Re: [RFC] 12266 Fallout

[Keith Seitz <keiths@redhat.com>]

On 08/26/2011 11:05 AM, Tom Tromey wrote:

> I don't understand why check_typedef would necessarily be involved.
> If we changed the canonicalizer to respect this ABI rule about std::
> names, then wouldn't std::string be the type's actual name?
> And so we wouldn't have to change check_typedef?

Yes, we could make that modification to cp-name-parser.y (and also add 
something to turn std::basic_{string, ostream, istream, iostream} back 
into std::{string, ostream, istream, iostream}.

> In any case, isn't #3 the best approach?  That is, separating the search
> key from the print name, and making the print name closer to what users
> expect.  I think we should always be considering what end state we want
> to be in, and this seems to be it.

I'm all about the "end result" (which is why I mentioned that gdb should
be printing "calltest(foo)"). But the crux of the matter is still 
whether we should store "calltest(std::string)" or 
"calltest(std::basic_string<...>)" in the symbol table. That's all. The 
later is a "quick fix," but I am attempting (poorly) to also argue that 
it is the correct way to do it.

> If you are looking for a quick fix, but plan to do #3, then #1 is fine
> by me.

The two go hand-in-hand IMO. Let me work up a patch and publish it on an 
archer branch. This will contain both the print name and DMGL_VERBOSE 
patches in one. Alas, the print name stuff is rather in limbo right now, 
so it may take me much of the day to fix it up and get this all 
published. But I will try to get something pushed before you leave EOD.

Keith