[Development] Q_DECLARE_TYPEINFO and namespaced Qt

[Olivier Goffart]

On Tuesday 10 July 2012 18:37:20 Marc Mutz wrote:
> Hi,
> 
> as seen on IRC:
> [18:18:43] <ossi|tt> thiago: how about we make Q_DECLARE_TYPEINFO
> automatically use the qt namespace, like Q_DECLARE_METATYPE does? it is
> massively SIC, except that probably not many outside qt/creator use it ...
> [18:20:09] <marc_kdab_> ossi|tt: errhm, I used it in many projects, and the
> Q_DECLARE_METATYPE solution is also not the best one
> [18:20:22] <marc_kdab_> though I'm all for consistency, too
> [18:20:34] <ossi|tt> marc_kdab_: but you probably didn't use qt in
> namespace? [18:21:08] <marc_kdab_> could someone with VS 2005 and/or 2008
> please test-drive https://codereview.qt-project.org/#/t/35/ ?
> [18:21:42] <marc_kdab_> ossi|tt: I made KDTools work with a namespaced Qt,
> and hit the problem of Q_DECLARE_TYPEINFO there
> [18:22:23] <ossi|tt> marc_kdab_: yeah. i'd instantly change it. not sure
> about the amount of opposition to expect, though
> [18:22:50] <marc_kdab_> ossi|tt: the question is: change which way?
> [18:23:26] <ossi|tt> marc_kdab_: have it in the namespace transparently.
> it's insane that user code needs to be aware of this in this context
> [18:24:21] <marc_kdab_> ossi|tt: but user code needs to be aware of a
> namespaced Qt in any case. Wrapping the QClass forward-declarations was the
> most work, and then wrapping Q_DECLARE_TYPEINFO, too, is just consistent
> [18:24:47] <marc_kdab_> (wrapping in QT_BEGIN_NAMESPACE /
> QT_END_NAMESPACE, that is)
> [18:25:06] <ossi|tt> marc_kdab_: yes, but it's stupid that i need to wrap a
> macro with another macro pair
> [18:26:12] <marc_kdab_> ossi|tt: depends how your take is on consistency, I
> guess [18:27:42] <marc_kdab_> I don't have an opinion myself; both ways
> have pros and cons... [18:27:59] <marc_kdab_> if in doubt, stay with the
> status quo, I guess [18:28:06] <ossi|tt> marc_kdab_: this macro is clearly
> optimized for the use in qt, where everything is wrapped in these macros
> anyway
> [18:28:35] <ossi|tt> marc_kdab_: the easiest way out would be
> Q_DECLARE_NAMESPACED_TYPEINFO
> 
> As I said, I'm undecided. On one hand, I always have to look at the
> implementation to see whether I need to wrap the macro in
> QT_BEGIN_NAMESPACE /
> QT_END_NAMESPACE, and I would like to see consistency here, on the other
> hand, in a way it's Q_DECLARE_METATYPE that's inconsistent, since it
> violates the "put Qt stuff into QT_BEGIN_NAMESPACE if you want to be
> compatible with a namespaced Qt, otherwise don't bother" assumption.
> 
> What's the general opinion on this?

My opinion is to fix Q_DECLARE_TYPEINFO to do the same as in 
Q_DECLATE_METATYPE (as ossi says)

It is source incompatible only with namespaced Qt, that are not that common 
anyway.