Re: [boost] ScopeExit

From: Alexander Nasonov (alnsn_at_[hidden])
Date: 2007-08-22 04:21:23

Steven Watanabe <watanabesj <at> gmail.com> writes:

> Also, I need to be able to specify the name of the function somehow
> and I am not sure how this can be accomplished with that interface.
> Here's what I have right now:
> 
>  ...
>
> So that
> 
> int BOOST_LOCAL_FUNCTION(name, (i)(j), (int x)(int y)) {
>     //body here
> } BOOST_LOCAL_FUNCTION_END
> 
> expands to
> 
> int result_tag;
> typedef BOOST_TYPEOF(result_tag) result_type;
> typedef BOOST_TYPEOF(i) type_0;
> typedef BOOST_TYPEOF(j) type_1;
> typedef boost::tuple<type_0, type_1> tuple_type;
> typedef local_function<tuple_type, result_type(tuple_type, int x, int 
> y)> function_type;
> function_type name(make_tuple(i, j));
> {
>     function_type* function_ptr = &name;
>     struct body {
>         result_type impl(tuple_type& args, int x, int y) {
>             type_0& i = boost::get<0>(args);
>             type_1 j = boost::get<1>(args);
>             {
>                 //body here
>             }
>         }
>     };
>     function_ptr->f = &body::impl;
> }

This is really impressive!

I'd moved function arguments closer to the function name

int BOOST_LOCAL_FUNCTION(name, (int x)(int y), (i)(j) )

Ideally, arguments should be listed naturally:

int BOOST_LOCAL_FUNCTION(name, (int x, int y), (i)(j) )

but I don't see how to pass the args variable to body::impl.

> Note that i and j are passed by value rather than by reference.  Otherwise,
> it is not safe to have a copy of the function around after control 
> leaves the scope
> it is defined in.

Passing by reference is often fine for ScopeExit but it's very dangerous
for local functions. Optionally passing by reference would be a nice
feature, though.
--
Alexander