expr.c

早期freebsd实现

/* expr.c -operands, expressions-
   Copyright (C) 1987 Free Software Foundation, Inc.

This file is part of GAS, the GNU Assembler.

GAS is free software; you can redistribute it and/or modify
it under the terms of the GNU General Public License as published by
the Free Software Foundation; either version 1, or (at your option)
any later version.

GAS is distributed in the hope that it will be useful,
but WITHOUT ANY WARRANTY; without even the implied warranty of
MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
GNU General Public License for more details.

You should have received a copy of the GNU General Public License
along with GAS; see the file COPYING.  If not, write to
the Free Software Foundation, 675 Mass Ave, Cambridge, MA 02139, USA.  */

/*
 * This is really a branch office of as-read.c. I split it out to clearly
 * distinguish the world of expressions from the world of statements.
 * (It also gives smaller files to re-compile.)
 * Here, "operand"s are of expressions, not instructions.
 */

#include <ctype.h>
#include "as.h"
#include "flonum.h"
#include "read.h"
#include "struc-symbol.h"
#include "expr.h"
#include "obstack.h"
#include "symbols.h"

static void clean_up_expression();	/* Internal. */
extern const char EXP_CHARS[];	/* JF hide MD floating pt stuff all the same place */
extern const char FLT_CHARS[];

#ifdef SUN_ASM_SYNTAX
extern int local_label_defined[];
#endif

/*
 * Build any floating-point literal here.
 * Also build any bignum literal here.
 */

/* LITTLENUM_TYPE	generic_buffer [6];	/* JF this is a hack */
/* Seems atof_machine can backscan through generic_bignum and hit whatever
   happens to be loaded before it in memory.  And its way too complicated
   for me to fix right.  Thus a hack.  JF:  Just make generic_bignum bigger,
   and never write into the early words, thus they'll always be zero.
   I hate Dean's floating-point code.  Bleh.
 */
LITTLENUM_TYPE	generic_bignum [SIZE_OF_LARGE_NUMBER+6];
FLONUM_TYPE	generic_floating_point_number =
{
  & generic_bignum [6],		/* low (JF: Was 0) */
  & generic_bignum [SIZE_OF_LARGE_NUMBER+6 - 1], /* high JF: (added +6) */
  0,				/* leader */
  0,				/* exponent */
  0				/* sign */
};
/* If nonzero, we've been asked to assemble nan, +inf or -inf */
int generic_floating_point_magic;

/*
 * Summary of operand().
 *
 * in:	Input_line_pointer points to 1st char of operand, which may
 *	be a space.
 *
 * out:	A expressionS. X_seg determines how to understand the rest of the
 *	expressionS.
 *	The operand may have been empty: in this case X_seg == SEG_NONE.
 *	Input_line_pointer -> (next non-blank) char after operand.
 *
 */

static segT
operand (expressionP)
     register expressionS *	expressionP;
{
  register char		c;
  register char *name;	/* points to name of symbol */
  register struct symbol *	symbolP; /* Points to symbol */

  extern  char hex_value[];	/* In hex_value.c */
  char	*local_label_name();

  SKIP_WHITESPACE();		/* Leading whitespace is part of operand. */
  c = * input_line_pointer ++;	/* Input_line_pointer -> past char in c. */
  if (isdigit(c))
    {
      register valueT	number;	/* offset or (absolute) value */
      register short int digit;	/* value of next digit in current radix */
				/* invented for humans only, hope */
				/* optimising compiler flushes it! */
      register short int radix;	/* 8, 10 or 16 */
				/* 0 means we saw start of a floating- */
				/* point constant. */
      register short int maxdig;/* Highest permitted digit value. */
      register int	too_many_digits; /* If we see >= this number of */
				/* digits, assume it is a bignum. */
      register char *	digit_2; /* -> 2nd digit of number. */
               int	small;	/* TRUE if fits in 32 bits. */

      if (c=='0')
	{			/* non-decimal radix */
	  if ((c = * input_line_pointer ++)=='x' || c=='X')
	    {
	      c = * input_line_pointer ++; /* read past "0x" or "0X" */
	      maxdig = radix = 16;
	      too_many_digits = 9;
	    }
	  else
	    {
	      /* If it says '0f' and the line ends or it DOESN'T look like
	         a floating point #, its a local label ref.  DTRT */
	      if(c=='f' && (! *input_line_pointer ||
			    (!index("+-.0123456789",*input_line_pointer) &&
 			    !index(EXP_CHARS,*input_line_pointer))))
		{
	          maxdig = radix = 10;
		  too_many_digits = 11;
		  c='0';
		  input_line_pointer-=2;
		}
	      else if (c && index (FLT_CHARS,c))
		{
		  radix = 0;	/* Start of floating-point constant. */
				/* input_line_pointer -> 1st char of number. */
		  expressionP -> X_add_number =  - (isupper(c) ? tolower(c) : c);
		}
	      else
		{		/* By elimination, assume octal radix. */
		  radix = 8;
		  maxdig = 10;	/* Un*x sux. Compatibility. */
		  too_many_digits = 11;
		}
	    }
	  /* c == char after "0" or "0x" or "0X" or "0e" etc.*/
	}
      else
	{
	  maxdig = radix = 10;
	  too_many_digits = 11;
	}
      if (radix)
	{			/* Fixed-point integer constant. */
				/* May be bignum, or may fit in 32 bits. */
/*
 * Most numbers fit into 32 bits, and we want this case to be fast.
 * So we pretend it will fit into 32 bits. If, after making up a 32
 * bit number, we realise that we have scanned more digits than
 * comfortably fit into 32 bits, we re-scan the digits coding
 * them into a bignum. For decimal and octal numbers we are conservative: some
 * numbers may be assumed bignums when in fact they do fit into 32 bits.
 * Numbers of any radix can have excess leading zeros: we strive
 * to recognise this and cast them back into 32 bits.
 * We must check that the bignum really is more than 32
 * bits, and change it back to a 32-bit number if it fits.
 * The number we are looking for is expected to be positive, but
 * if it fits into 32 bits as an unsigned number, we let it be a 32-bit
 * number. The cavalier approach is for speed in ordinary cases.
 */
	  digit_2 = input_line_pointer;
	  for (number=0;  (digit=hex_value[c])<maxdig;  c = * input_line_pointer ++)
	    {
	      number = number * radix + digit;
	    }
	  /* C contains character after number. */
	  /* Input_line_pointer -> char after C. */
	  small = input_line_pointer - digit_2 < too_many_digits;
	  if ( ! small)
	    {
	      /*
	       * We saw a lot of digits. Manufacture a bignum the hard way.
	       */
	      LITTLENUM_TYPE *	leader;	/* -> high order littlenum of the bignum. */
	      LITTLENUM_TYPE *	pointer; /* -> littlenum we are frobbing now. */
	      long int		carry;

	      leader = generic_bignum;
	      generic_bignum [0] = 0;
	      generic_bignum [1] = 0;
				/* We could just use digit_2, but lets be mnemonic. */
	      input_line_pointer = -- digit_2; /* -> 1st digit. */
	      c = *input_line_pointer ++;
	      for (;   (carry = hex_value [c]) < maxdig;   c = * input_line_pointer ++)
		{
		  for (pointer = generic_bignum;
		       pointer <= leader;
		       pointer ++)
		    {
		      long int	work;

		      work = carry + radix * * pointer;
		      * pointer = work & LITTLENUM_MASK;
		      carry = work >> LITTLENUM_NUMBER_OF_BITS;
		    }
		  if (carry)
		    {
		      if (leader < generic_bignum + SIZE_OF_LARGE_NUMBER - 1)
			{	/* Room to grow a longer bignum. */
			  * ++ leader = carry;
			}
		    }
		}
	      /* Again, C is char after number, */
	      /* input_line_pointer -> after C. */
	      know( BITS_PER_INT == 32 );
	      know( LITTLENUM_NUMBER_OF_BITS == 16 );
	      /* Hence the constant "2" in the next line. */
	      if (leader < generic_bignum + 2)
		{		/* Will fit into 32 bits. */
		  number =
		    ( (generic_bignum [1] & LITTLENUM_MASK) << LITTLENUM_NUMBER_OF_BITS )
		    | (generic_bignum [0] & LITTLENUM_MASK);
		  small = TRUE;
		}
	      else
		{
		  number = leader - generic_bignum + 1;	/* Number of littlenums in the bignum. */
		}
	    }
	  if (small)
	    {
	      /*
	       * Here with number, in correct radix. c is the next char.
	       * Note that unlike Un*x, we allow "011f" "0x9f" to
	       * both mean the same as the (conventional) "9f". This is simply easier
	       * than checking for strict canonical form. Syntax sux!
	       */
	      if (number<10)
		{
#ifdef SUN_ASM_SYNTAX
		  if (c=='b' || (c=='$' && local_label_defined[number]))
#else
		  if (c=='b')
#endif
		    {
		      /*
		       * Backward ref to local label.
		       * Because it is backward, expect it to be DEFINED.
		       */
		      /*
		       * Construct a local label.
		       */
		      name = local_label_name ((int)number, 0);
		      if ( (symbolP = symbol_table_lookup(name)) /* seen before */
			  && (symbolP -> sy_type & N_TYPE) != N_UNDF /* symbol is defined: OK */
			  )
			{		/* Expected path: symbol defined. */
			  /* Local labels are never absolute. Don't waste time checking absoluteness. */
			  know(   (symbolP -> sy_type & N_TYPE) == N_DATA
			       || (symbolP -> sy_type & N_TYPE) == N_TEXT );
			  expressionP -> X_add_symbol = symbolP;
			  expressionP -> X_add_number = 0;
			  expressionP -> X_seg	      = N_TYPE_seg [symbolP -> sy_type];
			}
		      else
			{		/* Either not seen or not defined. */
			  as_warn( "Backw. ref to unknown label \"%d:\", 0 assumed.",
				  number
				  );
			  expressionP -> X_add_number = 0;
			  expressionP -> X_seg        = SEG_ABSOLUTE;
			}
		    }
		  else
		    {
#ifdef SUN_ASM_SYNTAX
		      if (c=='f' || (c=='$' && !local_label_defined[number]))
#else
		      if (c=='f')
#endif
			{
			  /*
			   * Forward reference. Expect symbol to be undefined or
			   * unknown. Undefined: seen it before. Unknown: never seen
			   * it in this pass.
			   * Construct a local label name, then an undefined symbol.
			   * Don't create a XSEG frag for it: caller may do that.
			   * Just return it as never seen before.
			   */
			  name = local_label_name ((int)number, 1);
			  if ( symbolP = symbol_table_lookup( name ))
			    {
			      /* We have no need to check symbol properties. */
			      know(   (symbolP -> sy_type & N_TYPE) == N_UNDF
				   || (symbolP -> sy_type & N_TYPE) == N_DATA
				   || (symbolP -> sy_type & N_TYPE) == N_TEXT);
			    }
			  else
			    {
			      symbolP = symbol_new (name, N_UNDF, 0,0,0, & zero_address_frag);
			      symbol_table_insert (symbolP);
			    }
			  expressionP -> X_add_symbol      = symbolP;
			  expressionP -> X_seg             = SEG_UNKNOWN;
			  expressionP -> X_subtract_symbol = NULL;
			  expressionP -> X_add_number      = 0;
			}
		      else
			{		/* Really a number, not a local label. */
			  expressionP -> X_add_number = number;
			  expressionP -> X_seg        = SEG_ABSOLUTE;
			  input_line_pointer --; /* Restore following character. */
			}		/* if (c=='f') */
		    }			/* if (c=='b') */
		}
	      else
		{			/* Really a number. */
		  expressionP -> X_add_number = number;
		  expressionP -> X_seg        = SEG_ABSOLUTE;
		  input_line_pointer --; /* Restore following character. */
		}			/* if (number<10) */
	    }
	  else
	    {
	      expressionP -> X_add_number = number;
	      expressionP -> X_seg = SEG_BIG;
	      input_line_pointer --; /* -> char following number. */
	    }			/* if (small) */
	}			/* (If integer constant) */
      else
	{			/* input_line_pointer -> */
				/* floating-point constant. */
	  int error_code;

	  error_code = atof_generic
	    (& input_line_pointer, ".", EXP_CHARS,
	     & generic_floating_point_number);

	  if (error_code)
	    {
	      if (error_code == ERROR_EXPONENT_OVERFLOW)
		{
		  as_warn( "Bad floating-point constant: exponent overflow, probably assembling junk" );
		}
	      else
		{	      
		  as_warn( "Bad floating-point constant: unknown error code=%d.", error_code);
		}
	    }
	  expressionP -> X_seg = SEG_BIG;
				/* input_line_pointer -> just after constant, */
				/* which may point to whitespace. */
	  know( expressionP -> X_add_number < 0 ); /* < 0 means "floating point". */
	}			/* if (not floating-point constant) */
    }
  else if(c=='.' && !is_part_of_name(*input_line_pointer)) {
    extern struct obstack frags;

    /*
       JF:  '.' is pseudo symbol with value of current location in current
       segment. . .
     */
    symbolP = symbol_new("L0\001",
			 (unsigned char)(seg_N_TYPE[(int)now_seg]),
			 0,
			 0,
			 (valueT)(obstack_next_free(&frags)-frag_now->fr_literal),
			 frag_now);
    expressionP->X_add_number=0;
    expressionP->X_add_symbol=symbolP;
    expressionP->X_seg = now_seg;

  } else if ( is_name_beginner(c) ) /* here if did not begin with a digit */
    {
      /*
       * Identifier begins here.
       * This is kludged for speed, so code is repeated.
       */
      name =  -- input_line_pointer;
      c = get_symbol_end();
      symbolP = symbol_table_lookup(name);
      if (symbolP)
	    {
          /*
           * If we have an absolute symbol, then we know it's value now.
           */
          register segT    	seg;

          seg = N_TYPE_seg [(int) symbolP -> sy_type & N_TYPE];
          if ((expressionP -> X_seg = seg) == SEG_ABSOLUTE )
	    {
	      expressionP -> X_add_number = symbolP -> sy_value;
	    }
	  else
	    {
	      expressionP -> X_add_number  = 0;
	      expressionP -> X_add_symbol  = symbolP;
	    }
	}
      else
	{
	  expressionP -> X_add_symbol
		= symbolP
		= symbol_new (name, N_UNDF, 0,0,0, & zero_address_frag);

	  expressionP -> X_add_number  = 0;
	  expressionP -> X_seg         = SEG_UNKNOWN;
	  symbol_table_insert (symbolP);
	}
      * input_line_pointer = c;
      expressionP -> X_subtract_symbol = NULL;
    }
  else if (c=='(')/* didn't begin with digit & not a name */
    {
      (void)expression( expressionP );
      /* Expression() will pass trailing whitespace */
      if ( * input_line_pointer ++ != ')' )
	{
	  as_warn( "Missing ')' assumed");
	  input_line_pointer --;
	}
      /* here with input_line_pointer -> char after "(...)" */
    }
  else if ( c=='~' || c=='-' )
    {		/* unary operator: hope for SEG_ABSOLUTE */
      switch(operand (expressionP)) {
      case SEG_ABSOLUTE:
		    /* input_line_pointer -> char after operand */
	if ( c=='-' )
	  {
	    expressionP -> X_add_number = - expressionP -> X_add_number;
/*
 * Notice: '-' may  overflow: no warning is given. This is compatible
 * with other people's assemblers. Sigh.
 */
	  }
	else
	  {
	    expressionP -> X_add_number = ~ expressionP -> X_add_number;
	  }
	  break;

      case SEG_TEXT:
      case SEG_DATA:
      case SEG_BSS:
      case SEG_PASS1:
      case SEG_UNKNOWN:
	if(c=='-') {		/* JF I hope this hack works */
	  expressionP->X_subtract_symbol=expressionP->X_add_symbol;
	  expressionP->X_add_symbol=0;
	  expressionP->X_seg=SEG_DIFFERENCE;
	  break;
	}
      default:		/* unary on non-absolute is unsuported */
	as_warn("Unary operator %c ignored because bad operand follows", c);
	break;
	/* Expression undisturbed from operand(). */
      }
    }
  else if (c=='\'')
    {
/*
 * Warning: to conform to other people's assemblers NO ESCAPEMENT is permitted
 * for a single quote. The next character, parity errors and all, is taken
 * as the value of the operand. VERY KINKY.
 */
      expressionP -> X_add_number = * input_line_pointer ++;
      expressionP -> X_seg        = SEG_ABSOLUTE;
    }
  else
    {
		      /* can't imagine any other kind of operand */
      expressionP -> X_seg = SEG_NONE;
      input_line_pointer --;
    }
/*
 * It is more 'efficient' to clean up the expressions when they are created.
 * Doing it here saves lines of code.
 */
  clean_up_expression (expressionP);
  SKIP_WHITESPACE();		/* -> 1st char after operand. */
  know( * input_line_pointer != ' ' );
  return (expressionP -> X_seg);
}				/* operand */

/* Internal. Simplify a struct expression for use by expr() */

/*
 * In:	address of a expressionS.
 *	The X_seg field of the expressionS may only take certain values.
 *	Now, we permit SEG_PASS1 to make code smaller & faster.
 *	Elsewise we waste time special-case testing. Sigh. Ditto SEG_NONE.
 * Out:	expressionS may have been modified: