[Development] Experimental Qt 5 installers by Digia

lars.knoll at nokia.com lars.knoll at nokia.com
Tue Jul 31 11:27:11 CEST 2012


On Jul 31, 2012, at 11:10 AM, ext Thiago Macieira <thiago.macieira at intel.com> wrote:

> On terça-feira, 31 de julho de 2012 08.54.56, shane.kearns at accenture.com 
> wrote:
>> Or can we load it using dlopen like we do for openssl?
> 
> We used to, but it was also hardcoded to a single soname version. If it 
> couldn't find that one, it would print an annoying qWarning and disable ICU 
> functionality.
> 
> If this is the best solution, we can bring dlopening back. But first I'd like 
> to hear from the webkit people how they're dealing with the issue.

Currently webkit links to ICU directly (i.e., the issue is being ignored). WebKit seems to be using quite a bit of ICU functionality, but I would assume that the same will be true for Qt in the longer term.

One solution that came to my mind is to have a qicu.h header as follows:

namespace QIcu {
	UConverter *ucnv_open(…);
	…
}

#ifdef QT_USE_QICU_SYMBOLS
using namespace QIcu;
#endif

These methods would then dispatch to the real ICU methods that we resolved.

For WebKit to use this, we would need a solution where we can transparently remap the existing ICU symbols to the ones we export from Qt. One way might be to force the inclusion of the qicu.h header file and set the QT_USE_QICU_SYMBOLS on the compiler command line.

Cheers,
Lars




More information about the Development mailing list