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(struct lemon *xp)
{
  struct rule *rp;
  for(rp=xp->rule; rp; rp=rp->next){
    if( rp->precsym==0 ){
      int i, j;
      for(i=0; i<rp->nrhs && rp->precsym==0; i++){
        struct symbol *sp = rp->rhs[i];
        if( sp->type==MULTITERMINAL ){
          for(j=0; j<sp->nsubsym; j++){
            if( sp->subsym[j]->prec>=0 ){
              rp->precsym = sp->subsym[j];
              break;
            }
          }
        }else if( sp->prec>=0 ){
          rp->precsym = rp->rhs[i];
        }
      }
    }
  }
  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(struct lemon *lemp)
{
  int i, j;
  struct rule *rp;
  int progress;

  for(i=0; i<lemp->nsymbol; i++){
    lemp->symbols[i]->lambda = LEMON_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++){
        struct symbol *sp = rp->rhs[i];
        assert( sp->type==NONTERMINAL || sp->lambda==LEMON_FALSE );
        if( sp->lambda==LEMON_FALSE ) break;
      }
      if( i==rp->nrhs ){
        rp->lhs->lambda = LEMON_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( s2->type==MULTITERMINAL ){
          for(j=0; j<s2->nsubsym; j++){
            progress += SetAdd(s1->firstset,s2->subsym[j]->index);
          }
          break;
        }else if( s1==s2 ){
          if( s1->lambda==LEMON_FALSE ) break;
        }else{
          progress += SetUnion(s1->firstset,s2->firstset);
          if( s2->lambda==LEMON_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(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->startRule->lhs->name);
      lemp->errorcnt++;
      sp = lemp->startRule->lhs;
    }
  }else if( lemp->startRule ){
    sp = lemp->startRule->lhs;
  }else{
    ErrorMsg(lemp->filename,0,"Internal error - no start rule\n");
    exit(1);
  }

  /* 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 ){   /* FIX ME:  Deal with multiterminals */
        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;
    rp->lhsStart = 1;
    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(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->statenum = 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;
}

/*
** Return true if two symbols are the same.
*/
int same_symbol(struct symbol *a, struct symbol *b)
{
  int i;
  if( a==b ) return 1;
  if( a->type!=MULTITERMINAL ) return 0;
  if( b->type!=MULTITERMINAL ) return 0;
  if( a->nsubsym!=b->nsubsym ) return 0;
  for(i=0; i<a->nsubsym; i++){
    if( a->subsym[i]!=b->subsym[i] ) return 0;
  }
  return 1;
}

/* 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(struct lemon *lemp, struct state *stp)
{
  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 *newcfg;  /* */
  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 contributes 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( !same_symbol(bsp,sp) ) continue;      /* Must be same as for "cfp" */
      bcfp->status = COMPLETE;                  /* Mark this config as used */
      newcfg = Configlist_addbasis(bcfp->rp,bcfp->dot+1);
      Plink_add(&newcfg->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" */
    if( sp->type==MULTITERMINAL ){
      int i;
      for(i=0; i<sp->nsubsym; i++){
        Action_add(&stp->ap,SHIFT,sp->subsym[i],(char*)newstp);
      }
    }else{
      Action_add(&stp->ap,SHIFT,sp,(char *)newstp);
    }
  }
}

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

  /* Housekeeping detail:
  ** Add to every propagate link a pointer back to the state to
  ** which the link is attached. */
  for(i=0; i<lemp->nstate; i++){
    stp = lemp->sorted[i];
    for(cfp=stp?stp->cfp:0; cfp; cfp=cfp->next){
      cfp->stp = stp;
    }
  }

  /* Convert all backlinks into forward links.  Only the forward
  ** links are used in the follow-set computation. */
  for(i=0; i<lemp->nstate; i++){
    stp = lemp->sorted[i];
    for(cfp=stp?stp->cfp:0; cfp; cfp=cfp->next){
      for(plp=cfp->bplp; plp; plp=plp->next){
        other = plp->cfp;
        Plink_add(&other->fplp,cfp);
      }
    }
  }
}

/* Compute all followsets.
**
** A followset is the set of all symbols which can come immediately
** after a configuration.
*/
void FindFollowSets(struct lemon *lemp)
{
  int i;
  struct config *cfp;
  struct plink *plp;
  int progress;
  int change;

  for(i=0; i<lemp->nstate; i++){
    assert( lemp->sorted[i]!=0 );
    for(cfp=lemp->sorted[i]->cfp; cfp; cfp=cfp->next){
      cfp->status = INCOMPLETE;
    }
  }

  do{
    progress = 0;
    for(i=0; i<lemp->nstate; i++){
      assert( lemp->sorted[i]!=0 );
      for(cfp=lemp->sorted[i]->cfp; cfp; cfp=cfp->next){
        if( cfp->status==COMPLETE ) continue;
        for(plp=cfp->fplp; plp; plp=plp->next){
          change = SetUnion(plp->cfp->fws,cfp->fws);
          if( change ){
            plp->cfp->status = INCOMPLETE;
            progress = 1;
          }
        }
        cfp->status = COMPLETE;
      }
    }
  }while( progress );
}

static int resolve_conflict(struct action *,struct action *);

/* Compute the reduce actions, and resolve conflicts.
*/
void FindActions(struct lemon *lemp)
{
  int i,j;
  struct config *cfp;
  struct state *stp;
  struct symbol *sp;
  struct rule *rp;

  /* Add all of the reduce actions
  ** A reduce action is added for each element of the followset of
  ** a configuration which has its dot at the extreme right.
  */
  for(i=0; i<lemp->nstate; i++){   /* Loop over all states */
    stp = lemp->sorted[i];
    for(cfp=stp->cfp; cfp; cfp=cfp->next){  /* Loop over all configurations */
      if( cfp->rp->nrhs==cfp->dot ){        /* Is dot at extreme right? */
        for(j=0; j<lemp->nterminal; j++){
          if( SetFind(cfp->fws,j) ){
            /* Add a reduce action to the state "stp" which will reduce by the
            ** rule "cfp->rp" if the lookahead symbol is "lemp->symbols[j]" */
            Action_add(&stp->ap,REDUCE,lemp->symbols[j],(char *)cfp->rp);
          }
        }
      }
    }
  }

  /* Add the accepting token */
  if( lemp->start ){
    sp = Symbol_find(lemp->start);
    if( sp==0 ){
      if( lemp->startRule==0 ){
        fprintf(stderr, "internal error on source line %d: no start rule\n",
                __LINE__);
        exit(1);
      }
      sp = lemp->startRule->lhs;
    }
  }else{
    sp = lemp->startRule->lhs;
  }
  /* Add to the first state (which is always the starting state of the
  ** finite state machine) an action to ACCEPT if the lookahead is the
  ** start nonterminal.  */
  Action_add(&lemp->sorted[0]->ap,ACCEPT,sp,0);

  /* Resolve conflicts */
  for(i=0; i<lemp->nstate; i++){
    struct action *ap, *nap;
    stp = lemp->sorted[i];
    /* assert( stp->ap ); */
    stp->ap = Action_sort(stp->ap);
    for(ap=stp->ap; ap && ap->next; ap=ap->next){
      for(nap=ap->next; nap && nap->sp==ap->sp; nap=nap->next){
         /* The two actions "ap" and "nap" have the same lookahead.
         ** Figure out which one should be used */
         lemp->nconflict += resolve_conflict(ap,nap);
      }
    }
  }

  /* Report an error for each rule that can never be reduced. */
  for(rp=lemp->rule; rp; rp=rp->next) rp->canReduce = LEMON_FALSE;
  for(i=0; i<lemp->nstate; i++){
    struct action *ap;
    for(ap=lemp->sorted[i]->ap; ap; ap=ap->next){
      if( ap->type==REDUCE ) ap->x.rp->canReduce = LEMON_TRUE;
    }
  }
  for(rp=lemp->rule; rp; rp=rp->next){
    if( rp->canReduce ) continue;
    ErrorMsg(lemp->filename,rp->ruleline,"This rule can not be reduced.\n");
    lemp->errorcnt++;
  }
}

/* Resolve a conflict between the two given actions.  If the
** conflict can't be resolved, return non-zero.
**
** NO LONGER TRUE:
**   To resolve a conflict, first look to see if either action
**   is on an error rule.  In that case, take the action which
**   is not associated with the error rule.  If neither or both
**   actions are associated with an error rule, then try to
**   use precedence to resolve the conflict.
**
** If either action is a SHIFT, then it must be apx.  This
** function won't work if apx->type==REDUCE and apy->type==SHIFT.
*/
static int resolve_conflict(
  struct action *apx,
  struct action *apy
){
  struct symbol *spx, *spy;
  int errcnt = 0;
  assert( apx->sp==apy->sp );  /* Otherwise there would be no conflict */
  if( apx->type==SHIFT && apy->type==SHIFT ){
    apy->type = SSCONFLICT;
    errcnt++;
  }
  if( apx->type==SHIFT && apy->type==REDUCE ){
    spx = apx->sp;
    spy = apy->x.rp->precsym;
    if( spy==0 || spx->prec<0 || spy->prec<0 ){
      /* Not enough precedence information. */
      apy->type = SRCONFLICT;
      errcnt++;
    }else if( spx->prec>spy->prec ){    /* higher precedence wins */
      apy->type = RD_RESOLVED;
    }else if( spx->prec<spy->prec ){
      apx->type = SH_RESOLVED;
    }else if( spx->prec==spy->prec && spx->assoc==RIGHT ){ /* Use operator */
      apy->type = RD_RESOLVED;                             /* associativity */
    }else if( spx->prec==spy->prec && spx->assoc==LEFT ){  /* to break tie */
      apx->type = SH_RESOLVED;
    }else{
      assert( spx->prec==spy->prec && spx->assoc==NONE );
      apx->type = ERROR;
    }
  }else if( apx->type==REDUCE && apy->type==REDUCE ){
    spx = apx->x.rp->precsym;
    spy = apy->x.rp->precsym;
    if( spx==0 || spy==0 || spx->prec<0 ||
    spy->prec<0 || spx->prec==spy->prec ){
      apy->type = RRCONFLICT;
      errcnt++;
    }else if( spx->prec>spy->prec ){
      apy->type = RD_RESOLVED;
    }else if( spx->prec<spy->prec ){
      apx->type = RD_RESOLVED;
    }
  }else{
    assert(
      apx->type==SH_RESOLVED ||
      apx->type==RD_RESOLVED ||
      apx->type==SSCONFLICT ||
      apx->type==SRCONFLICT ||
      apx->type==RRCONFLICT ||
      apy->type==SH_RESOLVED ||
      apy->type==RD_RESOLVED ||
      apy->type==SSCONFLICT ||
      apy->type==SRCONFLICT ||
      apy->type==RRCONFLICT
    );
    /* The REDUCE/SHIFT case cannot happen because SHIFTs come before
    ** REDUCEs on the list.  If we reach this point it must be because
    ** the parser conflict had already been resolved. */
  }
  return errcnt;
}