CommandLineTools/SDKs/MacOSX.sdk -mmacosx-version-min=15.0 -Wall -Wextra $(DEFINES)

Fast optimisation does not work any more on 6.9.0. Is it Qt or Apple difficult to say.

Thank you anyway .

zenon

Well in fact the problem came from --ffast-math in -03 or Ofast

But I do not understand why on Mac Qt 6.9.0 is so sensitive to fast floating point compilation options when it is not with Qt 6.8.0 ?

zenon

IT'S SOLVED