[SOLVED] QTableView - custom sizeHint() implementation slow

[aran]

Hi guys,

I'm using a QTableView to display a bunch of information. Since I need rich text support in the cells I implemented a custom item delegate with QTextDocument and overwrote the sizeHint();
However, the sizeHint() function causes some problems: It slows everything down really badly. By that I mean, that e.g. moving the cursor over a few cells, the highlighting really lags behind.

Here's the code that I use for the Delegate and the sizeHint:
@void RichTextDelegate::paint(QPainter * painter, const QStyleOptionViewItem & option, const QModelIndex & index) const {

QStyleOptionViewItemV4 options = option;
initStyleOption(&options, index);

painter->save();

QTextDocument doc;
doc.setTextWidth(options.rect.width());
doc.setHtml(options.text);

options.text = "";
options.widget->style()->drawControl(QStyle::CE_ItemViewItem, &options, painter);

painter->translate(options.rect.left(), options.rect.top());
QRect clip(0, 0, options.rect.width(), options.rect.height());

QPen greyPen(Qt::lightGray);
painter->setPen(greyPen);
painter->drawLine(0,options.rect.height(),options.rect.width(),options.rect.height());
painter->drawLine(0,0,options.rect.width(),0);

painter->setClipRect(clip);
QAbstractTextDocumentLayout::PaintContext ctx;

ctx.clip = clip;
doc.documentLayout()->draw(painter, ctx);

painter->restore();

}

QSize RichTextDelegate::sizeHint ( const QStyleOptionViewItem & option, const QModelIndex & index ) const {

QStyleOptionViewItemV4 optionV4 = option;
initStyleOption(&optionV4, index);
QTextDocument doc;
doc.setHtml(optionV4.text);
doc.setTextWidth(optionV4.rect.width());
return QSize(doc.idealWidth(), doc.size().height());

}@

I know, the laggish behaviour is caused by sizeHint, because if I remove its implementation everything works just fine... Any idea, what I can do to improve the code, stopping it to be "laggish"?

Appreciate any help!

Ps: I tested it both under Qt 4.8.5 and Qt 5.1.1, same result.

[Agroni]

You are recreating QTextDocument doc everytime sizeHint is called. It may be called a million times on just one refresh of a screen.

I would recommend to ...

1. keep doc outside sizeHint.
2. call doc.setHtml(optionV4.text); only when content has changed.
3. call #2 outside sizeHint
4. call doc.setTextWidth(optionV4.rect.width()); also only when content has changed
5. call #4 outside sizeHint
6. you can keep the last line maybe with a little modification :)
   return QSize(m_doc.idealWidth(), m_doc.size().height());

[aran]

That makes sense, thanks! No wonder it's slow.

However, I can't figure out how to check if the content has changed. The problem is, that both functions are const, i.e. assigning anything from there to e.g. a global variable doesn't work:

(header)
QString txt;

(paint/sizehint)
txt = options.text

-> doesn't compile:
passing 'const QString' as 'this' argument of 'QString& QString::operator=(const QString&)' discards qualifiers

But how else could I keep track of changes of the content?

[aran]

Okay, I think I got it now, the (now deleted?) comment about the mutable keyword helped a lot.
No I check for changes in content and/or changes in the optionV4.rect, and only if at least one of them changed then I recalculate the sizeHint. Here's the code I came up with:

(header)
@private:
mutable QHash<QPair<int,int>, QString> txt;
mutable QHash<QPair<int,int>, QSize> docSize;
mutable QHash<QPair<int,int>, QRect> optionRect;@

(source)
@QSize RichTextDelegate::sizeHint ( const QStyleOptionViewItem & option, const QModelIndex & index ) const {

QStyleOptionViewItemV4 optionV4 = option;
initStyleOption(&optionV4, index);

QPair<int,int> p;
p.first = index.row();
p.second = index.column();

if(txt[p] != optionV4.text || optionRect[p] != optionV4.rect) {

QTextDocument doc;
doc.setHtml(optionV4.text);
doc.setTextWidth(optionV4.rect.width());

txt[p] = optionV4.text;

docSize[p] = QSize(doc.idealWidth(),doc.size().height());
optionRect[p] = optionV4.rect;

}

return docSize[p];

}@

I hope I haven't missed something important, but, as far as I can tell, that works quite well.

Thanks a million to Agroni and the other person of the deleted comment!