expr.c

早期freebsd实现

 *	'foo-foo' symbol references cancelled to 0,
 *		which changes X_seg from SEG_DIFFERENCE to SEG_ABSOLUTE;
 *	Unused fields zeroed to help expr().
 */

static void
clean_up_expression (expressionP)
     register expressionS * expressionP;
{
  switch (expressionP -> X_seg)
    {
    case SEG_NONE:
    case SEG_PASS1:
      expressionP -> X_add_symbol	= NULL;
      expressionP -> X_subtract_symbol	= NULL;
      expressionP -> X_add_number	= 0;
      break;

    case SEG_BIG:
    case SEG_ABSOLUTE:
      expressionP -> X_subtract_symbol	= NULL;
      expressionP -> X_add_symbol	= NULL;
      break;

    case SEG_TEXT:
    case SEG_DATA:
    case SEG_BSS:
    case SEG_UNKNOWN:
      expressionP -> X_subtract_symbol	= NULL;
      break;

    case SEG_DIFFERENCE:
      /*
       * It does not hurt to 'cancel' NULL==NULL
       * when comparing symbols for 'eq'ness.
       * It is faster to re-cancel them to NULL
       * than to check for this special case.
       */
      if (expressionP -> X_subtract_symbol == expressionP -> X_add_symbol
          || (   expressionP->X_subtract_symbol
	      && expressionP->X_add_symbol
  	      && expressionP->X_subtract_symbol->sy_frag==expressionP->X_add_symbol->sy_frag
	      && expressionP->X_subtract_symbol->sy_value==expressionP->X_add_symbol->sy_value))
	{
	  expressionP -> X_subtract_symbol	= NULL;
	  expressionP -> X_add_symbol		= NULL;
	  expressionP -> X_seg			= SEG_ABSOLUTE;
	}
      break;

    default:
      BAD_CASE( expressionP -> X_seg);
      break;
    }
}

/*
 *			expr_part ()
 *
 * Internal. Made a function because this code is used in 2 places.
 * Generate error or correct X_?????_symbol of expressionS.
 */

/*
 * symbol_1 += symbol_2 ... well ... sort of.
 */

static segT
expr_part (symbol_1_PP, symbol_2_P)
     struct symbol **	symbol_1_PP;
     struct symbol *	symbol_2_P;
{
  segT			return_value;

  know(    (* symbol_1_PP)           		== NULL
       || ((* symbol_1_PP) -> sy_type & N_TYPE) == N_TEXT
       || ((* symbol_1_PP) -> sy_type & N_TYPE) == N_DATA
       || ((* symbol_1_PP) -> sy_type & N_TYPE) == N_BSS
       || ((* symbol_1_PP) -> sy_type & N_TYPE) == N_UNDF
       );
  know(      symbol_2_P             == NULL
       ||    (symbol_2_P   -> sy_type & N_TYPE) == N_TEXT
       ||    (symbol_2_P   -> sy_type & N_TYPE) == N_DATA
       ||    (symbol_2_P   -> sy_type & N_TYPE) == N_BSS
       ||    (symbol_2_P   -> sy_type & N_TYPE) == N_UNDF
       );
  if (* symbol_1_PP)
    {
      if (((* symbol_1_PP) -> sy_type & N_TYPE) == N_UNDF)
	{
	  if (symbol_2_P)
	    {
	      return_value = SEG_PASS1;
	      * symbol_1_PP = NULL;
	    }
	  else
	    {
	      know( ((* symbol_1_PP) -> sy_type & N_TYPE) == N_UNDF)
	      return_value = SEG_UNKNOWN;
	    }
	}
      else
	{
	  if (symbol_2_P)
	    {
	      if ((symbol_2_P -> sy_type & N_TYPE) == N_UNDF)
		{
		  * symbol_1_PP = NULL;
		  return_value = SEG_PASS1;
		}
	      else
		{
		  /* {seg1} - {seg2} */
		  as_warn( "Expression too complex, 2 symbols forgotten: \"%s\" \"%s\"",
			  (* symbol_1_PP) -> sy_name, symbol_2_P -> sy_name );
		  * symbol_1_PP = NULL;
		  return_value = SEG_ABSOLUTE;
		}
	    }
	  else
	    {
	      return_value = N_TYPE_seg [(* symbol_1_PP) -> sy_type & N_TYPE];
	    }
	}
    }
  else
    {				/* (* symbol_1_PP) == NULL */
      if (symbol_2_P)
	{
	  * symbol_1_PP = symbol_2_P;
	  return_value = N_TYPE_seg [(symbol_2_P) -> sy_type & N_TYPE];
	}
      else
	{
	  * symbol_1_PP = NULL;
	  return_value = SEG_ABSOLUTE;
	}
    }
  know(   return_value == SEG_ABSOLUTE			
       || return_value == SEG_TEXT			
       || return_value == SEG_DATA			
       || return_value == SEG_BSS			
       || return_value == SEG_UNKNOWN			
       || return_value == SEG_PASS1			
       );
  know(   (* symbol_1_PP) == NULL				
       || ((* symbol_1_PP) -> sy_type & N_TYPE) == seg_N_TYPE [(int) return_value] );
  return (return_value);
}				/* expr_part() */

/* Expression parser. */

/*
 * We allow an empty expression, and just assume (absolute,0) silently.
 * Unary operators and parenthetical expressions are treated as operands.
 * As usual, Q==quantity==operand, O==operator, X==expression mnemonics.
 *
 * We used to do a aho/ullman shift-reduce parser, but the logic got so
 * warped that I flushed it and wrote a recursive-descent parser instead.
 * Now things are stable, would anybody like to write a fast parser?
 * Most expressions are either register (which does not even reach here)
 * or 1 symbol. Then "symbol+constant" and "symbol-symbol" are common.
 * So I guess it doesn't really matter how inefficient more complex expressions
 * are parsed.
 *
 * After expr(RANK,resultP) input_line_pointer -> operator of rank <= RANK.
 * Also, we have consumed any leading or trailing spaces (operand does that)
 * and done all intervening operators.
 */

typedef enum
{
O_illegal,			/* (0)  what we get for illegal op */

O_multiply,			/* (1)  * */
O_divide,			/* (2)  / */
O_modulus,			/* (3)  % */
O_left_shift,			/* (4)  < */
O_right_shift,			/* (5)  > */
O_bit_inclusive_or,		/* (6)  | */
O_bit_or_not,			/* (7)  ! */
O_bit_exclusive_or,		/* (8)  ^ */
O_bit_and,			/* (9)  & */
O_add,				/* (10) + */
O_subtract			/* (11) - */
}
operatorT;

#define __ O_illegal

static const operatorT op_encoding [256] = {	/* maps ASCII -> operators */

__, __, __, __, __, __, __, __, __, __, __, __, __, __, __, __,
__, __, __, __, __, __, __, __, __, __, __, __, __, __, __, __,

__, O_bit_or_not, __, __, __, O_modulus, O_bit_and, __,
__, __, O_multiply, O_add, __, O_subtract, __, O_divide,
__, __, __, __, __, __, __, __,
__, __, __, __, O_left_shift, __, O_right_shift, __,
__, __, __, __, __, __, __, __,
__, __, __, __, __, __, __, __,
__, __, __, __, __, __, __, __,
__, __, __, __, __, __, O_bit_exclusive_or, __,
__, __, __, __, __, __, __, __,
__, __, __, __, __, __, __, __,
__, __, __, __, __, __, __, __,
__, __, __, __, O_bit_inclusive_or, __, __, __,

__, __, __, __, __, __, __, __, __, __, __, __, __, __, __, __,
__, __, __, __, __, __, __, __, __, __, __, __, __, __, __, __,
__, __, __, __, __, __, __, __, __, __, __, __, __, __, __, __,
__, __, __, __, __, __, __, __, __, __, __, __, __, __, __, __,
__, __, __, __, __, __, __, __, __, __, __, __, __, __, __, __,
__, __, __, __, __, __, __, __, __, __, __, __, __, __, __, __,
__, __, __, __, __, __, __, __, __, __, __, __, __, __, __, __,
__, __, __, __, __, __, __, __, __, __, __, __, __, __, __, __
};


/*
 *	Rank	Examples
 *	0	operand, (expression)
 *	1	+ -
 *	2	& ^ ! |
 *	3	* / % < >
 */
typedef char operator_rankT;
static const operator_rankT
op_rank [] = { 0, 3, 3, 3, 3, 3, 2, 2, 2, 2, 1, 1 };

segT				/* Return resultP -> X_seg. */
expr (rank, resultP)
     register operator_rankT	rank; /* Larger # is higher rank. */
     register expressionS *	resultP; /* Deliver result here. */
{
  expressionS		right;
  register operatorT	op_left;
  register char		c_left;	/* 1st operator character. */
  register operatorT	op_right;
  register char		c_right;

  know( rank >= 0 );
  (void)operand (resultP);
  know( * input_line_pointer != ' ' ); /* Operand() gobbles spaces. */
  c_left = * input_line_pointer; /* Potential operator character. */
  op_left = op_encoding [c_left];
  while (op_left != O_illegal && op_rank [(int) op_left] > rank)
    {
      input_line_pointer ++;	/* -> after 1st character of operator. */
				/* Operators "<<" and ">>" have 2 characters. */
      if (* input_line_pointer == c_left && (c_left == '<' || c_left == '>') )
	{
	  input_line_pointer ++;
	}			/* -> after operator. */
      if (SEG_NONE == expr (op_rank[(int) op_left], &right))
	{
	  as_warn("Missing operand value assumed absolute 0.");
	  resultP -> X_add_number	= 0;
	  resultP -> X_subtract_symbol	= NULL;
	  resultP -> X_add_symbol	= NULL;
	  resultP -> X_seg = SEG_ABSOLUTE;
	}
      know( * input_line_pointer != ' ' );
      c_right = * input_line_pointer;
      op_right = op_encoding [c_right];
      if (* input_line_pointer == c_right && (c_right == '<' || c_right == '>') )
	{
	  input_line_pointer ++;
	}			/* -> after operator. */
      know(   (int) op_right == 0
	   || op_rank [(int) op_right] <= op_rank[(int) op_left] );
      /* input_line_pointer -> after right-hand quantity. */
      /* left-hand quantity in resultP */
      /* right-hand quantity in right. */
      /* operator in op_left. */
      if ( resultP -> X_seg == SEG_PASS1 || right . X_seg == SEG_PASS1 )
	{
	  resultP -> X_seg = SEG_PASS1;
	}
      else
	{
	  if ( resultP -> X_seg == SEG_BIG )
	    {
	      as_warn( "Left operand of %c is a %s.  Integer 0 assumed.",
		      c_left, resultP -> X_add_number > 0 ? "bignum" : "float");
	      resultP -> X_seg = SEG_ABSOLUTE;
	      resultP -> X_add_symbol = 0;
	      resultP -> X_subtract_symbol = 0;
	      resultP -> X_add_number = 0;
	    }
	  if ( right . X_seg == SEG_BIG )
	    {
	      as_warn( "Right operand of %c is a %s.  Integer 0 assumed.",
		      c_left, right . X_add_number > 0 ? "bignum" : "float");
	      right . X_seg = SEG_ABSOLUTE;
	      right . X_add_symbol = 0;
	      right . X_subtract_symbol = 0;
	      right . X_add_number = 0;
	    }
	  if ( op_left == O_subtract )
	    {
	      /*
	       * Convert - into + by exchanging symbols and negating number.
	       * I know -infinity can't be negated in 2's complement:
	       * but then it can't be subtracted either. This trick
	       * does not cause any further inaccuracy.
	       */

	      register struct symbol *	symbolP;

	      right . X_add_number      = - right . X_add_number;
	      symbolP                   = right . X_add_symbol;
	      right . X_add_symbol	= right . X_subtract_symbol;
	      right . X_subtract_symbol = symbolP;
	      if (symbolP)
		{
		  right . X_seg		= SEG_DIFFERENCE;
		}
	      op_left = O_add;
	    }

	  if ( op_left == O_add )
	    {
	      segT	seg1;
	      segT	seg2;
	      
	      know(   resultP -> X_seg == SEG_DATA		
		   || resultP -> X_seg == SEG_TEXT		
		   || resultP -> X_seg == SEG_BSS		
		   || resultP -> X_seg == SEG_UNKNOWN		
		   || resultP -> X_seg == SEG_DIFFERENCE	
		   || resultP -> X_seg == SEG_ABSOLUTE		
		   || resultP -> X_seg == SEG_PASS1		
		   );
	      know(     right .  X_seg == SEG_DATA		
		   ||   right .  X_seg == SEG_TEXT		
		   ||   right .  X_seg == SEG_BSS		
		   ||   right .  X_seg == SEG_UNKNOWN		
		   ||   right .  X_seg == SEG_DIFFERENCE	
		   ||   right .  X_seg == SEG_ABSOLUTE		
		   ||   right .  X_seg == SEG_PASS1		
		   );
	      
	      clean_up_expression (& right);
	      clean_up_expression (resultP);

	      seg1 = expr_part (& resultP -> X_add_symbol, right . X_add_symbol);
	      seg2 = expr_part (& resultP -> X_subtract_symbol, right . X_subtract_symbol);
	      if (seg1 == SEG_PASS1 || seg2 == SEG_PASS1) {
		  need_pass_2 = TRUE;
		  resultP -> X_seg = SEG_PASS1;
	      } else if (seg2 == SEG_ABSOLUTE)
		  resultP -> X_seg = seg1;
	      else if (   seg1 != SEG_UNKNOWN
			&& seg1 != SEG_ABSOLUTE
			&& seg2 != SEG_UNKNOWN
			&& seg1 != seg2) {
		  know( seg2 != SEG_ABSOLUTE );
		  know( resultP -> X_subtract_symbol );

		  know( seg1 == SEG_TEXT || seg1 == SEG_DATA || seg1== SEG_BSS );
		  know( seg2 == SEG_TEXT || seg2 == SEG_DATA || seg2== SEG_BSS );
		  know( resultP -> X_add_symbol      );
		  know( resultP -> X_subtract_symbol );
		  as_warn("Expression too complex: forgetting %s - %s",
			  resultP -> X_add_symbol      -> sy_name,
			  resultP -> X_subtract_symbol -> sy_name);
		  resultP -> X_seg = SEG_ABSOLUTE;
		  /* Clean_up_expression() will do the rest. */
		} else
		  resultP -> X_seg = SEG_DIFFERENCE;

	      resultP -> X_add_number += right . X_add_number;
	      clean_up_expression (resultP);
	    }
	  else
	    {			/* Not +. */
	      if ( resultP -> X_seg == SEG_UNKNOWN || right . X_seg == SEG_UNKNOWN )
		{
		  resultP -> X_seg = SEG_PASS1;
		  need_pass_2 = TRUE;
		}
	      else
		{
		  resultP -> X_subtract_symbol = NULL;
		  resultP -> X_add_symbol = NULL;
		  /* Will be SEG_ABSOLUTE. */
		  if ( resultP -> X_seg != SEG_ABSOLUTE || right . X_seg != SEG_ABSOLUTE )
		    {
		      as_warn( "Relocation error. Absolute 0 assumed.");
		      resultP -> X_seg        = SEG_ABSOLUTE;
		      resultP -> X_add_number = 0;
		    }
		  else
		    {
		      switch ( op_left )
			{
			case O_bit_inclusive_or:
			  resultP -> X_add_number |= right . X_add_number;
			  break;
			  
			case O_modulus:
			  if (right . X_add_number)
			    {
			      resultP -> X_add_number %= right . X_add_number;
			    }
			  else
			    {
			      as_warn( "Division by 0. 0 assumed." );
			      resultP -> X_add_number = 0;
			    }
			  break;
			  
			case O_bit_and:
			  resultP -> X_add_number &= right . X_add_number;
			  break;
			  
			case O_multiply:
			  resultP -> X_add_number *= right . X_add_number;
			  break;
			  
			case O_divide:
			  if (right . X_add_number)
			    {
			      resultP -> X_add_number /= right . X_add_number;
			    }
			  else
			    {
			      as_warn( "Division by 0. 0 assumed." );
			      resultP -> X_add_number = 0;
			    }
			  break;
			  
			case O_left_shift:
			  resultP -> X_add_number <<= right . X_add_number;
			  break;
			  
			case O_right_shift:
			  resultP -> X_add_number >>= right . X_add_number;
			  break;
			  
			case O_bit_exclusive_or:
			  resultP -> X_add_number ^= right . X_add_number;
			  break;
			  
			case O_bit_or_not:
			  resultP -> X_add_number |= ~ right . X_add_number;
			  break;
			  
			default:
			  BAD_CASE( op_left );
			  break;
			} /* switch(operator) */
		    }
		}		/* If we have to force need_pass_2. */
	    }			/* If operator was +. */
	}			/* If we didn't set need_pass_2. */
      op_left = op_right;
    }				/* While next operator is >= this rank. */
  return (resultP -> X_seg);
}

/*
 *			get_symbol_end()
 *
 * This lives here because it belongs equally in expr.c & read.c.
 * Expr.c is just a branch office read.c anyway, and putting it
 * here lessens the crowd at read.c.
 *
 * Assume input_line_pointer is at start of symbol name.
 * Advance input_line_pointer past symbol name.
 * Turn that character into a '\0', returning its former value.
 * This allows a string compare (RMS wants symbol names to be strings)
 * of the symbol name.
 * There will always be a char following symbol name, because all good
 * lines end in end-of-line.
 */
char
get_symbol_end()
{
  register char c;

  while ( is_part_of_name( c = * input_line_pointer ++ ) )
    ;
  * -- input_line_pointer = 0;
  return (c);
}

/* end: expr.c */