tools/specmaker/msmangle.c

   1 /*
   2  *  Demangle VC++ symbols into C function prototypes
   3  *
   4  *  Copyright 2000 Jon Griffiths
   5  */
   6 #include "specmaker.h"
   7
   8 /* Type for parsing mangled types */
   9 typedef struct _compound_type
  10 {
  11   char  dest_type;
  12   int   flags;
  13   int   have_qualifiers;
  14   char *expression;
  15 } compound_type;
  16
  17
  18 /* Initialise a compound type structure */
  19 #define INIT_CT(ct) do { memset (&ct, 0, sizeof (ct)); } while (0)
  20
  21 /* free the memory used by a compound structure */
  22 #define FREE_CT(ct) do { if (ct.expression) free (ct.expression); } while (0)
  23
  24
  25 /* Internal functions */
  26 static char *demangle_datatype (char **str, compound_type *ct,
  27                                 parsed_symbol* sym);
  28
  29 static char *get_constraints_convention_1 (char **str, compound_type *ct);
  30
  31 static char *get_constraints_convention_2 (char **str, compound_type *ct);
  32
  33 static char *get_type_string (const char c, const int constraints);
  34
  35 static int   get_type_constant (const char c, const int constraints);
  36
  37 static char *get_pointer_type_string (compound_type *ct,
  38                                       const char *expression);
  39
  40
  41 /*******************************************************************
  42  *         demangle_symbol
  43  *
  44  * Demangle a C++ linker symbol into a C prototype
  45  */
  46 int symbol_demangle (parsed_symbol *sym)
  47 {
  48   compound_type ct;
  49   int is_static = 0, is_const = 0;
  50   char *function_name = NULL;
  51   char *class_name = NULL;
  52   char *name;
  53   static unsigned int hash = 0; /* In case of overloaded functions */
  54
  55   assert (globals.do_code);
  56   assert (sym && sym->symbol);
  57
  58   hash++;
  59
  60   /* MS mangled names always begin with '?' */
  61   name = sym->symbol;
  62   if (*name++ != '?')
  63     return -1;
  64
  65   if (VERBOSE)
  66     puts ("Attempting to demangle symbol");
  67
  68   /* Then function name or operator code */
  69   if (*name == '?')
  70   {
  71     /* C++ operator code (one character, or two if the first is '_') */
  72     switch (*++name)
  73     {
  74     case '0': function_name = strdup ("ctor"); break;
  75     case '1': function_name = strdup ("dtor"); break;
  76     case '2': function_name = strdup ("operator_new"); break;
  77     case '3': function_name = strdup ("operator_delete"); break;
  78     case '4': function_name = strdup ("operator_equals"); break;
  79     case '5': function_name = strdup ("operator_5"); break;
  80     case '6': function_name = strdup ("operator_6"); break;
  81     case '7': function_name = strdup ("operator_7"); break;
  82     case '8': function_name = strdup ("operator_equals_equals"); break;
  83     case '9': function_name = strdup ("operator_not_equals"); break;
  84     case 'E': function_name = strdup ("operator_plus_plus"); break;
  85     case 'H': function_name = strdup ("operator_plus"); break;
  86     case '_':
  87       /* FIXME: Seems to be some kind of escape character - overloads? */
  88       switch (*++name)
  89       {
  90       case '7': /* FIXME: Compiler generated default copy/assignment ctor? */
  91         return -1;
  92       case 'E': function_name = strdup ("_unknown_E"); break;
  93       case 'G': function_name = strdup ("_unknown_G"); break;
  94       default:
  95         return -1;
  96       }
  97       break;
  98     default:
  99       /* FIXME: Other operators */
 100       return -1;
 101     }
 102     name++;
 103   }
 104   else
 105   {
 106     /* Type or function name terminated by '@' */
 107     function_name = name;
 108     while (*name && *name++ != '@') ;
 109     if (!*name)
 110       return -1;
 111     function_name = str_substring (function_name, name - 1);
 112   }
 113
 114   /* Either a class name, or '@' if the symbol is not a class member */
 115   if (*name == '@')
 116   {
 117     class_name = strdup ("global"); /* Non member function (or a datatype) */
 118     name++;
 119   }
 120   else
 121   {
 122     /* Class the function is associated with, terminated by '@@' */
 123     class_name = name;
 124     while (*name && *name++ != '@') ;
 125     if (*name++ != '@')
 126       return -1;
 127     class_name = str_substring (class_name, name - 2);
 128   }
 129
 130   /* Note: This is guesswork on my part, but it seems to work:
 131    * 'Q' Means the function is passed an implicit 'this' pointer.
 132    * 'S' Means static member function, i.e. no implicit 'this' pointer.
 133    * 'Y' Is used for datatypes and functions, so there is no 'this' pointer.
 134    * This character also implies some other things:
 135    * 'Y','S' = The character after the calling convention is always the
 136    *     start of the return type code.
 137    * 'Q' Character after the calling convention is 'const'ness code
 138    *     (only non static member functions can be const).
 139    * 'U' also occurs, it seems to behave like Q, but probably implies
 140    *     something else.
 141    */
 142   switch(*name++)
 143   {
 144   case 'U' :
 145   case 'Q' :
 146     /* Implicit 'this' pointer */
 147     sym->arg_text [sym->argc] = str_create (3, "struct ", class_name, " *");
 148     sym->arg_type [sym->argc] = ARG_POINTER;
 149     sym->arg_flag [sym->argc] = 0;
 150     sym->arg_name [sym->argc++] = strdup ("_this");
 151     /* New struct definitions can be 'grep'ed out for making a fixup header */
 152     if (VERBOSE)
 153       printf ("struct %s { int _FIXME; };\n", class_name);
 154     break;
 155   case 'S' :
 156     is_static = 1;
 157     break;
 158   case 'Y' :
 159     break;
 160   default:
 161     return -1;
 162   }
 163
 164   /* Next is the calling convention */
 165   switch (*name++)
 166   {
 167   case 'A':
 168     sym->calling_convention = strdup ("__cdecl");
 169     break;
 170   case 'B': /* FIXME: Something to do with __declspec(dllexport)? */
 171   case 'I': /* __fastcall */
 172   case 'G':
 173     sym->calling_convention = strdup ("__stdcall");
 174     break;
 175   default:
 176     return -1;
 177   }
 178
 179   /* If the symbol is associated with a class, its 'const' status follows */
 180   if (sym->argc)
 181   {
 182     if (*name == 'B')
 183       is_const = 1;
 184     else if (*name != 'E')
 185       return -1;
 186     name++;
 187   }
 188
 189   /* Return type, or @ if 'void' */
 190   if (*name == '@')
 191   {
 192     sym->return_text = strdup ("void");
 193     sym->return_type = ARG_VOID;
 194     name++;
 195   }
 196   else
 197   {
 198     INIT_CT (ct);
 199     if (!demangle_datatype (&name, &ct, sym))
 200       return -1;
 201     sym->return_text = ct.expression;
 202     sym->return_type = get_type_constant(ct.dest_type, ct.flags);
 203     ct.expression = NULL;
 204     FREE_CT (ct);
 205   }
 206
 207   /* Now come the function arguments */
 208   while (*name && *name != 'Z')
 209   {
 210     /* Decode each data type and append it to the argument list */
 211     if (*name != '@')
 212     {
 213       INIT_CT (ct);
 214       if (!demangle_datatype(&name, &ct, sym))
 215         return -1;
 216
 217       if (strcmp (ct.expression, "void"))
 218       {
 219         sym->arg_text [sym->argc] = ct.expression;
 220         ct.expression = NULL;
 221         sym->arg_type [sym->argc] = get_type_constant (ct.dest_type, ct.flags);
 222         sym->arg_flag [sym->argc] = ct.flags;
 223         sym->arg_name[sym->argc] = str_create_num (1, sym->argc, "arg");
 224         sym->argc++;
 225       }
 226       else
 227         break; /* 'void' terminates an argument list */
 228       FREE_CT (ct);
 229     }
 230     else
 231       name++;
 232   }
 233
 234   while (*name == '@')
 235     name++;
 236
 237   /* Functions are always terminated by 'Z'. If we made it this far and
 238    * Don't find it, we have incorrectly identified a data type.
 239    */
 240   if (*name != 'Z')
 241     return -1;
 242
 243   /* Note: '()' after 'Z' means 'throws', but we don't care here */
 244
 245   /* Create the function name. Include a unique number because otherwise
 246    * overloaded functions could have the same c signature.
 247    */
 248   sym->function_name = str_create_num (4, hash, class_name, "_",
 249        function_name, is_static ? "_static" : is_const ? "_const" : "_");
 250
 251   assert (sym->return_text);
 252   assert (sym->calling_convention);
 253   assert (sym->function_name);
 254
 255   free (class_name);
 256   free (function_name);
 257
 258   if (VERBOSE)
 259     puts ("Demangled symbol OK");
 260
 261   return 0;
 262 }
 263
 264
 265 /*******************************************************************
 266  *         demangle_datatype
 267  *
 268  * Attempt to demangle a C++ data type, which may be compound.
 269  * a compound type is made up of a number of simple types. e.g:
 270  * char** = (pointer to (pointer to (char)))
 271  *
 272  * Uses a simple recursive descent algorithm that is broken
 273  * and/or incomplete, without a doubt ;-)
 274  */
 275 static char *demangle_datatype (char **str, compound_type *ct,
 276                                 parsed_symbol* sym)
 277 {
 278   char *iter;
 279
 280   assert (str && *str);
 281   assert (ct);
 282
 283   iter = *str;
 284
 285   if (!get_constraints_convention_1 (&iter, ct))
 286     return NULL;
 287
 288   switch (*iter)
 289   {
 290     case 'C': case 'D': case 'E': case 'F': case 'G':
 291     case 'H': case 'I': case 'J': case 'K': case 'M':
 292     case 'N': case 'O': case 'X': case 'Z':
 293       /* Simple data types */
 294       ct->dest_type = *iter++;
 295       if (!get_constraints_convention_2 (&iter, ct))
 296         return NULL;
 297       ct->expression = get_type_string (ct->dest_type, ct->flags);
 298       break;
 299     case 'U':
 300     case 'V':
 301       /* Class/struct/union */
 302       ct->dest_type = *iter++;
 303       if (*iter == '0' || *iter == '1')
 304       {
 305         /* Referring to class type (implicit 'this') */
 306         char *stripped;
 307         if (!sym->argc)
 308           return NULL;
 309
 310         iter++;
 311         /* Apply our constraints to the base type (struct xxx *) */
 312         stripped = strdup (sym->arg_text [0]);
 313         if (!stripped)
 314           fatal ("Out of Memory");
 315
 316         /* If we're a reference, re-use the pointer already in the type */
 317         if (!ct->flags & CT_BY_REFERENCE)
 318           stripped[ strlen (stripped) - 2] = '\0'; /* otherwise, strip it */
 319
 320         ct->expression = str_create (2, ct->flags & CT_CONST ? "const " :
 321                          ct->flags & CT_VOLATILE ? "volatile " : "", stripped);
 322         free (stripped);
 323       }
 324       else if (*iter == '_')
 325       {
 326         /* The name of the class/struct, followed by '@@' */
 327         char *struct_name = ++iter;
 328         while (*iter && *iter++ != '@') ;
 329         if (*iter++ != '@')
 330           return NULL;
 331         struct_name = str_substring (struct_name, iter - 2);
 332         ct->expression = str_create (4, ct->flags & CT_CONST ? "const " :
 333                          ct->flags & CT_VOLATILE ? "volatile " : "", "struct ",
 334                          struct_name, ct->flags & CT_BY_REFERENCE ? " *" : "");
 335         free (struct_name);
 336       }
 337       break;
 338     case 'Q': /* FIXME: Array Just treated as pointer currently  */
 339     case 'P': /* Pointer */
 340       {
 341         compound_type sub_ct;
 342         INIT_CT (sub_ct);
 343
 344         ct->dest_type = *iter++;
 345         if (!get_constraints_convention_2 (&iter, ct))
 346           return NULL;
 347
 348         /* FIXME: P6 = Function pointer, others who knows.. */
 349         if (isdigit (*iter))
 350           return NULL;
 351
 352         /* Recurse to get the pointed-to type */
 353         if (!demangle_datatype (&iter, &sub_ct, sym))
 354           return NULL;
 355
 356         ct->expression = get_pointer_type_string (ct, sub_ct.expression);
 357
 358         FREE_CT (sub_ct);
 359       }
 360       break;
 361     case '0': case '1': case '2': case '3': case '4':
 362     case '5': case '6': case '7': case '8': case '9':
 363       /* Referring back to previously parsed type */
 364       if (sym->argc >= (size_t)('0' - *iter))
 365         return NULL;
 366       ct->dest_type = sym->arg_type ['0' - *iter];
 367       ct->expression = strdup (sym->arg_text ['0' - *iter]);
 368       iter++;
 369       break;
 370     default :
 371       return NULL;
 372   }
 373   if (!ct->expression)
 374     return NULL;
 375
 376   return (char *)(*str = iter);
 377 }
 378
 379
 380 /* Constraints:
 381  * There are two conventions for specifying data type constaints. I
 382  * don't know how the compiler chooses between them, but I suspect it
 383  * is based on ensuring that linker names are unique.
 384  * Convention 1. The data type modifier is given first, followed
 385  *   by the data type it operates on. '?' means passed by value,
 386  *   'A' means passed by reference. Note neither of these characters
 387  *   is a valid base data type. This is then followed by a character
 388  *   specifying constness or volatilty.
 389  * Convention 2. The base data type (which is never '?' or 'A') is
 390  *   given first. The character modifier is optionally given after
 391  *   the base type character. If a valid character mofifier is present,
 392  *   then it only applies to the current data type if the character
 393  *   after that is not 'A' 'B' or 'C' (Because this makes a convention 1
 394  *   constraint for the next data type).
 395  *
 396  * The conventions are usually mixed within the same symbol.
 397  * Since 'C' is both a qualifier and a data type, I suspect that
 398  * convention 1 allows specifying e.g. 'volatile signed char*'. In
 399  * convention 2 this would be 'CC' which is ambigious (i.e. Is it two
 400  * pointers, or a single pointer + modifier?). In convention 1 it
 401  * is encoded as '?CC' which is not ambigious. This probably
 402  * holds true for some other types as well.
 403  */
 404
 405 /*******************************************************************
 406  *         get_constraints_convention_1
 407  *
 408  * Get type constraint information for a data type
 409  */
 410 static char *get_constraints_convention_1 (char **str, compound_type *ct)
 411 {
 412   char *iter = *str, **retval = str;
 413
 414   if (ct->have_qualifiers)
 415     return (char *)*str; /* Previously got constraints for this type */
 416
 417   if (*iter == '?' || *iter == 'A')
 418   {
 419     ct->have_qualifiers = 1;
 420     ct->flags |= (*iter++ == '?' ? 0 : CT_BY_REFERENCE);
 421
 422     switch (*iter++)
 423     {
 424     case 'A' :
 425       break; /* non-const, non-volatile */
 426     case 'B' :
 427       ct->flags |= CT_CONST;
 428       break;
 429     case 'C' :
 430       ct->flags |= CT_VOLATILE;
 431       break;
 432     default  :
 433       return NULL;
 434     }
 435   }
 436
 437   return (char *)(*retval = iter);
 438 }
 439
 440
 441 /*******************************************************************
 442  *         get_constraints_convention_2
 443  *
 444  * Get type constraint information for a data type
 445  */
 446 static char *get_constraints_convention_2 (char **str, compound_type *ct)
 447 {
 448   char *iter = *str, **retval = str;
 449
 450   /* FIXME: Why do arrays have both convention 1 & 2 constraints? */
 451   if (ct->have_qualifiers && ct->dest_type != 'Q')
 452     return (char *)*str; /* Previously got constraints for this type */
 453
 454   ct->have_qualifiers = 1; /* Even if none, we've got all we're getting */
 455
 456   switch (*iter)
 457   {
 458   case 'A' :
 459     if (iter[1] != 'A' && iter[1] != 'B' && iter[1] != 'C')
 460       iter++;
 461     break;
 462   case 'B' :
 463     ct->flags |= CT_CONST;
 464     iter++;
 465     break;
 466   case 'C' :
 467     /* See note above, if we find 'C' it is _not_ a signed char */
 468     ct->flags |= CT_VOLATILE;
 469     iter++;
 470     break;
 471   }
 472
 473   return (char *)(*retval = iter);
 474 }
 475
 476
 477 /*******************************************************************
 478  *         get_type_string
 479  *
 480  * Return a string containing the name of a data type
 481  */
 482 static char *get_type_string (const char c, const int constraints)
 483 {
 484   char *type_string;
 485
 486   switch (c)
 487   {
 488   case 'C': /* Signed char, fall through */
 489   case 'D': type_string = "char"; break;
 490   case 'E': type_string = "unsigned char"; break;
 491   case 'F': type_string = "short int"; break;
 492   case 'G': type_string = "unsigned short int"; break;
 493   case 'H': type_string = "int"; break;
 494   case 'I': type_string = "unsigned int"; break;
 495   case 'J': type_string = "long"; break;
 496   case 'K': type_string = "unsigned long"; break;
 497   case 'M': type_string = "float"; break;
 498   case 'N': type_string = "double"; break;
 499   case 'O': type_string = "long double"; break;
 500   case 'U':
 501   case 'V': type_string = "struct"; break;
 502   case 'X': return strdup ("void");
 503   case 'Z': return strdup ("...");
 504   default:
 505     return NULL;
 506   }
 507
 508   return str_create (3, constraints & CT_CONST ? "const " :
 509                      constraints & CT_VOLATILE ? "volatile " : "", type_string,
 510                      constraints & CT_BY_REFERENCE ? " *" : "");
 511 }
 512
 513
 514 /*******************************************************************
 515  *         get_type_constant
 516  *
 517  * Get the ARG_* constant for this data type
 518  */
 519 static int get_type_constant (const char c, const int constraints)
 520 {
 521   /* Any reference type is really a pointer */
 522   if (constraints & CT_BY_REFERENCE)
 523      return ARG_POINTER;
 524
 525   switch (c)
 526   {
 527   case 'C': case 'D': case 'E': case 'F': case 'G': case 'H': case 'I':
 528   case 'J': case 'K':
 529     return ARG_LONG;
 530   case 'M':
 531     return -1; /* FIXME */
 532   case 'N': case 'O':
 533     return ARG_DOUBLE;
 534   case 'P': case 'Q':
 535     return ARG_POINTER;
 536   case 'U': case 'V':
 537     return ARG_STRUCT;
 538   case 'X':
 539     return ARG_VOID;
 540   case 'Z':
 541   default:
 542     return -1;
 543   }
 544 }
 545
 546
 547 /*******************************************************************
 548  *         get_pointer_type_string
 549  *
 550  * Return a string containing 'pointer to expression'
 551  */
 552 static char *get_pointer_type_string (compound_type *ct,
 553                                       const char *expression)
 554 {
 555   /* FIXME: set a compound flag for bracketing expression if needed */
 556   return str_create (3, ct->flags & CT_CONST ? "const " :
 557                      ct->flags & CT_VOLATILE ? "volatile " : "", expression,
 558                      ct->flags & CT_BY_REFERENCE ? " **" : " *");
 559
 560 }