Re: [PATCH][C++] Fix PR29433, make C++ use a lot less time/memory

[Michael Matz <matz at suse dot de>]

Hi,

On Tue, 20 Feb 2007, Daniel Jacobowitz wrote:

> On Tue, Feb 20, 2007 at 03:24:20PM +0100, Michael Matz wrote:
> > -       .string "_Ios_Openmode"
> > +       .string "std::_Ios_Openmode"
> > -       .string "new_allocator<char>"
> > +       .string "__gnu_cxx::new_allocator<char>"
> 
> These are more incorrect than the current names.  They're inside a
> DW_TAG_namespace scope, so GDB goes to a lot of trouble to prepend the
> namespace to the specified name.  It'll get std::std::_Ios_Openmode
> and I think that's the right interpretation of this debug info.

Hmpf.

> Is there some way we can lose the initial scope?

That's better, but is it really correct?  Think about a name like:

  N1::TA<N1::A>

(the DIE being nested in the one for namespace N1).  Current g++ would 
emit the name "TA<N1::A>", and I guess gdb doesn't try to fiddle with the 
template arguments itself, so that would work.

Hmm... I guess the diagnostic machinery could be persuaded to do that, but 
I'm not sure how (Gaby, any advice?).  Alternatively I could just reuse 
the old code from mangle_class_name_for_template, which does that by hand, 
not the nicest solution, though :-/

> > -       .string "._16"
> > +       .string "<anonymous struct>::<anonymous union>"
> 
> (dunno about this one, maybe it shouldn't have a DW_AT_name at all?)

That's what the dwarf2out code is doing normally when a 0 pointer is 
returned as name.


Ciao,
Michael.