1 /* PSPP - computes sample statistics.
2 Copyright (C) 1997-9, 2000 Free Software Foundation, Inc.
3 Written by Ben Pfaff <blp@gnu.org>.
5 This program is free software; you can redistribute it and/or
6 modify it under the terms of the GNU General Public License as
7 published by the Free Software Foundation; either version 2 of the
8 License, or (at your option) any later version.
10 This program is distributed in the hope that it will be useful, but
11 WITHOUT ANY WARRANTY; without even the implied warranty of
12 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
13 General Public License for more details.
15 You should have received a copy of the GNU General Public License
16 along with this program; if not, write to the Free Software
17 Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA
26 #include "algorithm.h"
29 #include "dictionary.h"
42 /* Recursive descent parser in order of increasing precedence. */
43 typedef union any_node *parse_recursively_func (struct expression *);
44 static parse_recursively_func parse_or, parse_and, parse_not;
45 static parse_recursively_func parse_rel, parse_add, parse_mul;
46 static parse_recursively_func parse_neg, parse_exp;
47 static parse_recursively_func parse_primary;
48 static parse_recursively_func parse_vector_element, parse_function;
50 /* Utility functions. */
51 static struct expression *expr_create (struct dictionary *);
52 atom_type expr_node_returns (const union any_node *);
54 static const char *atom_type_name (atom_type);
55 static struct expression *finish_expression (union any_node *,
57 static bool type_check (struct expression *, union any_node **,
58 enum expr_type expected_type);
59 static union any_node *allocate_unary_variable (struct expression *,
62 /* Public functions. */
64 /* Parses an expression of the given TYPE.
65 If DICT is nonnull then variables and vectors within it may be
66 referenced within the expression; otherwise, the expression
67 must not reference any variables or vectors.
68 Returns the new expression if successful or a null pointer
71 expr_parse (struct dictionary *dict, enum expr_type type)
76 assert (type == EXPR_NUMBER || type == EXPR_STRING || type == EXPR_BOOLEAN);
78 e = expr_create (dict);
80 if (n != NULL && type_check (e, &n, type))
81 return finish_expression (expr_optimize (n, e), e);
89 /* Free expression E. */
91 expr_free (struct expression *e)
94 pool_destroy (e->expr_pool);
98 expr_parse_any (struct dictionary *dict, bool optimize)
101 struct expression *e;
103 e = expr_create (dict);
112 n = expr_optimize (n, e);
113 return finish_expression (n, e);
116 /* Finishing up expression building. */
118 /* Height of an expression's stacks. */
121 int number_height; /* Height of number stack. */
122 int string_height; /* Height of string stack. */
125 /* Stack heights used by different kinds of arguments. */
126 static const struct stack_heights on_number_stack = {1, 0};
127 static const struct stack_heights on_string_stack = {0, 1};
128 static const struct stack_heights not_on_stack = {0, 0};
130 /* Returns the stack heights used by an atom of the given
132 static const struct stack_heights *
133 atom_type_stack (atom_type type)
135 assert (is_atom (type));
141 return &on_number_stack;
144 return &on_string_stack;
154 return ¬_on_stack;
161 /* Measures the stack height needed for node N, supposing that
162 the stack height is initially *HEIGHT and updating *HEIGHT to
163 the final stack height. Updates *MAX, if necessary, to
164 reflect the maximum intermediate or final height. */
166 measure_stack (const union any_node *n,
167 struct stack_heights *height, struct stack_heights *max)
169 const struct stack_heights *return_height;
171 if (is_composite (n->type))
173 struct stack_heights args;
177 for (i = 0; i < n->composite.arg_cnt; i++)
178 measure_stack (n->composite.args[i], &args, max);
180 return_height = atom_type_stack (operations[n->type].returns);
183 return_height = atom_type_stack (n->type);
185 height->number_height += return_height->number_height;
186 height->string_height += return_height->string_height;
188 if (height->number_height > max->number_height)
189 max->number_height = height->number_height;
190 if (height->string_height > max->string_height)
191 max->string_height = height->string_height;
194 /* Allocates stacks within E sufficient for evaluating node N. */
196 allocate_stacks (union any_node *n, struct expression *e)
198 struct stack_heights initial = {0, 0};
199 struct stack_heights max = {0, 0};
201 measure_stack (n, &initial, &max);
202 e->number_stack = pool_alloc (e->expr_pool,
203 sizeof *e->number_stack * max.number_height);
204 e->string_stack = pool_alloc (e->expr_pool,
205 sizeof *e->string_stack * max.string_height);
208 /* Finalizes expression E for evaluating node N. */
209 static struct expression *
210 finish_expression (union any_node *n, struct expression *e)
212 /* Allocate stacks. */
213 allocate_stacks (n, e);
215 /* Output postfix representation. */
218 /* The eval_pool might have been used for allocating strings
219 during optimization. We need to keep those strings around
220 for all subsequent evaluations, so start a new eval_pool. */
221 e->eval_pool = pool_create_subpool (e->expr_pool);
226 /* Verifies that expression E, whose root node is *N, can be
227 converted to type EXPECTED_TYPE, inserting a conversion at *N
228 if necessary. Returns true if successful, false on failure. */
230 type_check (struct expression *e,
231 union any_node **n, enum expr_type expected_type)
233 atom_type actual_type = expr_node_returns (*n);
235 switch (expected_type)
239 if (actual_type != OP_number && actual_type != OP_boolean)
241 msg (SE, _("Type mismatch: expression has %s type, "
242 "but a numeric value is required here."),
243 atom_type_name (actual_type));
246 if (actual_type == OP_number && expected_type == OP_boolean)
247 *n = expr_allocate_unary (e, OP_NUM_TO_BOOLEAN, *n);
251 if (actual_type != OP_string)
253 msg (SE, _("Type mismatch: expression has %s type, "
254 "but a string value is required here."),
255 atom_type_name (actual_type));
267 /* Recursive-descent expression parser. */
269 /* Considers whether *NODE may be coerced to type REQUIRED_TYPE.
270 Returns true if possible, false if disallowed.
272 If DO_COERCION is zero, then *NODE is not modified and there
275 Otherwise, DO_COERCION is nonzero. In this case, we perform
276 the coercion if possible, possibly modifying *NODE. If the
277 coercion is not possible then we free *NODE and set *NODE to
280 This function's interface is somewhat awkward. Use one of the
281 wrapper functions type_coercion(), type_coercion_assert(), or
282 is_coercible() instead. */
284 type_coercion_core (struct expression *e,
285 atom_type required_type,
286 union any_node **node,
287 const char *operator_name,
290 atom_type actual_type;
292 assert (!!do_coercion == (e != NULL));
295 /* Propagate error. Whatever caused the original error
296 already emitted an error message. */
300 actual_type = expr_node_returns (*node);
301 if (actual_type == required_type)
307 switch (required_type)
310 if (actual_type == OP_boolean)
312 /* To enforce strict typing rules, insert Boolean to
313 numeric "conversion". This conversion is a no-op,
314 so it will be removed later. */
316 *node = expr_allocate_unary (e, OP_BOOLEAN_TO_NUM, *node);
322 /* No coercion to string. */
326 if (actual_type == OP_number)
328 /* Convert numeric to boolean. */
330 *node = expr_allocate_unary (e, OP_NUM_TO_BOOLEAN, *node);
339 if ((*node)->type == OP_format
340 && check_input_specifier (&(*node)->format.f, 0))
343 (*node)->type = OP_ni_format;
349 if ((*node)->type == OP_format
350 && check_output_specifier (&(*node)->format.f, 0))
353 (*node)->type = OP_no_format;
359 if ((*node)->type == OP_NUM_VAR)
362 *node = (*node)->composite.args[0];
368 if ((*node)->type == OP_STR_VAR)
371 *node = (*node)->composite.args[0];
377 if ((*node)->type == OP_number
378 && floor ((*node)->number.n) == (*node)->number.n
379 && (*node)->number.n > 0 && (*node)->number.n < INT_MAX)
382 *node = expr_allocate_pos_int (e, (*node)->number.n);
392 msg (SE, _("Type mismatch while applying %s operator: "
393 "cannot convert %s to %s."),
395 atom_type_name (actual_type), atom_type_name (required_type));
399 /* Coerces *NODE to type REQUIRED_TYPE, and returns success. If
400 *NODE cannot be coerced to the desired type then we issue an
401 error message about operator OPERATOR_NAME and free *NODE. */
403 type_coercion (struct expression *e,
404 atom_type required_type, union any_node **node,
405 const char *operator_name)
407 return type_coercion_core (e, required_type, node, operator_name, true);
410 /* Coerces *NODE to type REQUIRED_TYPE.
411 Assert-fails if the coercion is disallowed. */
413 type_coercion_assert (struct expression *e,
414 atom_type required_type, union any_node **node)
416 int success = type_coercion_core (e, required_type, node, NULL, true);
420 /* Returns true if *NODE may be coerced to type REQUIRED_TYPE,
423 is_coercible (atom_type required_type, union any_node *const *node)
425 return type_coercion_core (NULL, required_type,
426 (union any_node **) node, NULL, false);
429 /* How to parse an operator. */
432 int token; /* Token representing operator. */
433 operation_type type; /* Operation type representing operation. */
434 const char *name; /* Name of operator. */
437 /* Attempts to match the current token against the tokens for the
438 OP_CNT operators in OPS[]. If successful, returns true
439 and, if OPERATOR is non-null, sets *OPERATOR to the operator.
440 On failure, returns false and, if OPERATOR is non-null, sets
441 *OPERATOR to a null pointer. */
443 match_operator (const struct operator ops[], size_t op_cnt,
444 const struct operator **operator)
446 const struct operator *op;
448 for (op = ops; op < ops + op_cnt; op++)
450 if (op->token == '-')
451 lex_negative_to_dash ();
452 if (lex_match (op->token))
454 if (operator != NULL)
459 if (operator != NULL)
465 check_operator (const struct operator *op, int arg_cnt, atom_type arg_type)
467 const struct operation *o;
471 o = &operations[op->type];
472 assert (o->arg_cnt == arg_cnt);
473 assert ((o->flags & OPF_ARRAY_OPERAND) == 0);
474 for (i = 0; i < arg_cnt; i++)
475 assert (o->args[i] == arg_type);
480 check_binary_operators (const struct operator ops[], size_t op_cnt,
485 for (i = 0; i < op_cnt; i++)
486 check_operator (&ops[i], 2, arg_type);
491 get_operand_type (const struct operator *op)
493 return operations[op->type].args[0];
496 /* Parses a chain of left-associative operator/operand pairs.
497 There are OP_CNT operators, specified in OPS[]. The
498 operators' operands must all be the same type. The next
499 higher level is parsed by PARSE_NEXT_LEVEL. If CHAIN_WARNING
500 is non-null, then it will be issued as a warning if more than
501 one operator/operand pair is parsed. */
502 static union any_node *
503 parse_binary_operators (struct expression *e, union any_node *node,
504 const struct operator ops[], size_t op_cnt,
505 parse_recursively_func *parse_next_level,
506 const char *chain_warning)
508 atom_type operand_type = get_operand_type (&ops[0]);
510 const struct operator *operator;
512 assert (check_binary_operators (ops, op_cnt, operand_type));
516 for (op_count = 0; match_operator (ops, op_cnt, &operator); op_count++)
520 /* Convert the left-hand side to type OPERAND_TYPE. */
521 if (!type_coercion (e, operand_type, &node, operator->name))
524 /* Parse the right-hand side and coerce to type
526 rhs = parse_next_level (e);
527 if (!type_coercion (e, operand_type, &rhs, operator->name))
529 node = expr_allocate_binary (e, operator->type, node, rhs);
532 if (op_count > 1 && chain_warning != NULL)
533 msg (SW, chain_warning);
538 static union any_node *
539 parse_inverting_unary_operator (struct expression *e,
540 const struct operator *op,
541 parse_recursively_func *parse_next_level)
543 union any_node *node;
546 check_operator (op, 1, get_operand_type (op));
549 while (match_operator (op, 1, NULL))
552 node = parse_next_level (e);
554 && type_coercion (e, get_operand_type (op), &node, op->name)
555 && op_count % 2 != 0)
556 return expr_allocate_unary (e, op->type, node);
561 /* Parses the OR level. */
562 static union any_node *
563 parse_or (struct expression *e)
565 static const struct operator op =
566 { T_OR, OP_OR, "logical disjunction (\"OR\")" };
568 return parse_binary_operators (e, parse_and (e), &op, 1, parse_and, NULL);
571 /* Parses the AND level. */
572 static union any_node *
573 parse_and (struct expression *e)
575 static const struct operator op =
576 { T_AND, OP_AND, "logical conjunction (\"AND\")" };
578 return parse_binary_operators (e, parse_not (e), &op, 1, parse_not, NULL);
581 /* Parses the NOT level. */
582 static union any_node *
583 parse_not (struct expression *e)
585 static const struct operator op
586 = { T_NOT, OP_NOT, "logical negation (\"NOT\")" };
587 return parse_inverting_unary_operator (e, &op, parse_rel);
590 /* Parse relational operators. */
591 static union any_node *
592 parse_rel (struct expression *e)
594 const char *chain_warning =
595 _("Chaining relational operators (e.g. \"a < b < c\") will "
596 "not produce the mathematically expected result. "
597 "Use the AND logical operator to fix the problem "
598 "(e.g. \"a < b AND b < c\"). "
599 "If chaining is really intended, parentheses will disable "
600 "this warning (e.g. \"(a < b) < c\".)");
602 union any_node *node = parse_add (e);
607 switch (expr_node_returns (node))
612 static const struct operator ops[] =
614 { '=', OP_EQ, "numeric equality (\"=\")" },
615 { T_EQ, OP_EQ, "numeric equality (\"EQ\")" },
616 { T_GE, OP_GE, "numeric greater-than-or-equal-to (\">=\")" },
617 { T_GT, OP_GT, "numeric greater than (\">\")" },
618 { T_LE, OP_LE, "numeric less-than-or-equal-to (\"<=\")" },
619 { T_LT, OP_LT, "numeric less than (\"<\")" },
620 { T_NE, OP_NE, "numeric inequality (\"<>\")" },
623 return parse_binary_operators (e, node, ops, sizeof ops / sizeof *ops,
624 parse_add, chain_warning);
629 static const struct operator ops[] =
631 { '=', OP_EQ_STRING, "string equality (\"=\")" },
632 { T_EQ, OP_EQ_STRING, "string equality (\"EQ\")" },
633 { T_GE, OP_GE_STRING, "string greater-than-or-equal-to (\">=\")" },
634 { T_GT, OP_GT_STRING, "string greater than (\">\")" },
635 { T_LE, OP_LE_STRING, "string less-than-or-equal-to (\"<=\")" },
636 { T_LT, OP_LT_STRING, "string less than (\"<\")" },
637 { T_NE, OP_NE_STRING, "string inequality (\"<>\")" },
640 return parse_binary_operators (e, node, ops, sizeof ops / sizeof *ops,
641 parse_add, chain_warning);
649 /* Parses the addition and subtraction level. */
650 static union any_node *
651 parse_add (struct expression *e)
653 static const struct operator ops[] =
655 { '+', OP_ADD, "addition (\"+\")" },
656 { '-', OP_SUB, "subtraction (\"-\")-" },
659 return parse_binary_operators (e, parse_mul (e),
660 ops, sizeof ops / sizeof *ops,
664 /* Parses the multiplication and division level. */
665 static union any_node *
666 parse_mul (struct expression *e)
668 static const struct operator ops[] =
670 { '*', OP_MUL, "multiplication (\"*\")" },
671 { '/', OP_DIV, "division (\"/\")" },
674 return parse_binary_operators (e, parse_neg (e),
675 ops, sizeof ops / sizeof *ops,
679 /* Parses the unary minus level. */
680 static union any_node *
681 parse_neg (struct expression *e)
683 static const struct operator op = { '-', OP_NEG, "negation (\"-\")" };
684 return parse_inverting_unary_operator (e, &op, parse_exp);
687 static union any_node *
688 parse_exp (struct expression *e)
690 static const struct operator op =
691 { T_EXP, OP_POW, "exponentiation (\"**\")" };
693 const char *chain_warning =
694 _("The exponentiation operator (\"**\") is left-associative, "
695 "even though right-associative semantics are more useful. "
696 "That is, \"a**b**c\" equals \"(a**b)**c\", not as \"a**(b**c)\". "
697 "To disable this warning, insert parentheses.");
699 return parse_binary_operators (e, parse_primary (e), &op, 1,
700 parse_primary, chain_warning);
703 /* Parses system variables. */
704 static union any_node *
705 parse_sysvar (struct expression *e)
707 if (lex_match_id ("$CASENUM"))
708 return expr_allocate_nullary (e, OP_CASENUM);
709 else if (lex_match_id ("$DATE"))
711 static const char *months[12] =
713 "JAN", "FEB", "MAR", "APR", "MAY", "JUN",
714 "JUL", "AUG", "SEP", "OCT", "NOV", "DEC",
720 time = localtime (&last_vfm_invocation);
721 sprintf (temp_buf, "%02d %s %02d", abs (time->tm_mday) % 100,
722 months[abs (time->tm_mon) % 12], abs (time->tm_year) % 100);
724 return expr_allocate_string_buffer (e, temp_buf, strlen (temp_buf));
726 else if (lex_match_id ("$TRUE"))
727 return expr_allocate_boolean (e, 1.0);
728 else if (lex_match_id ("$FALSE"))
729 return expr_allocate_boolean (e, 0.0);
730 else if (lex_match_id ("$SYSMIS"))
731 return expr_allocate_number (e, SYSMIS);
732 else if (lex_match_id ("$JDATE"))
734 struct tm *time = localtime (&last_vfm_invocation);
735 return expr_allocate_number (e, expr_ymd_to_ofs (time->tm_year + 1900,
739 else if (lex_match_id ("$TIME"))
741 struct tm *time = localtime (&last_vfm_invocation);
742 return expr_allocate_number (e,
743 expr_ymd_to_date (time->tm_year + 1900,
746 + time->tm_hour * 60 * 60.
750 else if (lex_match_id ("$LENGTH"))
751 return expr_allocate_number (e, get_viewlength ());
752 else if (lex_match_id ("$WIDTH"))
753 return expr_allocate_number (e, get_viewwidth ());
756 msg (SE, _("Unknown system variable %s."), tokid);
761 /* Parses numbers, varnames, etc. */
762 static union any_node *
763 parse_primary (struct expression *e)
768 if (lex_look_ahead () == '(')
770 /* An identifier followed by a left parenthesis may be
771 a vector element reference. If not, it's a function
773 if (e->dict != NULL && dict_lookup_vector (e->dict, tokid) != NULL)
774 return parse_vector_element (e);
776 return parse_function (e);
778 else if (tokid[0] == '$')
780 /* $ at the beginning indicates a system variable. */
781 return parse_sysvar (e);
783 else if (e->dict != NULL && dict_lookup_var (e->dict, tokid))
785 /* It looks like a user variable.
786 (It could be a format specifier, but we'll assume
787 it's a variable unless proven otherwise. */
788 return allocate_unary_variable (e, parse_dict_variable (e->dict));
792 /* Try to parse it as a format specifier. */
794 if (parse_format_specifier (&fmt, FMTP_SUPPRESS_ERRORS))
795 return expr_allocate_format (e, &fmt);
797 /* All attempts failed. */
798 msg (SE, _("Unknown identifier %s."), tokid);
805 union any_node *node = expr_allocate_number (e, tokval);
812 union any_node *node = expr_allocate_string_buffer (e, ds_c_str (&tokstr),
813 ds_length (&tokstr));
820 union any_node *node;
823 if (node != NULL && !lex_match (')'))
825 lex_error (_("expecting `)'"));
832 lex_error (_("in expression"));
837 static union any_node *
838 parse_vector_element (struct expression *e)
840 const struct vector *vector;
841 union any_node *element;
843 /* Find vector, skip token.
844 The caller must already have verified that the current token
845 is the name of a vector. */
846 vector = dict_lookup_vector (default_dict, tokid);
847 assert (vector != NULL);
850 /* Skip left parenthesis token.
851 The caller must have verified that the lookahead is a left
853 assert (token == '(');
856 element = parse_or (e);
857 if (!type_coercion (e, OP_number, &element, "vector indexing")
861 return expr_allocate_binary (e, (vector->var[0]->type == NUMERIC
862 ? OP_VEC_ELEM_NUM : OP_VEC_ELEM_STR),
863 expr_allocate_vector (e, vector), element);
866 /* Individual function parsing. */
868 struct operation operations[OP_first + OP_cnt] = {
873 word_matches (const char **test, const char **name)
875 size_t test_len = strcspn (*test, ".");
876 size_t name_len = strcspn (*name, ".");
877 if (test_len == name_len)
879 if (memcmp (*test, *name, test_len))
882 else if (test_len < 3 || test_len > name_len)
886 if (memcmp (*test, *name, test_len))
892 if (**test != **name)
904 compare_names (const char *test, const char *name)
908 if (!word_matches (&test, &name))
910 if (*name == '\0' && *test == '\0')
916 lookup_function_helper (const char *name,
917 int (*compare) (const char *test, const char *name),
918 const struct operation **first,
919 const struct operation **last)
923 for (f = operations + OP_function_first;
924 f <= operations + OP_function_last; f++)
925 if (!compare (name, f->name))
929 while (f <= operations + OP_function_last && !compare (name, f->name))
940 lookup_function (const char *name,
941 const struct operation **first,
942 const struct operation **last)
944 *first = *last = NULL;
945 return (lookup_function_helper (name, strcmp, first, last)
946 || lookup_function_helper (name, compare_names, first, last));
950 extract_min_valid (char *s)
952 char *p = strrchr (s, '.');
954 || p[1] < '0' || p[1] > '9'
955 || strspn (p + 1, "0123456789") != strlen (p + 1))
962 function_arg_type (const struct operation *f, size_t arg_idx)
964 assert (arg_idx < f->arg_cnt || (f->flags & OPF_ARRAY_OPERAND));
966 return f->args[arg_idx < f->arg_cnt ? arg_idx : f->arg_cnt - 1];
970 match_function (union any_node **args, int arg_cnt, const struct operation *f)
974 if (arg_cnt < f->arg_cnt
975 || (arg_cnt > f->arg_cnt && (f->flags & OPF_ARRAY_OPERAND) == 0)
976 || arg_cnt - (f->arg_cnt - 1) < f->array_min_elems)
979 for (i = 0; i < arg_cnt; i++)
980 if (!is_coercible (function_arg_type (f, i), &args[i]))
987 coerce_function_args (struct expression *e, const struct operation *f,
988 union any_node **args, size_t arg_cnt)
992 for (i = 0; i < arg_cnt; i++)
993 type_coercion_assert (e, function_arg_type (f, i), &args[i]);
997 validate_function_args (const struct operation *f, int arg_cnt, int min_valid)
999 int array_arg_cnt = arg_cnt - (f->arg_cnt - 1);
1000 if (array_arg_cnt < f->array_min_elems)
1002 msg (SE, _("%s must have at least %d arguments in list."),
1003 f->prototype, f->array_min_elems);
1007 if ((f->flags & OPF_ARRAY_OPERAND)
1008 && array_arg_cnt % f->array_granularity != 0)
1010 if (f->array_granularity == 2)
1011 msg (SE, _("%s must have even number of arguments in list."),
1014 msg (SE, _("%s must have multiple of %d arguments in list."),
1015 f->prototype, f->array_granularity);
1019 if (min_valid != -1)
1021 if (f->array_min_elems == 0)
1023 assert ((f->flags & OPF_MIN_VALID) == 0);
1024 msg (SE, _("%s function does not accept a minimum valid "
1025 "argument count."));
1030 assert (f->flags & OPF_MIN_VALID);
1031 if (array_arg_cnt < f->array_min_elems)
1033 msg (SE, _("%s requires at least %d valid arguments in list."),
1037 else if (min_valid > array_arg_cnt)
1039 msg (SE, _("With %s, "
1040 "using minimum valid argument count of %d "
1041 "does not make sense when passing only %d "
1042 "arguments in list."),
1043 f->prototype, min_valid, array_arg_cnt);
1053 add_arg (union any_node ***args, int *arg_cnt, int *arg_cap,
1054 union any_node *arg)
1056 if (*arg_cnt >= *arg_cap)
1059 *args = xrealloc (*args, sizeof **args * *arg_cap);
1062 (*args)[(*arg_cnt)++] = arg;
1066 put_invocation (struct string *s,
1067 const char *func_name, union any_node **args, size_t arg_cnt)
1071 ds_printf (s, "%s(", func_name);
1072 for (i = 0; i < arg_cnt; i++)
1076 ds_puts (s, operations[expr_node_returns (args[i])].prototype);
1082 no_match (const char *func_name,
1083 union any_node **args, size_t arg_cnt,
1084 const struct operation *first, const struct operation *last)
1087 const struct operation *f;
1091 if (last - first == 1)
1093 ds_printf (&s, _("Type mismatch invoking %s as "), first->prototype);
1094 put_invocation (&s, func_name, args, arg_cnt);
1098 ds_create (&s, _("Function invocation "));
1099 put_invocation (&s, func_name, args, arg_cnt);
1100 ds_puts (&s, _(" does not match any known function. Candidates are:"));
1102 for (f = first; f < last; f++)
1104 ds_printf (&s, "\n%s", f->prototype);
1109 msg (SE, "%s", ds_c_str (&s));
1114 static union any_node *
1115 parse_function (struct expression *e)
1118 const struct operation *f, *first, *last;
1120 union any_node **args = NULL;
1124 struct fixed_string func_name;
1128 ls_create (&func_name, ds_c_str (&tokstr));
1129 min_valid = extract_min_valid (ds_c_str (&tokstr));
1130 if (!lookup_function (ds_c_str (&tokstr), &first, &last))
1132 msg (SE, _("No function or vector named %s."), ds_c_str (&tokstr));
1133 ls_destroy (&func_name);
1138 if (!lex_force_match ('('))
1140 ls_destroy (&func_name);
1145 arg_cnt = arg_cap = 0;
1149 if (token == T_ID && lex_look_ahead () == 'T')
1151 struct variable **vars;
1155 if (!parse_variables (default_dict, &vars, &var_cnt, PV_SINGLE))
1157 for (i = 0; i < var_cnt; i++)
1158 add_arg (&args, &arg_cnt, &arg_cap,
1159 allocate_unary_variable (e, vars[i]));
1164 union any_node *arg = parse_or (e);
1168 add_arg (&args, &arg_cnt, &arg_cap, arg);
1170 if (lex_match (')'))
1172 else if (!lex_match (','))
1174 lex_error (_("expecting `,' or `)' invoking %s function"),
1180 for (f = first; f < last; f++)
1181 if (match_function (args, arg_cnt, f))
1185 no_match (ls_c_str (&func_name), args, arg_cnt, first, last);
1189 coerce_function_args (e, f, args, arg_cnt);
1190 if (!validate_function_args (f, arg_cnt, min_valid))
1193 if ((f->flags & OPF_EXTENSION) && get_syntax () == COMPATIBLE)
1194 msg (SW, _("%s is a PSPP extension."), f->prototype);
1195 if (f->flags & OPF_UNIMPLEMENTED)
1197 msg (SE, _("%s is not yet implemented."), f->prototype);
1201 n = expr_allocate_composite (e, f - operations, args, arg_cnt);
1202 n->composite.min_valid = min_valid != -1 ? min_valid : f->array_min_elems;
1204 if (n->type == OP_LAG_Vnn || n->type == OP_LAG_Vsn)
1207 assert (n->composite.arg_cnt == 2);
1208 assert (n->composite.args[1]->type == OP_pos_int);
1209 n_before = n->composite.args[1]->integer.i;
1210 if (n_before > n_lag)
1215 ls_destroy (&func_name);
1220 ls_destroy (&func_name);
1224 /* Utility functions. */
1226 static struct expression *
1227 expr_create (struct dictionary *dict)
1229 struct pool *pool = pool_create ();
1230 struct expression *e = pool_alloc (pool, sizeof *e);
1231 e->expr_pool = pool;
1233 e->eval_pool = pool_create_subpool (e->expr_pool);
1236 e->op_cnt = e->op_cap = 0;
1241 expr_node_returns (const union any_node *n)
1244 assert (is_operation (n->type));
1245 if (is_atom (n->type))
1247 else if (is_composite (n->type))
1248 return operations[n->type].returns;
1254 atom_type_name (atom_type type)
1256 assert (is_atom (type));
1257 return operations[type].name;
1261 expr_allocate_nullary (struct expression *e, operation_type op)
1263 return expr_allocate_composite (e, op, NULL, 0);
1267 expr_allocate_unary (struct expression *e, operation_type op,
1268 union any_node *arg0)
1270 return expr_allocate_composite (e, op, &arg0, 1);
1274 expr_allocate_binary (struct expression *e, operation_type op,
1275 union any_node *arg0, union any_node *arg1)
1277 union any_node *args[2];
1280 return expr_allocate_composite (e, op, args, 2);
1284 is_valid_node (union any_node *n)
1286 struct operation *op;
1290 assert (is_operation (n->type));
1291 op = &operations[n->type];
1293 if (!is_atom (n->type))
1295 struct composite_node *c = &n->composite;
1297 assert (is_composite (n->type));
1298 assert (c->arg_cnt >= op->arg_cnt);
1299 for (i = 0; i < op->arg_cnt; i++)
1300 assert (expr_node_returns (c->args[i]) == op->args[i]);
1301 if (c->arg_cnt > op->arg_cnt && !is_operator (n->type))
1303 assert (op->flags & OPF_ARRAY_OPERAND);
1304 for (i = 0; i < c->arg_cnt; i++)
1305 assert (operations[c->args[i]->type].returns
1306 == op->args[op->arg_cnt - 1]);
1314 expr_allocate_composite (struct expression *e, operation_type op,
1315 union any_node **args, size_t arg_cnt)
1320 n = pool_alloc (e->expr_pool, sizeof n->composite);
1322 n->composite.arg_cnt = arg_cnt;
1323 n->composite.args = pool_alloc (e->expr_pool,
1324 sizeof *n->composite.args * arg_cnt);
1325 for (i = 0; i < arg_cnt; i++)
1327 if (args[i] == NULL)
1329 n->composite.args[i] = args[i];
1331 memcpy (n->composite.args, args, sizeof *n->composite.args * arg_cnt);
1332 n->composite.min_valid = 0;
1333 assert (is_valid_node (n));
1338 expr_allocate_number (struct expression *e, double d)
1340 union any_node *n = pool_alloc (e->expr_pool, sizeof n->number);
1341 n->type = OP_number;
1347 expr_allocate_boolean (struct expression *e, double b)
1349 union any_node *n = pool_alloc (e->expr_pool, sizeof n->number);
1350 assert (b == 0.0 || b == 1.0 || b == SYSMIS);
1351 n->type = OP_boolean;
1357 expr_allocate_integer (struct expression *e, int i)
1359 union any_node *n = pool_alloc (e->expr_pool, sizeof n->integer);
1360 n->type = OP_integer;
1366 expr_allocate_pos_int (struct expression *e, int i)
1368 union any_node *n = pool_alloc (e->expr_pool, sizeof n->integer);
1370 n->type = OP_pos_int;
1376 expr_allocate_vector (struct expression *e, const struct vector *vector)
1378 union any_node *n = pool_alloc (e->expr_pool, sizeof n->vector);
1379 n->type = OP_vector;
1380 n->vector.v = vector;
1385 expr_allocate_string_buffer (struct expression *e,
1386 const char *string, size_t length)
1388 union any_node *n = pool_alloc (e->expr_pool, sizeof n->string);
1389 n->type = OP_string;
1392 n->string.s = copy_string (e, string, length);
1397 expr_allocate_string (struct expression *e, struct fixed_string s)
1399 union any_node *n = pool_alloc (e->expr_pool, sizeof n->string);
1400 n->type = OP_string;
1406 expr_allocate_variable (struct expression *e, struct variable *v)
1408 union any_node *n = pool_alloc (e->expr_pool, sizeof n->variable);
1409 n->type = v->type == NUMERIC ? OP_num_var : OP_str_var;
1415 expr_allocate_format (struct expression *e, const struct fmt_spec *format)
1417 union any_node *n = pool_alloc (e->expr_pool, sizeof n->format);
1418 n->type = OP_format;
1419 n->format.f = *format;
1423 /* Allocates a unary composite node that represents the value of
1424 variable V in expression E. */
1425 static union any_node *
1426 allocate_unary_variable (struct expression *e, struct variable *v)
1429 return expr_allocate_unary (e, v->type == NUMERIC ? OP_NUM_VAR : OP_STR_VAR,
1430 expr_allocate_variable (e, v));