Qt-interest Archive, April 2008
Qt 4.3.2 - Unable to load MySQL plugin on Win32
Message 1 in thread
1. I'm using Qt the 4.3.2 opensource package and have built
it with VSC++ 2005 Express edition.
2. The following program:
> int main (int argc, char *argv[])
> {
> QCoreApplication qapp(argc, argv);
> qDebug() << QSqlDatabase::drivers();
> }
emits the following when run with QT_DEBUG_PLUGINS=1:
> QFactoryLoader::QFactoryLoader() looking at "E:/qt/qt-win-opensource-src-4.3.2/plugins/sqldrivers/qsqlited4.dll"
> keys ("QSQLITE")
>
> QFactoryLoader::QFactoryLoader() looking at "E:/qt/qt-win-opensource-src-4.3.2/plugins/sqldrivers/qsqlmysqld4.dll"
> "The file 'E:/qt/qt-win-opensource-src-4.3.2/plugins/sqldrivers/qsqlmysqld4.dll' is not a valid Qt plugin."
> not a plugin
3. I've compiled the plugins both with and without
CONFIG -= embed_manifest_dll
in src/plugins/sqldrivers/mysql/mysql.pro. The results
are the same with each build.
4. I've taken a look at qsqlmysqld4.dll with the Windows dependency
walker tool, and AFAICS, it finds all required modules in the dependency
tree correctly (in particular, it finds the mysql.dll dependency), and
the symbols exported (2 of them) are identical to those in the SQLITE
driver, which is loaded fine.
5. I've debugged QSqlDatabase::drivers() to the point where I can
see that the code in QLibraryPrivate::isPlugin(), line 586 is the
culprit:
> 573 QString regkey = QString::fromLatin1("Qt Plugin Cache %1.%2.%3/%4")
> 574 .arg((QT_VERSION & 0xff0000) >> 16)
> 575 .arg((QT_VERSION & 0xff00) >> 8)
> 576 .arg(QLIBRARY_AS_DEBUG ? QLatin1String("debug") : QLatin1String("false"))
> 577 .arg(fileName);
> 578 QStringList reg;
> 579
> 580 QSettings settings(QSettings::UserScope, QLatin1String("Trolltech"));
> 581 reg = settings.value(regkey).toStringList();
> 582 if (reg.count() == 4 && lastModified == reg.at(3)) {
> 583 qt_version = reg.at(0).toUInt(0, 16);
> 584 debug = bool(reg.at(1).toInt());
> 585 key = reg.at(2).toLatin1();
> 586 success = qt_version != 0;
> 587 } else {
In my case, qt_version is 0 which is the point at which isPlugin()
gives up. This is something to do with the plugin cache, which I don't
understand fully; however, I get the impression that the plugin loader
has tried previously (when ?) to load the plugin and cached someinfo
about its failure in QSettings, so I'm no further forward in finding the
root cause of the failure to load.
So:
1) does anyone have any idea as to what I've screwed up ?
(I'm currently leaning towards the subtle-bug-in-Qt/VSC++
option, but I'm willing to be convinced otherwise)
2) is there anything more I can do to track this problem down ?
I'm currently stumped. I've found a single reference to
QT_DEBUG_COMPONENTS in Qt Assistant but I'm not sure what that's
for - should I be rebuilding everything with this defined, and
trying again ?
--
[ signature omitted ]
Message 2 in thread
qtabwidget with vertical tabs
this feature is registered as suggestion on task tracker with ID186458.
http://trolltech.com/customer/task-tracker/index_html?id=186458&method=entry
this link offer possibility ONLY for customers (commercial qt licence
owners?...) to vote.
this vote help Trolltech to take decision about doing or not or later a
suggested feature.
Don't know if today there is a public access to vote (it was planed by
trolls...)
so, customers if you're interested by vertical tab widgets, like this
http://www.nuos.edu.ua/~pavel/psi/psi1.jpg
or
http://www.nuos.edu.ua/~pavel/psi/psi-win2.PNG
the author of this screenshots sent a patch to trolltech so ....
vote!
--
[ signature omitted ]
Message 3 in thread
Stephen Collyer schrieb:
<snip>
> So:
>
> 1) does anyone have any idea as to what I've screwed up ?
> (I'm currently leaning towards the subtle-bug-in-Qt/VSC++
> option, but I'm willing to be convinced otherwise)
>
> 2) is there anything more I can do to track this problem down ?
> I'm currently stumped. I've found a single reference to
> QT_DEBUG_COMPONENTS in Qt Assistant but I'm not sure what that's
> for - should I be rebuilding everything with this defined, and
> trying again ?
>
Maybe it's worth taking a look / deleting
HKCU\Software\Trolltech\OrganizationDefaults
HTH
Christian
--
[ signature omitted ]
Message 4 in thread
Christian Ehrlicher wrote:
> Stephen Collyer schrieb:
> <snip>
>> So:
>>
>> 1) does anyone have any idea as to what I've screwed up ?
>> (I'm currently leaning towards the subtle-bug-in-Qt/VSC++
>> option, but I'm willing to be convinced otherwise)
>>
>> 2) is there anything more I can do to track this problem down ?
>> I'm currently stumped. I've found a single reference to
>> QT_DEBUG_COMPONENTS in Qt Assistant but I'm not sure what that's
>> for - should I be rebuilding everything with this defined, and
>> trying again ?
>>
> Maybe it's worth taking a look / deleting
> HKCU\Software\Trolltech\OrganizationDefaults
Toll, prima, ausgezeichnet !!!!!!
You are a star. That solved the problem. However, it raises
another question, since the value in the appropriate key
contained a date stamp identical to the time at which I built
the .dll so presumably, the library loader failed to load the
same .dll at least once, so why should deleting the plugin cache
fix it ?
--
[ signature omitted ]
Message 5 in thread
Stephen Collyer schrieb:
> Christian Ehrlicher wrote:
>> Stephen Collyer schrieb:
>> <snip>
>>> So:
>>>
>>> 1) does anyone have any idea as to what I've screwed up ?
>>> (I'm currently leaning towards the subtle-bug-in-Qt/VSC++
>>> option, but I'm willing to be convinced otherwise)
>>>
>>> 2) is there anything more I can do to track this problem down ?
>>> I'm currently stumped. I've found a single reference to
>>> QT_DEBUG_COMPONENTS in Qt Assistant but I'm not sure what that's
>>> for - should I be rebuilding everything with this defined, and
>>> trying again ?
>>>
>> Maybe it's worth taking a look / deleting
>> HKCU\Software\Trolltech\OrganizationDefaults
>
> Toll, prima, ausgezeichnet !!!!!!
>
> You are a star. That solved the problem. However, it raises
> another question, since the value in the appropriate key
> contained a date stamp identical to the time at which I built
> the .dll so presumably, the library loader failed to load the
> same .dll at least once, so why should deleting the plugin cache
> fix it ?
>
Maybe the lib could not be loaded because of a missing dep whic became
available later?
The cache is fine but imho failed plugins (which failed to load, not
which have a to low version number or other things) should not go into
the plugin cache / should be checked the next time (maybe with a counter
to stop loading after x times or so)...
Christian
--
[ signature omitted ]
Message 6 in thread
Christian Ehrlicher wrote:
> Stephen Collyer schrieb:
>> Christian Ehrlicher wrote:
>>> Stephen Collyer schrieb:
>>> <snip>
>>>> So:
>>>>
>>>> 1) does anyone have any idea as to what I've screwed up ?
>>>> (I'm currently leaning towards the subtle-bug-in-Qt/VSC++
>>>> option, but I'm willing to be convinced otherwise)
>>>>
>>>> 2) is there anything more I can do to track this problem down ?
>>>> I'm currently stumped. I've found a single reference to
>>>> QT_DEBUG_COMPONENTS in Qt Assistant but I'm not sure what that's
>>>> for - should I be rebuilding everything with this defined, and
>>>> trying again ?
>>>>
>>> Maybe it's worth taking a look / deleting
>>> HKCU\Software\Trolltech\OrganizationDefaults
>>
>> Toll, prima, ausgezeichnet !!!!!!
>>
>> You are a star. That solved the problem. However, it raises
>> another question, since the value in the appropriate key
>> contained a date stamp identical to the time at which I built
>> the .dll so presumably, the library loader failed to load the
>> same .dll at least once, so why should deleting the plugin cache
>> fix it ?
>>
> Maybe the lib could not be loaded because of a missing dep whic became
> available later?
> The cache is fine but imho failed plugins (which failed to load, not
> which have a to low version number or other things) should not go into
> the plugin cache / should be checked the next time (maybe with a counter
> to stop loading after x times or so)...
OK, I now understand. I had to correct the path to the MySQL lib
after I first tried to load the plugin. At the time, I wasn't
aware of the plugin caching mechanism, so I expected it to work
immediately.
I agree that the plugin caching logic should be changed, or
at least there should be some more more info available on
failed load attempts.
--
[ signature omitted ]
