Re: [PATCH] Check DECL_CONTEXT of new/delete operators.

["Martin Liška" <mliska@suse.cz>]

On 4/6/20 2:45 PM, Nathan Sidwell wrote:
> On 4/6/20 4:34 AM, Martin Liška wrote:
> 
>>
>> May I please ping Jason, Nathan and Jonathan who can help us here?
> 
> On IRC Martin clarified the question as essentially 'how do you pair up operator new and operator delete calls?' (so you may delete them if the object is not used).
> 
> I am not sure you're permitted to remove those calls in general.  All I can find is [expr.new]/12
> 'An implementation is allowed to omit a call to a replaceable global allocation function (17.6.2.1, 17.6.2.2). When it does so, the storage is instead provided by the implementation or provided by extending the allocation of another new-expression.'
> 
> But, I suspect the optimization is safe, in that either no one counts objects by their allocation, or if they do, they don't actually care that the std-conforming number of allocations happen.
> 
> The both operator new and operator delete are looked up in the same manner.  The std does not require a 'matching pair' be found.  but it is extremely poor form for a class to declare exactly one of operator {new,delete}.

Hi.

Thank you for the information.

> 
> The following is well formed:
> 
> struct PoorForm {
>    void *operator new (size_t s) {count++; return ::operator new (s)};
>    static unsigned count;
> };
> 
> Have you considered throwing ctors?
> 
> struct Obj {
>    Obj (); // might throw
> };
> 
> 'obj = new Obj (); ... delete obj;' sequence expand to something like ...
> 
> // obj = new Obj ();
> void *p = ::operator new (sizeof (Obj));
> try {
>    Obj::ctor(p);
> }
> catch (...) // cleanup code
> {
>    ::operator delete (p); // #1
>    throw;
> }
> 
> obj = (Obj*)p;
> 
> .... user code
> 
> // delete obj;
> Obj::dtor (obj);
> ::operator delete (obj); // #2
> 
> calls 1 & 2 matchup to the operator new call

Looking at the throwing ctors:

$ cat new-delete2.C
#include <stdio.h>

struct A
{
   __attribute__((always_inline)) A(int x)
   {
     if (x == 123)
       throw x;
   }
};

int main(int argc, char **argv) {
   A *a = new A (argc);
   delete a;

   return 0;
}

$ g++-9 new-delete2.C -O2 -c -fdump-tree-optimized=/dev/stdout
...
   <bb 2> [local count: 1073741824]:
   _3 = operator new (1);
   if (argc_4(D) == 123)
     goto <bb 3>; [0.00%]
   else
     goto <bb 4>; [100.00%]

   <bb 3> [count: 0]:
   _8 = __cxa_allocate_exception (4);
   MEM[(int *)_8] = 123;
   __cxa_throw (_8, &_ZTIi, 0B);

   <bb 4> [local count: 1073741824]:
   operator delete (_3, 1);
   return 0;
...

As seen cddce can still optimize out
   _3 = operator new (1);
...
   operator delete (_3, 1);

Martin

> 
> Notice that para I quoted allows one to coalesce allocations using the global operator new/deletes.  The rules are pretty much as you can guess -- one lifetime must be entirely within the other.  If inner one's ctor throws, the exception path must destroy the outer.
> 
> does that help?
> 
> nathan
>