Re: C++ PATCH: PR 23293

[Mark Mitchell <mark at codesourcery dot com>]

David Abrahams wrote:

>>http://gcc.gnu.org/bugzilla/show_bug.cgi?id=23293
> 
> This one would be better if you only preserved namespace-scope
> typedefs.
> 
>>http://gcc.gnu.org/bugzilla/show_bug.cgi?id=17763 is another example.
> 
> This one, too.

I considered that -- but for the second alternative I don't think it
helps; there, looking at Comment #7, the only typedefs are in fact in
namespace scope.

Your idea of dropping typedefs for simple types might help, but there's
still the risk that a distant typedef wandering around in a namespace
about which you know nothing will start showing up in messages
pertaining to your code.

I'm not opposed to people making refinements here.  I think the best
plan would be to treat "Outer<X>" as a typedef for "Outer<int>"
internally, and then handle it like we do typedefs elsewhere, which,
would I think meet both sets of requests.  However, to do that, we need
to overhaul internal infrastructure in a way that's not tractable.

The current problem is to do the least bad thing in the very short term.

> I fail to see how this very real problem:
> 
> theory.cc:22:                 double ch (dummy<0>, dummy<n>) [with unsigned int m = 0]
>                                                    ^^^^^^^^
>                                                        \__ INCORRECT
>  
> would be solved by using numbers instead of actual identifiers.  You'd
> end up with something like:
> 
> theory.cc:22:                 double ch (dummy<0>, dummy<#1>) [with unsigned int #2 = 0]
>                                                    ^^^^^^^^^
>                                                        \__ STILL INCORRECT

That's not what I meant.  But, hey, let's not get into this; it should
be my problem to come up with a proper proposal before wasting
everyone's time debating it.

-- 
Mark Mitchell
CodeSourcery, LLC
mark@codesourcery.com
(916) 791-8304