local names

[Alex Samuel]

Jason Merrill <jason at redhat.com> writes:

  Jason> Why not just put the foo inside the nested-name, rather than
  Jason> using the local-name production for local types?

I thought of that, but that leads to some funny-looking results.
Suppose you have

    class C {
      void foo (int) {
        class D {};
      }
    };

The mangling for C::foo() is `_ZN1C3fooEi'.  Notice that the the
parameter type (`i') is outside the `N...E'.  However, if you mangle
the local type D, you would get something like `N1CZ3fooiE1DE'.  Now
the parameter type has to go inside the `Z...E' since otherwise you
can't tell where the function's signature ends.

If the usual mangling for functions were delimited by `Z...E' instead
of starting with `Z' but ending at the end-of-string, I'd definitely
advocate encoding a function scope similarly to a class or namespace
scope, in a <nested-name>.  But given the above issue (and the fact
that we have currently two working mangling implementations and a
working demangler), I'd strongly recommend leaving it as it is, which
is correct and just as efficient.

Regards
Alex