Re: [RFA] Select a particular mangling of a demangled symbol in lookup_block_symbol

[Elena Zannoni <ezannoni@redhat.com>]

Daniel Jacobowitz writes:
 > I just described the problem this patch addresses in my testsuite patch;
 > perhaps not the best place :)  Here's the relevant bit:
 > 
 >   - Multiple symbols with the same demangled name.  We can work around this
 >    for stabs, because we have the physname.  We don't have that option for   
 >    DWARF-2, and we shouldn't need to for stabs.  I have a patch for the
 >    workaround.  We get the [not-in-charge] constructor by default,
 >    unfortunately.
 > 
 > 
 > What this means is that we call lookup_block_symbol on something like:
 >   _ZN3fooC1ERS_
 > but get the information for:
 >   _ZN3fooC2ERS_
 > 
 > The breakpoint ends up on the base-not-in-charge constructor.
 > 
 > What we really SHOULD do is set it on both constructors silently, without
 > even acknowledging that they are different functions, or else offer the user
 > the choice.  My preference is actually for the former.  That requires
 > support for a single function existing in multiple places, which will also
 > give us nice things like better support for inlined functions with DWARF-2
 > (which will always be somewhat shoddy due to the nature of inlining, in that
 > it only occurs with lots of other optimization - but we can do much better
 > than we do).
 > 
 > Is this patch OK, or is it deemed too gross?
 > 
 > -- 
 > Daniel Jacobowitz                           Carnegie Mellon University
 > MontaVista Software                         Debian GNU/Linux Developer
 > 
 > 2002-02-14  Daniel Jacobowitz  <drow@mvista.com>
 > 
 > 	* symtab.h (lookup_block_symbol): Add mangled_name argument
 > 	to prototype.
 > 
 > 	* symmisc.c (maintenance_check_symtabs): Call lookup_block_symbol
 > 	with new mangled_name argument.
 > 	* linespec.c (decode_line_1): Likewise.
 > 	* valops (value_of_this): Likewise.
 > 	* symtab.c (lookup_transparent_type): Likewise.
 > 	(lookup_symbol_aux): Likewise.  Accept new mangled_name argument.
 > 	(lookup_symbol): If we are given a mangled name, pass it down
 > 	to lookup_symbol_aux.
 > 	(lookup_block_symbol): If we are given a mangled name to check
 > 	against, only return symbols which match it.
 > 
 > @@ -567,6 +568,7 @@ lookup_symbol (const char *name, const s
 >  {
 >    char *modified_name = NULL;
 >    char *modified_name2 = NULL;
 > +  const char *mangled_name = NULL;
 >    int needtofreename = 0;
 >    struct symbol *returnval;
 >  
 > @@ -592,13 +594,14 @@ lookup_symbol (const char *name, const s
 >        modified_name2 = cplus_demangle (modified_name, DMGL_ANSI | DMGL_PARAMS);
 >        if (modified_name2)
 >  	{
 > +	  mangled_name = name;
 >  	  modified_name = modified_name2;
 >  	  needtofreename = 1;
 >  	}
 >      }
 >  
 > -  returnval = lookup_symbol_aux (modified_name, block, namespace,
 > -				 is_a_field_of_this, symtab);
 > +  returnval = lookup_symbol_aux (modified_name, mangled_name, block,
 > +				 namespace, is_a_field_of_this, symtab);
 >    if (needtofreename)
 >      xfree (modified_name2);
 >  


OK, approved. But I have my usual couple of questions: 

Was the corresponding testsuite patch sorted out?  Looks like it
wasn't. Does this patch have any effect on the testsuite results w/o
the testsuite patch?

In the above, should it be mangled_name = name or mangled_name =
modified_name?  It would seem more uniform with the rest of the
function if we just used modified_name. Unless there is some problem
with case sensitivity, in which case, calling cplus_demangle with
modified_name seems wrong anyway. I.e. is it guaranteed that
case_sensitive_off is NOT in effect?  Just out of curiosity What
would happen if the user sets the case sensitivity off?
Wouldn't it change _ZN3fooC1ERS_ to _zn3fooc1ers_ ? (of course the user
can always do a lot of things to screw himself up)

I guess what I am really asking is when is lookup_symbol called with a
mangled name. I tried to do "break foo::foo", and I never saw it called
with a mangled name.

Elena