preproc.c

开源的nasm编译器源码,研究编译器原理很有帮且

/* -*- mode: c; c-file-style: "bsd" -*- */
/* preproc.c   macro preprocessor for the Netwide Assembler
 *
 * The Netwide Assembler is copyright (C) 1996 Simon Tatham and
 * Julian Hall. All rights reserved. The software is
 * redistributable under the licence given in the file "Licence"
 * distributed in the NASM archive.
 *
 * initial version 18/iii/97 by Simon Tatham
 */

/* Typical flow of text through preproc
 *
 * pp_getline gets tokenised lines, either
 *
 *   from a macro expansion
 *
 * or
 *   {
 *   read_line  gets raw text from stdmacpos, or predef, or current input file
 *   tokenise   converts to tokens
 *   }
 *
 * expand_mmac_params is used to expand %1 etc., unless a macro is being
 * defined or a false conditional is being processed
 * (%0, %1, %+1, %-1, %%foo
 *
 * do_directive checks for directives
 *
 * expand_smacro is used to expand single line macros
 *
 * expand_mmacro is used to expand multi-line macros
 *
 * detoken is used to convert the line back to text
 */

#include <stdio.h>
#include <stdarg.h>
#include <stdlib.h>
#include <stddef.h>
#include <string.h>
#include <ctype.h>
#include <limits.h>

#include "nasm.h"
#include "nasmlib.h"

typedef struct SMacro SMacro;
typedef struct MMacro MMacro;
typedef struct Context Context;
typedef struct Token Token;
typedef struct Blocks Blocks;
typedef struct Line Line;
typedef struct Include Include;
typedef struct Cond Cond;
typedef struct IncPath IncPath;

/*
 * Store the definition of a single-line macro.
 */
struct SMacro
{
    SMacro *next;
    char *name;
    int casesense;
    int nparam;
    int in_progress;
    Token *expansion;
};

/*
 * Store the definition of a multi-line macro. This is also used to
 * store the interiors of `%rep...%endrep' blocks, which are
 * effectively self-re-invoking multi-line macros which simply
 * don't have a name or bother to appear in the hash tables. %rep
 * blocks are signified by having a NULL `name' field.
 *
 * In a MMacro describing a `%rep' block, the `in_progress' field
 * isn't merely boolean, but gives the number of repeats left to
 * run.
 *
 * The `next' field is used for storing MMacros in hash tables; the
 * `next_active' field is for stacking them on istk entries.
 *
 * When a MMacro is being expanded, `params', `iline', `nparam',
 * `paramlen', `rotate' and `unique' are local to the invocation.
 */
struct MMacro
{
    MMacro *next;
    char *name;
    int casesense;
    int nparam_min, nparam_max;
    int plus;			/* is the last parameter greedy? */
    int nolist;			/* is this macro listing-inhibited? */
    int in_progress;
    Token *dlist;		/* All defaults as one list */
    Token **defaults;		/* Parameter default pointers */
    int ndefs;			/* number of default parameters */
    Line *expansion;

    MMacro *next_active;
    MMacro *rep_nest;		/* used for nesting %rep */
    Token **params;		/* actual parameters */
    Token *iline;		/* invocation line */
    int nparam, rotate, *paramlen;
    unsigned long unique;
    int lineno;			/* Current line number on expansion */
};

/*
 * The context stack is composed of a linked list of these.
 */
struct Context
{
    Context *next;
    SMacro *localmac;
    char *name;
    unsigned long number;
};

/*
 * This is the internal form which we break input lines up into.
 * Typically stored in linked lists.
 *
 * Note that `type' serves a double meaning: TOK_SMAC_PARAM is not
 * necessarily used as-is, but is intended to denote the number of
 * the substituted parameter. So in the definition
 *
 *     %define a(x,y) ( (x) & ~(y) )
 * 
 * the token representing `x' will have its type changed to
 * TOK_SMAC_PARAM, but the one representing `y' will be
 * TOK_SMAC_PARAM+1.
 *
 * TOK_INTERNAL_STRING is a dirty hack: it's a single string token
 * which doesn't need quotes around it. Used in the pre-include
 * mechanism as an alternative to trying to find a sensible type of
 * quote to use on the filename we were passed.
 */
struct Token
{
    Token *next;
    char *text;
    SMacro *mac;		/* associated macro for TOK_SMAC_END */
    int type;
};
enum
{
    TOK_WHITESPACE = 1, TOK_COMMENT, TOK_ID, TOK_PREPROC_ID, TOK_STRING,
    TOK_NUMBER, TOK_SMAC_END, TOK_OTHER, TOK_SMAC_PARAM,
    TOK_INTERNAL_STRING
};

/*
 * Multi-line macro definitions are stored as a linked list of
 * these, which is essentially a container to allow several linked
 * lists of Tokens.
 * 
 * Note that in this module, linked lists are treated as stacks
 * wherever possible. For this reason, Lines are _pushed_ on to the
 * `expansion' field in MMacro structures, so that the linked list,
 * if walked, would give the macro lines in reverse order; this
 * means that we can walk the list when expanding a macro, and thus
 * push the lines on to the `expansion' field in _istk_ in reverse
 * order (so that when popped back off they are in the right
 * order). It may seem cockeyed, and it relies on my design having
 * an even number of steps in, but it works...
 *
 * Some of these structures, rather than being actual lines, are
 * markers delimiting the end of the expansion of a given macro.
 * This is for use in the cycle-tracking and %rep-handling code.
 * Such structures have `finishes' non-NULL, and `first' NULL. All
 * others have `finishes' NULL, but `first' may still be NULL if
 * the line is blank.
 */
struct Line
{
    Line *next;
    MMacro *finishes;
    Token *first;
};

/*
 * To handle an arbitrary level of file inclusion, we maintain a
 * stack (ie linked list) of these things.
 */
struct Include
{
    Include *next;
    FILE *fp;
    Cond *conds;
    Line *expansion;
    char *fname;
    int lineno, lineinc;
    MMacro *mstk;		/* stack of active macros/reps */
};

/*
 * Include search path. This is simply a list of strings which get
 * prepended, in turn, to the name of an include file, in an
 * attempt to find the file if it's not in the current directory.
 */
struct IncPath
{
    IncPath *next;
    char *path;
};

/*
 * Conditional assembly: we maintain a separate stack of these for
 * each level of file inclusion. (The only reason we keep the
 * stacks separate is to ensure that a stray `%endif' in a file
 * included from within the true branch of a `%if' won't terminate
 * it and cause confusion: instead, rightly, it'll cause an error.)
 */
struct Cond
{
    Cond *next;
    int state;
};
enum
{
    /*
     * These states are for use just after %if or %elif: IF_TRUE
     * means the condition has evaluated to truth so we are
     * currently emitting, whereas IF_FALSE means we are not
     * currently emitting but will start doing so if a %else comes
     * up. In these states, all directives are admissible: %elif,
     * %else and %endif. (And of course %if.)
     */
    COND_IF_TRUE, COND_IF_FALSE,
    /*
     * These states come up after a %else: ELSE_TRUE means we're
     * emitting, and ELSE_FALSE means we're not. In ELSE_* states,
     * any %elif or %else will cause an error.
     */
    COND_ELSE_TRUE, COND_ELSE_FALSE,
    /*
     * This state means that we're not emitting now, and also that
     * nothing until %endif will be emitted at all. It's for use in
     * two circumstances: (i) when we've had our moment of emission
     * and have now started seeing %elifs, and (ii) when the
     * condition construct in question is contained within a
     * non-emitting branch of a larger condition construct.
     */
    COND_NEVER
};
#define emitting(x) ( (x) == COND_IF_TRUE || (x) == COND_ELSE_TRUE )

/* 
 * These defines are used as the possible return values for do_directive
 */
#define NO_DIRECTIVE_FOUND  0
#define DIRECTIVE_FOUND	    1

/*
 * Condition codes. Note that we use c_ prefix not C_ because C_ is
 * used in nasm.h for the "real" condition codes. At _this_ level,
 * we treat CXZ and ECXZ as condition codes, albeit non-invertible
 * ones, so we need a different enum...
 */
static const char *conditions[] = {
    "a", "ae", "b", "be", "c", "cxz", "e", "ecxz", "g", "ge", "l", "le",
    "na", "nae", "nb", "nbe", "nc", "ne", "ng", "nge", "nl", "nle", "no",
    "np", "ns", "nz", "o", "p", "pe", "po", "s", "z"
};
enum
{
    c_A, c_AE, c_B, c_BE, c_C, c_CXZ, c_E, c_ECXZ, c_G, c_GE, c_L, c_LE,
    c_NA, c_NAE, c_NB, c_NBE, c_NC, c_NE, c_NG, c_NGE, c_NL, c_NLE, c_NO,
    c_NP, c_NS, c_NZ, c_O, c_P, c_PE, c_PO, c_S, c_Z
};
static int inverse_ccs[] = {
    c_NA, c_NAE, c_NB, c_NBE, c_NC, -1, c_NE, -1, c_NG, c_NGE, c_NL, c_NLE,
    c_A, c_AE, c_B, c_BE, c_C, c_E, c_G, c_GE, c_L, c_LE, c_O, c_P, c_S,
    c_Z, c_NO, c_NP, c_PO, c_PE, c_NS, c_NZ
};

/*
 * Directive names.
 */
static const char *directives[] = {
    "%arg",
    "%assign", "%clear", "%define", "%elif", "%elifctx", "%elifdef",
    "%elifid", "%elifidn", "%elifidni", "%elifmacro", "%elifnctx", "%elifndef",
    "%elifnid", "%elifnidn", "%elifnidni", "%elifnmacro", "%elifnnum", "%elifnstr",
    "%elifnum", "%elifstr", "%else", "%endif", "%endm", "%endmacro",
    "%endrep", "%error", "%exitrep", "%iassign", "%idefine", "%if",
    "%ifctx", "%ifdef", "%ifid", "%ifidn", "%ifidni", "%ifmacro", "%ifnctx",
    "%ifndef", "%ifnid", "%ifnidn", "%ifnidni", "%ifnmacro", "%ifnnum",
    "%ifnstr", "%ifnum", "%ifstr", "%imacro", "%include",
    "%ixdefine", "%line",
    "%local",
    "%macro", "%pop", "%push", "%rep", "%repl", "%rotate",
    "%stacksize",
    "%strlen", "%substr", "%undef", "%xdefine"
};
enum
{
    PP_ARG,
    PP_ASSIGN, PP_CLEAR, PP_DEFINE, PP_ELIF, PP_ELIFCTX, PP_ELIFDEF,
    PP_ELIFID, PP_ELIFIDN, PP_ELIFIDNI, PP_ELIFMACRO, PP_ELIFNCTX, PP_ELIFNDEF,
    PP_ELIFNID, PP_ELIFNIDN, PP_ELIFNIDNI, PP_ELIFNMACRO, PP_ELIFNNUM, PP_ELIFNSTR,
    PP_ELIFNUM, PP_ELIFSTR, PP_ELSE, PP_ENDIF, PP_ENDM, PP_ENDMACRO,
    PP_ENDREP, PP_ERROR, PP_EXITREP, PP_IASSIGN, PP_IDEFINE, PP_IF,
    PP_IFCTX, PP_IFDEF, PP_IFID, PP_IFIDN, PP_IFIDNI, PP_IFMACRO, PP_IFNCTX,
    PP_IFNDEF, PP_IFNID, PP_IFNIDN, PP_IFNIDNI, PP_IFNMACRO, PP_IFNNUM,
    PP_IFNSTR, PP_IFNUM, PP_IFSTR, PP_IMACRO, PP_INCLUDE,
    PP_IXDEFINE, PP_LINE,
    PP_LOCAL,
    PP_MACRO, PP_POP, PP_PUSH, PP_REP, PP_REPL, PP_ROTATE,
    PP_STACKSIZE,
    PP_STRLEN, PP_SUBSTR, PP_UNDEF, PP_XDEFINE
};

/* If this is a an IF, ELIF, ELSE or ENDIF keyword */
static int is_condition(int arg)
{
    return ((arg >= PP_ELIF) && (arg <= PP_ENDIF)) ||
	((arg >= PP_IF) && (arg <= PP_IFSTR));
}

/* For TASM compatibility we need to be able to recognise TASM compatible
 * conditional compilation directives. Using the NASM pre-processor does
 * not work, so we look for them specifically from the following list and
 * then jam in the equivalent NASM directive into the input stream.
 */

#ifndef MAX
#       define MAX(a,b) ( ((a) > (b)) ? (a) : (b))
#endif

enum
{
    TM_ARG, TM_ELIF, TM_ELSE, TM_ENDIF, TM_IF, TM_IFDEF, TM_IFDIFI,
    TM_IFNDEF, TM_INCLUDE, TM_LOCAL
};

static const char *tasm_directives[] = {
    "arg", "elif", "else", "endif", "if", "ifdef", "ifdifi",
    "ifndef", "include", "local"
};

static int StackSize = 4;
static char *StackPointer = "ebp";
static int ArgOffset = 8;
static int LocalOffset = 4;


static Context *cstk;
static Include *istk;
static IncPath *ipath = NULL;

static efunc _error;		/* Pointer to client-provided error reporting function */
static evalfunc evaluate;

static int pass;		/* HACK: pass 0 = generate dependencies only */

static unsigned long unique;	/* unique identifier numbers */

static Line *predef = NULL;

static ListGen *list;

/*
 * The number of hash values we use for the macro lookup tables.
 * FIXME: We should *really* be able to configure this at run time,
 * or even have the hash table automatically expanding when necessary.
 */
#define NHASH 31

/*
 * The current set of multi-line macros we have defined.
 */
static MMacro *mmacros[NHASH];

/*
 * The current set of single-line macros we have defined.
 */
static SMacro *smacros[NHASH];

/*
 * The multi-line macro we are currently defining, or the %rep
 * block we are currently reading, if any.
 */
static MMacro *defining;

/*
 * The number of macro parameters to allocate space for at a time.
 */
#define PARAM_DELTA 16

/*
 * The standard macro set: defined as `static char *stdmac[]'. Also
 * gives our position in the macro set, when we're processing it.
 */
#include "macros.c"
static const char **stdmacpos;

/*
 * The extra standard macros that come from the object format, if
 * any.
 */
static const char **extrastdmac = NULL;
int any_extrastdmac;

/*
 * Tokens are allocated in blocks to improve speed
 */
#define TOKEN_BLOCKSIZE 4096
static Token *freeTokens = NULL;
struct Blocks {
	Blocks *next;
	void *chunk;
};

static Blocks blocks = { NULL, NULL };

/*
 * Forward declarations.
 */
static Token *expand_mmac_params(Token * tline);
static Token *expand_smacro(Token * tline);
static Token *expand_id(Token * tline);
static Context *get_ctx(char *name, int all_contexts);
static void make_tok_num(Token * tok, long val);
static void error(int severity, const char *fmt, ...);
static void *new_Block(size_t size);
static void delete_Blocks(void);
static Token *new_Token(Token * next, int type, char *text, int txtlen);
static Token *delete_Token(Token * t);

/*
 * Macros for safe checking of token pointers, avoid *(NULL)
 */
#define tok_type_(x,t) ((x) && (x)->type == (t))
#define skip_white_(x) if (tok_type_((x), TOK_WHITESPACE)) (x)=(x)->next
#define tok_is_(x,v) (tok_type_((x), TOK_OTHER) && !strcmp((x)->text,(v)))
#define tok_isnt_(x,v) ((x) && ((x)->type!=TOK_OTHER || strcmp((x)->text,(v))))

/* Handle TASM specific directives, which do not contain a % in
 * front of them. We do it here because I could not find any other
 * place to do it for the moment, and it is a hack (ideally it would
 * be nice to be able to use the NASM pre-processor to do it).
 */
static char *
check_tasm_directive(char *line)
{
    int i, j, k, m, len;
    char *p = line, *oldline, oldchar;

    /* Skip whitespace */
    while (isspace(*p) && *p != 0)
	p++;

    /* Binary search for the directive name */
    i = -1;
    j = elements(tasm_directives);
    len = 0;
    while (!isspace(p[len]) && p[len] != 0)
	len++;
    if (len)
    {
	oldchar = p[len];
	p[len] = 0;
	while (j - i > 1)
	{
	    k = (j + i) / 2;
	    m = nasm_stricmp(p, tasm_directives[k]);
	    if (m == 0)
	    {
		/* We have found a directive, so jam a % in front of it
		 * so that NASM will then recognise it as one if it's own.
		 */
		p[len] = oldchar;
		len = strlen(p);
		oldline = line;
		line = nasm_malloc(len + 2);
		line[0] = '%';
		if (k == TM_IFDIFI)
		{
		    /* NASM does not recognise IFDIFI, so we convert it to
		     * %ifdef BOGUS. This is not used in NASM comaptible
		     * code, but does need to parse for the TASM macro
		     * package.
		     */
		    strcpy(line + 1, "ifdef BOGUS");
		}
		else
		{
		    memcpy(line + 1, p, len + 1);
		}
		nasm_free(oldline);
		return line;
	    }
	    else if (m < 0)
	    {
		j = k;
	    }
	    else
		i = k;
	}
	p[len] = oldchar;
    }
    return line;
}

/*
 * The pre-preprocessing stage... This function translates line
 * number indications as they emerge from GNU cpp (`# lineno "file"
 * flags') into NASM preprocessor line number indications (`%line
 * lineno file').
 */
static char *
prepreproc(char *line)
{
    int lineno, fnlen;
    char *fname, *oldline;

    if (line[0] == '#' && line[1] == ' ')
    {
	oldline = line;
	fname = oldline + 2;
	lineno = atoi(fname);
	fname += strspn(fname, "0123456789 ");
	if (*fname == '"')