Why does DIRECT_RW macro look as it does?

[Jan K]

Hello,

Browsing the code, I noticed that D_RW macro (for compilers other than microsoft visual) is defined as:

#define DIRECT_RW(o) (\
{__typeof__(o) _o; _o._type = NULL; (void)_o;\
(__typeof__(*(o)._type) *)pmemobj_direct((o).oid); })

While D_RO is just:

#define DIRECT_RO(o) ((const __typeof__(*(o)._type) *)pmemobj_direct((o).oid))

Now my question is: what is the purpose of the part: __typeof__(o) _o; _o._type = NULL; (void)_o; in D_RW macro?

Would it not be just be always optimized out?

What would go wrong if D_RW were implemented just as:

#define DIRECT_RW(o) ((__typeof__(*(o)._type) *)pmemobj_direct((o).oid))

thus avoiding non-standard ({ … }) syntax and seemingly redundant statements?

Plus: do apps with no optimization (-O0) create the _o and execute the assignment _o._type = NULL?

Link to github:

https://github.com/pmem/pmdk/blob/62d4150f73b3a112670c22cefa4a9b2c0a59b838/src/include/libpmemobj/types.h#L205

Regards,

Jan

[Piotr Balcer]

Hi,

It there so that you can't D_RW on a const TOID.

And yes, it does have some impact on non-optimized assembly.

Piotr

[OpenData Synth]

Yes