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. (This allows for a better error message if the
129 expression is not numeric.) Otherwise similar to expr_parse(). */
131 expr_parse_new_variable (struct lexer *lexer, struct dataset *ds,
132 const char *new_var_name)
134 struct expression *e = expr_create (ds);
135 struct expr_node *n = parse_expr (lexer, e);
142 atom_type actual_type = expr_node_returns (n);
143 if (actual_type != OP_number && actual_type != OP_boolean)
145 msg (SE, _("This command tries to create a new variable %s by assigning a "
146 "string value to it, but this is not supported. Use "
147 "the STRING command to create the new variable with the "
148 "correct width before assigning to it, e.g. STRING %s(A20)."),
149 new_var_name, new_var_name);
154 return finish_expression (expr_optimize (n, e), e);
157 /* Free expression E. */
159 expr_free (struct expression *e)
162 pool_destroy (e->expr_pool);
166 expr_parse_any (struct lexer *lexer, struct dataset *ds, bool optimize)
169 struct expression *e;
171 e = expr_create (ds);
172 n = parse_expr (lexer, e);
180 n = expr_optimize (n, e);
181 return finish_expression (n, e);
184 /* Finishing up expression building. */
186 /* Height of an expression's stacks. */
189 int number_height; /* Height of number stack. */
190 int string_height; /* Height of string stack. */
193 /* Stack heights used by different kinds of arguments. */
194 static const struct stack_heights on_number_stack = {1, 0};
195 static const struct stack_heights on_string_stack = {0, 1};
196 static const struct stack_heights not_on_stack = {0, 0};
198 /* Returns the stack heights used by an atom of the given
200 static const struct stack_heights *
201 atom_type_stack (atom_type type)
203 assert (is_atom (type));
209 case OP_num_vec_elem:
210 return &on_number_stack;
213 return &on_string_stack;
224 return ¬_on_stack;
231 /* Measures the stack height needed for node N, supposing that
232 the stack height is initially *HEIGHT and updating *HEIGHT to
233 the final stack height. Updates *MAX, if necessary, to
234 reflect the maximum intermediate or final height. */
236 measure_stack (const struct expr_node *n,
237 struct stack_heights *height, struct stack_heights *max)
239 const struct stack_heights *return_height;
241 if (is_composite (n->type))
243 struct stack_heights args;
247 for (i = 0; i < n->n_args; i++)
248 measure_stack (n->args[i], &args, max);
250 return_height = atom_type_stack (operations[n->type].returns);
253 return_height = atom_type_stack (n->type);
255 height->number_height += return_height->number_height;
256 height->string_height += return_height->string_height;
258 if (height->number_height > max->number_height)
259 max->number_height = height->number_height;
260 if (height->string_height > max->string_height)
261 max->string_height = height->string_height;
264 /* Allocates stacks within E sufficient for evaluating node N. */
266 allocate_stacks (struct expr_node *n, struct expression *e)
268 struct stack_heights initial = {0, 0};
269 struct stack_heights max = {0, 0};
271 measure_stack (n, &initial, &max);
272 e->number_stack = pool_alloc (e->expr_pool,
273 sizeof *e->number_stack * max.number_height);
274 e->string_stack = pool_alloc (e->expr_pool,
275 sizeof *e->string_stack * max.string_height);
278 /* Finalizes expression E for evaluating node N. */
279 static struct expression *
280 finish_expression (struct expr_node *n, struct expression *e)
282 /* Allocate stacks. */
283 allocate_stacks (n, e);
285 /* Output postfix representation. */
288 /* The eval_pool might have been used for allocating strings
289 during optimization. We need to keep those strings around
290 for all subsequent evaluations, so start a new eval_pool. */
291 e->eval_pool = pool_create_subpool (e->expr_pool);
296 /* Verifies that expression E, whose root node is *N, can be
297 converted to type EXPECTED_TYPE, inserting a conversion at *N
298 if necessary. Returns true if successful, false on failure. */
300 type_check (const struct expression *e, const struct expr_node *n,
301 enum val_type expected_type)
303 atom_type actual_type = expr_node_returns (n);
305 switch (expected_type)
308 if (actual_type != OP_number && actual_type != OP_boolean)
310 msg_at (SE, expr_location (e, n),
311 _("Type mismatch: expression has type '%s', "
312 "but a numeric value is required."),
313 atom_type_name (actual_type));
319 if (actual_type != OP_string)
321 msg_at (SE, expr_location (e, n),
322 _("Type mismatch: expression has type '%s', "
323 "but a string value is required."),
324 atom_type_name (actual_type));
336 /* Recursive-descent expression parser. */
339 free_msg_location (void *loc_)
341 struct msg_location *loc = loc_;
342 msg_location_destroy (loc);
346 expr_location__ (struct expression *e,
347 const struct expr_node *node,
348 const struct msg_location **minp,
349 const struct msg_location **maxp)
351 struct msg_location *loc = node->location;
354 const struct msg_location *min = *minp;
357 || loc->start.line < min->start.line
358 || (loc->start.line == min->start.line
359 && loc->start.column < min->start.column)))
362 const struct msg_location *max = *maxp;
365 || loc->end.line > max->end.line
366 || (loc->end.line == max->end.line
367 && loc->end.column > max->end.column)))
373 if (is_composite (node->type))
374 for (size_t i = 0; i < node->n_args; i++)
375 expr_location__ (e, node->args[i], minp, maxp);
378 /* Returns the source code location corresponding to expression NODE, computing
379 it lazily if needed. */
380 const struct msg_location *
381 expr_location (const struct expression *e_, const struct expr_node *node_)
383 struct expr_node *node = CONST_CAST (struct expr_node *, node_);
389 struct expression *e = CONST_CAST (struct expression *, e_);
390 const struct msg_location *min = NULL;
391 const struct msg_location *max = NULL;
392 expr_location__ (e, node, &min, &max);
395 node->location = msg_location_dup (min);
396 node->location->end = max->end;
397 pool_register (e->expr_pool, free_msg_location, node->location);
400 return node->location;
403 /* Sets e->location to the tokens in S's lexer from offset START_OFS to the
404 token before the current one. Has no effect if E already has a location or
407 expr_add_location (struct lexer *lexer, struct expression *e,
408 int start_ofs, struct expr_node *node)
410 if (node && !node->location)
412 node->location = lex_ofs_location (lexer, start_ofs, lex_ofs (lexer) - 1);
413 pool_register (e->expr_pool, free_msg_location, node->location);
418 type_coercion__ (struct expression *e, struct expr_node *node, size_t arg_idx,
421 assert (!!do_coercion == (e != NULL));
426 struct expr_node **argp = &node->args[arg_idx];
427 struct expr_node *arg = *argp;
431 const struct operation *op = &operations[node->type];
432 atom_type required_type = op->args[MIN (arg_idx, op->n_args - 1)];
433 atom_type actual_type = expr_node_returns (arg);
434 if (actual_type == required_type)
440 switch (required_type)
443 if (actual_type == OP_boolean)
445 /* To enforce strict typing rules, insert Boolean to
446 numeric "conversion". This conversion is a no-op,
447 so it will be removed later. */
449 *argp = expr_allocate_unary (e, OP_BOOLEAN_TO_NUM, arg);
452 else if (actual_type == OP_num_vec_elem)
455 arg->type = OP_VEC_ELEM_NUM;
461 /* No coercion to string. */
465 if (actual_type == OP_number)
467 /* Convert numeric to boolean. */
469 *argp = expr_allocate_binary (e, OP_OPERAND_TO_BOOLEAN, arg,
470 expr_allocate_expr_node (e, node));
476 if (actual_type == OP_number)
478 /* Convert number to integer. */
480 *argp = expr_allocate_unary (e, OP_NUM_TO_INTEGER, arg);
486 /* We never coerce to OP_format, only to OP_ni_format or OP_no_format. */
491 if (arg->type == OP_format
492 && fmt_check_input (&arg->format)
493 && fmt_check_type_compat (&arg->format, VAL_NUMERIC))
497 arg->type = OP_ni_format;
505 if (arg->type == OP_format
506 && fmt_check_output (&arg->format)
507 && fmt_check_type_compat (&arg->format, VAL_NUMERIC))
511 arg->type = OP_no_format;
518 if (arg->type == OP_NUM_VAR)
521 *argp = arg->args[0];
527 if (arg->type == OP_STR_VAR)
530 *argp = arg->args[0];
536 if (arg->type == OP_NUM_VAR || arg->type == OP_STR_VAR)
539 *argp = arg->args[0];
545 if (arg->type == OP_number
546 && floor (arg->number) == arg->number
547 && arg->number > 0 && arg->number < INT_MAX)
550 *argp = expr_allocate_pos_int (e, arg->number);
562 type_coercion (struct expression *e, struct expr_node *node, size_t arg_idx)
564 return type_coercion__ (e, node, arg_idx, true);
568 is_coercible (const struct expr_node *node_, size_t arg_idx)
570 struct expr_node *node = CONST_CAST (struct expr_node *, node_);
571 return type_coercion__ (NULL, node, arg_idx, false);
574 /* How to parse an operator.
576 Some operators support both numeric and string operators. For those,
577 'num_op' and 'str_op' are both nonzero. Otherwise, only one 'num_op' is
578 nonzero. (PSPP doesn't have any string-only operators.) */
581 enum token_type token; /* Operator token. */
582 operation_type num_op; /* Operation for numeric operands (or 0). */
583 operation_type str_op; /* Operation for string operands (or 0). */
586 static operation_type
587 match_operator (struct lexer *lexer, const struct operator ops[], size_t n_ops,
588 const struct expr_node *lhs)
590 bool lhs_is_numeric = operations[lhs->type].returns != OP_string;
591 for (const struct operator *op = ops; op < ops + n_ops; op++)
592 if (lex_token (lexer) == op->token)
594 if (op->token != T_NEG_NUM)
597 return op->str_op && !lhs_is_numeric ? op->str_op : op->num_op;
603 operator_name (enum token_type token)
605 return token == T_NEG_NUM ? "-" : token_type_to_string (token);
608 static struct expr_node *
609 parse_binary_operators__ (struct lexer *lexer, struct expression *e,
610 const struct operator ops[], size_t n_ops,
611 parse_recursively_func *parse_next_level,
612 const char *chain_warning, struct expr_node *lhs)
614 for (int op_count = 0; ; op_count++)
616 enum token_type token = lex_token (lexer);
617 operation_type optype = match_operator (lexer, ops, n_ops, lhs);
620 if (op_count > 1 && chain_warning)
621 msg_at (SW, expr_location (e, lhs), "%s", chain_warning);
626 struct expr_node *rhs = parse_next_level (lexer, e);
630 struct expr_node *node = expr_allocate_binary (e, optype, lhs, rhs);
631 if (!is_coercible (node, 0) || !is_coercible (node, 1))
634 for (size_t i = 0; i < n_ops; i++)
635 if (ops[i].token == token)
636 both = ops[i].num_op && ops[i].str_op;
638 const char *name = operator_name (token);
640 msg_at (SE, expr_location (e, node),
641 _("Both operands of %s must have the same type."), name);
642 else if (operations[node->type].args[0] != OP_string)
643 msg_at (SE, expr_location (e, node),
644 _("Both operands of %s must be numeric."), name);
648 msg_at (SN, expr_location (e, node->args[0]),
649 _("This operand has type '%s'."),
650 atom_type_name (expr_node_returns (node->args[0])));
651 msg_at (SN, expr_location (e, node->args[1]),
652 _("This operand has type '%s'."),
653 atom_type_name (expr_node_returns (node->args[1])));
658 if (!type_coercion (e, node, 0) || !type_coercion (e, node, 1))
665 static struct expr_node *
666 parse_binary_operators (struct lexer *lexer, struct expression *e,
667 const struct operator ops[], size_t n_ops,
668 parse_recursively_func *parse_next_level,
669 const char *chain_warning)
671 struct expr_node *lhs = parse_next_level (lexer, e);
675 return parse_binary_operators__ (lexer, e, ops, n_ops, parse_next_level,
679 static struct expr_node *
680 parse_inverting_unary_operator (struct lexer *lexer, struct expression *e,
681 const struct operator *op,
682 parse_recursively_func *parse_next_level)
684 int start_ofs = lex_ofs (lexer);
685 unsigned int op_count = 0;
686 while (lex_match (lexer, op->token))
689 struct expr_node *inner = parse_next_level (lexer, e);
690 if (!inner || !op_count)
693 struct expr_node *outer = expr_allocate_unary (e, op->num_op, inner);
694 expr_add_location (lexer, e, start_ofs, outer);
696 if (!type_coercion (e, outer, 0))
698 assert (operations[outer->type].args[0] != OP_string);
700 const char *name = operator_name (op->token);
701 msg_at (SE, expr_location (e, outer),
702 _("The unary %s operator requires a numeric operand."), name);
704 msg_at (SN, expr_location (e, outer->args[0]),
705 _("The operand of %s has type '%s'."),
706 name, atom_type_name (expr_node_returns (outer->args[0])));
711 return op_count % 2 ? outer : outer->args[0];
714 /* Parses the OR level. */
715 static struct expr_node *
716 parse_or (struct lexer *lexer, struct expression *e)
718 static const struct operator op = { .token = T_OR, .num_op = OP_OR };
719 return parse_binary_operators (lexer, e, &op, 1, parse_and, NULL);
722 /* Parses the AND level. */
723 static struct expr_node *
724 parse_and (struct lexer *lexer, struct expression *e)
726 static const struct operator op = { .token = T_AND, .num_op = OP_AND };
728 return parse_binary_operators (lexer, e, &op, 1, parse_not, NULL);
731 /* Parses the NOT level. */
732 static struct expr_node *
733 parse_not (struct lexer *lexer, struct expression *e)
735 static const struct operator op = { .token = T_NOT, .num_op = OP_NOT };
736 return parse_inverting_unary_operator (lexer, e, &op, parse_rel);
739 /* Parse relational operators. */
740 static struct expr_node *
741 parse_rel (struct lexer *lexer, struct expression *e)
743 const char *chain_warning =
744 _("Chaining relational operators (e.g. `a < b < c') will "
745 "not produce the mathematically expected result. "
746 "Use the AND logical operator to fix the problem "
747 "(e.g. `a < b AND b < c'). "
748 "To disable this warning, insert parentheses.");
750 static const struct operator ops[] =
752 { .token = T_EQUALS, .num_op = OP_EQ, .str_op = OP_EQ_STRING },
753 { .token = T_EQ, .num_op = OP_EQ, .str_op = OP_EQ_STRING },
754 { .token = T_GE, .num_op = OP_GE, .str_op = OP_GE_STRING },
755 { .token = T_GT, .num_op = OP_GT, .str_op = OP_GT_STRING },
756 { .token = T_LE, .num_op = OP_LE, .str_op = OP_LE_STRING },
757 { .token = T_LT, .num_op = OP_LT, .str_op = OP_LT_STRING },
758 { .token = T_NE, .num_op = OP_NE, .str_op = OP_NE_STRING },
761 return parse_binary_operators (lexer, e, ops, sizeof ops / sizeof *ops,
762 parse_add, chain_warning);
765 /* Parses the addition and subtraction level. */
766 static struct expr_node *
767 parse_add (struct lexer *lexer, struct expression *e)
769 static const struct operator ops[] =
771 { .token = T_PLUS, .num_op = OP_ADD },
772 { .token = T_DASH, .num_op = OP_SUB },
773 { .token = T_NEG_NUM, .num_op = OP_ADD },
776 return parse_binary_operators (lexer, e, ops, sizeof ops / sizeof *ops,
780 /* Parses the multiplication and division level. */
781 static struct expr_node *
782 parse_mul (struct lexer *lexer, struct expression *e)
784 static const struct operator ops[] =
786 { .token = T_ASTERISK, .num_op = OP_MUL },
787 { .token = T_SLASH, .num_op = OP_DIV },
790 return parse_binary_operators (lexer, e, ops, sizeof ops / sizeof *ops,
794 /* Parses the unary minus level. */
795 static struct expr_node *
796 parse_neg (struct lexer *lexer, struct expression *e)
798 static const struct operator op = { .token = T_DASH, .num_op = OP_NEG };
799 return parse_inverting_unary_operator (lexer, e, &op, parse_exp);
802 static struct expr_node *
803 parse_exp (struct lexer *lexer, struct expression *e)
805 static const struct operator op = { .token = T_EXP, .num_op = OP_POW };
807 const char *chain_warning =
808 _("The exponentiation operator (`**') is left-associative: "
809 "`a**b**c' equals `(a**b)**c', not `a**(b**c)'. "
810 "To disable this warning, insert parentheses.");
812 if (lex_token (lexer) != T_NEG_NUM || lex_next_token (lexer, 1) != T_EXP)
813 return parse_binary_operators (lexer, e, &op, 1,
814 parse_primary, chain_warning);
816 /* Special case for situations like "-5**6", which must be parsed as
819 int start_ofs = lex_ofs (lexer);
820 struct expr_node *lhs = expr_allocate_number (e, -lex_tokval (lexer));
822 expr_add_location (lexer, e, start_ofs, lhs);
824 struct expr_node *node = parse_binary_operators__ (
825 lexer, e, &op, 1, parse_primary, chain_warning, lhs);
829 node = expr_allocate_unary (e, OP_NEG, node);
830 expr_add_location (lexer, e, start_ofs, node);
835 ymd_to_offset (int y, int m, int d)
838 double retval = calendar_gregorian_to_offset (
839 y, m, d, settings_get_fmt_settings (), &error);
842 msg (SE, "%s", error);
848 static struct expr_node *
849 expr_date (struct expression *e, int year_digits)
851 static const char *months[12] =
853 "JAN", "FEB", "MAR", "APR", "MAY", "JUN",
854 "JUL", "AUG", "SEP", "OCT", "NOV", "DEC",
857 time_t last_proc_time = time_of_last_procedure (e->ds);
858 struct tm *time = localtime (&last_proc_time);
860 char *tmp = (year_digits == 2
861 ? xasprintf ("%02d-%s-%02d", time->tm_mday, months[time->tm_mon],
863 : xasprintf ("%02d-%s-%04d", time->tm_mday, months[time->tm_mon],
864 time->tm_year + 1900));
867 ss_alloc_substring_pool (&s, ss_cstr (tmp), e->expr_pool);
871 return expr_allocate_string (e, s);
874 /* Parses system variables. */
875 static struct expr_node *
876 parse_sysvar (struct lexer *lexer, struct expression *e)
878 if (lex_match_id (lexer, "$CASENUM"))
879 return expr_allocate_nullary (e, OP_CASENUM);
880 else if (lex_match_id (lexer, "$DATE"))
881 return expr_date (e, 2);
882 else if (lex_match_id (lexer, "$DATE11"))
883 return expr_date (e, 4);
884 else if (lex_match_id (lexer, "$TRUE"))
885 return expr_allocate_boolean (e, 1.0);
886 else if (lex_match_id (lexer, "$FALSE"))
887 return expr_allocate_boolean (e, 0.0);
888 else if (lex_match_id (lexer, "$SYSMIS"))
889 return expr_allocate_number (e, SYSMIS);
890 else if (lex_match_id (lexer, "$JDATE"))
892 time_t time = time_of_last_procedure (e->ds);
893 struct tm *tm = localtime (&time);
894 return expr_allocate_number (e, ymd_to_offset (tm->tm_year + 1900,
898 else if (lex_match_id (lexer, "$TIME"))
900 time_t time = time_of_last_procedure (e->ds);
901 struct tm *tm = localtime (&time);
902 return expr_allocate_number (e, ymd_to_offset (tm->tm_year + 1900,
905 + tm->tm_hour * 60 * 60.
909 else if (lex_match_id (lexer, "$LENGTH"))
910 return expr_allocate_number (e, settings_get_viewlength ());
911 else if (lex_match_id (lexer, "$WIDTH"))
912 return expr_allocate_number (e, settings_get_viewwidth ());
915 msg (SE, _("Unknown system variable %s."), lex_tokcstr (lexer));
920 /* Parses numbers, varnames, etc. */
921 static struct expr_node *
922 parse_primary__ (struct lexer *lexer, struct expression *e)
924 switch (lex_token (lexer))
927 if (lex_next_token (lexer, 1) == T_LPAREN)
929 /* An identifier followed by a left parenthesis may be
930 a vector element reference. If not, it's a function
932 if (e->ds != NULL && dict_lookup_vector (dataset_dict (e->ds), lex_tokcstr (lexer)) != NULL)
933 return parse_vector_element (lexer, e);
935 return parse_function (lexer, e);
937 else if (lex_tokcstr (lexer)[0] == '$')
939 /* $ at the beginning indicates a system variable. */
940 return parse_sysvar (lexer, e);
942 else if (e->ds != NULL && dict_lookup_var (dataset_dict (e->ds), lex_tokcstr (lexer)))
944 /* It looks like a user variable.
945 (It could be a format specifier, but we'll assume
946 it's a variable unless proven otherwise. */
947 return allocate_unary_variable (e, parse_variable (lexer, dataset_dict (e->ds)));
951 /* Try to parse it as a format specifier. */
956 ok = parse_format_specifier (lexer, &fmt);
960 return expr_allocate_format (e, &fmt);
962 /* All attempts failed. */
963 msg (SE, _("Unknown identifier %s."), lex_tokcstr (lexer));
971 struct expr_node *node = expr_allocate_number (e, lex_tokval (lexer));
978 const char *dict_encoding;
979 struct expr_node *node;
982 dict_encoding = (e->ds != NULL
983 ? dict_get_encoding (dataset_dict (e->ds))
985 s = recode_string_pool (dict_encoding, "UTF-8", lex_tokcstr (lexer),
986 ss_length (lex_tokss (lexer)), e->expr_pool);
987 node = expr_allocate_string (e, ss_cstr (s));
996 struct expr_node *node = parse_or (lexer, e);
997 return !node || !lex_force_match (lexer, T_RPAREN) ? NULL : node;
1001 lex_error (lexer, NULL);
1006 static struct expr_node *
1007 parse_primary (struct lexer *lexer, struct expression *e)
1009 int start_ofs = lex_ofs (lexer);
1010 struct expr_node *node = parse_primary__ (lexer, e);
1011 expr_add_location (lexer, e, start_ofs, node);
1015 static struct expr_node *
1016 parse_vector_element (struct lexer *lexer, struct expression *e)
1018 int vector_start_ofs = lex_ofs (lexer);
1020 /* Find vector, skip token.
1021 The caller must already have verified that the current token
1022 is the name of a vector. */
1023 const struct vector *vector = dict_lookup_vector (dataset_dict (e->ds),
1024 lex_tokcstr (lexer));
1025 assert (vector != NULL);
1028 /* Skip left parenthesis token.
1029 The caller must have verified that the lookahead is a left
1031 assert (lex_token (lexer) == T_LPAREN);
1034 int element_start_ofs = lex_ofs (lexer);
1035 struct expr_node *element = parse_or (lexer, e);
1038 expr_add_location (lexer, e, element_start_ofs, element);
1040 if (!lex_match (lexer, T_RPAREN))
1043 operation_type type = (vector_get_type (vector) == VAL_NUMERIC
1044 ? OP_VEC_ELEM_NUM_RAW : OP_VEC_ELEM_STR);
1045 struct expr_node *node = expr_allocate_binary (
1046 e, type, element, expr_allocate_vector (e, vector));
1047 expr_add_location (lexer, e, vector_start_ofs, node);
1049 if (!type_coercion (e, node, 0))
1051 msg_at (SE, expr_location (e, node),
1052 _("A vector index must be numeric."));
1054 msg_at (SN, expr_location (e, node->args[0]),
1055 _("This vector index has type '%s'."),
1056 atom_type_name (expr_node_returns (node->args[0])));
1064 /* Individual function parsing. */
1066 const struct operation operations[OP_first + n_OP] = {
1067 #include "parse.inc"
1071 word_matches (const char **test, const char **name)
1073 size_t test_len = strcspn (*test, ".");
1074 size_t name_len = strcspn (*name, ".");
1075 if (test_len == name_len)
1077 if (buf_compare_case (*test, *name, test_len))
1080 else if (test_len < 3 || test_len > name_len)
1084 if (buf_compare_case (*test, *name, test_len))
1090 if (**test != **name)
1101 /* Returns 0 if TOKEN and FUNC do not match,
1102 1 if TOKEN is an acceptable abbreviation for FUNC,
1103 2 if TOKEN equals FUNC. */
1105 compare_function_names (const char *token_, const char *func_)
1107 const char *token = token_;
1108 const char *func = func_;
1109 while (*token || *func)
1110 if (!word_matches (&token, &func))
1112 return !c_strcasecmp (token_, func_) ? 2 : 1;
1116 lookup_function (const char *token,
1117 const struct operation **first,
1118 const struct operation **last)
1120 *first = *last = NULL;
1121 const struct operation *best = NULL;
1123 for (const struct operation *f = operations + OP_function_first;
1124 f <= operations + OP_function_last; f++)
1126 int score = compare_function_names (token, f->name);
1132 else if (score == 1 && !(f->flags & OPF_NO_ABBREV) && !best)
1141 const struct operation *f = best;
1142 while (f <= operations + OP_function_last
1143 && !c_strcasecmp (f->name, best->name))
1151 extract_min_valid (const char *s)
1153 char *p = strrchr (s, '.');
1155 || p[1] < '0' || p[1] > '9'
1156 || strspn (p + 1, "0123456789") != strlen (p + 1))
1159 return atoi (p + 1);
1163 match_function__ (struct expr_node *node, const struct operation *f)
1165 if (node->n_args < f->n_args
1166 || (node->n_args > f->n_args && (f->flags & OPF_ARRAY_OPERAND) == 0)
1167 || node->n_args - (f->n_args - 1) < f->array_min_elems)
1170 node->type = f - operations;
1171 for (size_t i = 0; i < node->n_args; i++)
1172 if (!is_coercible (node, i))
1178 static const struct operation *
1179 match_function (struct expr_node *node,
1180 const struct operation *first, const struct operation *last)
1182 for (const struct operation *f = first; f < last; f++)
1183 if (match_function__ (node, f))
1189 validate_function_args (const struct expression *e, const struct expr_node *n,
1190 const struct operation *f, int n_args, int min_valid)
1192 /* Count the function arguments that go into the trailing array (if any). We
1193 know that there must be at least the minimum number because
1194 match_function() already checked. */
1195 int array_n_args = n_args - (f->n_args - 1);
1196 assert (array_n_args >= f->array_min_elems);
1198 if ((f->flags & OPF_ARRAY_OPERAND)
1199 && array_n_args % f->array_granularity != 0)
1201 /* RANGE is the only case we have so far. It has paired arguments with
1202 one initial argument, and that's the only special case we deal with
1204 assert (f->array_granularity == 2);
1205 assert (n_args % 2 == 0);
1206 msg_at (SE, expr_location (e, n),
1207 _("%s must have an odd number of arguments."), f->prototype);
1211 if (min_valid != -1)
1213 if (f->array_min_elems == 0)
1215 assert ((f->flags & OPF_MIN_VALID) == 0);
1216 msg_at (SE, expr_location (e, n),
1217 _("%s function cannot accept suffix .%d to specify the "
1218 "minimum number of valid arguments."),
1219 f->prototype, min_valid);
1224 assert (f->flags & OPF_MIN_VALID);
1225 if (min_valid > array_n_args)
1227 msg_at (SE, expr_location (e, n),
1228 _("For %s with %d arguments, at most %d (not %d) may be "
1229 "required to be valid."),
1230 f->prototype, n_args, array_n_args, min_valid);
1240 add_arg (struct expr_node ***args, size_t *n_args, size_t *allocated_args,
1241 struct expr_node *arg,
1242 struct expression *e, struct lexer *lexer, int arg_start_ofs)
1244 if (*n_args >= *allocated_args)
1245 *args = x2nrealloc (*args, allocated_args, sizeof **args);
1247 expr_add_location (lexer, e, arg_start_ofs, arg);
1248 (*args)[(*n_args)++] = arg;
1252 put_invocation (struct string *s,
1253 const char *func_name, struct expr_node *node)
1257 ds_put_format (s, "%s(", func_name);
1258 for (i = 0; i < node->n_args; i++)
1261 ds_put_cstr (s, ", ");
1262 ds_put_cstr (s, operations[expr_node_returns (node->args[i])].prototype);
1264 ds_put_byte (s, ')');
1268 no_match (struct expression *e, const char *func_name, struct expr_node *node,
1269 const struct operation *ops, size_t n)
1277 ds_put_format (&s, _("Type mismatch invoking %s as "), ops->prototype);
1278 put_invocation (&s, func_name, node);
1282 ds_put_cstr (&s, _("Function invocation "));
1283 put_invocation (&s, func_name, node);
1284 ds_put_cstr (&s, _(" does not match any known function. Candidates are:"));
1286 for (size_t i = 0; i < n; i++)
1287 ds_put_format (&s, "\n%s", ops[i].prototype);
1289 ds_put_byte (&s, '.');
1291 msg_at (SE, expr_location (e, node), "%s", ds_cstr (&s));
1293 if (n == 1 && ops->n_args == node->n_args)
1295 for (size_t i = 0; i < node->n_args; i++)
1296 if (!is_coercible (node, i))
1298 atom_type expected = ops->args[i];
1299 atom_type actual = expr_node_returns (node->args[i]);
1300 if ((expected == OP_ni_format || expected == OP_no_format)
1301 && actual == OP_format)
1303 const struct fmt_spec *f = &node->args[i]->format;
1304 char *error = fmt_check__ (f, (ops->args[i] == OP_ni_format
1305 ? FMT_FOR_INPUT : FMT_FOR_OUTPUT));
1307 error = fmt_check_type_compat__ (f, VAL_NUMERIC);
1310 msg_at (SN, expr_location (e, node->args[i]), "%s", error);
1315 msg_at (SN, expr_location (e, node->args[i]),
1316 _("This argument has type '%s' but '%s' is required."),
1317 atom_type_name (actual), atom_type_name (expected));
1324 static struct expr_node *
1325 parse_function (struct lexer *lexer, struct expression *e)
1327 struct string func_name;
1328 ds_init_substring (&func_name, lex_tokss (lexer));
1330 int min_valid = extract_min_valid (lex_tokcstr (lexer));
1332 const struct operation *first, *last;
1333 if (!lookup_function (lex_tokcstr (lexer), &first, &last))
1335 msg (SE, _("No function or vector named %s."), lex_tokcstr (lexer));
1336 ds_destroy (&func_name);
1340 int func_start_ofs = lex_ofs (lexer);
1342 if (!lex_force_match (lexer, T_LPAREN))
1344 ds_destroy (&func_name);
1348 struct expr_node **args = NULL;
1350 size_t allocated_args = 0;
1351 if (lex_token (lexer) != T_RPAREN)
1354 int arg_start_ofs = lex_ofs (lexer);
1355 if (lex_token (lexer) == T_ID
1356 && lex_next_token (lexer, 1) == T_TO)
1358 const struct variable **vars;
1361 if (!parse_variables_const (lexer, dataset_dict (e->ds),
1362 &vars, &n_vars, PV_SINGLE))
1364 for (size_t i = 0; i < n_vars; i++)
1365 add_arg (&args, &n_args, &allocated_args,
1366 allocate_unary_variable (e, vars[i]),
1367 e, lexer, arg_start_ofs);
1372 struct expr_node *arg = parse_or (lexer, e);
1376 add_arg (&args, &n_args, &allocated_args, arg,
1377 e, lexer, arg_start_ofs);
1379 if (lex_match (lexer, T_RPAREN))
1381 else if (!lex_match (lexer, T_COMMA))
1383 lex_error_expecting (lexer, "`,'", "`)'");
1388 struct expr_node *n = expr_allocate_composite (e, first - operations,
1390 expr_add_location (lexer, e, func_start_ofs, n);
1391 const struct operation *f = match_function (n, first, last);
1394 no_match (e, ds_cstr (&func_name), n, first, last - first);
1397 n->type = f - operations;
1398 n->min_valid = min_valid != -1 ? min_valid : f->array_min_elems;
1400 for (size_t i = 0; i < n_args; i++)
1401 if (!type_coercion (e, n, i))
1403 /* Unreachable because match_function already checked that the
1404 arguments were coercible. */
1407 if (!validate_function_args (e, n, f, n_args, min_valid))
1410 if ((f->flags & OPF_EXTENSION) && settings_get_syntax () == COMPATIBLE)
1411 msg_at (SW, expr_location (e, n),
1412 _("%s is a PSPP extension."), f->prototype);
1413 if (f->flags & OPF_UNIMPLEMENTED)
1415 msg_at (SE, expr_location (e, n),
1416 _("%s is not available in this version of PSPP."), f->prototype);
1419 if ((f->flags & OPF_PERM_ONLY) &&
1420 proc_in_temporary_transformations (e->ds))
1422 msg_at (SE, expr_location (e, n),
1423 _("%s may not appear after %s."), f->prototype, "TEMPORARY");
1427 if (n->type == OP_LAG_Vn || n->type == OP_LAG_Vs)
1428 dataset_need_lag (e->ds, 1);
1429 else if (n->type == OP_LAG_Vnn || n->type == OP_LAG_Vsn)
1431 assert (n->n_args == 2);
1432 assert (n->args[1]->type == OP_pos_int);
1433 dataset_need_lag (e->ds, n->args[1]->integer);
1437 ds_destroy (&func_name);
1442 ds_destroy (&func_name);
1446 /* Utility functions. */
1448 static struct expression *
1449 expr_create (struct dataset *ds)
1451 struct pool *pool = pool_create ();
1452 struct expression *e = pool_alloc (pool, sizeof *e);
1453 *e = (struct expression) {
1456 .eval_pool = pool_create_subpool (pool),
1462 expr_node_returns (const struct expr_node *n)
1465 assert (is_operation (n->type));
1466 if (is_atom (n->type))
1468 else if (is_composite (n->type))
1469 return operations[n->type].returns;
1475 atom_type_name (atom_type type)
1477 assert (is_atom (type));
1479 /* The Boolean type is purely an internal concept that the documentation
1480 doesn't mention, so it might confuse users if we talked about them in
1482 return type == OP_boolean ? "number" : operations[type].name;
1486 expr_allocate_nullary (struct expression *e, operation_type op)
1488 return expr_allocate_composite (e, op, NULL, 0);
1492 expr_allocate_unary (struct expression *e, operation_type op,
1493 struct expr_node *arg0)
1495 return expr_allocate_composite (e, op, &arg0, 1);
1499 expr_allocate_binary (struct expression *e, operation_type op,
1500 struct expr_node *arg0, struct expr_node *arg1)
1502 struct expr_node *args[2];
1505 return expr_allocate_composite (e, op, args, 2);
1509 expr_allocate_composite (struct expression *e, operation_type op,
1510 struct expr_node **args, size_t n_args)
1512 for (size_t i = 0; i < n_args; i++)
1516 struct expr_node *n = pool_alloc (e->expr_pool, sizeof *n);
1517 *n = (struct expr_node) {
1520 .args = pool_clone (e->expr_pool, args, sizeof *n->args * n_args),
1526 expr_allocate_number (struct expression *e, double d)
1528 struct expr_node *n = pool_alloc (e->expr_pool, sizeof *n);
1529 *n = (struct expr_node) { .type = OP_number, .number = d };
1534 expr_allocate_boolean (struct expression *e, double b)
1536 assert (b == 0.0 || b == 1.0 || b == SYSMIS);
1538 struct expr_node *n = pool_alloc (e->expr_pool, sizeof *n);
1539 *n = (struct expr_node) { .type = OP_boolean, .number = b };
1544 expr_allocate_integer (struct expression *e, int i)
1546 struct expr_node *n = pool_alloc (e->expr_pool, sizeof *n);
1547 *n = (struct expr_node) { .type = OP_integer, .integer = i };
1552 expr_allocate_pos_int (struct expression *e, int i)
1556 struct expr_node *n = pool_alloc (e->expr_pool, sizeof *n);
1557 *n = (struct expr_node) { .type = OP_pos_int, .integer = i };
1562 expr_allocate_vector (struct expression *e, const struct vector *vector)
1564 struct expr_node *n = pool_alloc (e->expr_pool, sizeof *n);
1565 *n = (struct expr_node) { .type = OP_vector, .vector = vector };
1570 expr_allocate_string (struct expression *e, struct substring s)
1572 struct expr_node *n = pool_alloc (e->expr_pool, sizeof *n);
1573 *n = (struct expr_node) { .type = OP_string, .string = s };
1578 expr_allocate_variable (struct expression *e, const struct variable *v)
1580 struct expr_node *n = pool_alloc (e->expr_pool, sizeof *n);
1581 *n = (struct expr_node) {
1582 .type = var_is_numeric (v) ? OP_num_var : OP_str_var,
1589 expr_allocate_format (struct expression *e, const struct fmt_spec *format)
1591 struct expr_node *n = pool_alloc (e->expr_pool, sizeof *n);
1592 *n = (struct expr_node) { .type = OP_format, .format = *format };
1597 expr_allocate_expr_node (struct expression *e,
1598 const struct expr_node *expr_node)
1600 struct expr_node *n = pool_alloc (e->expr_pool, sizeof *n);
1601 *n = (struct expr_node) { .type = OP_expr_node, .expr_node = expr_node };
1605 /* Allocates a unary composite node that represents the value of
1606 variable V in expression E. */
1607 static struct expr_node *
1608 allocate_unary_variable (struct expression *e, const struct variable *v)
1611 return expr_allocate_unary (e, var_is_numeric (v) ? OP_NUM_VAR : OP_STR_VAR,
1612 expr_allocate_variable (e, v));
1615 /* Export function details to other modules. */
1617 /* Returns the operation structure for the function with the
1619 const struct operation *
1620 expr_get_function (size_t idx)
1622 assert (idx < n_OP_function);
1623 return &operations[OP_function_first + idx];
1626 /* Returns the number of expression functions. */
1628 expr_get_n_functions (void)
1630 return n_OP_function;
1633 /* Returns the name of operation OP. */
1635 expr_operation_get_name (const struct operation *op)
1640 /* Returns the human-readable prototype for operation OP. */
1642 expr_operation_get_prototype (const struct operation *op)
1644 return op->prototype;
1647 /* Returns the number of arguments for operation OP. */
1649 expr_operation_get_n_args (const struct operation *op)