Re: GSIMap and gcc 2.95

[Fred Kiefer]

On 16.08.2011 13:06, Riccardo Mottola wrote:
> what goes wrong here?
>
>
> Compiling file GSObjCRuntime.m ...
> In file included from GSObjCRuntime.m:727:
> ../../Headers/GNUstepBase/GSIMap.h: In function `GSIMapFreeNode':
> ../../Headers/GNUstepBase/GSIMap.h:465: cast to union type from type not
> present in union
> ../../Headers/GNUstepBase/GSIMap.h:468: cast to union type from type not
> present in union
> ../../Headers/GNUstepBase/GSIMap.h: In function `GSIMapCleanMap':
> ../../Headers/GNUstepBase/GSIMap.h:1186: cast to union type from type
> not present in union
> ../../Headers/GNUstepBase/GSIMap.h:1190: cast to union type from type
> not present in union

I think this is caused by a compiler bug, that can be safely ignored.
The lines in question are similar to:
GSI_MAP_CLEAR_KEY(node);

And for the case GSI_MAP_KTYPES = GSUNION_PTR this is defined as:
#define GSI_MAP_CLEAR_KEY(node)  GSI_MAP_WRITE_KEY(map, &node->key, 
(GSIMapKey)NULL)

And GSIMapKey is a union that contains in that case void* as a possible 
type. I don't understand why NULL isn't detected as belonging to that 
type, but you could try to add another type cast to the line above and 
see that gets reported.
#define GSI_MAP_CLEAR_KEY(node)  GSI_MAP_WRITE_KEY(map, &node->key, 
(GSIMapKey)(void *)NULL)
Or rather use (char*)? I really don't see what goes on in the compiler 
to complain here.