/*	jgofs_selection_parser.c					*/
/*		from parse v 2.2					*/
/*    Separated test code from parse code to avoid precision issues	*/
/*  Turns out it was better to REALLY avoid the precision issues, and	*/
/*  things could have stayed as one module, but by that time other	*/
/*  changes had been made
/*    Rename source modules to avoid having a "test.c"			*/
/*    Start new versions at 2.5 for both modules			*/

/*	parse.c								*/
/*    parse and test sections from outer v 3.0a	 Jun 07			*/
/*    Separated out so that inners can parse selections.  Of interest	*/
/*  for optimization purposes to be sure that vars in selections are	*/
/*  not projected out prematurely.  Turns out this premature stuff	*/
/*  would help outer, too, and is much more critical there! (Apr 09)	*/

#define JG_SEL_PARSER_VERSION "jg_sel_parse v 2.5 15 May 2009"

/*  15 May 2009.  wjs							*/
/*	Check sizes as we add to a couple of arrays			*/
/*	Split tester from parser (see above)				*/
/*	No longer need per-precision variants				*/
/*		[Needs parse.h v 2.2]					*/
/*		[Begin v 2.5]						*/
/*  19 Apr 2009.  wjs							*/
/*	Bug fix: Type of var being returned needs to match func type	*/
/*   3 Apr 2009.  wjs							*/
/*	Modify jgofs_selection_test to return "Unknown" when it's 	*/
/*    unknown if record	matches selection criteria.  This happens	*/
/*    when criteria are at a lower level than the record being		*/
/*    tested.  Previous behavior was to try to return 			*/
/*    "accept record" in such cases.  The attempt did not succeed	*/
/*    in cases where the "not" logical operator was used		*/
/*	Increase size of stack in anticipation of more parenthesized	*/
/*    stuff from the OOserver subselection page				*/
/*	Change jgofs_selection_test function type to Logical from int	*/
/*	Move more functions to parse_function_struct to allow parse	*/
/*    to be used in a more standalone fashion				*/
/*	Accept commented out old parse code formerly in outer		*/
/*		[Needs parse.h v 2.1]					*/
/*		[Begin v 2.2]						*/

/*  11 Feb 2009.  wjs							*/
/*	Need single and double precision versions			*/
/*	As a consequence, move precision-independent stuff out		*/
/*	As a consequence of the implementation lexically redefining	*/
/*    entry names, change said names so "test" is not globally replaced	*/
/*    in some innocent source code					*/
/*		[Begin v 2.1]						*/
/*   2 May 2008.  wjs							*/
/*	Bug fix: uppercase copy of string operator was 1 char buf ovf	*/
/*		[Begin v 2.0a]						*/
/*  25 Oct 2007.  wjs							*/
/*	Declare some functions						*/
/*   1 Aug 2007.  wjs							*/
/*	V 2.0 is kind of a punt.  Code is certainly the "original", but	*/
/*    much has been added (string ops, for one big thing).		*/
/*		[Begin v 2.0]						*/

/*
P-code form:
128..255 exec test(n-128) [negative #'s on Sun]
			  [negativity not used in this code-fortunately.  WJS]
1=or,2=and,3=not
0: end
*/

#include "parse.h"

/*  Function in parse_utils						*/
char *analyze_wilds();

/*  To allow parse to be used standalone, we don't want inner or	*/
/*  outer functions (like err) to be used by parse.  Such functions	*/
/*  are in the parse_functions data structure				*/

char *jgofs_selection_parser_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[] = JG_SEL_PARSER_VERSION"/"FULL_PARSEH_VERSION;
  return version;
}

void jgofs_selection_parser(s,parse_data_struct,parse_function_struct)
char *s;
struct parse_data *parse_data_struct;
struct parse_functions *parse_function_struct;
  /*		Description is approximate.  Seems to me it can do	*/
  /*		weird things with not...				*/
  /*  Parser has 4 states: variable or open paren			*/
  /*			   comparison operator				*/
  /*			   thing being compared				*/
  /*			   logical operator or close paren or end	*/
  /*  Idea is to get next token, determine its type, the process it	*/
  /*  depending on the state the parser is in (possibly changing state)	*/
  /*	Odors observed or speculated about emanating from code		*/
  /*	  1) No escape mechanism for special characters.  Therefore	*/
  /*	     string ops cannot look for parens, etc.  Don't confuse	*/
  /*	     this w/escape mechanism in token that represents a 	*/
  /*	     "like" expression...					*/
  /*	  2) Suspect that we can't test for strings that conflict w/	*/
  /*	     our comparison operators (eg can't say op eq like)		*/
{

#define OPSTACKSIZE 100	/* Depth of operation stack; eg (a+b)*(c+d)	*/
			/*  needs a depth of 3				*/
  /*  Within each of next 4 arrays, "equivalent" strings (eg, != & <>)	*/
  /*  are grouped together.  "Group size" is constant (eg, see empty	*/
  /*  strings in testop where there are no equivalent strings).  Groups	*/
  /*  are ordered to drive "switch" statements in code, so order alter-	*/
  /*  ation means code alteration.  Infix array is ordered by operation	*/
  /*  priority as well.  testop is parallel to teststrop as far as it	*/
  /*  goes.  testop defines numeric comparisons; teststrop string	*/
  /*  comparisons.							*/
  /*  Each array ends with NULL						*/
  /*    Recoded not to use prefix array, since "not" prefix (and any	*/
  /*    future one like it) must be tested for in variable section,	*/
  /*    while ! is tested for in "normal" special character section	*/
static char *prefix[]={"!","not",NULL};
static char *infix[]={"||","or","|","&&","and","&",NULL};
static char *testop[]={"<","","=","==",">","","<=","","<>","!=",">=","",NULL};
static char *teststrop[]={"lt","eq","gt","le","ne","ge","contains",
		  "is_contained_in","begins_with","ends_with","like",NULL};
#define SIZE_PREFIX_GROUP 2
#define SIZE_INFIX_GROUP 3
#define SIZE_TESTOP_GROUP 2

  /*  Next is a non-null character we add to a selection string to	*/
  /*  indicate end-of-string.  Accordingly, choose a character that 	*/
  /*  will not normally be found in selection string (formerly used a	*/
  /*  $... either we got away w/it for a long time or my analysis is	*/
  /*  wrong)								*/
  /*    Note: we choose character at compile time so we can have switch	*/
  /*  statements use it							*/
#define SELECTION_STRING_END 1
  /*  Leading X is space to receive SELECTION_STRING_END.  Gave up	*/
  /*  trying to get string generated at compile time			*/
char selection_token_terminators[] = "X ()=<>|&!";

int opstkp;
double f;
int opstack[OPSTACKSIZE];
int opprior[OPSTACKSIZE];
char *s1,*t,*t1;
char tok[TOKEN+1],tmp[TOKEN+1];
int i,state,typ,finaland,type_string_op,paren_depth;
Logical case_sensitive;

  /*  Stuff from parse interface structures				*/
    /*  Data								*/
int tstcnt;
int tstproccnt;
struct parse_test_struct *tst;
int *tstproc;

tstcnt = parse_data_struct->tstcnt;
tstproccnt = parse_data_struct->tstproccnt;
tst = parse_data_struct->tst;
tstproc = parse_data_struct->tstproc;

selection_token_terminators[0] = SELECTION_STRING_END;

paren_depth = 0;
opstkp = 0;
opstack[0] = 0;
opprior[0] = -1;
state = 0;

finaland = (tstproccnt != 0);

  /*  Append SELECTION_STRING_END to input string as "end of string"	*/
  /*  indicator								*/
if (strchr(s,SELECTION_STRING_END) != NULL) 
    parse_function_struct->errn
	("Selection string contains character whose decimal value follows",
	 SELECTION_STRING_END);
i = strlen(s);
if (   ( s1 = (char *) malloc(i+2) ) == NULL   ) 
				parse_function_struct->malloc_err("s1",i+2);
strcpy(s1,s);
s1[i] = SELECTION_STRING_END;
s1[i+1] = '\0';

t = s1;
while (t != NULL) {

    /*  Scheme: If first char not one of the special ones, save the	*/
    /*  token.  Resulting alpha token can be "not" operator or one of	*/
    /*  the string operators.						*/
    /*  In original outer code, a token beginning with something like @	*/
    /*  would show up as "bad prefix operator".  In this code, I think	*/
    /*  it will show up as "error on conditions"			*/
    /*  The original outer code for this section is commented at the	*/
    /*  end of this file						*/
    /*  Skip leading blanks.  Whole string cannot be blank because of	*/
    /*  appended SELECTION_STRING_END, above				*/
  t += strspn(t," ");
  switch (*t) {
    case '(':
      paren_depth++;
      typ = 2;
      t++;
      break;
    case ')':
      if ((--paren_depth) < 0) 
		parse_function_struct->err("Illegally nested parens",s);
      typ = 3;
      t++;
      break;
    case SELECTION_STRING_END:
      typ = 4;
      t = NULL;
      break;
    case '=':  case '<':  case '>':  case '|':  case '&':  case '!':
      tok[0] = *t;
	/*  Check for 2nd char of 2-char ops like !=.  Validity		*/
	/*  of particular 2-char combo checked later			*/
      switch (*(t+1)) {
	case '&':  case '|':  case '=':  case '>':	
	  tok[1] = *(++t);
	  tok[2] = '\0';
	  break;
	case SELECTION_STRING_END:
	  parse_function_struct->err
				("No argument after last logical operator",s);
	  break;
	default:
	  tok[1] = '\0';
	  break;
      }
      typ = 5;
      t++;
      break;
    default:	/*  alpha token.  Could still be string operator...	*/
      if ( (i = strcspn(t,selection_token_terminators)) > TOKEN )
	parse_function_struct->err("Token too long",tok);
      strncpy(tok,t,i);
      tok[i] = '\0';
      t += i;

      type_string_op = -1;
      while (teststrop[++type_string_op] != NULL) {
	  /*  All upper case = "abnormal spelling" = "abnormal		*/
	  /*  processing" = not case sensitive				*/
	case_sensitive = TRUE;
	if (strcmp(teststrop[type_string_op],tok) == 0) break;
	else {
	  i = 0;
	  while (tok[i] == toupper(teststrop[type_string_op][i])) {
	    if (tok[i] == '\0') {
	      case_sensitive = FALSE;
	      break;
	    }
	    i++;
	  }
	    /*  If not case_sensitive, we found all upper case, so we	*/
	    /*  know it IS a string operator and we should leave string	*/
	    /*  loop searching string operator list			*/
	  if ( ! case_sensitive) break;
	}
      }
      if (teststrop[type_string_op] == NULL) {
	  /*  Token is not string operator.  See if it's numeric	*/
        f=strtod(tok,&t1);
        typ = (*t1 == '\0') ? 1 : 0;
      } else {
	type_string_op++;	/* Begins-w-0 vs begins-w-1 mod		*/
        typ = 6;
      }

      break;
  }

  switch (state*10+typ) {
    case 0: /* variable or the string "not" */
    case 6: /* string operator could be a variable name, too		*/
	    /*    NB: "syntactic" varname_lookup routines are,		*/
	    /* essentially, 'return legal unless tok="not"' routines	*/
	    /* They may require change if something here changes	*/
	    /* (syntactic routines are called in an environment where a	*/
	    /* varlist is not available, such as optimizing inner stuff	*/   
      i = parse_function_struct->varname_lookup(tok);
      if (i == ILLEGAL_VARNUM) {
        if (strcmp(tok,"not") != 0) 
			parse_function_struct->err("Error on conditions",tok);
        opstack[++opstkp]=3;
        opprior[opstkp]=10;
        break;
      }
      tst[tstcnt].testvar = i;
      tst[tstcnt].testlev = parse_function_struct->iovarlevel(&i);
      state=1;
      if (tstproccnt == N_JG_SEL_OPERATIONS) parse_function_struct->errn
	      ("Too many ops in selection.  Max =",N_JG_SEL_OPERATIONS);
      tstproc[tstproccnt++] = tstcnt+128;
      break;
    case 2: /* ( */
      opstack[++opstkp] = 4;
      opprior[opstkp] = 1;
      break;
    case 5: /* ! */
      if (tok[0] != '!') parse_function_struct->err("Bad prefix operator",tok);
      opstack[++opstkp] = 3;
      opprior[opstkp] = 10;
      break;
    case 15: /* ==, etc. */
      i = -1;
      while (testop[++i] != NULL)
	if (strcmp(tok,testop[i]) ==0 ) break;
      if (testop[i] == NULL) parse_function_struct->err("Bad condition",tok);
      tst[tstcnt].testcode = (i/SIZE_TESTOP_GROUP)+1;
      state=2;
      break;
    case 16: /* eq, ls, etc  */
      tst[tstcnt].testcode = - type_string_op;
      tst[tstcnt].case_sensitive = case_sensitive;
      state=2;
      break;
    case 20: 		/* token being compared to is not numeric 	*/
    case 26:
      if (tst[tstcnt].testcode > 0)	/*  Requested test is numeric	*/
	tst[tstcnt].testcode = - tst[tstcnt].testcode; /* Convert it	*/
	  /*  Deliberate fall-through here				*/
    case 21: 		/* token being compared to IS numeric 		*/
      if (tst[tstcnt].testcode < 0) {	/*  Requested test is string	*/
	tst[tstcnt].lenteststr = strlen(tok);
	i = tst[tstcnt].lenteststr + 1;
	if (  (tst[tstcnt].teststr = (char *) malloc(i)) == NULL  )
			parse_function_struct->malloc_err("teststr",i);
	strcpy(tst[tstcnt].teststr,tok);
	if (  (tst[tstcnt].upteststr = (char *) malloc(i)) == NULL  )
			parse_function_struct->malloc_err("upteststr",i);
	i--;
	while (i >= 0) {
	  tst[tstcnt].upteststr[i] = toupper(tok[i]);
	  i--;
	}
	f = 0.;		/*  Convenient for trick used later in test	*/
      }
      if (tstcnt == N_JG_SEL_TESTS) parse_function_struct->errn
	      ("Too many relational ops in selection.  Max =",N_JG_SEL_TESTS);
      tst[tstcnt++].testval=f;
      state=3;
      break;
    case 30:
    case 35: /* and or */
      i = -1;
      while (infix[++i] != NULL)
	if (strcmp(tok,infix[i]) ==0 ) break;
      if (infix[i] == NULL) 
			parse_function_struct->err("Bad infix operator",tok);
      i = (i/SIZE_INFIX_GROUP)+1;
      while (i+1 <= opprior[opstkp]) {
        if (tstproccnt == N_JG_SEL_OPERATIONS) parse_function_struct->errn
	      ("Too many ops in selection.  Max =",N_JG_SEL_OPERATIONS);
	tstproc[tstproccnt++] = opstack[opstkp--];
      }
      opstack[++opstkp] = i;
      opprior[opstkp] = i+1;
      state=0;
      break;
    case 33: /* ) */
      while (1 < opprior[opstkp]) {
        if (tstproccnt == N_JG_SEL_OPERATIONS) parse_function_struct->errn
	      ("Too many ops in selection.  Max =",N_JG_SEL_OPERATIONS);
        tstproc[tstproccnt++] = opstack[opstkp--];
      }
      if ((--opstkp) < 0) parse_function_struct->err("Op stack underflow","");
      break;
    case 34: /* SELECTION_STRING_END */
      while (0 <= opprior[opstkp]) {
        if (tstproccnt == N_JG_SEL_OPERATIONS) parse_function_struct->errn
	      ("Too many ops in selection.  Max =",N_JG_SEL_OPERATIONS);
        tstproc[tstproccnt++] = opstack[opstkp--];
      }
      if (finaland) {
        if (tstproccnt == N_JG_SEL_OPERATIONS) parse_function_struct->errn
	      ("Too many ops in selection.  Max =",N_JG_SEL_OPERATIONS);
	tstproc[tstproccnt++] = 2;
      }
#ifdef DEBUG
      printf("finaland = %d, tstproccnt = %d\n",finaland,tstproccnt);
      for (i=0; i<tstproccnt; i++)  {
	printf("tstproc %d ",tstproc[i]);
	if(tstproc[i]>=128)
	  printf("%d %d %f %s\n",tst[tstproc[i]-128].testvar,
		 tst[tstproc[i]-128].testcode,tst[tstproc[i]-128].testval,
		 tst[tstproc[i]-128].teststr);
	else printf("\n");
      };
#endif
      free (s1);
      parse_data_struct->tstcnt = tstcnt;
      parse_data_struct->tstproccnt = tstproccnt;
      return;
    default:
      parse_function_struct->err("Parsing error",s);
  }
}

if (paren_depth != 0) parse_function_struct->err("Unpaired parentheses",s);
parse_function_struct->err("Internal problem: ",
  "End-of-string while parsing selection, but normal return not taken");
}
/********************			**********************		*/
/*		Replaced token processing code from parse in original outer
  if(isdigit(*t) || *t=='.' || *t=='+' || *t=='-'){
    f=strtod(t,&t1);
    if(strchr(" ()$=<>|&!",*t1)){
      typ=1;
      t=t1;
    } else {
      i=strcspn(t," ()$=<>|&!");
      strncpy(tok,t,i);
      tok[i]='\0';
      t += i;
      typ=0;
    };
  } else if(isalpha(*t) || *t=='#'){
    i=strcspn(t," ()$=<>|&!");
    strncpy(tok,t,i);
    tok[i]='\0';
    t += i;
    typ=0;
  } else if(*t == '('){
    typ=2;
    t++;
  } else if(*t == ')'){
    typ=3;
    t++;
  } else if(*t == '$'){
    typ=4;
    t=NULL;
  } else {
    tok[0]= *t;
    if((int)strspn(t+1,"&|=>")>0){
      tok[1]= *(++t);
      tok[2]='\0';
    } else
      tok[1]='\0';
    t++;
    typ=5;
  };
*/