Global variable for the world?

[prissi]

Sorry, I am a little at loss of the many stuff here.

So why npot make  a static member of karte_t called w. karte_t::w will be either NULL (bei definition in the header file) or will be set in init (the first point karte_t contains useful content) to this. Simworld.h is anyway include everywhere where the karte_t this is needed.

Code Select Expand


in simworld.h

#define WORLD (karte_t::w)

class karte_t
{
static karte_t *w;
...
};

in simworld.cc

karte_t karte_t::w = NULL;
....

void karte_t::init()
{
...
karte_t::w = this;
}

void karte_t::destroy()
{
....
karte_t::w = NULL;
}

everywhere else
welt -> karte_t::w
or
WORLD->

Using a define would allow for a different funtion for debug levels>3 ...

This is threadsafe by desgin, would also allow for testing karte_t::w for NULL (map may exist but not properly no initialized or is already destroyed), and you need to include no new header. Accessing a static is a fast as it could be. I fail to see the advantage of a special pointer class, when there should be no test on accessing it anyway. But I admid I may be thick, and happily hear more explaination.

[Dwachs]

The idea was to implement something that needs as less modification of other files as possible. This means that the static members 'karte_t* welt' in the code still are static members (now of a different type).

Macros may or may not lead to different problems if 'welt' is replaced by 'karte_t::w' and the word 'welt' is used in a different context, for instance as name of a function parameter. I will try.

[Markohs]

I'd stay away from macros, they are allways a source of problems.

[prissi]

Ok, I can understand to stay away from macros.

karte_prt_t could achieve the same if it returns karte_t::welt and still karte_t  decides when it is time for initialisation.

Code Select Expand


class karte_ptr_t
{
karte_t* operator->() { return karte_t::world; }
...
}

and init then in karte_t::init() and NULL in destroy.

The extra list of all instances is not needed, respective only needed if there ever going to be more than one world (even then a static access would work). Also the list seems not very thread save. Or am I overlooking something again?

EDIT: forgot the patch

[Dwachs]

this is now incorporated, r685x.

[Dwachs]

Here is a follow-up patch that removes the 'karte_t*' parameters in functions of fabrik_t, some related gui windows lost karte_t* parameters as well. Should we proceed in this direction? Ie transform the code to not pass any karte_t parameters around?

[kierongreen]

Well if there's no need to then seems a bit pointless passing them around.

[prissi]

I agree not passing around. However, MSVC complains that struct cannot be freind, class can. So I declare karte_ptr_t a class then ...

[Dwachs]

The big patch is committed r6871