/*******************************************************************
* demangle_symbol
*
* Demangle a C++ linker symbol into a C prototype
*/
int symbol_demangle (parsed_symbol *sym)
{
compound_type ct;
int is_static = 0, is_const = 0;
char *function_name = NULL;
char *class_name = NULL;
char *name;
const char *const_status;
static unsigned int hash = 0; /* In case of overloaded functions */
unsigned int data_flags = 0;
assert (globals.do_code);
assert (sym && sym->symbol);
hash++;
/* MS mangled names always begin with '?' */
name = sym->symbol;
if (*name++ != '?')
return -1;
if (VERBOSE)
puts ("Attempting to demangle symbol");
/* Then function name or operator code */
if (*name == '?')
{
/* C++ operator code (one character, or two if the first is '_') */
switch (*++name)
{
case '0': function_name = strdup ("ctor"); break;
case '1': function_name = strdup ("dtor"); break;
case '2': function_name = strdup ("operator_new"); break;
case '3': function_name = strdup ("operator_delete"); break;
case '4': function_name = strdup ("operator_equals"); break;
case '5': function_name = strdup ("operator_shiftright"); break;
case '6': function_name = strdup ("operator_shiftleft"); break;
case '7': function_name = strdup ("operator_not"); break;
case '8': function_name = strdup ("operator_equalsequals"); break;
case '9': function_name = strdup ("operator_notequals"); break;
case 'A': function_name = strdup ("operator_array"); break;
case 'C': function_name = strdup ("operator_dereference"); break;
case 'D': function_name = strdup ("operator_multiply"); break;
case 'E': function_name = strdup ("operator_plusplus"); break;
case 'F': function_name = strdup ("operator_minusminus"); break;
case 'G': function_name = strdup ("operator_minus"); break;
case 'H': function_name = strdup ("operator_plus"); break;
case 'I': function_name = strdup ("operator_address"); break;
case 'J': function_name = strdup ("operator_dereferencememberptr"); break;
case 'K': function_name = strdup ("operator_divide"); break;
case 'L': function_name = strdup ("operator_modulo"); break;
case 'M': function_name = strdup ("operator_lessthan"); break;
case 'N': function_name = strdup ("operator_lessthanequal"); break;
case 'O': function_name = strdup ("operator_greaterthan"); break;
case 'P': function_name = strdup ("operator_greaterthanequal"); break;
case 'Q': function_name = strdup ("operator_comma"); break;
case 'R': function_name = strdup ("operator_functioncall"); break;
case 'S': function_name = strdup ("operator_complement"); break;
case 'T': function_name = strdup ("operator_xor"); break;
case 'U': function_name = strdup ("operator_logicalor"); break;
case 'V': function_name = strdup ("operator_logicaland"); break;
case 'W': function_name = strdup ("operator_or"); break;
case 'X': function_name = strdup ("operator_multiplyequals"); break;
case 'Y': function_name = strdup ("operator_plusequals"); break;
case 'Z': function_name = strdup ("operator_minusequals"); break;
case '_':
switch (*++name)
{
case '0': function_name = strdup ("operator_divideequals"); break;
case '1': function_name = strdup ("operator_moduloequals"); break;
case '2': function_name = strdup ("operator_shiftrightequals"); break;
case '3': function_name = strdup ("operator_shiftleftequals"); break;
case '4': function_name = strdup ("operator_andequals"); break;
case '5': function_name = strdup ("operator_orequals"); break;
case '6': function_name = strdup ("operator_xorequals"); break;
case '7': function_name = strdup ("vftable"); data_flags = DATA_VTABLE; break;
case '8': function_name = strdup ("vbtable"); data_flags = DATA_VTABLE; break;
case '9': function_name = strdup ("vcall"); data_flags = DATA_VTABLE; break;
case 'A': function_name = strdup ("typeof"); data_flags = DATA_VTABLE; break;
case 'B': function_name = strdup ("local_static_guard"); data_flags = DATA_VTABLE; break;
case 'C': function_name = strdup ("string"); data_flags = DATA_VTABLE; break;
case 'D': function_name = strdup ("vbase_dtor"); data_flags = DATA_VTABLE; break;
case 'E': function_name = strdup ("vector_dtor"); break;
case 'G': function_name = strdup ("scalar_dtor"); break;
case 'H': function_name = strdup ("vector_ctor_iter"); break;
case 'I': function_name = strdup ("vector_dtor_iter"); break;
case 'J': function_name = strdup ("vector_vbase_ctor_iter"); break;
case 'L': function_name = strdup ("eh_vector_ctor_iter"); break;
case 'M': function_name = strdup ("eh_vector_dtor_iter"); break;
case 'N': function_name = strdup ("eh_vector_vbase_ctor_iter"); break;
case 'O': function_name = strdup ("copy_ctor_closure"); break;
case 'S': function_name = strdup ("local_vftable"); data_flags = DATA_VTABLE; break;
case 'T': function_name = strdup ("local_vftable_ctor_closure"); break;
case 'U': function_name = strdup ("operator_new_vector"); break;
case 'V': function_name = strdup ("operator_delete_vector"); break;
case 'X': function_name = strdup ("placement_new_closure"); break;
case 'Y': function_name = strdup ("placement_delete_closure"); break;
default:
return -1;
}
break;
default:
/* FIXME: Other operators */
return -1;
}
name++;
}
else
{
/* Type or function name terminated by '@' */
function_name = name;
while (*name && *name++ != '@') ;
if (!*name)
return -1;
function_name = str_substring (function_name, name - 1);
}
/* Either a class name, or '@' if the symbol is not a class member */
if (*name == '@')
{
class_name = strdup ("global"); /* Non member function (or a datatype) */
name++;
}
else
{
/* Class the function is associated with, terminated by '@@' */
class_name = name;
while (*name && *name++ != '@') ;
if (*name++ != '@') {
free (function_name);
return -1;
}
class_name = str_substring (class_name, name - 2); /* Allocates a new string */
}
/* Function/Data type and access level */
/* FIXME: why 2 possible letters for each option? */
switch(*name++)
{
/* Data */
case '0' : /* private static */
case '1' : /* protected static */
case '2' : /* public static */
is_static = 1;
/* Fall through */
case '3' : /* non static */
case '4' : /* non static */
/* Data members need to be implemented: report */
INIT_CT (ct);
if (!demangle_datatype (&name, &ct, sym))
{
if (VERBOSE)
printf ("/*FIXME: %s: unknown data*/\n", sym->symbol);
free (function_name);
free (class_name);
return -1;
}
sym->flags |= SYM_DATA;
sym->argc = 1;
sym->arg_name[0] = str_create (5, OUTPUT_UC_DLL_NAME, "_", class_name,
is_static ? "static_" : "_", function_name);
sym->arg_text[0] = str_create (3, ct.expression, " ", sym->arg_name[0]);
FREE_CT (ct);
free (function_name);
free (class_name);
return 0;
case '6' : /* compiler generated static */
case '7' : /* compiler generated static */
if (data_flags & DATA_VTABLE)
{
sym->flags |= SYM_DATA;
sym->argc = 1;
sym->arg_name[0] = str_create (5, OUTPUT_UC_DLL_NAME, "_", class_name,
"_", function_name);
sym->arg_text[0] = str_create (2, "void *", sym->arg_name[0]);
if (VERBOSE)
puts ("Demangled symbol OK [vtable]");
free (function_name);
free (class_name);
return 0;
}
free (function_name);
free (class_name);
return -1;
/* Functions */
case 'E' : /* private virtual */
case 'F' : /* private virtual */
case 'M' : /* protected virtual */
case 'N' : /* protected virtual */
case 'U' : /* public virtual */
case 'V' : /* public virtual */
/* Virtual functions need to be added to the exported vtable: report */
if (VERBOSE)
printf ("/*FIXME %s: %s::%s is virtual-add to vftable*/\n", sym->symbol,
class_name, function_name);
/* Fall through */
case 'A' : /* private */
case 'B' : /* private */
case 'I' : /* protected */
case 'J' : /* protected */
case 'Q' : /* public */
case 'R' : /* public */
/* Implicit 'this' pointer */
sym->arg_text [sym->argc] = str_create (3, "struct ", class_name, " *");
sym->arg_type [sym->argc] = ARG_POINTER;
sym->arg_flag [sym->argc] = 0;
sym->arg_name [sym->argc++] = strdup ("_this");
/* New struct definitions can be 'grep'ed out for making a fixup header */
if (VERBOSE)
printf ("struct %s { void **vtable; /*FIXME: class definition */ };\n", class_name);
break;
case 'C' : /* private: static */
case 'D' : /* private: static */
case 'K' : /* protected: static */
case 'L' : /* protected: static */
case 'S' : /* public: static */
case 'T' : /* public: static */
is_static = 1; /* No implicit this pointer */
break;
case 'Y' :
case 'Z' :
break;
/* FIXME: G,H / O,P / W,X are private / protected / public thunks */
default:
free (function_name);
free (class_name);
return -1;
}
/* If there is an implicit this pointer, const status follows */
if (sym->argc)
{
switch (*name++)
{
case 'A': break; /* non-const */
case 'B': is_const = CT_CONST; break;
case 'C': is_const = CT_VOLATILE; break;
case 'D': is_const = (CT_CONST | CT_VOLATILE); break;
default:
free (function_name);
free (class_name);
return -1;
}
}
/* Next is the calling convention */
switch (*name++)
{
case 'A': /* __cdecl */
case 'B': /* __cdecl __declspec(dllexport) */
if (!sym->argc)
{
sym->flags |= SYM_CDECL;
break;
}
/* Else fall through */
case 'C': /* __pascal */
case 'D': /* __pascal __declspec(dllexport) */
case 'E': /* __thiscall */
case 'F': /* __thiscall __declspec(dllexport) */
case 'G': /* __stdcall */
case 'H': /* __stdcall __declspec(dllexport) */
case 'I': /* __fastcall */
case 'J': /* __fastcall __declspec(dllexport)*/
case 'K': /* default (none given) */
if (sym->argc)
sym->flags |= SYM_THISCALL;
else
sym->flags |= SYM_STDCALL;
break;
default:
free (function_name);
free (class_name);
return -1;
}
/* Return type, or @ if 'void' */
if (*name == '@')
{
sym->return_text = strdup ("void");
sym->return_type = ARG_VOID;
name++;
}
else
{
INIT_CT (ct);
if (!demangle_datatype (&name, &ct, sym)) {
free (function_name);
free (class_name);
return -1;
}
sym->return_text = ct.expression;
sym->return_type = get_type_constant(ct.dest_type, ct.flags);
ct.expression = NULL;
FREE_CT (ct);
}
/* Now come the function arguments */
while (*name && *name != 'Z')
{
/* Decode each data type and append it to the argument list */
if (*name != '@')
{
INIT_CT (ct);
if (!demangle_datatype(&name, &ct, sym)) {
free (function_name);
free (class_name);
return -1;
}
if (strcmp (ct.expression, "void"))
{
sym->arg_text [sym->argc] = ct.expression;
ct.expression = NULL;
sym->arg_type [sym->argc] = get_type_constant (ct.dest_type, ct.flags);
sym->arg_flag [sym->argc] = ct.flags;
sym->arg_name[sym->argc] = str_create_num (1, sym->argc, "arg");
sym->argc++;
}
else
break; /* 'void' terminates an argument list */
FREE_CT (ct);
}
else
name++;
}
while (*name == '@')
name++;
/* Functions are always terminated by 'Z'. If we made it this far and
* Don't find it, we have incorrectly identified a data type.
*/
if (*name != 'Z') {
free (function_name);
free (class_name);
return -1;
}
/* Note: '()' after 'Z' means 'throws', but we don't care here */
/* Create the function name. Include a unique number because otherwise
* overloaded functions could have the same c signature.
*/
switch (is_const)
{
case (CT_CONST | CT_VOLATILE): const_status = "_const_volatile"; break;
case CT_CONST: const_status = "_const"; break;
case CT_VOLATILE: const_status = "_volatile"; break;
default: const_status = "_"; break;
}
sym->function_name = str_create_num (4, hash, class_name, "_",
function_name, is_static ? "_static" : const_status);
assert (sym->return_text);
assert (sym->flags);
assert (sym->function_name);
free (class_name);
free (function_name);
if (VERBOSE)
puts ("Demangled symbol OK");
return 0;
}
/*******************************************************************
* get_type
*
* Read a type from a prototype
*/
static const char *get_type (parsed_symbol *sym, const char *proto, int arg)
{
int is_const, is_volatile, is_struct, is_signed, is_unsigned, ptrs = 0;
const char *iter, *type_str, *base_type, *catch_unsigned;
char dest_type;
assert (sym && sym->symbol);
assert (proto && *proto);
assert (arg < 0 || (unsigned)arg == sym->argc);
type_str = proto;
proto = str_match (proto, "const", &is_const);
proto = str_match (proto, "volatile", &is_volatile);
proto = str_match (proto, "struct", &is_struct);
if (!is_struct)
proto = str_match (proto, "union", &is_struct);
catch_unsigned = proto;
proto = str_match (proto, "unsigned", &is_unsigned);
proto = str_match (proto, "signed", &is_signed);
/* Can have 'unsigned const' or 'const unsigned' etc */
if (!is_const)
proto = str_match (proto, "const", &is_const);
if (!is_volatile)
proto = str_match (proto, "volatile", &is_volatile);
base_type = proto;
iter = str_find_set (proto, " ,*)");
if (!iter)
return NULL;
if (arg < 0 && (is_signed || is_unsigned))
{
/* Prevent calling convention from being swallowed by 'un/signed' alone */
if (strncmp (base_type, "int", 3) && strncmp (base_type, "long", 4) &&
strncmp (base_type, "short", 5) && strncmp (base_type, "char", 4))
{
iter = proto;
base_type = catch_unsigned;
} else
catch_unsigned = NULL;
}
else
catch_unsigned = NULL;
/* FIXME: skip const/volatile here too */
for (proto = iter; *proto; proto++)
if (*proto == '*')
ptrs++;
else if (*proto != ' ')
break;
if (!*proto)
return NULL;
type_str = str_substring (type_str, proto);
if (iter == base_type || catch_unsigned)
{
/* 'unsigned' with no type */
char *tmp = str_create (2, type_str, " int");
free ((char*)type_str);
type_str = tmp;
}
symbol_clean_string (type_str);
dest_type = symbol_get_type (type_str);
if (arg < 0)
{
sym->return_text = (char*)type_str;
sym->return_type = dest_type;
}
else
{
sym->arg_type [arg] = dest_type;
sym->arg_flag [arg] = is_const ? CT_CONST : is_volatile ? CT_VOLATILE : 0;
if (*proto == ',' || *proto == ')')
sym->arg_name [arg] = str_create_num (1, arg, "arg");
else
{
iter = str_find_set (proto, " ,)");
if (!iter)
{
free ((char*)type_str);
return NULL;
}
sym->arg_name [arg] = str_substring (proto, iter);
proto = iter;
}
sym->arg_text [arg] = (char*)type_str;
}
return proto;
}
