Re: type-punning warnings with optimizations turned on?

[jcupitt gmail com]

On 7/5/07, Alan M. Evans <gtkappdevellist alanevans org> wrote:
> But with optimizations on (actually, at or above -O2), I get warnings
> about type-punning on the calls to g_static_mutex_lock/unlock:
>
>         g++ -O2 -Wall -c `pkg-config --cflags glib-2.0` test.cpp
>         test.cpp: In function 'void MyMutex_lock(MyMutex*)':
>         test.cpp:19: warning: type-punning to incomplete type might break strict-aliasing rules

I think this is a gcc 4.1 issue. I found I couldn't cast from a Thing
*x[] to a void ** in one step without triggering a warning like this.
If I cast to void* and then to void **, all was well.

Anyway, insert a mysterious extra cast and the warnings vanish. Try
compiling this with -O2 -Wall and you'll see only one line triggers a
warning.

============
void
test (void **base)
{
}

int
main ()
{
 int *thing[3];

 test ((void **) thing);
 test ((void **) ((void *) thing));

 return (0);
}
==============

(now someone will tell me the compiler is correct and casting "int
*x[]" to "void **" is indeed wrong :-)

John