Re: [RFA 2/4] dwarf2_physname

[Keith Seitz <keiths at redhat dot com>]

On 02/01/2010 11:39 AM, Daniel Jacobowitz wrote:
> DW_AT_specification is present for many variables in namespaces
> (at least in current GCC output).  But what significance does it
> actually have?  Can you explain the check above in terms of the DWARF
> standard?  If not, I don't think it's right.

I don't know if what gcc is doing is proper DWARF or not, but I don't 
see any indication on reading the DWARF3 spec that it is.

If I may use namespace.exp as an example:

namespace
{
  namespace G
  {
    int Xg;
  };
};

For this code, gcc will output a type-like information DIE tree, telling 
us ONLY about the composition of the anonymous namespace with namespace 
G and variable Xg. Later gcc gives us a DIE with DW_AT_location (to 
continue my (poor) type analogy: an instance of the variable). To 
describe this DIE, we also get DW_AT_specification and nothing more. 
Just DW_AT_location and DW_AT_specification.

If we do not follow DW_AT_specification, dwarf2_physname will put the 
variable Xg in the global namespace instead of "(anonymous 
namespace)::G::Xg", which is where it really is defined.

> DW_AT_specification just says that there is a specification of this
> variable in another DIE.  The same variable could be written with
> or without it.

In the case where DW_AT_specification is omitted, the physname 
information is contained within the DIE's immediate ancestry. I would 
expect to see, e.g.,

DW_TAG_namespace
-- DW_AT_name
-- DW_TAG_namespace
   -- DW_AT_name
   -- DW_TAG_variable
      -- DW_AT_name
      -- DW_AT_location

as opposed to (the below is what we actually get from gcc)

DW_TAG_namespace
-- DW_AT_name
-- DW_TAG_namespace
   -- DW_AT_name
   -- DW_TAG_variable
      -- DW_AT_name
...
DW_AT_variable
-- DW_AT_specification (points to DW_TAG_variable above)
-- DW_AT_location

Keith