Why is it important to dethunk or to recognize as a thunk?

 

For comparison/function object or address identity?

 

For speed?
 

-----Ursprüngliche Nachricht-----
Von: Frederick Virchanza Gotham via Std-Proposals <std-proposals@lists.isocpp.org>
Gesendet: Mo 08.01.2024 21:43
Betreff: Re: [std-proposals] Allow conversion of memfunc pointers to func pointers
An: std-proposals@lists.isocpp.org;
CC: Frederick Virchanza Gotham <cauldwell.thomas@gmail.com>;


On Monday, January 8, 2024, Thiago Macieira wrote:
But again pointer-equality problems because
the entry point of the function you de-virtualised may be that of the thunk
instead of the actual function.
 
 
    So maybe we could use 'std::dethunk' to de-thunk the function pointer.
 
    q = std::dethunk( p );
    if ( q == p )  cout << "It wasn't a thunk\n";
    else cout << "Actual function address: 0x" << hex << (void*)q << endl;
 
So then the question is how do we identify thunks at runtime? Well we could binary-search for the address in a global constexpr array of thunk addresses. Or alternatively we could precede every thunk with 8 identifying bytes something like:
 
    't' 'h' 'u' 'n' 'k' '\0' '\0' '\0'
 
Or maybe just identify the machine code of a thunk generated from the following assembler:
 
    add rdi, 8
    jmp ActualFunction
 
If it adds a constant to RDI and immediately jumps to a constant address then assume it to be a thunk (although this could be problematic if a non-thunk function which begins with a loop (and is misidentified as a thunk)).
 
I would really like to see a function to retrieve the function pointer from the vtable:
 
    class MyClass {  . . . };
    MyClass myobj;
    void (*p)(void) = std::devirtualise( &MyClass::SomeMethod, &myobj );
 
so then you could chain them:
 
    void (*p)(void) = std::dethunk( std::devirtualise( &MyClass::SomeMethod, &myobj ) );
 
On computers that don't use thunks, "std::dethunk" shall be optimised away to a nop.
 
 
-- 
 Std-Proposals mailing list
 Std-Proposals@lists.isocpp.org
 https://lists.isocpp.org/mailman/listinfo.cgi/std-proposals