QPolygonF error

Christian Ehrlicher

This is a testcase similar to what we want to see:

int main(int argc, char **argv)
{
  QCoreApplication app(argc, argv);

  QPolygonF p1 = QVector<QPointF>({{1., 0.}, {2., 1.}, {3., 1.}, {4., 1.}});
  QPolygonF p2 = QVector<QPointF>({{2., 2.}, {3., 2.}, {4., 5.}, {3., 2.}});
  QPolygonF united = p1.united(p2);
  QPolygonF expected = QVector<QPointF>({{1., 0.}, {2., 1.}, {3., 1.}, {4., 1.}, {1., 0.}, {2., 2.}, {3., 2.}, {4., 5.}, {3., 2.}, {2., 2.}, {1., 0.}});
  qDebug() << (united == expected);
  qDebug() << united;
  qDebug() << expected;

  return 0;
}

Daniil S.

@Christian-Ehrlicher

int main(int argc, char *argv[])
{
  QPolygonF high, low;
  high << QPointF(293, 0) << QPointF(406.102, 225) << QPointF(556, 0);
  low << QPointF(79, -1) << QPointF(182.303, 60) << QPointF(673, -1);



  QPolygonF low1(low.united(high).united(high))
      , expected({ QPointF(79,-1), QPointF(182.303,60), QPointF(314.876,43.5194), QPointF(406.102,225), QPointF(546.162,14.7677), QPointF(673,-1), QPointF(79,-1) });

  qDebug() << low1;
  qDebug() << expected;
  qDebug() << (expected == low1);

  return 0;
}

qDebug()

QPolygonF(QPointF(79,-1)QPointF(182.303,60)QPointF(314.876,43.5194)QPointF(293,0)QPointF(556,0)QPointF(406.102,225)QPointF(314.876,43.5194)QPointF(182.303,60)QPointF(79,-1)QPointF(314.876,43.5194)QPointF(406.102,225)QPointF(546.162,14.7677)QPointF(673,-1)QPointF(79,-1)QPointF(182.303,60)QPointF(314.876,43.5194)QPointF(293,0)QPointF(556,0)QPointF(406.102,225)QPointF(314.876,43.5194)QPointF(79,-1))

QPolygonF(QPointF(79,-1)QPointF(182.303,60)QPointF(314.876,43.5194)QPointF(406.102,225)QPointF(546.162,14.7677)QPointF(673,-1)QPointF(79,-1))

false

Daniil S.

@Christian-Ehrlicher still not what you need?

kshegunov

low.united(high).united(high)?

Why the second call? What does calling it only once produce?

Daniil S.

@kshegunov I have an application that combines graphs of changing values, nothing prevents different "scenes" with same graphs that can combined more times. now I have 3 scenes, on 3 lines, the top two scenes have exactly the same graph, and if i combine them all, this leads to this error

kshegunov

Okay, this answers the first question, what about the second? Does calling .union only once produce the correct result?

Christian Ehrlicher

@kshegunov said in QPolygonF error:

Does calling .union only once produce the correct result?

Yes, but with slightly different numbers (not visible in a plain debug output). I would guess somewhere in the clipping logic a rounding error triggers this bug (e.g. 314.87615740 / 43.51943745 <-> 314.87600000 / 43.51940000)

Daniil S.

@kshegunov yes, but I cant check top polygons are equal, because they must be combined in order in the loop and can have different overlay types, and checking high by points before united is not entirely correct, because a polygon can contain a point, but not a union

kshegunov

@Christian-Ehrlicher said in QPolygonF error:

Yes, but with slightly different numbers (not visible in a plain debug output). I would guess somewhere in the clipping logic a rounding error triggers this bug (e.g. 314.87615740 / 43.51943745 <-> 314.87600000 / 43.51940000)

Looks truncated more than a rounding error.

@Daniil-S said in QPolygonF error:

yes, but I cant check top polygons are equal, because they must be combined in order in the loop and can have different overlay types, and checking high by points before united is not entirely correct, because a polygon can contain a point, but not a union

I was asking to discern whether the problem is that the two unions apply the same polygon only, there's a lot of edge cases whenever you're dealing with lines which are almost parallel (e.g. https://codereview.qt-project.org/c/qt/qtbase/+/292807).

Please search the tracker and if this isn't reported, do so. Attach the minimal code to reproduce as well. Whenever I finish with that QLineF thing, given time I'll take a look.

Christian Ehrlicher

@kshegunov said in QPolygonF error:

Looks truncated more than a rounding error.

No error - qDebug() prints not the complete double value, I guess the values are taken from there. Mine are the exact ones :)

So basically a polygon is united with a polygon which nearly matches parts of the first polygon which fails sometimes.

kshegunov

@Christian-Ehrlicher said in QPolygonF error:

So basically a polygon is united with a polygon which nearly matches parts of the first polygon which fails sometimes.

Yes, what I imagine happens. Probably somewhere in the clip/intersection code there's some numerical instability that breaks it silently.

Daniil S.

@kshegunov yes, it only happens if two identical polygons are combine.
here i think, is what is needed

kshegunov

lol. "Out of scope" ... what can I say ...

I suggest you open a new one, attach the reproducing code (the minimal one that @Christian-Ehrlicher insisted on), and put in the:
"Relates to ..." field in JIRA QTBUG-38037, as it does indeed seem related.

Daniil S.

https://bugreports.qt.io/browse/QTBUG-100172

kshegunov

As I was looking into this, I also opened:

https://bugreports.qt.io/browse/QTBUG-100302

Daniil S.

@kshegunov I noticed it, but thought that I did something wrong

Daniil S.

@kshegunov probably, as advised in one of the reports, I will use the clipper library

kshegunov

@Daniil-S said in QPolygonF error:

@kshegunov probably, as advised in one of the reports, I will use the clipper library

Yes, that's prudent as even if I have the time and will to fix at least some of this, assuming it gets accepted to begin with, it can be a long time.