Mangling: Special names

[Jim Dehnert]

> Date: Wed, 22 Mar 2000 15:46:59 +0100
> From: Martin von Loewis <loewis at informatik.hu-berlin.de>
> 
> According to the mangling syntax,
> 
>    Foo::Foo(int)
> 
> would be mangled as
> 
>    _ZN3FooC1Ei
> 
> i.e. the constructor <special-name> would syntactically fit only after
> a qualified name. Is this intentional?

Good question.  I interpreted C1/C2/C3/C4 as being the default
constructors, with the following derivation for Foo::Foo(), which is
the constructor for type Foo:

    <encoding>	-> _Z <name> <type>
		-> _Z <special-name> <class-enum-type>
		-> _Z C1 <source-name>
		-> _Z C1 3Foo

Your example, no longer the default constructor, becomes:

	_ZN3Foo3FooEi

Is that what you intended, Daveed?

> Likewise, _ZN3Foo3BarTVE would be the Foo::Bar virtual table.

No, it goes:

    <encoding>	-> _Z <name>	     <type>
		-> _Z <special-name> <class-enum-type>
		-> _Z TV	     <nested-name>
		-> _Z TV	     N 3Foo 3Bar E

> I don't understand the Th and Tv special names: Instead of a
> function-type, would you not need an <name> somewhere? Or even an
> <encoding>?

There is one.  Suppose we have A::foo(), and then derive B from A
with a B::foo() overriding.  Then B::foo() is

	_ZN1B3fooE

which serves as the base in the derivation for the override thunk from
A.  The non-virtual one, assuming the A->B adjustment is -36, is:

	Thn36_ <base name> <base function-type>
	Thn36_ N1B3fooE	   FE

The virtual one (i.e. A is a virtual base), assuming the vptr
displacement of the vcall offset is -12, is:

	Tv12 <base name> <base function-type> <base class type>
	Tv12 N1B3fooE	 FE		      1A

So in both cases, <base name> <base function-type> is essentially an
encoding, but the type is delimited (because it needs to be to separate
it from the base class type in the virtual case, though I guess it
could be eliminated in the non-virtual case).

Any other interpretations out there?  Must be...

Jim

-	    Jim Dehnert		dehnert at sgi.com
				(650)933-4272