Re: Problems with breakpoints in constructors

[Daniel Jacobowitz <drow at mvista dot com>]

On Tue, Jul 06, 2004 at 02:33:00AM -0400, Jason Merrill wrote:
> On Tue, 6 Jul 2004 11:24:37 +0530, "Amit S. Kale" <amitkale@linsyssoft.com> wrote:
> 
> > Mangled names are emitted in debugging information generated in the function 
> > add_name_and_src_coords_attributes. They are emitted as attribute 
> > DW_AT_MIPS_linkage_name. This attribute is emitted for all member functions 
> > except for constructors. It's not emitted when DECL_ABSTRACT is true, which 
> > applies to constructors.
> 
> ...which applies to the abstract instance of constructors.  This should in
> turn be referenced by two concrete instances, which should have
> DW_AT_MIPS_linkage_name.
> 
> > This hack generates DW_AT_MIPS_linkage_name for only one attribute. 
> > add_name_and_src_coords_attributes is not called for second instance of the 
> > constructor since both these constructors are instance of an abstract 
> > declaration, for which add_name_and_src_coords_attributes has been called at 
> > the time of first instance.
> 
> That's a bug; since we (correctly) aren't attaching the linkage name to the
> abstract instance, we need to attach it to the concrete instances.

I thought we used to do this.... but I can not find an example in my
archives, so I'm probably wrong.  I agree it should be fixed - for now,
until GDB is finally fixed to stop relying on DW_AT_MIPS_linkage_name.

In any case, that's not the problem.  GDB manages to work OK with the
debugging information it has now, with a little awkwardness.  You can't
say "break A::A", but "break 'A::A(int)'" works.  The only real problem
is the inability to handle cloned constructors, which is more of a user
interface problem than a technical problem.

-- 
Daniel Jacobowitz