queryparser/lemon.c

Go to the documentation of this file.

/*
** This file contains all sources (including headers) to the LEMON
** LALR(1) parser generator.  The sources have been combined into a
** single file to make it easy to include LEMON in the source tree
** and Makefile of another program.
**
** The authors of this program disclaim copyright.
**
** Modified to add "-o" and "-h" command line options.  Olly Betts 2005-02-14
** Modified to fix a number of compiler warnings.  Olly Betts 2007-02-20
*/
#include <stdio.h>
#include <stdarg.h>
#include <string.h>
#include <ctype.h>
#include <stdlib.h>

#ifndef __WIN32__
#   if defined(_WIN32) || defined(WIN32)
#       define __WIN32__
#   endif
#endif

#ifndef __WIN32__
#   include <unistd.h>
#endif

/* #define PRIVATE static */
#define PRIVATE

#ifdef TEST
#define MAXRHS 5       /* Set low to exercise exception code */
#else
#define MAXRHS 1000
#endif

char *msort();
/* ISO C prototypes malloc in stdlib.h: extern void *malloc(); */

/******** From the file "action.h" *************************************/
struct action *Action_new();
struct action *Action_sort();

/********* From the file "assert.h" ************************************/
void myassert();
#ifndef NDEBUG
#  define assert(X) if(!(X))myassert(__FILE__,__LINE__)
#else
#  define assert(X)
#endif

/********** From the file "build.h" ************************************/
void FindRulePrecedences();
void FindFirstSets();
void FindStates();
void FindLinks();
void FindFollowSets();
void FindActions();

/********* From the file "configlist.h" *********************************/
void Configlist_init(/* void */);
struct config *Configlist_add(/* struct rule *, int */);
struct config *Configlist_addbasis(/* struct rule *, int */);
void Configlist_closure(/* void */);
void Configlist_sort(/* void */);
void Configlist_sortbasis(/* void */);
struct config *Configlist_return(/* void */);
struct config *Configlist_basis(/* void */);
void Configlist_eat(/* struct config * */);
void Configlist_reset(/* void */);

/********* From the file "error.h" ***************************************/
void ErrorMsg(const char *, int,const char *, ...);

/****** From the file "option.h" ******************************************/
struct s_options {
  enum { OPT_FLAG=1,  OPT_INT,  OPT_DBL,  OPT_STR,
         OPT_FFLAG, OPT_FINT, OPT_FDBL, OPT_FSTR} type;
  char *label;
  void *arg;
  void(*func)();

  char *message;
};
int    OptInit(/* char**,struct s_options*,FILE* */);
int    OptNArgs(/* void */);
char  *OptArg(/* int */);
void   OptErr(/* int */);
void   OptPrint(/* void */);

/******** From the file "parse.h" *****************************************/
void Parse(/* struct lemon *lemp */);

/********* From the file "plink.h" ***************************************/
struct plink *Plink_new(/* void */);
void Plink_add(/* struct plink **, struct config * */);
void Plink_copy(/* struct plink **, struct plink * */);
void Plink_delete(/* struct plink * */);

/********** From the file "report.h" *************************************/
void Reprint(/* struct lemon * */);
void ReportOutput(/* struct lemon * */);
void ReportTable(/* struct lemon * */);
void ReportHeader(/* struct lemon * */);
void CompressTables(/* struct lemon * */);

/********** From the file "set.h" ****************************************/
void  SetSize(/* int N */);             /* All sets will be of size N */
char *SetNew(/* void */);               /* A new set for element 0..N */
void  SetFree(/* char* */);             /* Deallocate a set */

int SetAdd(/* char*,int */);            /* Add element to a set */
int SetUnion(/* char *A,char *B */);    /* A <- A U B, thru element N */

#define SetFind(X,Y) (X[Y])       /* True if Y is in set X */

/********** From the file "struct.h" *************************************/
/*
** Principal data structures for the LEMON parser generator.
*/

typedef enum {B_FALSE=0, B_TRUE} Boolean;

/* Symbols (terminals and nonterminals) of the grammar are stored
** in the following: */
struct symbol {
  char *name;              /* Name of the symbol */
  int index;               /* Index number for this symbol */
  enum {
    TERMINAL,
    NONTERMINAL
  } type;                  /* Symbols are all either TERMINALS or NTs */
  struct rule *rule;       /* Linked list of rules of this (if an NT) */
  struct symbol *fallback; /* fallback token in case this token doesn't parse */
  int prec;                /* Precedence if defined (-1 otherwise) */
  enum e_assoc {
    LEFT,
    RIGHT,
    NONE,
    UNK
  } assoc;                 /* Associativity if predecence is defined */
  char *firstset;          /* First-set for all rules of this symbol */
  Boolean lambda;          /* True if NT and can generate an empty string */
  char *destructor;        /* Code which executes whenever this symbol is
                           ** popped from the stack during error processing */
  int destructorln;        /* Line number of destructor code */
  char *datatype;          /* The data type of information held by this
                           ** object. Only used if type==NONTERMINAL */
  int dtnum;               /* The data type number.  In the parser, the value
                           ** stack is a union.  The .yy%d element of this
                           ** union is the correct data type for this object */
};

/* Each production rule in the grammar is stored in the following
** structure.  */
struct rule {
  struct symbol *lhs;      /* Left-hand side of the rule */
  char *lhsalias;          /* Alias for the LHS (NULL if none) */
  int ruleline;            /* Line number for the rule */
  int nrhs;                /* Number of RHS symbols */
  struct symbol **rhs;     /* The RHS symbols */
  char **rhsalias;         /* An alias for each RHS symbol (NULL if none) */
  int line;                /* Line number at which code begins */
  char *code;              /* The code executed when this rule is reduced */
  struct symbol *precsym;  /* Precedence symbol for this rule */
  int index;               /* An index number for this rule */
  Boolean canReduce;       /* True if this rule is ever reduced */
  struct rule *nextlhs;    /* Next rule with the same LHS */
  struct rule *next;       /* Next rule in the global list */
};

/* A configuration is a production rule of the grammar together with
** a mark (dot) showing how much of that rule has been processed so far.
** Configurations also contain a follow-set which is a list of terminal
** symbols which are allowed to immediately follow the end of the rule.
** Every configuration is recorded as an instance of the following: */
struct config {
  struct rule *rp;         /* The rule upon which the configuration is based */
  int dot;                 /* The parse point */
  char *fws;               /* Follow-set for this configuration only */
  struct plink *fplp;      /* Follow-set forward propagation links */
  struct plink *bplp;      /* Follow-set backwards propagation links */
  struct state *stp;       /* Pointer to state which contains this */
  enum {
    COMPLETE,              /* The status is used during followset and */
    INCOMPLETE             /*    shift computations */
  } status;
  struct config *next;     /* Next configuration in the state */
  struct config *bp;       /* The next basis configuration */
};

/* Every shift or reduce operation is stored as one of the following */
struct action {
  struct symbol *sp;       /* The look-ahead symbol */
  enum e_action {
    SHIFT,
    ACCEPT,
    REDUCE,
    ERROR,
    CONFLICT,                /* Was a reduce, but part of a conflict */
    SH_RESOLVED,             /* Was a shift.  Precedence resolved conflict */
    RD_RESOLVED,             /* Was reduce.  Precedence resolved conflict */
    NOT_USED                 /* Deleted by compression */
  } type;
  union {
    struct state *stp;     /* The new state, if a shift */
    struct rule *rp;       /* The rule, if a reduce */
  } x;
  struct action *next;     /* Next action for this state */
  struct action *collide;  /* Next action with the same hash */
};

/* Each state of the generated parser's finite state machine
** is encoded as an instance of the following structure. */
struct state {
  struct config *bp;       /* The basis configurations for this state */
  struct config *cfp;      /* All configurations in this set */
  int index;               /* Sequencial number for this state */
  struct action *ap;       /* Array of actions for this state */
  int nTknAct, nNtAct;     /* Number of actions on terminals and nonterminals */
  int iTknOfst, iNtOfst;   /* yy_action[] offset for terminals and nonterms */
  int iDflt;               /* Default action */
};
#define NO_OFFSET (-2147483647)

/* A followset propagation link indicates that the contents of one
** configuration followset should be propagated to another whenever
** the first changes. */
struct plink {
  struct config *cfp;      /* The configuration to which linked */
  struct plink *next;      /* The next propagate link */
};

/* The state vector for the entire parser generator is recorded as
** follows.  (LEMON uses no global variables and makes little use of
** static variables.  Fields in the following structure can be thought
** of as begin global variables in the program.) */
struct lemon {
  struct state **sorted;   /* Table of states sorted by state number */
  struct rule *rule;       /* List of all rules */
  int nstate;              /* Number of states */
  int nrule;               /* Number of rules */
  int nsymbol;             /* Number of terminal and nonterminal symbols */
  int nterminal;           /* Number of terminal symbols */
  struct symbol **symbols; /* Sorted array of pointers to symbols */
  int errorcnt;            /* Number of errors */
  struct symbol *errsym;   /* The error symbol */
  char *name;              /* Name of the generated parser */
  char *arg;               /* Declaration of the 3th argument to parser */
  char *tokentype;         /* Type of terminal symbols in the parser stack */
  char *vartype;           /* The default type of non-terminal symbols */
  char *start;             /* Name of the start symbol for the grammar */
  char *stacksize;         /* Size of the parser stack */
  char *include;           /* Code to put at the start of the C file */
  int  includeln;          /* Line number for start of include code */
  char *error;             /* Code to execute when an error is seen */
  int  errorln;            /* Line number for start of error code */
  char *overflow;          /* Code to execute on a stack overflow */
  int  overflowln;         /* Line number for start of overflow code */
  char *failure;           /* Code to execute on parser failure */
  int  failureln;          /* Line number for start of failure code */
  char *accept;            /* Code to execute when the parser excepts */
  int  acceptln;           /* Line number for the start of accept code */
  char *extracode;         /* Code appended to the generated file */
  int  extracodeln;        /* Line number for the start of the extra code */
  char *tokendest;         /* Code to execute to destroy token data */
  int  tokendestln;        /* Line number for token destroyer code */
  char *vardest;           /* Code for the default non-terminal destructor */
  int  vardestln;          /* Line number for default non-term destructor code*/
  char *filename;          /* Name of the input file */
  char *outname;           /* Name of the current output file */
  char *tokenprefix;       /* A prefix added to token names in the .h file */
  int nconflict;           /* Number of parsing conflicts */
  int tablesize;           /* Size of the parse tables */
  int basisflag;           /* Print only basis configurations */
  int has_fallback;        /* True if any %fallback is seen in the grammer */
  char *argv0;             /* Name of the program */
};

#define MemoryCheck(X) if((X)==0){ \
  extern void memory_error(); \
  memory_error(); \
}

/**************** From the file "table.h" *********************************/
/*
** All code in this file has been automatically generated
** from a specification in the file
**              "table.q"
** by the associative array code building program "aagen".
** Do not edit this file!  Instead, edit the specification
** file, then rerun aagen.
*/
/*
** Code for processing tables in the LEMON parser generator.
*/

/* Routines for handling a strings */

char *Strsafe();

void Strsafe_init(/* void */);
int Strsafe_insert(/* char * */);
char *Strsafe_find(/* char * */);

/* Routines for handling symbols of the grammar */

struct symbol *Symbol_new();
int Symbolcmpp(const void *void_a, const void *void_b);
void Symbol_init(/* void */);
int Symbol_insert(/* struct symbol *, char * */);
struct symbol *Symbol_find(/* char * */);
struct symbol *Symbol_Nth(/* int */);
int Symbol_count(/*  */);
struct symbol **Symbol_arrayof(/*  */);

/* Routines to manage the state table */

int Configcmp(/* struct config *, struct config * */);
struct state *State_new();
void State_init(/* void */);
int State_insert(/* struct state *, struct config * */);
struct state *State_find(/* struct config * */);
struct state **State_arrayof(/*  */);

/* Routines used for efficiency in Configlist_add */

void Configtable_init(/* void */);
int Configtable_insert(/* struct config * */);
struct config *Configtable_find(/* struct config * */);
void Configtable_clear(/* int(*)(struct config *) */);
/****************** From the file "action.c" *******************************/
/*
** Routines processing parser actions in the LEMON parser generator.
*/

/* Allocate a new parser action */
struct action *Action_new(){
  static struct action *freelist = 0;
  struct action *new;

  if( freelist==0 ){
    int i;
    int amt = 100;
    freelist = (struct action *)malloc( sizeof(struct action)*amt );
    if( freelist==0 ){
      fprintf(stderr,"Unable to allocate memory for a new parser action.");
      exit(1);
    }
    for(i=0; i<amt-1; i++) freelist[i].next = &freelist[i+1];
    freelist[amt-1].next = 0;
  }
  new = freelist;
  freelist = freelist->next;
  return new;
}

/* Compare two actions */
static int actioncmp(ap1,ap2)
struct action *ap1;
struct action *ap2;
{
  int rc;
  rc = ap1->sp->index - ap2->sp->index;
  if( rc==0 ) rc = (int)ap1->type - (int)ap2->type;
  if( rc==0 ){
    assert( ap1->type==REDUCE || ap1->type==RD_RESOLVED || ap1->type==CONFLICT);
    assert( ap2->type==REDUCE || ap2->type==RD_RESOLVED || ap2->type==CONFLICT);
    rc = ap1->x.rp->index - ap2->x.rp->index;
  }
  return rc;
}

/* Sort parser actions */
struct action *Action_sort(ap)
struct action *ap;
{
  /* Cast to "char **" via "void *" to avoid aliasing problems. */
  ap = (struct action *)msort((char *)ap,(char **)(void *)&ap->next,actioncmp);
  return ap;
}

void Action_add(app,type,sp,arg)
struct action **app;
enum e_action type;
struct symbol *sp;
char *arg;
{
  struct action *new;
  new = Action_new();
  new->next = *app;
  *app = new;
  new->type = type;
  new->sp = sp;
  if( type==SHIFT ){
    new->x.stp = (struct state *)arg;
  }else{
    new->x.rp = (struct rule *)arg;
  }
}
/********************** New code to implement the "acttab" module ***********/
/*
** This module implements routines use to construct the yy_action[] table.
*/

/*
** The state of the yy_action table under construction is an instance of
** the following structure
*/
typedef struct acttab acttab;
struct acttab {
  int nAction;                 /* Number of used slots in aAction[] */
  int nActionAlloc;            /* Slots allocated for aAction[] */
  struct {
    int lookahead;             /* Value of the lookahead token */
    int action;                /* Action to take on the given lookahead */
  } *aAction,                  /* The yy_action[] table under construction */
    *aLookahead;               /* A single new transaction set */
  int mnLookahead;             /* Minimum aLookahead[].lookahead */
  int mnAction;                /* Action associated with mnLookahead */
  int mxLookahead;             /* Maximum aLookahead[].lookahead */
  int nLookahead;              /* Used slots in aLookahead[] */
  int nLookaheadAlloc;         /* Slots allocated in aLookahead[] */
};

/* Return the number of entries in the yy_action table */
#define acttab_size(X) ((X)->nAction)

/* The value for the N-th entry in yy_action */
#define acttab_yyaction(X,N)  ((X)->aAction[N].action)

/* The value for the N-th entry in yy_lookahead */
#define acttab_yylookahead(X,N)  ((X)->aAction[N].lookahead)

/* Free all memory associated with the given acttab */
void acttab_free(acttab *p){
  free( p->aAction );
  free( p->aLookahead );
  free( p );
}

/* Allocate a new acttab structure */
acttab *acttab_alloc(void){
  acttab *p = malloc( sizeof(*p) );
  if( p==0 ){
    fprintf(stderr,"Unable to allocate memory for a new acttab.");
    exit(1);
  }
  memset(p, 0, sizeof(*p));
  return p;
}

/* Add a new action to the current transaction set
*/
void acttab_action(acttab *p, int lookahead, int action){
  if( p->nLookahead>=p->nLookaheadAlloc ){
    p->nLookaheadAlloc += 25;
    p->aLookahead = realloc( p->aLookahead,
                             sizeof(p->aLookahead[0])*p->nLookaheadAlloc );
    if( p->aLookahead==0 ){
      fprintf(stderr,"malloc failed\n");
      exit(1);
    }
  }
  if( p->nLookahead==0 ){
    p->mxLookahead = lookahead;
    p->mnLookahead = lookahead;
    p->mnAction = action;
  }else{
    if( p->mxLookahead<lookahead ) p->mxLookahead = lookahead;
    if( p->mnLookahead>lookahead ){
      p->mnLookahead = lookahead;
      p->mnAction = action;
    }
  }
  p->aLookahead[p->nLookahead].lookahead = lookahead;
  p->aLookahead[p->nLookahead].action = action;
  p->nLookahead++;
}

/*
** Add the transaction set built up with prior calls to acttab_action()
** into the current action table.  Then reset the transaction set back
** to an empty set in preparation for a new round of acttab_action() calls.
**
** Return the offset into the action table of the new transaction.
*/
int acttab_insert(acttab *p){
  int i, j, k, n;
  assert( p->nLookahead>0 );

  /* Make sure we have enough space to hold the expanded action table
  ** in the worst case.  The worst case occurs if the transaction set
  ** must be appended to the current action table
  */
  n = p->mxLookahead + 1;
  if( p->nAction + n >= p->nActionAlloc ){
    int oldAlloc = p->nActionAlloc;
    p->nActionAlloc = p->nAction + n + p->nActionAlloc + 20;
    p->aAction = realloc( p->aAction,
                          sizeof(p->aAction[0])*p->nActionAlloc);
    if( p->aAction==0 ){
      fprintf(stderr,"malloc failed\n");
      exit(1);
    }
    for(i=oldAlloc; i<p->nActionAlloc; i++){
      p->aAction[i].lookahead = -1;
      p->aAction[i].action = -1;
    }
  }

  /* Scan the existing action table looking for an offset where we can
  ** insert the current transaction set.  Fall out of the loop when that
  ** offset is found.  In the worst case, we fall out of the loop when
  ** i reaches p->nAction, which means we append the new transaction set.
  **
  ** i is the index in p->aAction[] where p->mnLookahead is inserted.
  */
  for(i=0; i<p->nAction+p->mnLookahead; i++){
    if( p->aAction[i].lookahead<0 ){
      for(j=0; j<p->nLookahead; j++){
        k = p->aLookahead[j].lookahead - p->mnLookahead + i;
        if( k<0 ) break;
        if( p->aAction[k].lookahead>=0 ) break;
      }
      if( j<p->nLookahead ) continue;
      for(j=0; j<p->nAction; j++){
        if( p->aAction[j].lookahead==j+p->mnLookahead-i ) break;
      }
      if( j==p->nAction ){
        break;  /* Fits in empty slots */
      }
    }else if( p->aAction[i].lookahead==p->mnLookahead ){
      if( p->aAction[i].action!=p->mnAction ) continue;
      for(j=0; j<p->nLookahead; j++){
        k = p->aLookahead[j].lookahead - p->mnLookahead + i;
        if( k<0 || k>=p->nAction ) break;
        if( p->aLookahead[j].lookahead!=p->aAction[k].lookahead ) break;
        if( p->aLookahead[j].action!=p->aAction[k].action ) break;
      }
      if( j<p->nLookahead ) continue;
      n = 0;
      for(j=0; j<p->nAction; j++){
        if( p->aAction[j].lookahead<0 ) continue;
        if( p->aAction[j].lookahead==j+p->mnLookahead-i ) n++;
      }
      if( n==p->nLookahead ){
        break;  /* Same as a prior transaction set */
      }
    }
  }
  /* Insert transaction set at index i. */
  for(j=0; j<p->nLookahead; j++){
    k = p->aLookahead[j].lookahead - p->mnLookahead + i;
    p->aAction[k] = p->aLookahead[j];
    if( k>=p->nAction ) p->nAction = k+1;
  }
  p->nLookahead = 0;

  /* Return the offset that is added to the lookahead in order to get the
  ** index into yy_action of the action */
  return i - p->mnLookahead;
}

/********************** From the file "assert.c" ****************************/
/*
** A more efficient way of handling assertions.
*/
void myassert(file,line)
char *file;
int line;
{
  fprintf(stderr,"Assertion failed on line %d of file \"%s\"\n",line,file);
  exit(1);
}
/********************** From the file "build.c" *****************************/
/*
** Routines to construction the finite state machine for the LEMON
** parser generator.
*/

/* Find a precedence symbol of every rule in the grammar.
** 
** Those rules which have a precedence symbol coded in the input
** grammar using the "[symbol]" construct will already have the
** rp->precsym field filled.  Other rules take as their precedence
** symbol the first RHS symbol with a defined precedence.  If there
** are not RHS symbols with a defined precedence, the precedence
** symbol field is left blank.
*/
void FindRulePrecedences(xp)
struct lemon *xp;
{
  struct rule *rp;
  for(rp=xp->rule; rp; rp=rp->next){
    if( rp->precsym==0 ){
      int i;
      for(i=0; i<rp->nrhs; i++){
        if( rp->rhs[i]->prec>=0 ){
          rp->precsym = rp->rhs[i];
          break;
        }
      }
    }
  }
  return;
}

/* Find all nonterminals which will generate the empty string.
** Then go back and compute the first sets of every nonterminal.
** The first set is the set of all terminal symbols which can begin
** a string generated by that nonterminal.
*/
void FindFirstSets(lemp)
struct lemon *lemp;
{
  int i;
  struct rule *rp;
  int progress;

  for(i=0; i<lemp->nsymbol; i++){
    lemp->symbols[i]->lambda = B_FALSE;
  }
  for(i=lemp->nterminal; i<lemp->nsymbol; i++){
    lemp->symbols[i]->firstset = SetNew();
  }

  /* First compute all lambdas */
  do{
    progress = 0;
    for(rp=lemp->rule; rp; rp=rp->next){
      if( rp->lhs->lambda ) continue;
      for(i=0; i<rp->nrhs; i++){
         if( rp->rhs[i]->lambda==B_FALSE ) break;
      }
      if( i==rp->nrhs ){
        rp->lhs->lambda = B_TRUE;
        progress = 1;
      }
    }
  }while( progress );

  /* Now compute all first sets */
  do{
    struct symbol *s1, *s2;
    progress = 0;
    for(rp=lemp->rule; rp; rp=rp->next){
      s1 = rp->lhs;
      for(i=0; i<rp->nrhs; i++){
        s2 = rp->rhs[i];
        if( s2->type==TERMINAL ){
          progress += SetAdd(s1->firstset,s2->index);
          break;
        }else if( s1==s2 ){
          if( s1->lambda==B_FALSE ) break;
        }else{
          progress += SetUnion(s1->firstset,s2->firstset);
          if( s2->lambda==B_FALSE ) break;
        }
      }
    }
  }while( progress );
  return;
}

/* Compute all LR(0) states for the grammar.  Links
** are added to between some states so that the LR(1) follow sets
** can be computed later.
*/
PRIVATE struct state *getstate(/* struct lemon * */);  /* forward reference */
void FindStates(lemp)
struct lemon *lemp;
{
  struct symbol *sp;
  struct rule *rp;

  Configlist_init();

  /* Find the start symbol */
  if( lemp->start ){
    sp = Symbol_find(lemp->start);
    if( sp==0 ){
      ErrorMsg(lemp->filename,0,
"The specified start symbol \"%s\" is not "
"in a nonterminal of the grammar.  \"%s\" will be used as the start "
"symbol instead.",lemp->start,lemp->rule->lhs->name);
      lemp->errorcnt++;
      sp = lemp->rule->lhs;
    }
  }else{
    sp = lemp->rule->lhs;
  }

  /* Make sure the start symbol doesn't occur on the right-hand side of
  ** any rule.  Report an error if it does.  (YACC would generate a new
  ** start symbol in this case.) */
  for(rp=lemp->rule; rp; rp=rp->next){
    int i;
    for(i=0; i<rp->nrhs; i++){
      if( rp->rhs[i]==sp ){
        ErrorMsg(lemp->filename,0,
"The start symbol \"%s\" occurs on the "
"right-hand side of a rule. This will result in a parser which "
"does not work properly.",sp->name);
        lemp->errorcnt++;
      }
    }
  }

  /* The basis configuration set for the first state
  ** is all rules which have the start symbol as their
  ** left-hand side */
  for(rp=sp->rule; rp; rp=rp->nextlhs){
    struct config *newcfp;
    newcfp = Configlist_addbasis(rp,0);
    SetAdd(newcfp->fws,0);
  }

  /* Compute the first state.  All other states will be
  ** computed automatically during the computation of the first one.
  ** The returned pointer to the first state is not used. */
  (void)getstate(lemp);
  return;
}

/* Return a pointer to a state which is described by the configuration
** list which has been built from calls to Configlist_add.
*/
PRIVATE void buildshifts(/* struct lemon *, struct state * */); /* Forwd ref */
PRIVATE struct state *getstate(lemp)
struct lemon *lemp;
{
  struct config *cfp, *bp;
  struct state *stp;

  /* Extract the sorted basis of the new state.  The basis was constructed
  ** by prior calls to "Configlist_addbasis()". */
  Configlist_sortbasis();
  bp = Configlist_basis();

  /* Get a state with the same basis */
  stp = State_find(bp);
  if( stp ){
    /* A state with the same basis already exists!  Copy all the follow-set
    ** propagation links from the state under construction into the
    ** preexisting state, then return a pointer to the preexisting state */
    struct config *x, *y;
    for(x=bp, y=stp->bp; x && y; x=x->bp, y=y->bp){
      Plink_copy(&y->bplp,x->bplp);
      Plink_delete(x->fplp);
      x->fplp = x->bplp = 0;
    }
    cfp = Configlist_return();
    Configlist_eat(cfp);
  }else{
    /* This really is a new state.  Construct all the details */
    Configlist_closure(lemp);    /* Compute the configuration closure */
    Configlist_sort();           /* Sort the configuration closure */
    cfp = Configlist_return();   /* Get a pointer to the config list */
    stp = State_new();           /* A new state structure */
    MemoryCheck(stp);
    stp->bp = bp;                /* Remember the configuration basis */
    stp->cfp = cfp;              /* Remember the configuration closure */
    stp->index = lemp->nstate++; /* Every state gets a sequence number */
    stp->ap = 0;                 /* No actions, yet. */
    State_insert(stp,stp->bp);   /* Add to the state table */
    buildshifts(lemp,stp);       /* Recursively compute successor states */
  }
  return stp;
}

/* Construct all successor states to the given state.  A "successor"
** state is any state which can be reached by a shift action.
*/
PRIVATE void buildshifts(lemp,stp)
struct lemon *lemp;
struct state *stp;     /* The state from which successors are computed */
{
  struct config *cfp;  /* For looping thru the config closure of "stp" */
  struct config *bcfp; /* For the inner loop on config closure of "stp" */
  struct config *new;  /* */
  struct symbol *sp;   /* Symbol following the dot in configuration "cfp" */
  struct symbol *bsp;  /* Symbol following the dot in configuration "bcfp" */
  struct state *newstp; /* A pointer to a successor state */

  /* Each configuration becomes complete after it contibutes to a successor
  ** state.  Initially, all configurations are incomplete */
  for(cfp=stp->cfp; cfp; cfp=cfp->next) cfp->status = INCOMPLETE;

  /* Loop through all configurations of the state "stp" */
  for(cfp=stp->cfp; cfp; cfp=cfp->next){
    if( cfp->status==COMPLETE ) continue;    /* Already used by inner loop */
    if( cfp->dot>=cfp->rp->nrhs ) continue;  /* Can't shift this config */
    Configlist_reset();                      /* Reset the new config set */
    sp = cfp->rp->rhs[cfp->dot];             /* Symbol after the dot */

    /* For every configuration in the state "stp" which has the symbol "sp"
    ** following its dot, add the same configuration to the basis set under
    ** construction but with the dot shifted one symbol to the right. */
    for(bcfp=cfp; bcfp; bcfp=bcfp->next){
      if( bcfp->status==COMPLETE ) continue;    /* Already used */
      if( bcfp->dot>=bcfp->rp->nrhs ) continue; /* Can't shift this one */
      bsp = bcfp->rp->rhs[bcfp->dot];           /* Get symbol after dot */
      if( bsp!=sp ) continue;                   /* Must be same as for "cfp" */
      bcfp->status = COMPLETE;                  /* Mark this config as used */
      new = Configlist_addbasis(bcfp->rp,bcfp->dot+1);
      Plink_add(&new->bplp,bcfp);
    }

    /* Get a pointer to the state described by the basis configuration set
    ** constructed in the preceding loop */
    newstp = getstate(lemp);

    /* The state "newstp" is reached from the state "stp" by a shift action
    ** on the symbol "sp" */
    Action_add(&stp->ap,SHIFT,sp,(char *)newstp);
  }
}

/*
** Construct the propagation links
*/
void FindLinks(lemp)
struct lemon *lemp;
{
  int i;
  struct config *cfp, *other;
  struct state *stp;
  struct plink *plp;

  /* Housekeeping detail