Preprocessor bug

[Martin Sebor]

Neil Booth wrote:
> 
> John Hinke wrote:-
> 
> > #define name2(a,b)  tmpname2(a,b)
> > #define tmpname2(a,b) a##b
> > #define declare(Class,type)  name2(Class,declare)(type)
> > #define MyClassdeclare(type) name2(type,MyClass)
> > declare(MyClass,int)
> 
> Yes, this is a longstanding bug fix in the current preprocessor.
> 
> > Any comments?  If I'm doing something wrong, please let me know.  This code
> > has been working for many years now.  Also, other compilers that I tested
> > produce the correct (2.95.2) output.
> 
> That doesn't make the other compilers, including 2.95.2, correct <g>.
> It's easy to get this wrong with the stack-based expansion approach that
> I imagine most macro expanders use.  However, if you can show me what's
> wrong with my reasoning, I'll revert the behaviour.
> 
> Here is the expansion stack as I and GNU CPP see it, where an extra
> level of indentation indicates an extra level of nestedness:-
> 
> declare(MyClass, int)
>   name2(MyClass,declare)(int)
>     tmpname2(MyClass,declare)(int)
>       MyClassDeclare(int)
>         name2(int,MyClass)
> 
> As you can see, name2 is nested 3 levels below a previous expansion of
> itself, so it shouldn't expand.  I don't think this is at all
> ambiguous in the standard.
> 
> This is a more complicated example of
> 
> #define foo(x) foo
> foo(foo)(x)
> 
> which I believe most would agree expands to "foo(x)" and not "foo".

Yes, although I don't agree that the two examples exploit the same
feature. You're talking about 6.10.3.4, p2, which is not relevant in
John's example. Here's a simplified testcase which I'm pretty sure is
well-formed and should produce, (what else?), foobar:

#define __paste(a,b)      a##b
#define paste(a,b)        __paste(a, b)

#define declfoo(a)        paste(foo, a)
#define declare(a, b)     paste(decl, a)(b)

declare(foo, bar)


Regards
Martin