Bug 76657 – [Qt] Fix the build with the newes Qt5 hashes

• Component: Tools / Tests
• Status: RESOLVED FIXED

[Andras Becsi]

Few things changed in qtbase, which causes failures with the newest Qt5 hash.
- QGLContext::currentContext now returns QFunctionPointer
- Because of the new automatic touch -> mouse mocking our touch mocking breaks in MiniBrowser and results in an infinite loop.

[Andras Becsi]

Created attachment 123179 [details]
proposed patch

[Andras Becsi]

Comment on attachment 123179 [details]
proposed patch

The MiniBrowser part is not correct. I'm going to look into this tomorrow.

[Csaba Osztrogonác]

I tried Andras' patch with the latest Qt5, layout tests work fine, 
but unfortunately qmltests API test is broken:

$ xvfb-run -a --server-args="-screen 0 1024x768x24" WebKitBuild/Release/Source/WebKit2/UIProcess/API/qt/tests/qmltests/tst_qmltests
Qml debugging is enabled. Only use this in a safe environment!
********* Start testing of qmltests *********
Config: Using QTest library 5.0.0, Qt 5.0.0
PASS   : qmltests::DesktopWebViewLinkHovered::initTestCase()
PASS   : qmltests::DesktopWebViewLinkHovered::test_linkHovered()
PASS   : qmltests::DesktopWebViewLinkHovered::test_linkHoveredDoesntEmitRepeated()
PASS   : qmltests::DesktopWebViewLinkHovered::cleanupTestCase()
PASS   : qmltests::DesktopWebViewLoadHtml::initTestCase()
PASS   : qmltests::DesktopWebViewLoadHtml::test_baseUrlAfterLoadHtml()
PASS   : qmltests::DesktopWebViewLoadHtml::cleanupTestCase()
PASS   : qmltests::DesktopWebViewMessaging::initTestCase()
Segmentation fault

[Csaba Osztrogonác]

Here is a debug backtrace with c++filt-ed output:

ASSERTION FAILED: !result
/home/oszi/WebKit/Source/JavaScriptCore/wtf/ThreadingPthreads.cpp(282) : void WTF::Mutex::lock()
1   0x7fb2b8e3f9e9 /home/oszi/WebKit/WebKitBuild/Debug/lib/libQtWebKit.so.4(WTF::Mutex::lock()+0x45) [0x7fb2b8e3f9e9]
2   0x7fb2b74338a6 /home/oszi/WebKit/WebKitBuild/Debug/lib/libQtWebKit.so.4(+0x4368a6) [0x7fb2b74338a6]
3   0x7fb2b762f6ef /home/oszi/WebKit/WebKitBuild/Debug/lib/libQtWebKit.so.4(+0x6326ef) [0x7fb2b762f6ef]
4   0x7fb2b762f95c /home/oszi/WebKit/WebKitBuild/Debug/lib/libQtWebKit.so.4(+0x63295c) [0x7fb2b762f95c]
5   0x7fb2b762f687 /home/oszi/WebKit/WebKitBuild/Debug/lib/libQtWebKit.so.4(+0x632687) [0x7fb2b762f687]
6   0x7fb2b759a635 /home/oszi/WebKit/WebKitBuild/Debug/lib/libQtWebKit.so.4(+0x59d635) [0x7fb2b759a635]
7   0x7fb2b7597c0d /home/oszi/WebKit/WebKitBuild/Debug/lib/libQtWebKit.so.4(+0x59ac0d) [0x7fb2b7597c0d]
8   0x7fb2b83cd8c3 /home/oszi/WebKit/WebKitBuild/Debug/lib/libQtWebKit.so.4(+0x13d08c3) [0x7fb2b83cd8c3]
9   0x7fb2b759755f /home/oszi/WebKit/WebKitBuild/Debug/lib/libQtWebKit.so.4(+0x59a55f) [0x7fb2b759755f]
10  0x7fb2b7738058 /home/oszi/WebKit/WebKitBuild/Debug/lib/libQtWebKit.so.4(+0x73b058) [0x7fb2b7738058]
11  0x7fb2b7737ad5 /home/oszi/WebKit/WebKitBuild/Debug/lib/libQtWebKit.so.4(+0x73aad5) [0x7fb2b7737ad5]
12  0x7fb2b7737616 /home/oszi/WebKit/WebKitBuild/Debug/lib/libQtWebKit.so.4(+0x73a616) [0x7fb2b7737616]
13  0x7fb2b7597fae /home/oszi/WebKit/WebKitBuild/Debug/lib/libQtWebKit.so.4(+0x59afae) [0x7fb2b7597fae]
14  0x7fb2b757a500 /home/oszi/WebKit/WebKitBuild/Debug/lib/libQtWebKit.so.4(+0x57d500) [0x7fb2b757a500]
15  0x7fb2b75ebbe9 /home/oszi/WebKit/WebKitBuild/Debug/lib/libQtWebKit.so.4(+0x5eebe9) [0x7fb2b75ebbe9]
16  0x7fb2b7570fd9 /home/oszi/WebKit/WebKitBuild/Debug/lib/libQtWebKit.so.4(+0x573fd9) [0x7fb2b7570fd9]
17  0x7fb2b74ca0ad /home/oszi/WebKit/WebKitBuild/Debug/lib/libQtWebKit.so.4(+0x4cd0ad) [0x7fb2b74ca0ad]
18  0x7fb2b74ca28f /home/oszi/WebKit/WebKitBuild/Debug/lib/libQtWebKit.so.4(+0x4cd28f) [0x7fb2b74ca28f]
19  0x7fb2b74d4647 /home/oszi/WebKit/WebKitBuild/Debug/lib/libQtWebKit.so.4(+0x4d7647) [0x7fb2b74d4647]
20  0x7fb2b74d4350 /home/oszi/WebKit/WebKitBuild/Debug/lib/libQtWebKit.so.4(+0x4d7350) [0x7fb2b74d4350]
21  0x7fb2b772d942 /home/oszi/WebKit/WebKitBuild/Debug/lib/libQtWebKit.so.4(+0x730942) [0x7fb2b772d942]
22  0x7fb2b8098937 /home/oszi/WebKit/WebKitBuild/Debug/lib/libQtWebKit.so.4(+0x109b937) [0x7fb2b8098937]
23  0x7fb2b8364ad6 /home/oszi/WebKit/WebKitBuild/Debug/lib/libQtWebKit.so.4(+0x1367ad6) [0x7fb2b8364ad6]
24  0x7fb2b8365911 /home/oszi/WebKit/WebKitBuild/Debug/lib/libQtWebKit.so.4(+0x1368911) [0x7fb2b8365911]
25  0x7fb2b382dca6 /usr/local/Trolltech/Qt5/Qt-5.0.0-r19/lib/libQtCore.so.5(QObject::event(QEvent*)+0x396) [0x7fb2b382dca6]
26  0x7fb2b43749dc /usr/local/Trolltech/Qt5/Qt-5.0.0-r19/lib/libQtWidgets.so.5(QApplicationPrivate::notify_helper(QObject*, QEvent*)+0xac) [0x7fb2b43749dc]
27  0x7fb2b4379df2 /usr/local/Trolltech/Qt5/Qt-5.0.0-r19/lib/libQtWidgets.so.5(QApplication::notify(QObject*, QEvent*)+0x152) [0x7fb2b4379df2]
28  0x7fb2b3812344 /usr/local/Trolltech/Qt5/Qt-5.0.0-r19/lib/libQtCore.so.5(QCoreApplication::notifyInternal(QObject*, QEvent*)+0x84) [0x7fb2b3812344]
29  0x7fb2b3816452 /usr/local/Trolltech/Qt5/Qt-5.0.0-r19/lib/libQtCore.so.5(QCoreApplicationPrivate::sendPostedEvents(QObject*, int, QThreadData*)+0x402) [0x7fb2b3816452]
30  0x7fb2b38520a3 /usr/local/Trolltech/Qt5/Qt-5.0.0-r19/lib/libQtCore.so.5(+0x1ff0a3) [0x7fb2b38520a3]
31  0x7fb2b05cc6f2 /lib/libglib-2.0.so.0(g_main_context_dispatch+0x1f2) [0x7fb2b05cc6f2]
Segmentation fault

[Andras Becsi]

(In reply to comment #4)
> Here is a debug backtrace with c++filt-ed output:
> 
> ASSERTION FAILED: !result
> /home/oszi/WebKit/Source/JavaScriptCore/wtf/ThreadingPthreads.cpp(282) : void WTF::Mutex::lock()
> Segmentation fault
Tor Arne, do you eventually know whether something changed in the Qt5 QMutex code which could cause this crash?
The backtrace does not say much :(

[Tor Arne Vestbø]

(In reply to comment #5)
> (In reply to comment #4)
> > Here is a debug backtrace with c++filt-ed output:
> > 
> > ASSERTION FAILED: !result
> > /home/oszi/WebKit/Source/JavaScriptCore/wtf/ThreadingPthreads.cpp(282) : void WTF::Mutex::lock()
> > Segmentation fault
> Tor Arne, do you eventually know whether something changed in the Qt5 QMutex code which could cause this crash?
> The backtrace does not say much :(

Try checking with Joao or Brad

[Rafael Brandao]

(In reply to comment #5)
> (In reply to comment #4)
> > Here is a debug backtrace with c++filt-ed output:
> > 
> > ASSERTION FAILED: !result
> > /home/oszi/WebKit/Source/JavaScriptCore/wtf/ThreadingPthreads.cpp(282) : void WTF::Mutex::lock()
> > Segmentation fault
> Tor Arne, do you eventually know whether something changed in the Qt5 QMutex code which could cause this crash?
> The backtrace does not say much :(

This is a problem with our icon client. I have a fix pending for review that will fix this.

[Anselmo L. S. Melo]

Created attachment 123337 [details]
Patch

[Anselmo L. S. Melo]

Qt5/QtBase 9702400e modified QGLContext::getProcAddress(), which now returns QFunctionPointer. The proposed patch fixes the build.

[Anselmo L. S. Melo]

Comment on attachment 123337 [details]
Patch

Nevermind, the previous patch already does this :-)
And in fact, my patch should say QGLContext::currentContext(), not getProcAddress()...

[Andras Becsi]

Created attachment 123350 [details]
proposed patch

After a discussion with Tor Arne and Zeno the decision was to temporarily disable the automatic touch->mouse synthesis in Qt to avoid the infinite loop with QGuiApplicationPrivate::processTouchEvent and MiniBrowser's touch mocking.

After we make it possible to set the synthesized flag (which is currently internally used to prevent infinite loops) on our touch mock events we can enable it again and mark our touch events synthesized.

This decision was made because using QWindowSystemInterface::handleTouchEvent to sent the mock events is the most straightforward solution.
If we would create the QTouchEvent's ourselves, we would need to set all the positions and cache the last positions, which would result in more code and would make the implementation more error prone to changes in QQuickCanvasPrivate::translateTouchEvent and other places where touch points are adjusted.

[Balazs Kelemen]

I am not sure how it is related but we experienced another bug with the touch mocking: seems like it is also active on touch devices. When I built WebKit to N9 and tried MiniBrowser it also produced the blue bubbles on the touch points and double touch was not working. However Zoltán Árvai informed me his last week build was working correctly. Maybe it's a side effect of the changes in Qt?

[Andras Becsi]

(In reply to comment #12)
> I am not sure how it is related but we experienced another bug with the touch mocking: seems like it is also active on touch devices. When I built WebKit to N9 and tried MiniBrowser it also produced the blue bubbles on the touch points and double touch was not working. However Zoltán Árvai informed me his last week build was working correctly. Maybe it's a side effect of the changes in Qt?

I did not try it on the device but it is the same issue indeed.
I found the bug in Qt and am on my way to fix it. Thanks.

[Csaba Osztrogonác]

(In reply to comment #4)
> Here is a debug backtrace with c++filt-ed output:
> 
> ASSERTION FAILED: !result
> /home/oszi/WebKit/Source/JavaScriptCore/wtf/ThreadingPthreads.cpp(282) : void WTF::Mutex::lock()

This crash is absolutely unrelated to Qt5 revision update. It is an old bug
but unfortunately we didn't notive it before because a serious bug in the API
test runner script: https://bugs.webkit.org/show_bug.cgi?id=76906

I filed a new bug to fix this crash: https://bugs.webkit.org/show_bug.cgi?id=76906

Now the latest Qt5 works fine for me, so update is coming in an hour.

[Csaba Osztrogonác]

Comment on attachment 123350 [details]
proposed patch

Clearing flags on attachment: 123350

Committed r105733: <http://trac.webkit.org/changeset/105733>

[Csaba Osztrogonác]

All reviewed patches have been landed.  Closing bug.

[Rafael Brandao]

Anyone else into this?

Source/WebKit2/WebProcess/qt/QtNetworkAccessManager.cpp:115:46: error: no matching function for call to 'QNetworkReply::ignoreSslErrors(const QList<QSslError>&)'

I've just verified with bbandix's hashes https://gist.github.com/1670199 and they're all the same. WebKit head is at 3c79efc755327 and I'm building with build-webkit --qt --release --makeargs=-j30

[Csaba Osztrogonác]

(In reply to comment #17)
> Anyone else into this?
> 
> Source/WebKit2/WebProcess/qt/QtNetworkAccessManager.cpp:115:46: error: no matching function for call to 'QNetworkReply::ignoreSslErrors(const QList<QSslError>&)'
> 
> I've just verified with bbandix's hashes https://gist.github.com/1670199 and they're all the same. WebKit head is at 3c79efc755327 and I'm building with build-webkit --qt --release --makeargs=-j30

Have you got libssl-dev package installed? It seems no. :)
Because QNetworkReply::ignoreSslErrors is defined in #ifndef QT_NO_OPENSSL guard.

[Rafael Brandao]

(In reply to comment #18)
> (In reply to comment #17)
> > Anyone else into this?
> > 
> > Source/WebKit2/WebProcess/qt/QtNetworkAccessManager.cpp:115:46: error: no matching function for call to 'QNetworkReply::ignoreSslErrors(const QList<QSslError>&)'
> > 
> > I've just verified with bbandix's hashes https://gist.github.com/1670199 and they're all the same. WebKit head is at 3c79efc755327 and I'm building with build-webkit --qt --release --makeargs=-j30
> 
> Have you got libssl-dev package installed? It seems no. :)
> Because QNetworkReply::ignoreSslErrors is defined in #ifndef QT_NO_OPENSSL guard.

It did the trick! Thanks. :-)