This is the mail archive of the gcc@gcc.gnu.org mailing list for the GCC project.


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]
Other format: [Raw text]

Re: less dwarf2 debug info stored in .o on ia64


On Mon, Jan 12, 2004 at 10:15:06PM -0800, Jim Wilson wrote:
> On Mon, 2004-01-12 at 21:01, Rens Sankar wrote:
> > And with the .o file having dwarf2 info, we are unable to display the
> > info
> > on the unnamed padding fields. This leads to difference in the
> > declaration
> > of the structure in the header file and displayed by the tool.
> 
> Is this a significant difference?  Does it impair operation of the tool?
> 
> > We tried going through gcc source code but could not get much headway in 
> > dwarf2out.c. If you could kindly point to the general direction 
> > (e.g. function names etc.) where this decision is taken, we can try to 
> > study the code more.
> 
> I believe the bit in question is in gen_decl_die, at the FIELD_DECL
> case, and the check DECL_NAME (decl) != NULL_TREE.  This could perhaps
> depend on the existing flag_eliminate_unused_debug_types.  You should
> check to see how much extra debug info this emits.  If this starts
> emitting debug info for unnamed C++ vtable fields that no one cares
> about, then we might need additional checks to get just the C language
> unnamed structure fields that you want.  If this doesn't emit much extra
> info, then it might be reasonable to do it by default.

I don't think it is; the unnamed fields don't serve any purpose for
most consumers in the debug info.  If all they want is to recreate a
source representation of the struct, it's not hard to infer what
padding a specific ABI would require based on the existing field
locations.

-- 
Daniel Jacobowitz
MontaVista Software                         Debian GNU/Linux Developer


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]