1 /* PSPP - a program for statistical analysis.
2 Copyright (C) 1997-9, 2000, 2006, 2010, 2011, 2012, 2014 Free Software Foundation, Inc.
4 This program is free software: you can redistribute it and/or modify
5 it under the terms of the GNU General Public License as published by
6 the Free Software Foundation, either version 3 of the License, or
7 (at your option) any later version.
9 This program is distributed in the hope that it will be useful,
10 but WITHOUT ANY WARRANTY; without even the implied warranty of
11 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
12 GNU General Public License for more details.
14 You should have received a copy of the GNU General Public License
15 along with this program. If not, see <http://www.gnu.org/licenses/>. */
26 #include "data/case.h"
27 #include "data/dictionary.h"
28 #include "data/settings.h"
29 #include "data/variable.h"
30 #include "language/expressions/helpers.h"
31 #include "language/lexer/format-parser.h"
32 #include "language/lexer/lexer.h"
33 #include "language/lexer/variable-parser.h"
34 #include "libpspp/array.h"
35 #include "libpspp/assertion.h"
36 #include "libpspp/i18n.h"
37 #include "libpspp/message.h"
38 #include "libpspp/misc.h"
39 #include "libpspp/pool.h"
40 #include "libpspp/str.h"
42 #include "gl/c-strcase.h"
43 #include "gl/minmax.h"
44 #include "gl/xalloc.h"
48 /* Recursive descent parser in order of increasing precedence. */
49 typedef struct expr_node *parse_recursively_func (struct lexer *, struct expression *);
50 static parse_recursively_func parse_or, parse_and, parse_not;
51 static parse_recursively_func parse_rel, parse_add, parse_mul;
52 static parse_recursively_func parse_neg, parse_exp;
53 static parse_recursively_func parse_primary;
54 static parse_recursively_func parse_vector_element, parse_function;
56 /* Utility functions. */
57 static struct expression *expr_create (struct dataset *ds);
58 atom_type expr_node_returns (const struct expr_node *);
60 static const char *atom_type_name (atom_type);
61 static struct expression *finish_expression (struct expr_node *,
63 static bool type_check (const struct expression *, const struct expr_node *,
64 enum val_type expected_type);
65 static struct expr_node *allocate_unary_variable (struct expression *,
66 const struct variable *);
68 /* Public functions. */
70 static struct expr_node *
71 parse_expr (struct lexer *lexer, struct expression *e)
73 struct expr_node *n = parse_or (lexer, e);
74 if (n && n->type == OP_VEC_ELEM_NUM_RAW)
75 n->type = OP_VEC_ELEM_NUM;
79 /* Parses an expression of the given TYPE. If DS is nonnull then variables and
80 vectors within it may be referenced within the expression; otherwise, the
81 expression must not reference any variables or vectors. Returns the new
82 expression if successful or a null pointer otherwise. */
84 expr_parse (struct lexer *lexer, struct dataset *ds, enum val_type type)
86 assert (val_type_is_valid (type));
88 struct expression *e = expr_create (ds);
89 struct expr_node *n = parse_expr (lexer, e);
90 if (!n || !type_check (e, n, type))
96 return finish_expression (expr_optimize (n, e), e);
99 /* Parses a boolean expression, otherwise similar to expr_parse(). */
101 expr_parse_bool (struct lexer *lexer, struct dataset *ds)
103 struct expression *e = expr_create (ds);
104 struct expr_node *n = parse_expr (lexer, e);
111 atom_type actual_type = expr_node_returns (n);
112 if (actual_type == OP_number)
113 n = expr_allocate_unary (e, OP_EXPR_TO_BOOLEAN, n);
114 else if (actual_type != OP_boolean)
116 msg_at (SE, expr_location (e, n),
117 _("Type mismatch: expression has %s type, "
118 "but a boolean value is required here."),
119 atom_type_name (actual_type));
124 return finish_expression (expr_optimize (n, e), e);
127 /* Parses a numeric expression that is intended to be assigned to newly created
128 variable NEW_VAR_NAME at NEW_VAR_LOCATION. (This allows for a better error
129 message if the expression is not numeric.) Otherwise similar to
132 expr_parse_new_variable (struct lexer *lexer, struct dataset *ds,
133 const char *new_var_name,
134 const struct msg_location *new_var_location)
136 struct expression *e = expr_create (ds);
137 struct expr_node *n = parse_expr (lexer, e);
144 atom_type actual_type = expr_node_returns (n);
145 if (actual_type != OP_number && actual_type != OP_boolean)
147 msg_at (SE, new_var_location,
148 _("This command tries to create a new variable %s by assigning a "
149 "string value to it, but this is not supported. Use "
150 "the STRING command to create the new variable with the "
151 "correct width before assigning to it, e.g. STRING %s(A20)."),
152 new_var_name, new_var_name);
157 return finish_expression (expr_optimize (n, e), e);
160 /* Free expression E. */
162 expr_free (struct expression *e)
165 pool_destroy (e->expr_pool);
169 expr_parse_any (struct lexer *lexer, struct dataset *ds, bool optimize)
172 struct expression *e;
174 e = expr_create (ds);
175 n = parse_expr (lexer, e);
183 n = expr_optimize (n, e);
184 return finish_expression (n, e);
187 /* Finishing up expression building. */
189 /* Height of an expression's stacks. */
192 int number_height; /* Height of number stack. */
193 int string_height; /* Height of string stack. */
196 /* Stack heights used by different kinds of arguments. */
197 static const struct stack_heights on_number_stack = {1, 0};
198 static const struct stack_heights on_string_stack = {0, 1};
199 static const struct stack_heights not_on_stack = {0, 0};
201 /* Returns the stack heights used by an atom of the given
203 static const struct stack_heights *
204 atom_type_stack (atom_type type)
206 assert (is_atom (type));
212 case OP_num_vec_elem:
213 return &on_number_stack;
216 return &on_string_stack;
227 return ¬_on_stack;
234 /* Measures the stack height needed for node N, supposing that
235 the stack height is initially *HEIGHT and updating *HEIGHT to
236 the final stack height. Updates *MAX, if necessary, to
237 reflect the maximum intermediate or final height. */
239 measure_stack (const struct expr_node *n,
240 struct stack_heights *height, struct stack_heights *max)
242 const struct stack_heights *return_height;
244 if (is_composite (n->type))
246 struct stack_heights args;
250 for (i = 0; i < n->n_args; i++)
251 measure_stack (n->args[i], &args, max);
253 return_height = atom_type_stack (operations[n->type].returns);
256 return_height = atom_type_stack (n->type);
258 height->number_height += return_height->number_height;
259 height->string_height += return_height->string_height;
261 if (height->number_height > max->number_height)
262 max->number_height = height->number_height;
263 if (height->string_height > max->string_height)
264 max->string_height = height->string_height;
267 /* Allocates stacks within E sufficient for evaluating node N. */
269 allocate_stacks (struct expr_node *n, struct expression *e)
271 struct stack_heights initial = {0, 0};
272 struct stack_heights max = {0, 0};
274 measure_stack (n, &initial, &max);
275 e->number_stack = pool_alloc (e->expr_pool,
276 sizeof *e->number_stack * max.number_height);
277 e->string_stack = pool_alloc (e->expr_pool,
278 sizeof *e->string_stack * max.string_height);
281 /* Finalizes expression E for evaluating node N. */
282 static struct expression *
283 finish_expression (struct expr_node *n, struct expression *e)
285 /* Allocate stacks. */
286 allocate_stacks (n, e);
288 /* Output postfix representation. */
291 /* The eval_pool might have been used for allocating strings
292 during optimization. We need to keep those strings around
293 for all subsequent evaluations, so start a new eval_pool. */
294 e->eval_pool = pool_create_subpool (e->expr_pool);
299 /* Verifies that expression E, whose root node is *N, can be
300 converted to type EXPECTED_TYPE, inserting a conversion at *N
301 if necessary. Returns true if successful, false on failure. */
303 type_check (const struct expression *e, const struct expr_node *n,
304 enum val_type expected_type)
306 atom_type actual_type = expr_node_returns (n);
308 switch (expected_type)
311 if (actual_type != OP_number && actual_type != OP_boolean)
313 msg_at (SE, expr_location (e, n),
314 _("Type mismatch: expression has type '%s', "
315 "but a numeric value is required."),
316 atom_type_name (actual_type));
322 if (actual_type != OP_string)
324 msg_at (SE, expr_location (e, n),
325 _("Type mismatch: expression has type '%s', "
326 "but a string value is required."),
327 atom_type_name (actual_type));
339 /* Recursive-descent expression parser. */
342 free_msg_location (void *loc_)
344 struct msg_location *loc = loc_;
345 msg_location_destroy (loc);
349 expr_location__ (struct expression *e,
350 const struct expr_node *node,
351 const struct msg_location **minp,
352 const struct msg_location **maxp)
354 struct msg_location *loc = node->location;
357 const struct msg_location *min = *minp;
360 || loc->start.line < min->start.line
361 || (loc->start.line == min->start.line
362 && loc->start.column < min->start.column)))
365 const struct msg_location *max = *maxp;
368 || loc->end.line > max->end.line
369 || (loc->end.line == max->end.line
370 && loc->end.column > max->end.column)))
376 if (is_composite (node->type))
377 for (size_t i = 0; i < node->n_args; i++)
378 expr_location__ (e, node->args[i], minp, maxp);
381 /* Returns the source code location corresponding to expression NODE, computing
382 it lazily if needed. */
383 const struct msg_location *
384 expr_location (const struct expression *e_, const struct expr_node *node_)
386 struct expr_node *node = CONST_CAST (struct expr_node *, node_);
392 struct expression *e = CONST_CAST (struct expression *, e_);
393 const struct msg_location *min = NULL;
394 const struct msg_location *max = NULL;
395 expr_location__ (e, node, &min, &max);
398 node->location = msg_location_dup (min);
399 node->location->end = max->end;
400 pool_register (e->expr_pool, free_msg_location, node->location);
403 return node->location;
406 /* Sets e->location to the tokens in S's lexer from offset START_OFS to the
407 token before the current one. Has no effect if E already has a location or
410 expr_add_location (struct lexer *lexer, struct expression *e,
411 int start_ofs, struct expr_node *node)
413 if (node && !node->location)
415 node->location = lex_ofs_location (lexer, start_ofs, lex_ofs (lexer) - 1);
416 pool_register (e->expr_pool, free_msg_location, node->location);
421 type_coercion__ (struct expression *e, struct expr_node *node, size_t arg_idx,
424 assert (!!do_coercion == (e != NULL));
429 struct expr_node **argp = &node->args[arg_idx];
430 struct expr_node *arg = *argp;
434 const struct operation *op = &operations[node->type];
435 atom_type required_type = op->args[MIN (arg_idx, op->n_args - 1)];
436 atom_type actual_type = expr_node_returns (arg);
437 if (actual_type == required_type)
443 switch (required_type)
446 if (actual_type == OP_boolean)
448 /* To enforce strict typing rules, insert Boolean to
449 numeric "conversion". This conversion is a no-op,
450 so it will be removed later. */
452 *argp = expr_allocate_unary (e, OP_BOOLEAN_TO_NUM, arg);
455 else if (actual_type == OP_num_vec_elem)
458 arg->type = OP_VEC_ELEM_NUM;
464 /* No coercion to string. */
468 if (actual_type == OP_number)
470 /* Convert numeric to boolean. */
472 *argp = expr_allocate_binary (e, OP_OPERAND_TO_BOOLEAN, arg,
473 expr_allocate_expr_node (e, node));
479 if (actual_type == OP_number)
481 /* Convert number to integer. */
483 *argp = expr_allocate_unary (e, OP_NUM_TO_INTEGER, arg);
489 /* We never coerce to OP_format, only to OP_ni_format or OP_no_format. */
493 if (arg->type == OP_format
494 && fmt_check_input (arg->format)
495 && fmt_check_type_compat (arg->format, VAL_NUMERIC))
498 arg->type = OP_ni_format;
504 if (arg->type == OP_format
505 && fmt_check_output (arg->format)
506 && fmt_check_type_compat (arg->format, VAL_NUMERIC))
509 arg->type = OP_no_format;
515 if (arg->type == OP_NUM_VAR)
518 *argp = arg->args[0];
524 if (arg->type == OP_STR_VAR)
527 *argp = arg->args[0];
533 if (arg->type == OP_NUM_VAR || arg->type == OP_STR_VAR)
536 *argp = arg->args[0];
542 if (arg->type == OP_number
543 && floor (arg->number) == arg->number
544 && arg->number > 0 && arg->number < INT_MAX)
547 *argp = expr_allocate_pos_int (e, arg->number);
559 type_coercion (struct expression *e, struct expr_node *node, size_t arg_idx)
561 return type_coercion__ (e, node, arg_idx, true);
565 is_coercible (const struct expr_node *node_, size_t arg_idx)
567 struct expr_node *node = CONST_CAST (struct expr_node *, node_);
568 return type_coercion__ (NULL, node, arg_idx, false);
571 /* How to parse an operator.
573 Some operators support both numeric and string operators. For those,
574 'num_op' and 'str_op' are both nonzero. Otherwise, only one 'num_op' is
575 nonzero. (PSPP doesn't have any string-only operators.) */
578 enum token_type token; /* Operator token. */
579 operation_type num_op; /* Operation for numeric operands (or 0). */
580 operation_type str_op; /* Operation for string operands (or 0). */
583 static operation_type
584 match_operator (struct lexer *lexer, const struct operator ops[], size_t n_ops,
585 const struct expr_node *lhs)
587 bool lhs_is_numeric = operations[lhs->type].returns != OP_string;
588 for (const struct operator *op = ops; op < ops + n_ops; op++)
589 if (lex_token (lexer) == op->token)
591 if (op->token != T_NEG_NUM)
594 return op->str_op && !lhs_is_numeric ? op->str_op : op->num_op;
600 operator_name (enum token_type token)
602 return token == T_NEG_NUM ? "-" : token_type_to_string (token);
605 static struct expr_node *
606 parse_binary_operators__ (struct lexer *lexer, struct expression *e,
607 const struct operator ops[], size_t n_ops,
608 parse_recursively_func *parse_next_level,
609 const char *chain_warning, struct expr_node *lhs)
611 for (int op_count = 0; ; op_count++)
613 enum token_type token = lex_token (lexer);
614 operation_type optype = match_operator (lexer, ops, n_ops, lhs);
617 if (op_count > 1 && chain_warning)
618 msg_at (SW, expr_location (e, lhs), "%s", chain_warning);
623 struct expr_node *rhs = parse_next_level (lexer, e);
627 struct expr_node *node = expr_allocate_binary (e, optype, lhs, rhs);
628 if (!is_coercible (node, 0) || !is_coercible (node, 1))
631 for (size_t i = 0; i < n_ops; i++)
632 if (ops[i].token == token)
633 both = ops[i].num_op && ops[i].str_op;
635 const char *name = operator_name (token);
637 msg_at (SE, expr_location (e, node),
638 _("Both operands of %s must have the same type."), name);
639 else if (operations[node->type].args[0] != OP_string)
640 msg_at (SE, expr_location (e, node),
641 _("Both operands of %s must be numeric."), name);
645 msg_at (SN, expr_location (e, node->args[0]),
646 _("This operand has type '%s'."),
647 atom_type_name (expr_node_returns (node->args[0])));
648 msg_at (SN, expr_location (e, node->args[1]),
649 _("This operand has type '%s'."),
650 atom_type_name (expr_node_returns (node->args[1])));
655 if (!type_coercion (e, node, 0) || !type_coercion (e, node, 1))
662 static struct expr_node *
663 parse_binary_operators (struct lexer *lexer, struct expression *e,
664 const struct operator ops[], size_t n_ops,
665 parse_recursively_func *parse_next_level,
666 const char *chain_warning)
668 struct expr_node *lhs = parse_next_level (lexer, e);
672 return parse_binary_operators__ (lexer, e, ops, n_ops, parse_next_level,
676 static struct expr_node *
677 parse_inverting_unary_operator (struct lexer *lexer, struct expression *e,
678 const struct operator *op,
679 parse_recursively_func *parse_next_level)
681 int start_ofs = lex_ofs (lexer);
682 unsigned int op_count = 0;
683 while (lex_match (lexer, op->token))
686 struct expr_node *inner = parse_next_level (lexer, e);
687 if (!inner || !op_count)
690 struct expr_node *outer = expr_allocate_unary (e, op->num_op, inner);
691 expr_add_location (lexer, e, start_ofs, outer);
693 if (!type_coercion (e, outer, 0))
695 assert (operations[outer->type].args[0] != OP_string);
697 const char *name = operator_name (op->token);
698 msg_at (SE, expr_location (e, outer),
699 _("The unary %s operator requires a numeric operand."), name);
701 msg_at (SN, expr_location (e, outer->args[0]),
702 _("The operand of %s has type '%s'."),
703 name, atom_type_name (expr_node_returns (outer->args[0])));
708 return op_count % 2 ? outer : outer->args[0];
711 /* Parses the OR level. */
712 static struct expr_node *
713 parse_or (struct lexer *lexer, struct expression *e)
715 static const struct operator op = { .token = T_OR, .num_op = OP_OR };
716 return parse_binary_operators (lexer, e, &op, 1, parse_and, NULL);
719 /* Parses the AND level. */
720 static struct expr_node *
721 parse_and (struct lexer *lexer, struct expression *e)
723 static const struct operator op = { .token = T_AND, .num_op = OP_AND };
725 return parse_binary_operators (lexer, e, &op, 1, parse_not, NULL);
728 /* Parses the NOT level. */
729 static struct expr_node *
730 parse_not (struct lexer *lexer, struct expression *e)
732 static const struct operator op = { .token = T_NOT, .num_op = OP_NOT };
733 return parse_inverting_unary_operator (lexer, e, &op, parse_rel);
736 /* Parse relational operators. */
737 static struct expr_node *
738 parse_rel (struct lexer *lexer, struct expression *e)
740 const char *chain_warning =
741 _("Chaining relational operators (e.g. `a < b < c') will "
742 "not produce the mathematically expected result. "
743 "Use the AND logical operator to fix the problem "
744 "(e.g. `a < b AND b < c'). "
745 "To disable this warning, insert parentheses.");
747 static const struct operator ops[] =
749 { .token = T_EQUALS, .num_op = OP_EQ, .str_op = OP_EQ_STRING },
750 { .token = T_EQ, .num_op = OP_EQ, .str_op = OP_EQ_STRING },
751 { .token = T_GE, .num_op = OP_GE, .str_op = OP_GE_STRING },
752 { .token = T_GT, .num_op = OP_GT, .str_op = OP_GT_STRING },
753 { .token = T_LE, .num_op = OP_LE, .str_op = OP_LE_STRING },
754 { .token = T_LT, .num_op = OP_LT, .str_op = OP_LT_STRING },
755 { .token = T_NE, .num_op = OP_NE, .str_op = OP_NE_STRING },
758 return parse_binary_operators (lexer, e, ops, sizeof ops / sizeof *ops,
759 parse_add, chain_warning);
762 /* Parses the addition and subtraction level. */
763 static struct expr_node *
764 parse_add (struct lexer *lexer, struct expression *e)
766 static const struct operator ops[] =
768 { .token = T_PLUS, .num_op = OP_ADD },
769 { .token = T_DASH, .num_op = OP_SUB },
770 { .token = T_NEG_NUM, .num_op = OP_ADD },
773 return parse_binary_operators (lexer, e, ops, sizeof ops / sizeof *ops,
777 /* Parses the multiplication and division level. */
778 static struct expr_node *
779 parse_mul (struct lexer *lexer, struct expression *e)
781 static const struct operator ops[] =
783 { .token = T_ASTERISK, .num_op = OP_MUL },
784 { .token = T_SLASH, .num_op = OP_DIV },
787 return parse_binary_operators (lexer, e, ops, sizeof ops / sizeof *ops,
791 /* Parses the unary minus level. */
792 static struct expr_node *
793 parse_neg (struct lexer *lexer, struct expression *e)
795 static const struct operator op = { .token = T_DASH, .num_op = OP_NEG };
796 return parse_inverting_unary_operator (lexer, e, &op, parse_exp);
799 static struct expr_node *
800 parse_exp (struct lexer *lexer, struct expression *e)
802 static const struct operator op = { .token = T_EXP, .num_op = OP_POW };
804 const char *chain_warning =
805 _("The exponentiation operator (`**') is left-associative: "
806 "`a**b**c' equals `(a**b)**c', not `a**(b**c)'. "
807 "To disable this warning, insert parentheses.");
809 if (lex_token (lexer) != T_NEG_NUM || lex_next_token (lexer, 1) != T_EXP)
810 return parse_binary_operators (lexer, e, &op, 1,
811 parse_primary, chain_warning);
813 /* Special case for situations like "-5**6", which must be parsed as
816 int start_ofs = lex_ofs (lexer);
817 struct expr_node *lhs = expr_allocate_number (e, -lex_tokval (lexer));
819 expr_add_location (lexer, e, start_ofs, lhs);
821 struct expr_node *node = parse_binary_operators__ (
822 lexer, e, &op, 1, parse_primary, chain_warning, lhs);
826 node = expr_allocate_unary (e, OP_NEG, node);
827 expr_add_location (lexer, e, start_ofs, node);
832 ymd_to_offset (int y, int m, int d)
835 double retval = calendar_gregorian_to_offset (
836 y, m, d, settings_get_fmt_settings (), &error);
839 msg (SE, "%s", error);
845 static struct expr_node *
846 expr_date (struct expression *e, int year_digits)
848 static const char *months[12] =
850 "JAN", "FEB", "MAR", "APR", "MAY", "JUN",
851 "JUL", "AUG", "SEP", "OCT", "NOV", "DEC",
854 time_t last_proc_time = time_of_last_procedure (e->ds);
855 struct tm *time = localtime (&last_proc_time);
857 char *tmp = (year_digits == 2
858 ? xasprintf ("%02d-%s-%02d", time->tm_mday, months[time->tm_mon],
860 : xasprintf ("%02d-%s-%04d", time->tm_mday, months[time->tm_mon],
861 time->tm_year + 1900));
863 struct substring s = ss_clone_pool (ss_cstr (tmp), e->expr_pool);
866 return expr_allocate_string (e, s);
869 /* Parses system variables. */
870 static struct expr_node *
871 parse_sysvar (struct lexer *lexer, struct expression *e)
873 if (lex_match_id (lexer, "$CASENUM"))
874 return expr_allocate_nullary (e, OP_CASENUM);
875 else if (lex_match_id (lexer, "$DATE"))
876 return expr_date (e, 2);
877 else if (lex_match_id (lexer, "$DATE11"))
878 return expr_date (e, 4);
879 else if (lex_match_id (lexer, "$TRUE"))
880 return expr_allocate_boolean (e, 1.0);
881 else if (lex_match_id (lexer, "$FALSE"))
882 return expr_allocate_boolean (e, 0.0);
883 else if (lex_match_id (lexer, "$SYSMIS"))
884 return expr_allocate_number (e, SYSMIS);
885 else if (lex_match_id (lexer, "$JDATE"))
887 time_t time = time_of_last_procedure (e->ds);
888 struct tm *tm = localtime (&time);
889 return expr_allocate_number (e, ymd_to_offset (tm->tm_year + 1900,
893 else if (lex_match_id (lexer, "$TIME"))
895 time_t time = time_of_last_procedure (e->ds);
896 struct tm *tm = localtime (&time);
897 return expr_allocate_number (e, ymd_to_offset (tm->tm_year + 1900,
900 + tm->tm_hour * 60 * 60.
904 else if (lex_match_id (lexer, "$LENGTH"))
905 return expr_allocate_number (e, settings_get_viewlength ());
906 else if (lex_match_id (lexer, "$WIDTH"))
907 return expr_allocate_number (e, settings_get_viewwidth ());
910 lex_error (lexer, _("Unknown system variable %s."), lex_tokcstr (lexer));
915 /* Parses numbers, varnames, etc. */
916 static struct expr_node *
917 parse_primary__ (struct lexer *lexer, struct expression *e)
919 switch (lex_token (lexer))
922 if (lex_next_token (lexer, 1) == T_LPAREN)
924 /* An identifier followed by a left parenthesis may be
925 a vector element reference. If not, it's a function
927 if (e->ds != NULL && dict_lookup_vector (dataset_dict (e->ds), lex_tokcstr (lexer)) != NULL)
928 return parse_vector_element (lexer, e);
930 return parse_function (lexer, e);
932 else if (lex_tokcstr (lexer)[0] == '$')
934 /* $ at the beginning indicates a system variable. */
935 return parse_sysvar (lexer, e);
937 else if (e->ds != NULL && dict_lookup_var (dataset_dict (e->ds), lex_tokcstr (lexer)))
939 /* It looks like a user variable.
940 (It could be a format specifier, but we'll assume
941 it's a variable unless proven otherwise. */
942 return allocate_unary_variable (e, parse_variable (lexer, dataset_dict (e->ds)));
946 /* Try to parse it as a format specifier. */
951 ok = parse_format_specifier (lexer, &fmt);
955 return expr_allocate_format (e, fmt);
957 /* All attempts failed. */
958 lex_error (lexer, _("Unknown identifier %s."), lex_tokcstr (lexer));
966 struct expr_node *node = expr_allocate_number (e, lex_tokval (lexer));
973 const char *dict_encoding;
974 struct expr_node *node;
977 dict_encoding = (e->ds != NULL
978 ? dict_get_encoding (dataset_dict (e->ds))
980 s = recode_string_pool (dict_encoding, "UTF-8", lex_tokcstr (lexer),
981 ss_length (lex_tokss (lexer)), e->expr_pool);
982 node = expr_allocate_string (e, ss_cstr (s));
991 struct expr_node *node = parse_or (lexer, e);
992 return !node || !lex_force_match (lexer, T_RPAREN) ? NULL : node;
996 lex_error (lexer, _("Syntax error parsing expression."));
1001 static struct expr_node *
1002 parse_primary (struct lexer *lexer, struct expression *e)
1004 int start_ofs = lex_ofs (lexer);
1005 struct expr_node *node = parse_primary__ (lexer, e);
1006 expr_add_location (lexer, e, start_ofs, node);
1010 static struct expr_node *
1011 parse_vector_element (struct lexer *lexer, struct expression *e)
1013 int vector_start_ofs = lex_ofs (lexer);
1015 /* Find vector, skip token.
1016 The caller must already have verified that the current token
1017 is the name of a vector. */
1018 const struct vector *vector = dict_lookup_vector (dataset_dict (e->ds),
1019 lex_tokcstr (lexer));
1020 assert (vector != NULL);
1023 /* Skip left parenthesis token.
1024 The caller must have verified that the lookahead is a left
1026 assert (lex_token (lexer) == T_LPAREN);
1029 int element_start_ofs = lex_ofs (lexer);
1030 struct expr_node *element = parse_or (lexer, e);
1033 expr_add_location (lexer, e, element_start_ofs, element);
1035 if (!lex_match (lexer, T_RPAREN))
1038 operation_type type = (vector_get_type (vector) == VAL_NUMERIC
1039 ? OP_VEC_ELEM_NUM_RAW : OP_VEC_ELEM_STR);
1040 struct expr_node *node = expr_allocate_binary (
1041 e, type, element, expr_allocate_vector (e, vector));
1042 expr_add_location (lexer, e, vector_start_ofs, node);
1044 if (!type_coercion (e, node, 0))
1046 msg_at (SE, expr_location (e, node),
1047 _("A vector index must be numeric."));
1049 msg_at (SN, expr_location (e, node->args[0]),
1050 _("This vector index has type '%s'."),
1051 atom_type_name (expr_node_returns (node->args[0])));
1059 /* Individual function parsing. */
1061 const struct operation operations[OP_first + n_OP] = {
1062 #include "parse.inc"
1066 word_matches (const char **test, const char **name)
1068 size_t test_len = strcspn (*test, ".");
1069 size_t name_len = strcspn (*name, ".");
1070 if (test_len == name_len)
1072 if (buf_compare_case (*test, *name, test_len))
1075 else if (test_len < 3 || test_len > name_len)
1079 if (buf_compare_case (*test, *name, test_len))
1085 if (**test != **name)
1096 /* Returns 0 if TOKEN and FUNC do not match,
1097 1 if TOKEN is an acceptable abbreviation for FUNC,
1098 2 if TOKEN equals FUNC. */
1100 compare_function_names (const char *token_, const char *func_)
1102 const char *token = token_;
1103 const char *func = func_;
1104 while (*token || *func)
1105 if (!word_matches (&token, &func))
1107 return !c_strcasecmp (token_, func_) ? 2 : 1;
1111 lookup_function (const char *token,
1112 const struct operation **first,
1113 const struct operation **last)
1115 *first = *last = NULL;
1116 const struct operation *best = NULL;
1118 for (const struct operation *f = operations + OP_function_first;
1119 f <= operations + OP_function_last; f++)
1121 int score = compare_function_names (token, f->name);
1127 else if (score == 1 && !(f->flags & OPF_NO_ABBREV) && !best)
1136 const struct operation *f = best;
1137 while (f <= operations + OP_function_last
1138 && !c_strcasecmp (f->name, best->name))
1146 extract_min_valid (const char *s)
1148 char *p = strrchr (s, '.');
1150 || p[1] < '0' || p[1] > '9'
1151 || strspn (p + 1, "0123456789") != strlen (p + 1))
1154 return atoi (p + 1);
1158 match_function__ (struct expr_node *node, const struct operation *f)
1160 if (node->n_args < f->n_args
1161 || (node->n_args > f->n_args && (f->flags & OPF_ARRAY_OPERAND) == 0)
1162 || node->n_args - (f->n_args - 1) < f->array_min_elems)
1165 node->type = f - operations;
1166 for (size_t i = 0; i < node->n_args; i++)
1167 if (!is_coercible (node, i))
1173 static const struct operation *
1174 match_function (struct expr_node *node,
1175 const struct operation *first, const struct operation *last)
1177 for (const struct operation *f = first; f < last; f++)
1178 if (match_function__ (node, f))
1184 validate_function_args (const struct expression *e, const struct expr_node *n,
1185 const struct operation *f, int n_args, int min_valid)
1187 /* Count the function arguments that go into the trailing array (if any). We
1188 know that there must be at least the minimum number because
1189 match_function() already checked. */
1190 int array_n_args = n_args - (f->n_args - 1);
1191 assert (array_n_args >= f->array_min_elems);
1193 if ((f->flags & OPF_ARRAY_OPERAND)
1194 && array_n_args % f->array_granularity != 0)
1196 /* RANGE is the only case we have so far. It has paired arguments with
1197 one initial argument, and that's the only special case we deal with
1199 assert (f->array_granularity == 2);
1200 assert (n_args % 2 == 0);
1201 msg_at (SE, expr_location (e, n),
1202 _("%s must have an odd number of arguments."), f->prototype);
1206 if (min_valid != -1)
1208 if (f->array_min_elems == 0)
1210 assert ((f->flags & OPF_MIN_VALID) == 0);
1211 msg_at (SE, expr_location (e, n),
1212 _("%s function cannot accept suffix .%d to specify the "
1213 "minimum number of valid arguments."),
1214 f->prototype, min_valid);
1219 assert (f->flags & OPF_MIN_VALID);
1220 if (min_valid > array_n_args)
1222 msg_at (SE, expr_location (e, n),
1223 _("For %s with %d arguments, at most %d (not %d) may be "
1224 "required to be valid."),
1225 f->prototype, n_args, array_n_args, min_valid);
1235 add_arg (struct expr_node ***args, size_t *n_args, size_t *allocated_args,
1236 struct expr_node *arg,
1237 struct expression *e, struct lexer *lexer, int arg_start_ofs)
1239 if (*n_args >= *allocated_args)
1240 *args = x2nrealloc (*args, allocated_args, sizeof **args);
1242 expr_add_location (lexer, e, arg_start_ofs, arg);
1243 (*args)[(*n_args)++] = arg;
1247 put_invocation (struct string *s,
1248 const char *func_name, struct expr_node *node)
1252 ds_put_format (s, "%s(", func_name);
1253 for (i = 0; i < node->n_args; i++)
1256 ds_put_cstr (s, ", ");
1257 ds_put_cstr (s, operations[expr_node_returns (node->args[i])].prototype);
1259 ds_put_byte (s, ')');
1263 no_match (struct expression *e, const char *func_name, struct expr_node *node,
1264 const struct operation *ops, size_t n)
1272 ds_put_format (&s, _("Type mismatch invoking %s as "), ops->prototype);
1273 put_invocation (&s, func_name, node);
1277 ds_put_cstr (&s, _("Function invocation "));
1278 put_invocation (&s, func_name, node);
1279 ds_put_cstr (&s, _(" does not match any known function. Candidates are:"));
1281 for (size_t i = 0; i < n; i++)
1282 ds_put_format (&s, "\n%s", ops[i].prototype);
1284 ds_put_byte (&s, '.');
1286 msg_at (SE, expr_location (e, node), "%s", ds_cstr (&s));
1288 if (n == 1 && ops->n_args == node->n_args)
1290 for (size_t i = 0; i < node->n_args; i++)
1291 if (!is_coercible (node, i))
1293 atom_type expected = ops->args[i];
1294 atom_type actual = expr_node_returns (node->args[i]);
1295 if ((expected == OP_ni_format || expected == OP_no_format)
1296 && actual == OP_format)
1298 struct fmt_spec f = node->args[i]->format;
1299 char *error = fmt_check__ (f, (ops->args[i] == OP_ni_format
1300 ? FMT_FOR_INPUT : FMT_FOR_OUTPUT));
1302 error = fmt_check_type_compat__ (f, NULL, VAL_NUMERIC);
1305 msg_at (SN, expr_location (e, node->args[i]), "%s", error);
1310 msg_at (SN, expr_location (e, node->args[i]),
1311 _("This argument has type '%s' but '%s' is required."),
1312 atom_type_name (actual), atom_type_name (expected));
1319 static struct expr_node *
1320 parse_function (struct lexer *lexer, struct expression *e)
1322 struct string func_name;
1323 ds_init_substring (&func_name, lex_tokss (lexer));
1325 int min_valid = extract_min_valid (lex_tokcstr (lexer));
1327 const struct operation *first, *last;
1328 if (!lookup_function (lex_tokcstr (lexer), &first, &last))
1330 lex_error (lexer, _("No function or vector named %s."),
1331 lex_tokcstr (lexer));
1332 ds_destroy (&func_name);
1336 int func_start_ofs = lex_ofs (lexer);
1338 if (!lex_force_match (lexer, T_LPAREN))
1340 ds_destroy (&func_name);
1344 struct expr_node **args = NULL;
1346 size_t allocated_args = 0;
1347 if (lex_token (lexer) != T_RPAREN)
1350 int arg_start_ofs = lex_ofs (lexer);
1351 if (lex_token (lexer) == T_ID
1352 && lex_next_token (lexer, 1) == T_TO)
1354 const struct variable **vars;
1357 if (!parse_variables_const (lexer, dataset_dict (e->ds),
1358 &vars, &n_vars, PV_SINGLE))
1360 for (size_t i = 0; i < n_vars; i++)
1361 add_arg (&args, &n_args, &allocated_args,
1362 allocate_unary_variable (e, vars[i]),
1363 e, lexer, arg_start_ofs);
1368 struct expr_node *arg = parse_or (lexer, e);
1372 add_arg (&args, &n_args, &allocated_args, arg,
1373 e, lexer, arg_start_ofs);
1375 if (lex_match (lexer, T_RPAREN))
1377 else if (!lex_match (lexer, T_COMMA))
1379 lex_error_expecting (lexer, "`,'", "`)'");
1384 struct expr_node *n = expr_allocate_composite (e, first - operations,
1386 expr_add_location (lexer, e, func_start_ofs, n);
1387 const struct operation *f = match_function (n, first, last);
1390 no_match (e, ds_cstr (&func_name), n, first, last - first);
1393 n->type = f - operations;
1394 n->min_valid = min_valid != -1 ? min_valid : f->array_min_elems;
1396 for (size_t i = 0; i < n_args; i++)
1397 if (!type_coercion (e, n, i))
1399 /* Unreachable because match_function already checked that the
1400 arguments were coercible. */
1403 if (!validate_function_args (e, n, f, n_args, min_valid))
1406 if ((f->flags & OPF_EXTENSION) && settings_get_syntax () == COMPATIBLE)
1407 msg_at (SW, expr_location (e, n),
1408 _("%s is a PSPP extension."), f->prototype);
1409 if (f->flags & OPF_UNIMPLEMENTED)
1411 msg_at (SE, expr_location (e, n),
1412 _("%s is not available in this version of PSPP."), f->prototype);
1415 if ((f->flags & OPF_PERM_ONLY) &&
1416 proc_in_temporary_transformations (e->ds))
1418 msg_at (SE, expr_location (e, n),
1419 _("%s may not appear after %s."), f->prototype, "TEMPORARY");
1423 if (n->type == OP_LAG_Vn || n->type == OP_LAG_Vs)
1424 dataset_need_lag (e->ds, 1);
1425 else if (n->type == OP_LAG_Vnn || n->type == OP_LAG_Vsn)
1427 assert (n->n_args == 2);
1428 assert (n->args[1]->type == OP_pos_int);
1429 dataset_need_lag (e->ds, n->args[1]->integer);
1433 ds_destroy (&func_name);
1438 ds_destroy (&func_name);
1442 /* Utility functions. */
1444 static struct expression *
1445 expr_create (struct dataset *ds)
1447 struct pool *pool = pool_create ();
1448 struct expression *e = pool_alloc (pool, sizeof *e);
1449 *e = (struct expression) {
1452 .eval_pool = pool_create_subpool (pool),
1458 expr_node_returns (const struct expr_node *n)
1461 assert (is_operation (n->type));
1462 if (is_atom (n->type))
1464 else if (is_composite (n->type))
1465 return operations[n->type].returns;
1471 atom_type_name (atom_type type)
1473 assert (is_atom (type));
1475 /* The Boolean type is purely an internal concept that the documentation
1476 doesn't mention, so it might confuse users if we talked about them in
1478 return type == OP_boolean ? "number" : operations[type].name;
1482 expr_allocate_nullary (struct expression *e, operation_type op)
1484 return expr_allocate_composite (e, op, NULL, 0);
1488 expr_allocate_unary (struct expression *e, operation_type op,
1489 struct expr_node *arg0)
1491 return expr_allocate_composite (e, op, &arg0, 1);
1495 expr_allocate_binary (struct expression *e, operation_type op,
1496 struct expr_node *arg0, struct expr_node *arg1)
1498 struct expr_node *args[2];
1501 return expr_allocate_composite (e, op, args, 2);
1505 expr_allocate_composite (struct expression *e, operation_type op,
1506 struct expr_node **args, size_t n_args)
1508 for (size_t i = 0; i < n_args; i++)
1512 struct expr_node *n = pool_alloc (e->expr_pool, sizeof *n);
1513 *n = (struct expr_node) {
1516 .args = pool_clone (e->expr_pool, args, sizeof *n->args * n_args),
1522 expr_allocate_number (struct expression *e, double d)
1524 struct expr_node *n = pool_alloc (e->expr_pool, sizeof *n);
1525 *n = (struct expr_node) { .type = OP_number, .number = d };
1530 expr_allocate_boolean (struct expression *e, double b)
1532 assert (b == 0.0 || b == 1.0 || b == SYSMIS);
1534 struct expr_node *n = pool_alloc (e->expr_pool, sizeof *n);
1535 *n = (struct expr_node) { .type = OP_boolean, .number = b };
1540 expr_allocate_integer (struct expression *e, int i)
1542 struct expr_node *n = pool_alloc (e->expr_pool, sizeof *n);
1543 *n = (struct expr_node) { .type = OP_integer, .integer = i };
1548 expr_allocate_pos_int (struct expression *e, int i)
1552 struct expr_node *n = pool_alloc (e->expr_pool, sizeof *n);
1553 *n = (struct expr_node) { .type = OP_pos_int, .integer = i };
1558 expr_allocate_vector (struct expression *e, const struct vector *vector)
1560 struct expr_node *n = pool_alloc (e->expr_pool, sizeof *n);
1561 *n = (struct expr_node) { .type = OP_vector, .vector = vector };
1566 expr_allocate_string (struct expression *e, struct substring s)
1568 struct expr_node *n = pool_alloc (e->expr_pool, sizeof *n);
1569 *n = (struct expr_node) { .type = OP_string, .string = s };
1574 expr_allocate_variable (struct expression *e, const struct variable *v)
1576 struct expr_node *n = pool_alloc (e->expr_pool, sizeof *n);
1577 *n = (struct expr_node) {
1578 .type = var_is_numeric (v) ? OP_num_var : OP_str_var,
1585 expr_allocate_format (struct expression *e, struct fmt_spec format)
1587 struct expr_node *n = pool_alloc (e->expr_pool, sizeof *n);
1588 *n = (struct expr_node) { .type = OP_format, .format = format };
1593 expr_allocate_expr_node (struct expression *e,
1594 const struct expr_node *expr_node)
1596 struct expr_node *n = pool_alloc (e->expr_pool, sizeof *n);
1597 *n = (struct expr_node) { .type = OP_expr_node, .expr_node = expr_node };
1601 /* Allocates a unary composite node that represents the value of
1602 variable V in expression E. */
1603 static struct expr_node *
1604 allocate_unary_variable (struct expression *e, const struct variable *v)
1607 return expr_allocate_unary (e, var_is_numeric (v) ? OP_NUM_VAR : OP_STR_VAR,
1608 expr_allocate_variable (e, v));
1611 /* Export function details to other modules. */
1613 /* Returns the operation structure for the function with the
1615 const struct operation *
1616 expr_get_function (size_t idx)
1618 assert (idx < n_OP_function);
1619 return &operations[OP_function_first + idx];
1622 /* Returns the number of expression functions. */
1624 expr_get_n_functions (void)
1626 return n_OP_function;
1629 /* Returns the name of operation OP. */
1631 expr_operation_get_name (const struct operation *op)
1636 /* Returns the human-readable prototype for operation OP. */
1638 expr_operation_get_prototype (const struct operation *op)
1640 return op->prototype;
1643 /* Returns the number of arguments for operation OP. */
1645 expr_operation_get_n_args (const struct operation *op)