Re: STL and Glibmm

[Yann LEYDIER <yann leydier info>]

Hi,

indeed, the piece of API I'm refering to is an abstract class that the 
end-user might use to create a custom parser. It's probable that nobody 
ever tried.

The other uses of std::map in Glibmm are CppOptionEntry, which I don't 
known what it does and a template class Variant. As it is a template, 
the implementation is available to the end user's compiler and it would 
only crash if an object of this type were returned by a function in Glibmm.

These 3 classes potentially don't work on Visual Studio. Since I don't 
really understand what their use is, I cannot test them, sorry for not 
being more useful.

Thanks,
yann

On 09/04/2014 12:46, Murray Cumming wrote:
> On Tue, 2014-04-08 at 17:39 +0200, Yann Leydier wrote:
> > Hi,
> >
> > when I first saw that Gtkmm was using wrappers to handle conversions to
> > STL containers, I thought that it was expected, because Visual C++'s STL
> > implementation is not fully compatible with the DLL import/export system.
> >
> > Up to now, I was using STLport to enable the use of STL in my DLLs (thus
> > I had to recompile Gtkmm with STLport…) and I'm currently looking for
> > another solution. I tried to look at how Glibmm does it and I'm quite
> > confused.
> >
> > It should be OK to work with vectors, provided you export the right
> > symbols. I suppose gendef does this, but it's difficult to check since
> > the symbols are mangled.
> >
> > According to MSDN, it should be totally impossible to use maps in the
> > API, however in markup.h (version 2.28.2), there is a typedef on a map:
> > typedef std::map<Glib::ustring, Glib::ustring,
> > Glib::Markup::AttributeKeyLess> AttributeMap;
> > This type is then used in a method of the class Parser:
> > virtual void on_start_element(ParseContext& context,
> > const Glib::ustring& element_name,
> > const AttributeMap& attributes);
> >
> > This is the same as passing a reference to a map from the main program
> > to the DLL, which should cause an undefined behaviour (in my past
> > experience it always caused a crash).
> >
> > Is it a bug? If not, does someone know why/how it works?
>
> Maybe that particular API has never been used on Windows?