Le 2011-11-12 à 17:13:00, Claude Heiland-Allen a écrit :

1. This does not work in C++ because implicit casts from void* are all assumed to be mistakes ;

2. In any case, this requires that you did type f in the malloc. The problem we have here is because the wrong type name was typed in a sizeof. If a wrong type name can be typed there, so can a wrong variable name.

Therefore, either use «operator new», or if you're stuck with C, use a macro that does the job for you :

#define NEW(T) ((T*)malloc(sizeof(T)))

so that if you do a wrong assignment, you will get an error for it :

struct foo *e = NEW(foo); struct foo *f = NEW(typeof(*f)); struct bar *g = NEW(foo); // wrong struct bar *h = NEW(typeof(*f)); // wrong

error: cannot convert ‘foo*’ to ‘bar*’ in initialization error: cannot convert ‘foo*’ to ‘bar*’ in initialization

But anyway, those mallocs in [fux_kinect] are superfluous because you can reserve that amount of RAM directly in fux_kinect by removing the «*».

That goes for nearly any struct that you need to create at the same time as another struct, and destroy at the same time as that same struct, and never need to change the size of. In order to do that, you need to be able to init and finalise separately from malloc/free. That's why the function is named pthread_cond_init and not pthread_cond_new, for example.