Re: [PATCH] [PR 9514] Fixing parse error for "pointer to a function pointer"

[Keith Seitz <keiths at redhat dot com>]

On 10/03/2011 09:45 AM, Tom Tromey wrote:
> Also, please look at my earlier patch for this bug.  I think it shows
> some cases that your patch does not address:
>
>      http://sourceware.org/ml/gdb-patches/2008-08/msg00539.html
>
> Perhaps I ought to simply commit that patch.  I am not sure why I never
> have.  What do you think?

I've looked over both of these patches. Abhijit's original patch is much 
simpler, but there are still regressions using it w/CVS HEAD:

+FAIL: gdb.base/code-expr.exp: (int ** @code)
+FAIL: gdb.base/cvexpr.exp: (int ** const)

I think Tom's approach is more generic, however more (marginally) 
complicated, but it, too, suffers from some problems. Most specifically, 
the ptr_operator production conflicts with the conversion operator 
production. Consider "ptype &foo::operator char* (void)" from 
cplusfuncs.exp.

We end up in the "OPERATOR ptype" production, but because of the new 
ptr_operator rules, this is parsed as OPERATOR, nonempty_typelist, 
func_mod instead of OPERATOR, nonempty_typelist, '(', nonempty_typelist, ')'

So we end up with "operator char (*" with Tom's patch. I keep thinking 
there must be a way to force the parser through the OPERATOR ptype 
production and then the TYPE_INSTANCE production, but I have not been 
successful. More savvy bison-ers might be able to do it, though. Or 
maybe I'll dedicate some time to this and figure it out.

Maybe some crafty massaging of these three productions will yield a 
"simpler" answer.

Keith