Re: [ast-optimizer-branch] simplifying C++

[Jason Merrill <jason at redhat dot com>]

>>>>> "Diego" == Diego Novillo <dnovillo@redhat.com> writes:

> On Mon, 2002-06-17 at 12:07, Jason Merrill wrote:
>> I've been thinking about simplifying C++, and the complex scoping that C++
>> loops have.  In a for loop, a decl in the for-init-stmt is in scope for the
>> duration of the loop; a decl in the condition is in scope within an
>> iteration, goes out of scope, and back in.  The first can be dealt with
>> simply by wrapping the loop in a scope.
>> 
> Are these the same rules as for C99?  When I ran into this one last
> week, the front end was already enclosing the loop in a scope, so the
> simplifier didn't have to do much more than recognizing DECL_STMTs
> inside FOR_INIT_STMT.

Yep, it's the same.

>>   {
>>     int i;
>>     for (i = 42; ; i = i - 1)
>>       {
>>         A* T.1;
>>         bool T.2;
>>         bool T.3;
>>         A a;
>>         T.1 = &a;
>>         T.2 = i == 24;
>>         A::ctor (T.1, T.2);
>>         __with_cleanup (A::dtor (T.1))
>>           { 
>>             T.3 = a.operator bool();
>>             if (T.3 == 0)
>>               break;
>>             ...
>>           }
>>       }
>>   }
>> 
>> I think it's best to leave the cleanup in this structured form initially,
>> and deal with adding it to the various block exits later.

> Not pretty, but that's how we're dealing with statement expressions
> now.  Ideally, we should inline expand all these constructs.

What do you mean?  What would you expand inline?

> Would it be too inconvenient to replace the FOR_EXPR with T.3 in the
> above code instead of having a break inside __with_cleanup?

Do you mean the FOR_COND, i.e. something like this?

  {
    int i;
    bool T.3;
    T.3 = true;
    for (i = 42; T.3; i = i - 1)
      {
        A* T.1;
        bool T.2;
        A a;
        T.1 = &a;
        T.2 = i == 24;
        A::ctor (T.1, T.2);
        __with_cleanup (A::dtor (T.1))
          { 
            T.3 = a.operator bool();
            if (T.3)
              {
                ...
              }
          }
      }
  }

I suppose that would work, too.

Jason