This is the mail archive of the
gdb-patches@sources.redhat.com
mailing list for the GDB project.
Re: RFA: Correct field names for class methods
- From: Elena Zannoni <ezannoni at redhat dot com>
- To: Daniel Jacobowitz <drow at mvista dot com>
- Cc: Elena Zannoni <ezannoni at redhat dot com>,gdb-patches at sources dot redhat dot com
- Date: Tue, 10 Sep 2002 18:13:41 -0400
- Subject: Re: RFA: Correct field names for class methods
- References: <20020827031346.GA16591@nevyn.them.org><15741.12051.396773.949592@localhost.redhat.com><20020910163610.GA10920@nevyn.them.org>
Daniel Jacobowitz writes:
> On Mon, Sep 09, 2002 at 07:30:27PM -0400, Elena Zannoni wrote:
> > Daniel Jacobowitz writes:
> > > Right now, the "name" field of a method is not always reliable. There's
> > > special-cased code all over the linespec, function calling, struct lookup,
> > > etc. code to handle this. Among the problems:
> > >
> > > - v3 stabs emit __comp_ctor etc.
> > > - v2 stabs emit constructors and destructors in the same fieldlist
> > > - v2 stabs emit mangled operator names
> > >
> > > This patch does not remove any of the special cases, but renders it
> > > all obsolete, to be cleaned up in a forthcoming patch. This will greatly
> > > simplify the revised method printing code that I'm working on, a necessary
> > > cleanup as we move towards namespace support. It pushes the stabs special
> > > casing back into stabs related code as much as practical.
> > >
> > > Elena, this does a bit of its ugliness in read_member_functions, so I'd like
> > > your approval before I go ahead with it.
> >
> >
> > See below. I have some questions and some general comments on the
> > structure of the patch.
> >
> > >
> > > Comments, anyone?
> > >
> > > --
> > > Daniel Jacobowitz
> > > MontaVista Software Debian GNU/Linux Developer
> > >
> > > 2002-08-26 Daniel Jacobowitz <drow@mvista.com>
> > >
> > > * gdbtypes.c (check_stub_method): Make static.
> > > (update_method_from_physname, check_stub_method_group)
> > > (find_last_component, class_name_from_physname)
> > > (method_name_from_physname): New functions.
> > > * gdbtypes.h: Update prototypes.
> > >
> > > * stabsread.c: Include "cp-abi.h".
> >
> > Makefile?
>
> Oops, thank you.
>
> >
> > > (read_member_functions): Correct method names for operators
> > > and v3 constructors/destructors. Separate v2 constructors and
> > > destructors.
> > >
> > > * cp-valprint.c (cp_print_class_method): Call
> > > check_stub_method_group instead of check_stub_method.
> > > * p-valprint.c (pascal_object_print_class_method): Likewise.
> > > * valops.c (search_struct_method): Likewise.
> > > (find_method_list, value_struct_elt_for_reference): Likewise.
> > >
> > > Index: cp-valprint.c
> > > ===================================================================
> > > RCS file: /cvs/src/src/gdb/cp-valprint.c,v
> > > retrieving revision 1.13
> > > diff -u -p -r1.13 cp-valprint.c
> > > --- cp-valprint.c 29 Jul 2002 22:55:26 -0000 1.13
> > > +++ cp-valprint.c 27 Aug 2002 02:02:55 -0000
> > > @@ -97,13 +97,12 @@ cp_print_class_method (char *valaddr,
> > > f = TYPE_FN_FIELDLIST1 (domain, i);
> > > len2 = TYPE_FN_FIELDLIST_LENGTH (domain, i);
> > >
> > > + check_stub_method_group (domain, i);
> > > for (j = 0; j < len2; j++)
> > > {
> > > QUIT;
> >
> > Did you mean to leave this QUIT?
>
> No, it should probably go, for the same reasons as the others.
OK.
>
> > > if (TYPE_FN_FIELD_VOFFSET (f, j) == offset)
> > > {
> > > - if (TYPE_FN_FIELD_STUB (f, j))
> > > - check_stub_method (domain, i, j);
> > > kind = "virtual ";
> > > goto common;
> > > }
> > > @@ -129,15 +128,11 @@ cp_print_class_method (char *valaddr,
> > > f = TYPE_FN_FIELDLIST1 (domain, i);
> > > len2 = TYPE_FN_FIELDLIST_LENGTH (domain, i);
> > >
> > > + check_stub_method_group (f, j);
> > > for (j = 0; j < len2; j++)
> > > {
> > > - QUIT;
> > > - if (TYPE_FN_FIELD_STUB (f, j))
> > > - check_stub_method (domain, i, j);
> > > if (STREQ (SYMBOL_NAME (sym), TYPE_FN_FIELD_PHYSNAME (f, j)))
> > > - {
> > > - goto common;
> > > - }
> > > + goto common;
> > > }
> > > }
> > > }
> > > Index: gdbtypes.c
> > > ===================================================================
> > > RCS file: /cvs/src/src/gdb/gdbtypes.c,v
> > > retrieving revision 1.56
> > > diff -u -p -r1.56 gdbtypes.c
> > > --- gdbtypes.c 20 Aug 2002 19:57:32 -0000 1.56
> > > +++ gdbtypes.c 27 Aug 2002 02:02:55 -0000
> > > @@ -1672,7 +1672,7 @@ safe_parse_type (char *p, int length)
> > > which info used to be in the stab's but was removed to hack back
> > > the space required for them. */
> > >
> > > -void
> > > +static void
> > > check_stub_method (struct type *type, int method_id, int signature_id)
> > > {
> > > struct fn_field *f;
> > > @@ -1781,6 +1781,54 @@ check_stub_method (struct type *type, in
> > > xfree (demangled_name);
> > > }
> > >
> >
> > Could you add some comments about what this function does?
> > Actually, since it is used only in stabsread.c, why not move it there?
> > I think a lot of this c++ specific stuff really doesn't belong in gdbtypes.c.
> > You can move all the new functions, except for check_stub_method_group to
> > stabsread.c. They are called from there only.
>
> update_method_name_from_physname? OK - at first I thought I would need
> it in more places but that doesn't seem to be true. DWARF-2 has its
> own quirks, but it always gets the method's _name_ right. I'll move it
> to stabsread.c and make it static. That even lets me plug the memory
> leak.
>
Yes! could you also pull the other new functions in stabsread.c?
They seem to be used there only.
> > > +void
> > > +update_method_name_from_physname (char **old_name, char *physname)
> > > +{
> > > + char *method_name;
> > > +
> > > + method_name = method_name_from_physname (physname);
> > > +
> > > + if (method_name == NULL)
> > > + error ("bad physname %s\n", physname);
> > > +
> > > + if (strcmp (*old_name, method_name) != 0)
> > > + *old_name = method_name;
> > > + else
> > > + xfree (method_name);
> > > +}
> > > +
> >
> > Could you add some comments here too?
>
> Sure.
>
> >
> > > +void
> > > +check_stub_method_group (struct type *type, int method_id)
> > > +{
>
>
> > Did you forget to post a piece of the patch? I don't see the function
> > below being called anywhere.
>
> My ulterior motives are leaking again! I added
> class_name_from_physname because it is the exact parallel to
> method_name_from_physname, and because I will need it in future
> patches. I can hold it till later if you prefer, but I'd rather just
> slip it in now. It's a pretty obvious function.
If you put it in stabsread.c.
>
> > > +/* Return the name of the class containing method PHYSNAME. */
> > > +
> > > +char *
> > > +class_name_from_physname (const char *physname)
> > > +{
>
>
> > Since you are at it, could you insert examples of mangled names,
> > physnames, etc in the comment?
>
> Sure. I wrote up a big table for it and then I even fixed the bug the
> table pointed out to me (needed a case to handle just destructors).
>
ok
> >
> >
> > > + tmp_sublist = sublist;
> > > + while (tmp_sublist != NULL)
> > > + {
> > > + if (tmp_sublist->fn_field.is_stub)
> > > + has_stub = 1;
> > > + if (tmp_sublist->fn_field.physname[0] == '_'
> > > + && tmp_sublist->fn_field.physname[1] == 'Z')
> > > + is_v3 = 1;
> > > +
> > > + if (is_destructor_name (tmp_sublist->fn_field.physname))
> > > + has_destructor++;
> > > + else
> >
> > > + has_nondestructor++;
> >
> > Dumb question, but, what is a nondestructor? Maybe a different variable
> > name would be a bit more enlightening.
>
> Anything that isn't a destructor. I've changed it to has_other - is
> that clearer?
>
Yes thanks.
> > > +
> > > + tmp_sublist = tmp_sublist->next;
> > > + }
> > > +
> > > + if (has_destructor && has_nondestructor)
> >
> >
> > Comments needed, for the c++ challenged. What does it mean to have both
> > a destructor and a nondestructor?
>
> It's explained in the big comment up above; I've clarified the wording.
> Basically, what happens is... wait... let me paste the revised comment:
>
> The caveat: GCC 2.95.x (and earlier?) put constructors and
> destructors in the same method group. We need to split this into
> two groups, because they should have different names. So for
> each method group we check whether it contains both routines
> whose physname appears to be a destructor (the physnames for
> destructors are always provided, due to quirks in v2 mangling)
> and routines whose physname does not appear to be a destructor.
> If so then we break up the list into two halves. Even if the
> constructors and destructors aren't in the same group the
> destructor will still lack the leading tilde, so that also needs
> to be fixed.
>
Ahhhh.
> >
> > > + {
> > > + struct next_fnfieldlist *destr_fnlist;
> > > + struct next_fnfield *last_sublist;
> > > +
> > > + /* Create a new fn_fieldlist for the destructors. */;
> > > + destr_fnlist = (struct next_fnfieldlist *)
> > > + xmalloc (sizeof (struct next_fnfieldlist));
> > > + make_cleanup (xfree, destr_fnlist);
> > > + memset (destr_fnlist, 0, sizeof (struct next_fnfieldlist));
> > > + destr_fnlist->fn_fieldlist.name
> > > + = obconcat (&objfile->type_obstack, "", "~",
> > > + new_fnlist->fn_fieldlist.name);
> > > +
> > > + destr_fnlist->fn_fieldlist.fn_fields = (struct fn_field *)
> > > + obstack_alloc (&objfile->type_obstack,
> > > + sizeof (struct fn_field) * has_destructor);
> > > + memset (destr_fnlist->fn_fieldlist.fn_fields, 0,
> > > + sizeof (struct fn_field) * has_destructor);
> > > + tmp_sublist = sublist;
> > > + last_sublist = NULL;
> >
> >
> > > + i = 0;
> >
> > I am confused. Why is i always 0? Or it isn't?
>
> Oops! It doesn't really matter, since there's always one destructor
> (until we support in-charge/not-in-charge destructors eventually). But
> I've fixed that. Should have been i++ below.
>
> Thanks for all the comments; here's a revised version.
>
Approved, except for moving the functions to stabsread.c.
Elena
> --
> Daniel Jacobowitz
> MontaVista Software Debian GNU/Linux Developer
>
> 2002-09-10 Daniel Jacobowitz <drow@mvista.com>
>
> * gdbtypes.c (check_stub_method): Make static.
> (check_stub_method_group, find_last_component)
> (class_name_from_physname, method_name_from_physname): New functions.
> * gdbtypes.h: Update prototypes.
>
> * stabsread.c: Include "cp-abi.h".
> (update_method_name_from_physname): New function.
> (read_member_functions): Correct method names for operators
> and v3 constructors/destructors. Separate v2 constructors and
> destructors.
> * Makefile.in (stabsread.o): Update dependencies.
>
> * cp-valprint.c (cp_print_class_method): Call
> check_stub_method_group instead of check_stub_method. Remove
> extraneous QUITs.
> * p-valprint.c (pascal_object_print_class_method): Likewise.
> * valops.c (search_struct_method): Likewise.
> (find_method_list, value_struct_elt_for_reference): Likewise.
>
> Index: Makefile.in
> ===================================================================
> RCS file: /cvs/src/src/gdb/Makefile.in,v
> retrieving revision 1.257
> diff -u -p -r1.257 Makefile.in
> --- Makefile.in 2 Sep 2002 18:09:06 -0000 1.257
> +++ Makefile.in 10 Sep 2002 16:35:15 -0000
> @@ -2174,7 +2174,7 @@ stabsread.o: stabsread.c $(defs_h) $(gdb
> $(symtab_h) $(gdbtypes_h) $(expression_h) $(symfile_h) $(objfiles_h) \
> $(aout_stab_gnu_h) $(libaout_h) $(aout_aout64_h) $(gdb_stabs_h) \
> $(buildsym_h) $(complaints_h) $(demangle_h) $(language_h) \
> - $(doublest_h) $(stabsread_h)
> + $(doublest_h) $(stabsread_h) $(cp_abi_h)
> stack.o: stack.c $(defs_h) $(gdb_string_h) $(value_h) $(symtab_h) \
> $(gdbtypes_h) $(expression_h) $(language_h) $(frame_h) $(gdbcmd_h) \
> $(gdbcore_h) $(target_h) $(breakpoint_h) $(demangle_h) $(inferior_h) \
> Index: cp-valprint.c
> ===================================================================
> RCS file: /cvs/src/src/gdb/cp-valprint.c,v
> retrieving revision 1.13
> diff -u -p -r1.13 cp-valprint.c
> --- cp-valprint.c 29 Jul 2002 22:55:26 -0000 1.13
> +++ cp-valprint.c 10 Sep 2002 16:35:15 -0000
> @@ -97,13 +97,11 @@ cp_print_class_method (char *valaddr,
> f = TYPE_FN_FIELDLIST1 (domain, i);
> len2 = TYPE_FN_FIELDLIST_LENGTH (domain, i);
>
> + check_stub_method_group (domain, i);
> for (j = 0; j < len2; j++)
> {
> - QUIT;
> if (TYPE_FN_FIELD_VOFFSET (f, j) == offset)
> {
> - if (TYPE_FN_FIELD_STUB (f, j))
> - check_stub_method (domain, i, j);
> kind = "virtual ";
> goto common;
> }
> @@ -129,15 +127,11 @@ cp_print_class_method (char *valaddr,
> f = TYPE_FN_FIELDLIST1 (domain, i);
> len2 = TYPE_FN_FIELDLIST_LENGTH (domain, i);
>
> + check_stub_method_group (f, j);
> for (j = 0; j < len2; j++)
> {
> - QUIT;
> - if (TYPE_FN_FIELD_STUB (f, j))
> - check_stub_method (domain, i, j);
> if (STREQ (SYMBOL_NAME (sym), TYPE_FN_FIELD_PHYSNAME (f, j)))
> - {
> - goto common;
> - }
> + goto common;
> }
> }
> }
> Index: gdbtypes.c
> ===================================================================
> RCS file: /cvs/src/src/gdb/gdbtypes.c,v
> retrieving revision 1.56
> diff -u -p -r1.56 gdbtypes.c
> --- gdbtypes.c 20 Aug 2002 19:57:32 -0000 1.56
> +++ gdbtypes.c 10 Sep 2002 16:35:16 -0000
> @@ -1672,7 +1672,7 @@ safe_parse_type (char *p, int length)
> which info used to be in the stab's but was removed to hack back
> the space required for them. */
>
> -void
> +static void
> check_stub_method (struct type *type, int method_id, int signature_id)
> {
> struct fn_field *f;
> @@ -1781,6 +1781,49 @@ check_stub_method (struct type *type, in
> xfree (demangled_name);
> }
>
> +/* This is the external interface to check_stub_method, above. This function
> + unstubs all of the signatures for TYPE's METHOD_ID method name. After
> + calling this function TYPE_FN_FIELD_STUB will be cleared for each signature
> + and TYPE_FN_FIELDLIST_NAME will be correct.
> +
> + This function unfortunately can not die until stabs do. */
> +
> +void
> +check_stub_method_group (struct type *type, int method_id)
> +{
> + int len = TYPE_FN_FIELDLIST_LENGTH (type, method_id);
> + struct fn_field *f = TYPE_FN_FIELDLIST1 (type, method_id);
> + int j, found_stub;
> +
> + for (j = 0; j < len; j++)
> + if (TYPE_FN_FIELD_STUB (f, j))
> + {
> + found_stub = 1;
> + check_stub_method (type, method_id, j);
> + }
> +
> + /* GNU v3 methods with incorrect names were corrected when we read in
> + type information, because it was cheaper to do it then. The only GNU v2
> + methods with incorrect method names are operators and destructors;
> + destructors were also corrected when we read in type information.
> +
> + Therefore the only thing we need to handle here are v2 operator
> + names. */
> + if (found_stub && strncmp (TYPE_FN_FIELD_PHYSNAME (f, 0), "_Z", 2) != 0)
> + {
> + int ret;
> + char dem_opname[256];
> +
> + ret = cplus_demangle_opname (TYPE_FN_FIELDLIST_NAME (type, method_id),
> + dem_opname, DMGL_ANSI);
> + if (!ret)
> + ret = cplus_demangle_opname (TYPE_FN_FIELDLIST_NAME (type, method_id),
> + dem_opname, 0);
> + if (ret)
> + TYPE_FN_FIELDLIST_NAME (type, method_id) = xstrdup (dem_opname);
> + }
> +}
> +
> const struct cplus_struct_type cplus_struct_default;
>
> void
> @@ -3435,6 +3478,120 @@ build_gdbtypes (void)
> "__bfd_vma", (struct objfile *) NULL);
> }
>
> +/* Find the last component of the demangled C++ name NAME. NAME
> + must be a method name including arguments, in order to correctly
> + locate the last component.
> +
> + This function return a pointer to the first colon before the
> + last component, or NULL if the name had only one component. */
> +
> +static const char *
> +find_last_component (const char *name)
> +{
> + const char *p;
> + int depth;
> +
> + /* Functions can have local classes, so we need to find the
> + beginning of the last argument list, not the end of the first
> + one. */
> + p = name + strlen (name) - 1;
> + while (p > name && *p != ')')
> + p--;
> +
> + if (p == name)
> + return NULL;
> +
> + /* P now points at the `)' at the end of the argument list. Walk
> + back to the beginning. */
> + p--;
> + depth = 1;
> + while (p > name && depth > 0)
> + {
> + if (*p == '<' || *p == '(')
> + depth--;
> + else if (*p == '>' || *p == ')')
> + depth++;
> + p--;
> + }
> +
> + if (p == name)
> + return NULL;
> +
> + while (p > name && *p != ':')
> + p--;
> +
> + if (p == name || p == name + 1 || p[-1] != ':')
> + return NULL;
> +
> + return p - 1;
> +}
> +
> +/* Return the name of the class containing method PHYSNAME. */
> +
> +char *
> +class_name_from_physname (const char *physname)
> +{
> + char *ret = NULL;
> + const char *end;
> + int depth = 0;
> + char *demangled_name = cplus_demangle (physname, DMGL_ANSI);
> +
> + if (demangled_name == NULL)
> + return NULL;
> +
> + end = find_last_component (demangled_name);
> + if (end != NULL)
> + {
> + ret = xmalloc (end - demangled_name + 1);
> + memcpy (ret, demangled_name, end - demangled_name);
> + ret[end - demangled_name] = '\0';
> + }
> +
> + xfree (demangled_name);
> + return ret;
> +}
> +
> +/* Return the name of the method whose linkage name is PHYSNAME. */
> +
> +char *
> +method_name_from_physname (const char *physname)
> +{
> + char *ret = NULL;
> + const char *end;
> + int depth = 0;
> + char *demangled_name = cplus_demangle (physname, DMGL_ANSI);
> +
> + if (demangled_name == NULL)
> + return NULL;
> +
> + end = find_last_component (demangled_name);
> + if (end != NULL)
> + {
> + char *args;
> + int len;
> +
> + /* Skip "::". */
> + end = end + 2;
> +
> + /* Find the argument list, if any. */
> + args = strchr (end, '(');
> + if (args == NULL)
> + len = strlen (end + 2);
> + else
> + {
> + args --;
> + while (*args == ' ')
> + args --;
> + len = args - end + 1;
> + }
> + ret = xmalloc (len + 1);
> + memcpy (ret, end, len);
> + ret[len] = 0;
> + }
> +
> + xfree (demangled_name);
> + return ret;
> +}
>
> extern void _initialize_gdbtypes (void);
> void
> Index: gdbtypes.h
> ===================================================================
> RCS file: /cvs/src/src/gdb/gdbtypes.h,v
> retrieving revision 1.35
> diff -u -p -r1.35 gdbtypes.h
> --- gdbtypes.h 10 Aug 2002 05:12:40 -0000 1.35
> +++ gdbtypes.h 10 Sep 2002 16:35:16 -0000
> @@ -1124,11 +1124,15 @@ extern struct type *check_typedef (struc
>
> #define CHECK_TYPEDEF(TYPE) (TYPE) = check_typedef (TYPE)
>
> -extern void check_stub_method (struct type *, int, int);
> +extern void check_stub_method_group (struct type *, int);
>
> extern struct type *lookup_primitive_typename (char *);
>
> extern char *gdb_mangle_name (struct type *, int, int);
> +
> +extern char *class_name_from_physname (const char *physname);
> +
> +extern char *method_name_from_physname (const char *physname);
>
> extern struct type *builtin_type (char **);
>
> Index: p-valprint.c
> ===================================================================
> RCS file: /cvs/src/src/gdb/p-valprint.c,v
> retrieving revision 1.13
> diff -u -p -r1.13 p-valprint.c
> --- p-valprint.c 19 Aug 2002 13:12:09 -0000 1.13
> +++ p-valprint.c 10 Sep 2002 16:35:16 -0000
> @@ -620,13 +620,11 @@ pascal_object_print_class_method (char *
> f = TYPE_FN_FIELDLIST1 (domain, i);
> len2 = TYPE_FN_FIELDLIST_LENGTH (domain, i);
>
> + check_stub_method_group (domain, i);
> for (j = 0; j < len2; j++)
> {
> - QUIT;
> if (TYPE_FN_FIELD_VOFFSET (f, j) == offset)
> {
> - if (TYPE_FN_FIELD_STUB (f, j))
> - check_stub_method (domain, i, j);
> kind = "virtual ";
> goto common;
> }
> @@ -646,15 +644,11 @@ pascal_object_print_class_method (char *
> f = TYPE_FN_FIELDLIST1 (domain, i);
> len2 = TYPE_FN_FIELDLIST_LENGTH (domain, i);
>
> + check_stub_method_group (domain, i);
> for (j = 0; j < len2; j++)
> {
> - QUIT;
> - if (TYPE_FN_FIELD_STUB (f, j))
> - check_stub_method (domain, i, j);
> if (STREQ (SYMBOL_NAME (sym), TYPE_FN_FIELD_PHYSNAME (f, j)))
> - {
> - goto common;
> - }
> + goto common;
> }
> }
> }
> Index: stabsread.c
> ===================================================================
> RCS file: /cvs/src/src/gdb/stabsread.c,v
> retrieving revision 1.38
> diff -u -p -r1.38 stabsread.c
> --- stabsread.c 1 Aug 2002 17:18:32 -0000 1.38
> +++ stabsread.c 10 Sep 2002 16:35:17 -0000
> @@ -44,6 +44,7 @@
> #include "demangle.h"
> #include "language.h"
> #include "doublest.h"
> +#include "cp-abi.h"
>
> #include <ctype.h>
>
> @@ -3080,6 +3081,27 @@ rs6000_builtin_type (int typenum)
>
> /* This page contains subroutines of read_type. */
>
> +/* Replace *OLD_NAME with the method name portion of PHYSNAME. */
> +
> +static void
> +update_method_name_from_physname (char **old_name, char *physname)
> +{
> + char *method_name;
> +
> + method_name = method_name_from_physname (physname);
> +
> + if (method_name == NULL)
> + error ("bad physname %s\n", physname);
> +
> + if (strcmp (*old_name, method_name) != 0)
> + {
> + xfree (*old_name);
> + *old_name = method_name;
> + }
> + else
> + xfree (method_name);
> +}
> +
> /* Read member function stabs info for C++ classes. The form of each member
> function data is:
>
> @@ -3377,6 +3399,164 @@ read_member_functions (struct field_info
> }
> else
> {
> + int has_stub = 0;
> + int has_destructor = 0, has_other = 0;
> + int is_v3 = 0;
> + struct next_fnfield *tmp_sublist;
> +
> + /* Various versions of GCC emit various mostly-useless
> + strings in the name field for special member functions.
> +
> + For stub methods, we need to defer correcting the name
> + until we are ready to unstub the method, because the current
> + name string is used by gdb_mangle_name. The only stub methods
> + of concern here are GNU v2 operators; other methods have their
> + names correct (see caveat below).
> +
> + For non-stub methods, in GNU v3, we have a complete physname.
> + Therefore we can safely correct the name now. This primarily
> + affects constructors and destructors, whose name will be
> + __comp_ctor or __comp_dtor instead of Foo or ~Foo. Cast
> + operators will also have incorrect names; for instance,
> + "operator int" will be named "operator i" (i.e. the type is
> + mangled).
> +
> + For non-stub methods in GNU v2, we have no easy way to
> + know if we have a complete physname or not. For most
> + methods the result depends on the platform (if CPLUS_MARKER
> + can be `$' or `.', it will use minimal debug information, or
> + otherwise the full physname will be included).
> +
> + Rather than dealing with this, we take a different approach.
> + For v3 mangled names, we can use the full physname; for v2,
> + we use cplus_demangle_opname (which is actually v2 specific),
> + because the only interesting names are all operators - once again
> + barring the caveat below. Skip this process if any method in the
> + group is a stub, to prevent our fouling up the workings of
> + gdb_mangle_name.
> +
> + The caveat: GCC 2.95.x (and earlier?) put constructors and
> + destructors in the same method group. We need to split this
> + into two groups, because they should have different names.
> + So for each method group we check whether it contains both
> + routines whose physname appears to be a destructor (the physnames
> + for and destructors are always provided, due to quirks in v2
> + mangling) and routines whose physname does not appear to be a
> + destructor. If so then we break up the list into two halves.
> + Even if the constructors and destructors aren't in the same group
> + the destructor will still lack the leading tilde, so that also
> + needs to be fixed.
> +
> + So, to summarize what we expect and handle here:
> +
> + Given Given Real Real Action
> + method name physname physname method name
> +
> + __opi [none] __opi__3Foo operator int opname
> + [now or later]
> + Foo _._3Foo _._3Foo ~Foo separate and
> + rename
> + operator i _ZN3FoocviEv _ZN3FoocviEv operator int demangle
> + __comp_ctor _ZN3FooC1ERKS_ _ZN3FooC1ERKS_ Foo demangle
> + */
> +
> + tmp_sublist = sublist;
> + while (tmp_sublist != NULL)
> + {
> + if (tmp_sublist->fn_field.is_stub)
> + has_stub = 1;
> + if (tmp_sublist->fn_field.physname[0] == '_'
> + && tmp_sublist->fn_field.physname[1] == 'Z')
> + is_v3 = 1;
> +
> + if (is_destructor_name (tmp_sublist->fn_field.physname))
> + has_destructor++;
> + else
> + has_other++;
> +
> + tmp_sublist = tmp_sublist->next;
> + }
> +
> + if (has_destructor && has_other)
> + {
> + struct next_fnfieldlist *destr_fnlist;
> + struct next_fnfield *last_sublist;
> +
> + /* Create a new fn_fieldlist for the destructors. */
> +
> + destr_fnlist = (struct next_fnfieldlist *)
> + xmalloc (sizeof (struct next_fnfieldlist));
> + make_cleanup (xfree, destr_fnlist);
> + memset (destr_fnlist, 0, sizeof (struct next_fnfieldlist));
> + destr_fnlist->fn_fieldlist.name
> + = obconcat (&objfile->type_obstack, "", "~",
> + new_fnlist->fn_fieldlist.name);
> +
> + destr_fnlist->fn_fieldlist.fn_fields = (struct fn_field *)
> + obstack_alloc (&objfile->type_obstack,
> + sizeof (struct fn_field) * has_destructor);
> + memset (destr_fnlist->fn_fieldlist.fn_fields, 0,
> + sizeof (struct fn_field) * has_destructor);
> + tmp_sublist = sublist;
> + last_sublist = NULL;
> + i = 0;
> + while (tmp_sublist != NULL)
> + {
> + if (!is_destructor_name (tmp_sublist->fn_field.physname))
> + {
> + tmp_sublist = tmp_sublist->next;
> + continue;
> + }
> +
> + destr_fnlist->fn_fieldlist.fn_fields[i++]
> + = tmp_sublist->fn_field;
> + if (last_sublist)
> + last_sublist->next = tmp_sublist->next;
> + else
> + sublist = tmp_sublist->next;
> + last_sublist = tmp_sublist;
> + tmp_sublist = tmp_sublist->next;
> + }
> +
> + destr_fnlist->fn_fieldlist.length = has_destructor;
> + destr_fnlist->next = fip->fnlist;
> + fip->fnlist = destr_fnlist;
> + nfn_fields++;
> + total_length += has_destructor;
> + length -= has_destructor;
> + }
> + else if (is_v3)
> + {
> + /* v3 mangling prevents the use of abbreviated physnames,
> + so we can do this here. There are stubbed methods in v3
> + only:
> + - in -gstabs instead of -gstabs+
> + - or for static methods, which are output as a function type
> + instead of a method type. */
> +
> + update_method_name_from_physname (&new_fnlist->fn_fieldlist.name,
> + sublist->fn_field.physname);
> + }
> + else if (has_destructor && new_fnlist->fn_fieldlist.name[0] != '~')
> + {
> + new_fnlist->fn_fieldlist.name = concat ("~", main_fn_name, NULL);
> + xfree (main_fn_name);
> + }
> + else if (!has_stub)
> + {
> + char dem_opname[256];
> + int ret;
> + ret = cplus_demangle_opname (new_fnlist->fn_fieldlist.name,
> + dem_opname, DMGL_ANSI);
> + if (!ret)
> + ret = cplus_demangle_opname (new_fnlist->fn_fieldlist.name,
> + dem_opname, 0);
> + if (ret)
> + new_fnlist->fn_fieldlist.name
> + = obsavestring (dem_opname, strlen (dem_opname),
> + &objfile->type_obstack);
> + }
> +
> new_fnlist->fn_fieldlist.fn_fields = (struct fn_field *)
> obstack_alloc (&objfile->type_obstack,
> sizeof (struct fn_field) * length);
> Index: valops.c
> ===================================================================
> RCS file: /cvs/src/src/gdb/valops.c,v
> retrieving revision 1.69
> diff -u -p -r1.69 valops.c
> --- valops.c 21 Aug 2002 17:24:31 -0000 1.69
> +++ valops.c 10 Sep 2002 16:35:18 -0000
> @@ -2302,12 +2302,11 @@ search_struct_method (char *name, struct
> struct fn_field *f = TYPE_FN_FIELDLIST1 (type, i);
> name_matched = 1;
>
> + check_stub_method_group (type, i);
> if (j > 0 && args == 0)
> error ("cannot resolve overloaded method `%s': no arguments supplied", name);
> else if (j == 0 && args == 0)
> {
> - if (TYPE_FN_FIELD_STUB (f, j))
> - check_stub_method (type, i, j);
> v = value_fn_field (arg1p, f, j, type, offset);
> if (v != NULL)
> return v;
> @@ -2315,8 +2314,6 @@ search_struct_method (char *name, struct
> else
> while (j >= 0)
> {
> - if (TYPE_FN_FIELD_STUB (f, j))
> - check_stub_method (type, i, j);
> if (!typecmp (TYPE_FN_FIELD_STATIC_P (f, j),
> TYPE_VARARGS (TYPE_FN_FIELD_TYPE (f, j)),
> TYPE_NFIELDS (TYPE_FN_FIELD_TYPE (f, j)),
> @@ -2555,20 +2552,15 @@ find_method_list (struct value **argp, c
> char *fn_field_name = TYPE_FN_FIELDLIST_NAME (type, i);
> if (fn_field_name && (strcmp_iw (fn_field_name, method) == 0))
> {
> - /* Resolve any stub methods. */
> int len = TYPE_FN_FIELDLIST_LENGTH (type, i);
> struct fn_field *f = TYPE_FN_FIELDLIST1 (type, i);
> - int j;
>
> *num_fns = len;
> *basetype = type;
> *boffset = offset;
>
> - for (j = 0; j < len; j++)
> - {
> - if (TYPE_FN_FIELD_STUB (f, j))
> - check_stub_method (type, i, j);
> - }
> + /* Resolve any stub methods. */
> + check_stub_method_group (type, i);
>
> return f;
> }
> @@ -3094,6 +3086,8 @@ value_struct_elt_for_reference (struct t
> int j = TYPE_FN_FIELDLIST_LENGTH (t, i);
> struct fn_field *f = TYPE_FN_FIELDLIST1 (t, i);
>
> + check_stub_method_group (t, i);
> +
> if (intype == 0 && j > 1)
> error ("non-unique member `%s' requires type instantiation", name);
> if (intype)
> @@ -3107,8 +3101,6 @@ value_struct_elt_for_reference (struct t
> else
> j = 0;
>
> - if (TYPE_FN_FIELD_STUB (f, j))
> - check_stub_method (t, i, j);
> if (TYPE_FN_FIELD_VIRTUAL_P (f, j))
> {
> return value_from_longest