Qt 4.3.0 lots of Bugs!

[VireX]

Ok, I upgraded to 4.3 today, and basically most of my skin is now messed up.

Example, border-image property now has "Stretch and repeat", and instead of stretching it normally and in an amazing way like in 4.2, now they messed it up completely for the sake of a tiny millisecond optimization by "dividing the image into 9 parts and repeating the middle-parts for more optimization". This rule just pixelates all my treewidgets and other things. I can actually see the lines of color changes, its not even a gradient anymore just a bunch of lines of awful graphic rendering.

Why did they make skinning 100x worse in 4.3, did they forget to test it??? Or did they think the pixelation bugs are just better looking?

Doesn't make any logical sense, why they would fix something that wasn't broken.

In addition, normal PNG images are now blurry and awful. Somehow, some of my backgrounds just felt like disappearing. My TreeWidget now has a horizontal scrollbar that wasn't there in 4.2.3, plus a bunch of other random bugs that make absolutely no sense. Really did they beta test for 1 week, because I don't see how they can go from the awesomeness of 4.2.3 to 4.3's crappy untested graphics enhancements. The only thing they did well in 4.3 is the new stylable widgets in stylesheets.

[marcel]

I don't think they messed anything up.
Most likely you didn't do everything by the book .

As for blurry things, etc... check if you have any antialiasing enabled.
And for things that weren't there in the prev version, most likely there have been added functions and parameters that let you take them out.

Regards

[VireX]

How do I check if anti-aliasing exists? And why didn't tell inform me of changes in the change logs???

[VireX]

Here's another AWFUL 4.3.0 bug, really dumb too...

If you make a table view, and put some buttons inside the cellwidgets, and at same time put some text under the buttons (so you can sort the buttons appropriately), it will RANDOMLY delete/hide some buttons for absolutely no reason, if you keep clicking sort, it will continue to hide/show random buttons.

Wow its like they released 4.3.0 without ANY testing... never seen this many bugs from Qt before. So very un-qt-like.

OH and let's add one more to the list, QApplication::alert() is just awful, it flashes even when the windows is activated, my own Win32 API Flash() function is so much better. And alert doesnt even allow you to change the flags.

[jacek]

Really did they beta test for 1 week, because...

Beta tests are when the product is released to the public for everyone to test. Did you test 4.3 beta (it was released almost three months ago)?

You metion "lots of bugs", but so far you gave only six examples of differences you have encountered after the upgrade. Just like Marcel noted, first make sure that they are really caused by bugs and not because you have relied on some undocumented features. We will happily help you to verify that, but we will need some code to be able to reproduce that behaviour.

[VireX]

No they're definitely bugs, because, when you change from 4.2.3 to 4.3, it works good in 4.2.3, but in 4.3 it makes it worse than it is. This is called a bug because its an unforeseen effect of code changes in 4.3 that isn't for the better.

I didn't use any "undocumented" anything, I used correct ways to do everything according to 4.2.3, and I have attempted to fix it using the 4.3 docs, but its unfixable because of internal bugs.

Example, the QAbstractItemView is buggy, if you add buttons to a TableWidget to like a column of cells, (large one), and then also add some text to that cell, SetItemText, and SetWidgetItem etc, and if you keep pressing sort a bunch of times, buttons will start disappearing. This is truly a bug... I would report it, but it may already been reported, I don't see how it can be missed by 3 months of testing, unless people are not testing stuff.

[jacek]

Example, the QAbstractItemView is buggy, if you add buttons to a TableWidget to like a column of cells, (large one), and then also add some text to that cell, SetItemText, and SetWidgetItem etc, and if you keep pressing sort a bunch of times, buttons will start disappearing. This is truly a bug...

Could you post a minimal compilable example for us to test?

I would report it, but it may already been reported,

Then check it: http://trolltech.com/developer/task-tracker
The Trolls can't fix bugs they aren't aware of.

...unless people are not testing stuff.

Exactly. You complain about the outcomes of beta tests, but did you participate in them?

[VireX]

No because Beta doesn't give out a compiled installer with all the DLLs etc... I couldn't participate because I cannot waste my time to figure out how to compile Qt from source code.

I'm trying to recreate it, but I can't recreate it in a small app, only in my own app something is missing. Maybe only happens when the thing is overloaded like my main app. With lots of buttons and rows. I will keep trying.

I also checked tracker i didn't see anything but I dont know whether its been reported under a different name or title etc.

Meanwhile I think Qt changed its TransformationMode, so that when Border-Image is resized, using QImage, it uses TransformationMode = Quick instead of smooth like it usually does, so I think that's the cause of 99% of my blurry images. What do you think?

[VireX]

Oh btw here I reproduced this bug... so easy... Sorting is broken.

#include <QtGui>
 
int main(int argc, char *argv[])
{
    QApplication app(argc, argv);
    QTableWidget* qt = new QTableWidget;
    qt->verticalHeader()->setVisible(false);
    qt->setSortingEnabled(true);
    qt->sortItems(0, Qt::AscendingOrder);
    qt->setRowCount(75);
    qt->setColumnCount(2);
    QTableWidgetItem *__colItem = new QTableWidgetItem();
    __colItem->setText(QApplication::translate("t", "test", 0, QApplication::UnicodeUTF8));
    qt->setHorizontalHeaderItem(0, __colItem);
    QTableWidgetItem* ax[75];
    for(int i = 0; i < 75; i++){
     ax[i] = new QTableWidgetItem(QString("Ebeeef%1").arg(i));
     qt->setItem(i, 0, ax[i]);
     qt->setCellWidget(i, 0, new QPushButton("ff"));
    }
    qt->show();
    return app.exec();
}

#include <QtGui>

int main(int argc, char *argv[])
{
    QApplication app(argc, argv);
    QTableWidget* qt = new QTableWidget;
    qt->verticalHeader()->setVisible(false);
    qt->setSortingEnabled(true);
    qt->sortItems(0, Qt::AscendingOrder);
    qt->setRowCount(75);
    qt->setColumnCount(2);
    QTableWidgetItem *__colItem = new QTableWidgetItem();
    __colItem->setText(QApplication::translate("t", "test", 0, QApplication::UnicodeUTF8));
    qt->setHorizontalHeaderItem(0, __colItem);
    QTableWidgetItem* ax[75];
    for(int i = 0; i < 75; i++){
     ax[i] = new QTableWidgetItem(QString("Ebeeef%1").arg(i));
     qt->setItem(i, 0, ax[i]);
     qt->setCellWidget(i, 0, new QPushButton("ff"));
    }
    qt->show();
    return app.exec();
}

To copy to clipboard, switch view to plain text mode 

I reported it to TaskTracker, but they said "mail sent" and didn't put it on yet....
This is definitely a bug.

[wysota]

No because Beta doesn't give out a compiled installer with all the DLLs etc... I couldn't participate because I cannot waste my time to figure out how to compile Qt from source code.

So don't complain it's not tested. Others are wasting their time so that you can use a well written product for free (both as in free beer and free speech). Why not return the favour?

I'm trying to recreate it, but I can't recreate it in a small app, only in my own app something is missing. Maybe only happens when the thing is overloaded like my main app. With lots of buttons and rows.

Or maybe the bug is in your code and not Qt's I think that if that happens to be the case, you should officially promise to compile and test Qt4.4 beta when it comes out and report any bugs you should find.

I also checked tracker i didn't see anything but I dont know whether its been reported under a different name or title etc.

You can try using the search engine...

Meanwhile I think Qt changed its TransformationMode, so that when Border-Image is resized, using QImage, it uses TransformationMode = Quick instead of smooth like it usually does, so I think that's the cause of 99% of my blurry images. What do you think?

Actually Qt by default uses FastTransformation and it has always been like that. Please attach the picture along the QSS you have in your code that causes it to be blurry. Better yet provide a simple single function code for us to help you test it.

Oh btw here I reproduced this bug... so easy... Sorting is broken.

In what way broken? What did you expect to see and what did you receive?

I reported it to TaskTracker, but they said "mail sent" and didn't put it on yet....

It's middle of the night in Norway, let them come to work, read your mail and verify the bug first... You should receive an automatic reply stating your report has been received and somebody should get back to you on the report within upcoming days.

[VireX]

1. It's easy for them to compile since they compile it constantly to test, so why can't they just include everything the way it is in installer like lib folder and bin folder?? Is that so hard, forcing me to compile it myself just makes me run away as well as thousands of other potentially good testers. That's their fault not mine. And I will complain if it's not tested well because that is what all programmers do with their client. I tried to participate, but i couldn't find any compiled files.
2. It's not in my code, why would my code suddenly decide its not gonna work when I switch from 4.2.3 to 4.3.0, that makes no logical sense.
3. I did use the search engine.
4. Then why is my images messed up now? It's blurry, I don't have a screenshot of old version, but I am telling you it's blurry, all I had in CSS was:
border-image: url(Stuff.png) 3;
border-width: 3;

It's a round circle button. Something changed, and now all the border-image stuff is wrong, example when it stretches instead of a smooth transition in a gradient, where you dont see the lines of color changes, now you can actually see the gradient changing color every 5 lines in big images. In addition, buttons that are just not repeating or stretched using border-image are blurred or choppy/pixelated. Test it, it's simple, just make a round button image in a Qapp, then save as transparent png, and then on QPushButton{ border-image: put the image here. It will do the same to your button.

I will try and get you one...
5. I told you, buttons are randomly disappearing from the tablewidget, test the code and you'll see. Only because of sorting etc.
6. I didn't get an autoresponse... weird... Didn't appear either in tasktracker.

[jacek]

Oh btw here I reproduced this bug... so easy... Sorting is broken.

I've compiled your example using Qt 4.2.3 and got this:
ss1.png

Are you sure that the problem is really with the transition from 4.2.3 to 4.3.0?

[VireX]

Yep, see you can see the bug too.
This didn't happen in 4.2.3, in 4.2.3 this would have all the buttons there, no Ebeef etc.

And I'm not gonna resubmit this bug to tasktracker again, I typed a lot of stuff in that form, submitted it and i got a screen saying "mail sent", and I never got an email, nor did my bug appear anywhere...

[Methedrine]

Oh btw here I reproduced this bug... so easy... Sorting is broken.

I compiled this code with Qt 4.2.2, Qt 4.2.3 and Qt 4.3.0 and it always looks and behaves the same. What should the bug be, specifically?

OS: Windows XP Pro SP2
Compiler: MSVC 2005

[jacek]

Yep, see you can see the bug too.

But that particular bug is in your code.

Look:

...
qt->setSortingEnabled(true);
qt->sortItems(0, Qt::AscendingOrder);
...
for(int i = 0; i < 75; i++){
     ax[i] = new QTableWidgetItem(QString("Ebeeef%1").arg(i));
     qt->setItem(i, 0, ax[i]);
     qt->setCellWidget(i, 0, new QPushButton("ff"));
}
...

...
qt->setSortingEnabled(true);
qt->sortItems(0, Qt::AscendingOrder);
...
for(int i = 0; i < 75; i++){
     ax[i] = new QTableWidgetItem(QString("Ebeeef%1").arg(i));
     qt->setItem(i, 0, ax[i]);
     qt->setCellWidget(i, 0, new QPushButton("ff"));
}
...

To copy to clipboard, switch view to plain text mode 

First you tell Qt to sort all of the items alphabetically and then you add them in non-alphabetical order, so Qt has to sort the items after every setItem(). In result you invoke setCellWidget() either zero or more than once on certain cells.

The proper way to do it is:

for(int i = 0; i < 75; i++){
     ax[i] = new QTableWidgetItem(QString("Ebeeef%1").arg(i));
     qt->setItem(i, 0, ax[i]);
     qt->setCellWidget(ax[i]->row(), 0, new QPushButton("ff"));
}

for(int i = 0; i < 75; i++){
     ax[i] = new QTableWidgetItem(QString("Ebeeef%1").arg(i));
     qt->setItem(i, 0, ax[i]);
     qt->setCellWidget(ax[i]->row(), 0, new QPushButton("ff"));
}

To copy to clipboard, switch view to plain text mode 

It works fine on Qt 4.2.3, but on Qt 4.3.0 there's indeed some problem with cell widgets and sorting.

[VireX]

The bug is that it doesnt sort it properly, it makes random buttons disappear, this makes absolutely no sense. I dont understand your question.

Yes you are right jacek, but it is just weird that it somehow worked in 4.2.3 and not in 4.3.0

[VireX]

Ok I found a way to compare the two and show you... One image is from PHotoshop, the other is the result from border-image... See how it pixelates the borders? Started in Qt 4.3, because of the new "divide into 9 equal parts and only stretch middle pieces appropriately", whatever this was done for, I guess maybe for optimization? Or just the fact that they poorly tested this or documented it. And it's not just me saying this, many of my Qt friends are saying this so dont just blame me. Either way, they didn't need to fix what was NOT broken.




See the pixely borders? Why, it was looking so sweet in 4.2.3, why did they need to do this? Feel free to post this on TaskTracker for me, because I dont think TaskTracker likes my email.

If you cannot see it, zoom in on the image, the inner button, the inner-circle is the one used in border-image, not the outer circle, outer circle is FINE. Notice the places its pixelated? Basically It looks like they cut up that image into 9 pieces, and stretched the middle parts, exactly as the docs say... its a horrible idea, nothing matches now. Always messed up corners in almost ALL my border-images.
>>---

As for the Sorting problem, this also worked perfectly in 4.2.3, why would they change sorting in 4.3, when it worked fine in 4.2.3? I dont even see this in the change logs. And Jacek showed me how to fix my SAMPLE APP, buttttt, I cannot do this in my own application because of the way I programmed it, it would take a lot of rewriting of major classes to fix it that way.

Also, what I've noticed is, that when a button disappears it replaces the wrong area.. Like if you load the list VIA sortItems, and enable sorting, everything looks fine, but when you click sort, u see a blank button in row 3... but now row 73 (3 from last), has row 3's button... so weird.

>--- EDIT: Nother bug to our big list.. should 4.3.0 be recalled? :P ----<
deleteLater() on any window, then you try exiting App... QApplication doesn't exist.

[wysota]

Yes you are right jacek, but it is just weird that it somehow worked in 4.2.3 and not in 4.3.0

Your code was incorrect so you can't assume it worked in 4.2.3. The sorting issue might be caused because of the fact that the items with cell widgets don't contain any DisplayRole data. You could check what happens if you add some text to it.

And I'm not gonna resubmit this bug to tasktracker again, I typed a lot of stuff in that form, submitted it and i got a screen saying "mail sent", and I never got an email, nor did my bug appear anywhere...

It was 2 AM in Norway when you sent the mail! Please let people working in Trolltech come to work and download your email. When they do, the info email about the report being received will be sent to you. If you have a commercial licence, then email or phone your support and ask for help. If you're using the free licence - don't use arguments that contain the word "client" in them, because you're not Trolltech's client, so they don't have to give you anything. And there is a perfectly good reason why unstable code is not shipped in binary form.

Started in Qt 4.3, because of the new "divide into 9 equal parts and only stretch middle pieces appropriately", whatever this was done for, I guess maybe for optimization?

Correct me if I'm wrong, but border-image used to cut the image in 9 pieces in Qt 4.2 as well and it complies with what the CSS standard says. It's what it was actually meant to do...

Or just the fact that they poorly tested this or documented it. And it's not just me saying this, many of my Qt friends are saying this so dont just blame me. Either way, they didn't need to fix what was NOT broken.

Please reproduce the error using code and not using Photoshop.

Feel free to post this on TaskTracker for me, because I dont think TaskTracker likes my email.

At 2am I wouldn't like your email either

Basically It looks like they cut up that image into 9 pieces, and stretched the middle parts, exactly as the docs say...

So that's exactly what it's meant to do, right?

And Jacek showed me how to fix my SAMPLE APP, buttttt, I cannot do this in my own application because of the way I programmed it, it would take a lot of rewriting of major classes to fix it that way.

So after all it is you who has buggy code

deleteLater() on any window, then you try exiting App... QApplication doesn't exist.

Example code, please.

[VireX]

Your code was incorrect so you can't assume it worked in 4.2.3. The sorting issue might be caused because of the fact that the items with cell widgets don't contain any DisplayRole data. You could check what happens if you add some text to it.

My code was not incorrect in 4.2.3, as it worked at that time, and you yourself (or perhaps it was jacek cant remember) suggested that I do setItems under the setCellWidget.

It was 2 AM in Norway when you sent the mail! Please let people working in Trolltech come to work and download your email. When they do, the info email about the report being received will be sent to you. If you have a commercial licence, then email or phone your support and ask for help. If you're using the free licence - don't use arguments that contain the word "client" in them, because you're not Trolltech's client, so they don't have to give you anything. And there is a perfectly good reason why unstable code is not shipped in binary form.

I guess you're trying to tell me that trolltech does not know how to use server-side coding or something, because I don't believe CODE looks at time and goes "Oh well I am not going to send an auto-response at this time of night"

Correct me if I'm wrong, but border-image used to cut the image in 9 pieces in Qt 4.2 as well and it complies with what the CSS standard says. It's what it was actually meant to do...

Yes that's true, I compared the 4.2 and 4.3 docs, but for some reason, now, it doesn't work well.. Try it yourself, make a gradient about this size from black to gray, [________]
its a small image, about 100px width by 30px height; ALSO make sure you put a small 1 px border around this image in a light gray color, and make sure the gradient looks like a rounded border-radius box. Now, make a laaaaarge TreeWidget, make it about 100widthX700 height. Now set the border-image on that, border-width 2. Run the program, and you will see how horribly Qt likes to render your gradient.

But perhaps this has something to do with the "experimental DX code" that I read somewhere in the changelogs.

Please reproduce the error using code and not using Photoshop.

How many times do I have to explain to you how to reproduce it, have you tried to reproduce it with code? This is an image rendering problem, not a code problem, its a code problem inside Qt not outside.

At 2am I wouldn't like your email either

Code doesn't care about time.

So that's exactly what it's meant to do, right?

Yes but in Qt 4.2.3 it did this smoothly, but in 4.3.0, it pixelates it like as if saving your image in a bad quality.

So after all it is you who has buggy code

No, because if a code works fine in one version, then next version your code is somehow now labeled buggy, that means either the documentation for 4.2.3 was wrong, or 4.3.0 documentation is wrong. Because they both dont say anything about how I coded it as being wrong. If in 4.2.3 it worked, and 4.3.0 it doesn't it means someone in Qt changed the code for that in a bad way, and downgraded Qt's code for that purpose.

Example code, please.

I explained to you the problem, you can try it yourself, I can't post code for every time one of my friends comes to me and tells me that Qt is bugged because of whatever reason.

My question to you is this... How come I couldn't find a single bug in 4.2.3, yet I can find at least 10 in 4.3.0.... There is obviously a reason for this.

[Methedrine]

*stuff*

No code - no help. Stop wasting our time with posts like the ones you made in this thread. If you have a problem, supply a minimum compilable example (or if not possible, at least the code around the area which is causing trouble).