/*  ******************************************************************
 *                                                                   *
 * File             : rs.c                                           *
 *                                                                   *
 * Purpose          :                                                *         
 *                                                                   *
 * Version Number   : 2.0b                                           *
 *                                                                   *
 * Revision History :                                                *
 *                                                                   *
 * Date                              Developer                       *
 * ----                              ---------                       *
 *  used to establish levels in a flat file                          *
*/

#define RS_VERSION "rs  version 2.0b  7 Oct 2009"
/*
 *  7 Oct 09                          wjs                            *
 *	Bug fix: Levelization argument parsing was being done	     *
 *	  against the optimized, reduced varlist.  Result was re-    *
 *	  jection of args referencing vars that wouldn't be needed   *
 *	Bug fix: Off-by-1 error counting # elements in list (!)	     *
 *	Bug fix: Off-by-1 error failing to account for trailing NULL *
 *	Reformat some error msgs				     *
 *	    [Begin 2.0b]					     *
 *  2 Jul 09                          wjs                            *
 *	Bug fix: when restructuring orig, no-bracket if statements,  *
 *	  blew a biggie.  Guess brackets are a good thing!	     *
 *	    [Begin 2.0a]					     *
 *								     *
 *  8 Jun 09                          wjs                            *
 *	Remove direct access to parse in favor of varnames_in_sel    *
 *	    [needs varnames_in_sel 1.0]				     *
 *	    [does NOT need parse.c - no longer exists w/that name]   *
 *	    [does NOT need parse.h]				     *
 *		THIS should have been the beginning of 2.0a	     *
 *		2.0 version below was released w/JGOFS 2.1	     *
 * 14 Feb 09                          wjs                            *
 *	Reference a couple of utils functions			     *
 *	Change entry point to parse.				     *
 *	Note that correct rs behavior requires sufficiently new	     *
 *	  versions of the parse package				     *
 * 23 Oct 08                          wjs                            *
 *	ILLEGAL_LEVEL now defined in core.h.  Pull from here	     *
 * 12 Oct 07                          wjs                            *
 *	Rework selection/projection optimization		     *
 *  5 Jul 07                          wjs                            *
 *	Change name of switch that controls optimization	     *
 * 14 Jun 07                          wjs                            *
 *	New operator to signify "levelize but do not compress"	     *
 *	  Needed "forever" since otherwise, 2 consecutive records    *
 *	  with identical, say, temperature values might be com-	     *
 *	  pressed to 1 record.					     *
 *	Similarly, need to deal with consecutive "nd"s per above     *
 *	Rework input arg processing to diagnose and reject multiple  *
 *	  specs of an input variable				     *
 *	Honor level part of input arg even if input args are not     *
 *	  spec'ed in level order				     *
 *	Some common code moved to libraries - use it from there	     *
 *      Remove array that was filled in but never used.  Checked     *
 *	  that it hadn't been used in rs 1.5 ("before our time")     *
 *	    [needs parse 2.1]					     *
 *	    [needs parse.h 2.0]					     *
 *	    [needs parse_utils 1.0]				     *
 *	    [needs path_info_routines 1.3a]			     *
 *	    [Needs utils 2.3]					     *
 *	    [Begin 2.0]						     *
 *								     *
 *  9 Nov 06                          wjs                            *
 *	Add switch for 1.9 optimization feature (along w/option to   *
 *	  establish default state of switch)			     *
 *	Change max # levels to depend on MAXLEVELS (core.h) rather   *
 *	  than on char - '0' (and a hardcoded 5!)		     *
 *	Parametrize the sizes of a couple of arrays		     *
 *  1 Nov 06                          wjs                            *
 *	Pass outer args to rs' input object			     *
 *	    [Begin 1.9]						     *
 *								     *
 *  6 Oct 06                          wjs                            *
 *	Parametrize MISSING_VALUE_REAL				     *
 *	    [Begin 1.8b]					     *
 *  5 Jul 05                          wjs                            *
 *	Get int get_integer_attribute from library		     *
 *	    [Needs utils 1.9a]					     *
 * 30 Jun 05                          wjs                            *
 *	Bug fix: v 1.7 code did not handle situation where object    *
 *	  had no comments.					     *
 *  4 Jun 05                          wjs                            *
 *	Needs jdb function defns				     *
 *	    [Needs jdbfuncdefns.h]				     *
 *	    [Begin 1.8a]					     *
 *  9 Jul 04                          wjs                            *
 *	Typo fix						     *
 *	Add iovaldouble_ entry					     *
 * 23 Apr 04                          wjs                            *
 *	Mods to "version-returning function".			     *
 *	  a) change from local to global to ensure it can't be       *
 *	     "optimized out"					     *
 *	  b) don't want "version" in any function name since such    *
 * 	     names appear when grep'ping for "version"		     *
 * 18 Feb 04                          wjs                            *
 *	Error if variable in levelizing list is not in object being  *
 *	  levelized						     *
 *	Check some potential buffer overflows			     *
 *	#include core.h.  Allows us to throw out stuff, and also     *
 *	  gets rid of some compiler warnings			     *
 *	"Full" error messages					     *
 *	    [Needs utils; path_info_routines]			     *
 *	    [Begin 1.8]						     *
 * 25 Jul 03                          wjs                            *
 *	Dynamically allocate comment buffer size		     *
 *	    [Needs path_info_routines 1.1 or later (.c & .h)]	     *
 *	    [Begin 1.7]						     *
 * 28 Jan 99                          wjs                            *
 *	Use path_info_routines to control PATH_INFO; change          *
 *	  PATH_INFO just before needed; change it back when done     *
 *	Add version string; reorder comments	                     *
 *	    [Needs path_info_routines 1.1 or later (.c & .h)]	     *
 *	    [Begin 1.6]						     *
 * June 1997 grf                                                     *
 *       changed sub names to in_xxxx, fixes for v 1.5               *
 * May 1997                         Glenn Flierl                     *
 *       changed error calls to use error_ in outer.c                *
 * May 21, 1996                       clh                            *
 *  iovalreal change to allow alpha vars to begin with digits        *
 * March 1996                         clh                            *
 *  use minbufsize.h to determine size of buffers                    *
 * January 1996                       clh                            *
 *  fixed rsw to respect the re-ordering of variables                *
 * July 1995                          clh                            *
 *  add subroutine to return column widths - iowidth_()              *
 *********************************************************************
*/
#include INNEROPTIONS
#include "jdbfuncdefns.h"
#include "path_info_routines.h"

double strtod();

/* #define DEBUG */

  /*  Next 2 are the operators rs works with.  LEVELIZE means "put	*/
  /*  named variable at the level following the operator".  CONDENSE,	*/
  /*  the traditional operator, means "LEVELIZE, and, in addition, if	*/
  /*  consecutive records at this level are identical, get rid of 2nd"	*/
  /*    Be careful w/choice of characters, since they are the keys for	*/
  /*  rs to determine if a method argument is for inner or outer.  In	*/
  /*  particular, watch out for outer's selection operators.  See utils	*/
  /*  keyword_pairs_to_wjstbl/set_option_string (used by decimator, for	*/
  /*  one) for an alternate parsing approach if push comes to shove	*/
#define CONDENSE ':'
#define LEVELIZE '#'
  /*  Character that means "all un-named variables" in levelization	*/
  /*  parameters							*/
#define UNNAMED_VARS '*'

  /*  See SELPROJ_OPTIMIZE doc released in outer's distribution		*/
#ifndef DEFAULT_SELPROJ_OPTIMIZE
#define DEFAULT_SELPROJ_OPTIMIZE TRUE
#endif
#if DEFAULT_SELPROJ_OPTIMIZE
static char *optimize_state = "Compiled with DEFAULT_SELPROJ_OPTIMIZE on";
#else
static char *optimize_state = "Compiled with DEFAULT_SELPROJ_OPTIMIZE off";
#endif
char *SELPROJ_OPTIMIZE_env_var_name = "SELPROJ_OPTIMIZE";
Logical do_optimize;

#ifndef DEFAULT_RS_OPTIMIZE
#define DEFAULT_RS_OPTIMIZE TRUE
#endif
#if DEFAULT_RS_OPTIMIZE
static char *rs_optimize_state = "Compiled with DEFAULT_RS_OPTIMIZE on";
#else
static char *rs_optimize_state = "Compiled with DEFAULT_RS_OPTIMIZE off";
#endif
char *RS_OPTIMIZE_env_var_name = "RS_OPTIMIZE";
Logical rs_optimize;

int nlevels,nnames;
  /*  Pre v 1.9, dimension in next line was 6.  Taking guess at new size */
int firstvar[MAXLEVELS+1];
char names[NVAR][VARNAMESIZE];
int fldwidths[NVAR];
int namesize=VARNAMESIZE;
char values[NVAR][DATUMSIZE];
char comp[NVAR][DATUMSIZE];
int valuesize=DATUMSIZE;
int handle;
int minlevelread;
char *comments;

  /* levelization info; derived from level:varname args to program	*/
struct varlevel {
  int level;
  char *var;
} *ss;
int num_ss;

int newlev[NVAR];
int newpntr[NVAR];
char comp_type[NVAR];
int maxoutlev;
  /* removed newlpntr[MAXLEVELS+1];  - see Jun 07 comments		*/

void error_();			/* in outer				*/
char *un_trigram();		/* in outer_utils.c(in jgofs.a library)	*/
  /*  Functions below in utils.c (in jgofs.a library)			*/
Logical add_id_to_err();
char *buildstring();
void errn();
void free_lengthened_str();
int get_integer_attribute();
Logical get_logical_from_string();
Logical is_a_varname();
char *lengthen_str_and_free();
char *strdupl();
Logical string_in_list();
char **string_sets_intersection();
char **string_sets_union();
int varnames_in_sel();

/************************************************************************/

char *rs_return_vers()
/*  Dummy routine.  Exists only to force .h file version string into	*/
/*  this module.  Note string must not be global or we'll have con-	*/
/*  flicts if another routine similarly includes the version string	*/
{
  static char version[] = \
RS_VERSION"/"FULL_PATH_INFO_ROUTINESH_VERSION"/"FULL_JDBFUNCDEFNSH_VERSION;
  return version;
}

void err(s,t)
char *s,*t;
{
  char *ss,*tt;
  add_id_to_err(&ss,&tt,s,t,RS_VERSION);
  error_(ss,tt);
  return;		/*  Not that it should ever get here...		*/
}

int iovarlevel_(vn)
int *vn;
{
  return newlev[newpntr[*vn]];
}

int ioattrout_(vn,str)
int *vn;
char *str;
{
   int j;

   j = newpntr[*vn];

   if ((jdbattributes_(&handle,&j,str)) == 0)
       return 0;
   else {
       if (strncmp(str,"width=",6) == 0)
           fldwidths[j] = get_integer_attribute(str);
       return 1;
   }
}

void iovaldouble_(vn,df)
int *vn;
double *df;
/*	See iovalreal_							*/
{

  int i;
  char *end_char_ptr;

  i=newpntr[*vn];
  if(i<0) {
    *df=MISSING_VALUE_REAL;
    return;
  }


  *df=strtod(values[i],&end_char_ptr);
  if (*end_char_ptr != '\0') *df= MISSING_VALUE_REAL;

return;
}

void iovalreal_(vn,f)
int *vn;
float *f;
/*	"Return real value (f) for variable indexed by vn. -9999"	*/
/*	"for strings"							*/
{
  double df;
  iovaldouble_(*vn,&df);
  *f = df;
  return;
 }

void iovalstr_(vn,tmp)
int *vn;
char *tmp;
{
  char *s;

  s=values[newpntr[*vn]];
  s=s+strspn(s," ");
  strcpy(tmp,s);
}

void ioname_(vn,s)
int *vn;
char *s;
{
  strcpy(s,names[newpntr[*vn]]);
}

int iocommout_(str)
char *str;
{
char *at;

if (comments == NULL) return 0;

if (comments[0]){
  at=strchr(comments,'\n');
  if(at){
    *at = '\0';
    strcpy(str,comments);
    strcpy(comments,at+1);
  } else {
    strcpy(str,comments);
    comments[0]='\0';
  };
  return 1;
} else return 0;
}

int iowidth_(vn)
int *vn;
{
   return  fldwidths[newpntr[*vn]];
}

int ioreadrec_(level)
int *level;
{
  int clevel,i;

  if (*level == minlevelread) {
    minlevelread = *level+1;
    return 1;
  } else if (*level > minlevelread) {
    return 0;
  }

  while (TRUE) {
    clevel = jdbreada_(&handle,values,&valuesize);
    					/* printf("clevel %d\n",clevel); */
    if (clevel < 0)  {
      minlevelread= ILLEGAL_LEVEL;
      return 0;
    }
    
    for  (i=firstvar[clevel]; i<nnames; i++)  {
      if (newlev[i] < minlevelread) { 
	if (comp_type[i] == CONDENSE) {
	  if (strcmp(values[i],comp[i]) != 0) { minlevelread = newlev[i]; }
	} else if (comp_type[i] == LEVELIZE) {
	  minlevelread = newlev[i];
	} else {
	  char s[2];
	  s[0] = comp_type[i];
	  s[1] = '\0';
	  err ("Internal error - illegal comp_type.  Bad value = ",s);
        }
      }
      COPY_INTO_FIXED_LEN_BUFFER(comp[i],values[i],
						"copying into comp buffer");
    }
    
    if (minlevelread < *level) {
      return 0;
    } else if  (minlevelread == *level)  {
      minlevelread++;
      return 1;
    }
  }
}

void ioclose_()
{
  jdbclose_(&handle);
}

Logical is_rs_levelization_arg(str)
char *str;
  /*  rs levelization args are of form level:varname (or level#varname)	*/
{
  char *ptr;
  Logical is_rs_operator;
  if ((*str == '+') || (*str == '-') || isspace(*str)) return FALSE;
  strtol(str,&ptr,10);
  is_rs_operator = ((*ptr == CONDENSE) || (*ptr == LEVELIZE));
    /*  Last check in next line is to ensure that there is a varname	*/
    /*  after the :.  If not, it's probably an error, but we will just	*/
    /*  punt the arg back to outer for processing.  Who knows how/where	*/
    /*  the abnormality will show up?					*/
  return (  is_rs_operator  &&  (ptr != str)  &&  (*(ptr+1) != '\0')  );
}

char *optimize_objspec(in_objspec,
			u_proj,num_u_proj,
			l_proj,num_l_proj,
			sel_list,num_sels,
			vars_in_sels,num_vars_in_sels)
char *in_objspec;
char **u_proj,**l_proj,**sel_list,**vars_in_sels;
int num_u_proj,num_l_proj,num_sels,num_vars_in_sels;
  /*    Returns NULL in memory allocation failure scenarios and the	*/
  /*  far more common scenario where no optimization is being performed	*/
  /*  Idea is to let calling program use the original object spec in	*/
  /*  either case, but since some memory allocation failures auto-	*/
  /*  matically generate error msgs and die, so much for consistency	*/
  /*    Non-NULL returns can be freed via the free_lengthened_str, etc	*/
  /*  calls - see utils							*/
{
  Logical any_u_proj,do_level_optimization;
  char *objspec;
  char **proj_list,**free_it;
  int i;

  any_u_proj = (num_u_proj != 0);
  do_level_optimization = (num_l_proj != 0);
  if ( ! any_u_proj  &&  ! do_level_optimization) return NULL;

  strdupl(&objspec,in_objspec,"...duplicating object spec");
  i = strlen(objspec) - 1;
  if (objspec[i] == ')') {
    objspec[i] = ',';
  } else { 
    objspec = lengthen_str_and_free(objspec,"(","",0,"adding ( to objspec");
  }

    /*  There are 2 possible projection lists we might want to add to	*/
    /*  the object spec to reduce the amount of data being transferred	*/
    /*  The first is the list of projections that the user put in the	*/
    /*  objspec.  The second is a level-based list of projections used	*/
    /*  only in html display situations.				*/
    /*    If either of these projection lists is used, we also want	*/
    /*  to project in all the variables used in user selections.	*/
    /*  (Wrestle with this if you want (sample: obj(a,b,b<c); do we	*/
    /*  need c? Answer: yes.  2 falls out of 3))			*/
    /*  Note we do NOT want to add a projection list due to selections	*/
    /*  if we have no other projections, but we don't get here in that	*/
    /*  case.								*/
    /*    If both projection lists are being used, project in the	*/
    /*  intersection of the lists					*/
  if (any_u_proj && do_level_optimization) {
    proj_list = string_sets_intersection(u_proj,num_u_proj,l_proj,num_l_proj);
    if (proj_list == NULL) return NULL;
    free_it = proj_list;
  } else {
    proj_list = (any_u_proj) ? u_proj : l_proj;
    free_it = NULL;
  }
    /*  Add projections for variables involved in selections to the	*/
    /*  list of variables specifically projected in			*/
  i = 0;
  while (proj_list[i] != NULL) i++;	/* count # entries in proj_list	*/
  proj_list = string_sets_union(proj_list,i,vars_in_sels,num_vars_in_sels);
  if (proj_list == NULL) return NULL;
  if (free_it != NULL) free(free_it);

  i = 0;
  while (proj_list[i] != NULL)
    objspec = lengthen_str_and_free
		(objspec,proj_list[i++],",",0,"adding projection to objspec");

  for (i=0; i<num_sels; i++)
      /*  trigramming is NOT needed (empirically determined).		*/
      /*  Now I wonder if the whole trigram business is due to		*/
      /*  some interface that involves a shell ... httpd -> cgi?	*/
    objspec = lengthen_str_and_free
		(objspec,sel_list[i],",",0,"adding selection to objspec");

  i = strlen(objspec) - 1;
  objspec[i] = ')';   
  return objspec;
}

int ioopen_(s,nparams,ntotal)
char *s[];
int *nparams;
int *ntotal;
{
  char tmp[INBUFSIZE],*sp,*objspec,*sel_proj_list_start;
  char *varname_ptr;
  char default_comp_type;
  int i,j,k,m,maxclev;
  int position_in_comments,size_comments;
  int default_level,html_max_level;
  Logical default_specified,any_levelization_args;
  Logical var_at_level[MAXLEVELS+1];

    /* u_proj = user projections; specified as args to program		*/
    /* l_proj = implied projections due to level restrictions; eg,	*/
    /*	  .html0 implies only level 0 vars.  Populated from varlevel	*/
    /*	  struct.  Only used for .hmtl output.				*/
    /* sel_list = user selections; specified as args to program		*/
    /* vars_in_sels = list of variables used in selections		*/
  char **u_proj,**l_proj,**sel_list,**vars_in_sels;
  int num_u_proj,num_l_proj,num_sels,num_vars_in_sels;

    /*  PATH_INFO function and strings.					*/
  int get_level();
  char *get_protocol();
  int new_and_old_path_infos();
      /* "PATH_INFO=" + getenv("PATH_INFO").  Must be static		*/
      /*  since it "survives" in process table after ioopen_ exits)	*/
  static char *PATH_INFO_orig_putenv;
      /* "PATH_INFO=" + getenv("PATH_INFO") w/ protocol = "jgof"	*/
  char *PATH_INFO_jgof_putenv;

  if (  (sp = getenv(SELPROJ_OPTIMIZE_env_var_name)) == NULL  ) {
    do_optimize = DEFAULT_SELPROJ_OPTIMIZE;
  } else {
    do_optimize = get_logical_from_string
	(sp,"SELPROJ_OPTIMIZE env var",SELPROJ_OPTIMIZE_env_var_name);
  }

    /*  As of rs 2.0, NOT using rs_optimize.  rs_optimize is a form of	*/
    /*  sel/proj optimization that may be OK even if full sel/proj	*/
    /*  optimization is not.  It allows pass-through projection		*/
    /*  optimization for LEVELIZE vars while not allowing it for	*/
    /*  CONDENSE vars.  Not implemented because it requires pre-jdbopen	*/
    /*  processing of the levelization parameters, which is not hap-	*/
    /*  pening in 2.0.  Let's see how many times we need to turn off	*/
    /*  full optimization first... Once to workaround 20b issue,	*/
    /*  pending its fix (Oct 09) ...					*/
  if (  (sp = getenv(RS_OPTIMIZE_env_var_name)) == NULL  ) {
    rs_optimize = DEFAULT_RS_OPTIMIZE;
  } else {
    rs_optimize = get_logical_from_string
	(sp,"RS_OPTIMIZE env var",RS_OPTIMIZE_env_var_name);
  }

    /*  If in html mode, save maximum level for optimization purposes	*/
    /*  If problems w/PATH_INFO, just bag optimization and let whatever	*/
    /*  used to deal w/this problem continue to deal w/it		*/
  html_max_level = ILLEGAL_LEVEL;
  if (do_optimize) {
    sp = get_protocol(NULL);
    if (sp != NULL) {
      if (   (strcmp(sp,"html") == 0) || (strcmp(sp,"brev") == 0)   )  { 
	i = get_level(NULL);
	if ( (i != LEVEL_NOT_SPECIFIED) && (i != LEVEL_ERROR)   )
							html_max_level = i;
      }
      free (sp);
    }
  }

    /*  Must be sure that methods down the line produce JGOFS	*/
    /*  protocol (or whatever jdbopen requires), and not html,	*/
    /*  flat, or whatever.  Do this by replacing proto string	*/
    /*  with "jgof" (WJS decision at this point... jgof has no	*/
    /*  meaning to system).					*/
    /*    Make 2 strings for putenv.  1 sets up existing 	*/
    /*    PATH_INFO; the other sets up PATH_INFO with "jgof"	*/
    /*    protocol						*/  
    /*  Make sure outers "down the line" don't produce flat, html, etc	*/
  if (new_and_old_path_infos(&PATH_INFO_jgof_putenv,&PATH_INFO_orig_putenv)
	!= PATH_INFO_NEW_AND_OLD_OK) err 
("problem processing PATH_INFO env var.  Memory issues or bad fmt. env var=",
						  getenv(PATH_INFO_ENV_VAR)  );

  if (PATH_INFO_jgof_putenv != NULL) 
    if (  (i = putenv(PATH_INFO_jgof_putenv)) != 0  )
      err ("putenv failure for PATH_INFO",strerror(i));

    /*  Process command line args.					*/
    /*    "Rules" are that inner copies its args and replaces them w/	*/
    /*  null strings so outer can process the rest			*/
    /*  1st arg to rs is object specifier.  Other args to rs are of the	*/
    /*  form level:varlist (or level#varlist), and can (hopefully!) be	*/
    /*  intermixed w/outer args						*/
    /*    In addition to processing the inner args, we process the	*/
    /*  outer args, anticipating optimization is in effect.  The pro-	*/
    /*  cessing consists of making a list of variables known to be	*/
    /*  needed.  These variables are projections + the variables used	*/
    /*  in selections							*/
  if (s[0] == NULL) err ("No input object spec","");
  if (s[0][0] == '\0') err ("Empty input object spec","");
  objspec = NULL;

  i = (*nparams) * sizeof(struct varlevel);
  ss = (struct varlevel *)malloc(i);
  if (ss == NULL) errn ("Could not get memory for varlevel structs. nbytes=",i);
  num_ss = 0;
  any_levelization_args = FALSE;
  default_level = ILLEGAL_LEVEL;
  maxoutlev = ILLEGAL_LEVEL;
  default_comp_type = '\0';

    /*  Since u_proj is NULL-terminated, need possible # proj +1.	*/
    /*  Since argv[0] is obj spec, nparams IS 1 bigger			*/
  i = (*nparams) * sizeof(char *);
  u_proj = (char **)malloc(i);
  if (u_proj == NULL) errn ("Could not get memory for u_proj list. nbytes=",i);
  num_u_proj = 0;

  sel_list = (char **)malloc(i);
  if (sel_list == NULL) errn 
			("Could not get memory for selection list. nbytes=",i);
  num_sels = 0;

  num_vars_in_sels = 0;
  vars_in_sels = NULL;

  for (i=1; i<*nparams; i++) {
    if (is_rs_levelization_arg(s[i])) {
      j = (int) strtol(s[i],&sp,10);
      if (j >= MAXLEVELS) err ("level too big in levelization parameter ",s[i]);
      if (j < 0) err("level is negative in levelization parameter ",s[i]);
      if ((j > maxoutlev) || (maxoutlev == ILLEGAL_LEVEL)) maxoutlev = j;
      varname_ptr = sp+1;
      if (*varname_ptr == UNNAMED_VARS) {
        if (default_level == ILLEGAL_LEVEL) {
	  default_level = j;
	  default_comp_type = *sp;
	} else {
	  err ("Default (*) set more than once","");
        }
      } else {
	for (k=0; k<num_ss; k++)
	  if (strcmp(varname_ptr,ss[k].var) == 0)
err ("Variable spec'd in more than one levelization parameter.  Var is",
								varname_ptr);
	ss[num_ss].level = j;
	ss[num_ss++].var = varname_ptr;
      }
      any_levelization_args = TRUE;
      s[i][0] = '\0';	/* This wipes a piece of the level, but		*/
			/* should leave varname_ptr's string alone	*/
    } else {
      if ( is_a_varname(s[i]) )  {
	u_proj[num_u_proj++] = s[i];
      } else {
	sp = un_trigram(s[i],DEFAULT_TRIGRAM_CHAR,NULL);
	num_vars_in_sels += varnames_in_sel(&vars_in_sels,sp,NULL,NULL,NULL);
	sel_list[num_sels++] = sp;
      }
    }
  }
  if ( ! any_levelization_args) 
			err ("No levelization arguments passed to rs","");
  u_proj[num_u_proj] = NULL;
  sel_list[num_sels] = NULL;

  if (do_optimize) {
      /*    If default_level is illegal, then it was not specified; ie,	*/
      /*  all variables are named.  If this is true, we can proceed	*/
      /*  with level-based projection optimization.			*/
      /*    If default_level is legal, then we must turn off		*/
      /*  optimization if the default level is requested because we	*/
      /*  don't know what vars are on default_level to be projected in.	*/
      /*    If html_max_level is illegal, we aren't doing optimization	*/
      /*  anyway.							*/
      /*    ACCORDINGLY, only issue is if both are legal		*/
    if ((default_level != ILLEGAL_LEVEL) && (html_max_level != ILLEGAL_LEVEL))
      if (html_max_level >= default_level) html_max_level = ILLEGAL_LEVEL;

    num_l_proj = 0;
    i = (num_ss + 1) * sizeof (char *);		/* +1 for trailing NULL	*/
    l_proj = (char **)malloc(i);
    if (l_proj == NULL) errn ("Could not get memory for l_proj list. nbytes=",i);
    if (html_max_level != ILLEGAL_LEVEL) {
      for (i=0; i<num_ss; i++)
	if (ss[i].level <= html_max_level) l_proj[num_l_proj++] = ss[i].var;
    }
    l_proj[num_l_proj] = NULL;

    objspec = optimize_objspec(s[0],
				u_proj,num_u_proj,
				l_proj,num_l_proj,
				sel_list,num_sels,
				vars_in_sels,num_vars_in_sels);
  }

  if (objspec == NULL) objspec = s[0];
  nnames = -NVAR;
  maxclev = jdbopen_(&handle,objspec,names,&namesize,&nnames);
    /*  next error message will probably be in an environment where	*/
    /*  jdbopen has already printed one of its error messages		*/
  if (maxclev < 0) err("Trouble with object specifier ",objspec);
  if (objspec != s[0]) free_lengthened_str(objspec);
  s[0][0] = '\0';

    /*  Restore original PATH_INFO if we changed it		*/
  if (PATH_INFO_jgof_putenv != NULL) {
    if (  (i = putenv(PATH_INFO_orig_putenv)) != 0  )
       err ("putenv failure restoring PATH_INFO",strerror(i));
    free(PATH_INFO_jgof_putenv);
  }

  comments = NULL;
  size_comments = 0;
  position_in_comments = 0;
    /*  Note that jdbcomments could overflow tmp - there is no consis-	*/
    /*  tency check between the buffer in jdbcomments and the parameter	*/
    /*  we use to size tmp						*/
    /*  Note we could jigger this to have jdbcomments_ put stuff di-	*/
    /*  rectly in comments, too						*/
  while (jdbcomments_(&handle,tmp)) {
    i = strlen(tmp);
      /*  +2 because we need a \n per tmp as well as a final \0		*/
    if (size_comments < position_in_comments + i + 2) {
	/*  COMMENTSIZE in next line could be "anything"		*/
      size_comments += (i+2 > COMMENTSIZE) ? i+2 : COMMENTSIZE;
      comments = (char *)realloc(comments,size_comments);
      if (comments == NULL)
	err("Could not allocate space for comments buffer","");
    }
    strncpy (comments + position_in_comments, tmp, i);
    position_in_comments += i;
    *(comments + position_in_comments++) = '\n';
    *(comments + position_in_comments) = '\0';
  }
  
  j = 0;
  firstvar[j] = 0;
  for  (i=0; i<nnames; i++)  {
    newlev[i] = ILLEGAL_LEVEL;
    if (iovarlevel_(&handle,&i) > j) firstvar[++j] = i;
  }

  m = 0;
  for  (k=0; k<num_ss; k++)  {
    varname_ptr = ss[k].var;
    for  (j=0; j<nnames; j++)  {
      if (  strcmp(names[j],varname_ptr) == 0  )   break;
    }
    if (j < nnames)  {
      newlev[j] = ss[k].level;
      comp_type[j] = *(varname_ptr - 1);
      newpntr[m++] = j;
    } else if ( (html_max_level == ILLEGAL_LEVEL) 
				||
		(ss[i].level <= html_max_level)    ) {
	/*  Levelization parameter specifies variable that is not in	*/
	/*  input object.  Error unless said variable was actually IN	*/
	/*  input object but was projected out.				*/
	/*	The "unless" clause is in 2 pieces.			*/
	/*	The for loop, below, takes care of explicit projection.	*/
	/*	The if, above, addresses the implicit			*/
	/*  projection via level by ignoring lev params that aren't	*/
	/*  "in" the level(s) we are about to display.  ILLEGAL_LEVEL	*/
	/*  means no optimization, means check all lev params		*/
	/*	Note that if we are optimizing, bad vars in lev params	*/
	/*  will appear in the proj list in the jdbopen above, causing	*/
	/*  THAT to fail, and we'll never get here (except at bottom	*/
	/*  level (maybe))						*/
      for (i=0; i<num_u_proj; i++)
	if (  strcmp(u_proj[i],varname_ptr) == 0  )   break;
      if (i == num_u_proj) err
        ("Variable in levelization parameter not in input object.  Var is",
								   varname_ptr);
    }
  }
  
  if (default_level == ILLEGAL_LEVEL) {
    default_specified = FALSE;
    default_level = maxoutlev;	/* Don't think default_level is ever	*/
				/* used if not specified		*/
    default_comp_type = CONDENSE;
  } else {
    default_specified = TRUE;
    if (default_level > maxoutlev) maxoutlev = default_level;
  }
  i = 0;
  for (j=0; j<nnames; j++) {
    if (newlev[j] == ILLEGAL_LEVEL)  {
      if ( ! default_specified) err ("No output level for variable ",names[j]);
      i++;
      newlev[j] = default_level;
      comp_type[j] = default_comp_type;
      newpntr[m++] = j;
    }
    var_at_level[newlev[j]] = TRUE;
  }

  /*  Both of the commented out tests below are correct, but not clear	*/
  /*  that detected situations are errors when projections are involved	*/
  /*  outer 3.0 should handle the "no variables on level" anyway...	*/
/*
  if (  (i == 0) && default_specified  )
		err ("No variables affected by * levelization parameter","");
  for (i=0;  i <= maxoutlev;  i++)  
    if ( ! var_at_level[i]) errn ("No variables end up on level ",i);
*/

  *ntotal=nnames;

  minlevelread = maxoutlev+1;
  return maxoutlev;
}