Re: [boost] change to weak_ptr.hpp to compile under Builder C++ 5,etc

From: Peter Dimov (pdimov_at_[hidden])
Date: 2002-05-15 06:21:40

From: "Thomas Witt" <witt_at_[hidden]>
>
> On Tuesday 14 May 2002 20:02, Peter Dimov wrote:
> > This will no longer be necessary with 1.28.0; we've removed operator*
and
> > operator-> from weak_ptr as unsafe. Now the only way to do something
with a
> > weak_ptr is to construct a shared_ptr first, via its constructor or
> > make_shared.
>
> Peter,
>
> can you explain why make_shared adds safety. Currently I am unable to see
why.
> IIUC what happens is a NULL ptr is dereferenced with or without
make_shared.
> On the first look it seems as if make_shared pretends to be safe but
isn't.
> When make shared is used the object might be gone already.

Consider this example:

weak_ptr<T> wp;

// ...

if(wp.get() != 0)
{
    wp->f();
    wp->g();
}

Looks innocent enough, but wp->f() may invalidate wp (by resetting the
shared_ptr holding the object), and wp->g() will crash.

Compare with the idiomatic make_shared approach:

if(shared_ptr<T> sp = make_shared(wp))
{
    sp->f();
    sp->g();
}

Now you have another shared_ptr, sp, keeping the object alive, preventing
the error.

In a multithreaded program, even wp->f() can crash; between the check and
the invocation, another thread may have destroyed the object.