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_binary (e, OP_EXPR_TO_BOOLEAN, n,
114 expr_allocate_expr_node (e, n));
115 else if (actual_type != OP_boolean)
117 msg_at (SE, expr_location (e, n),
118 _("Type mismatch: expression has %s type, "
119 "but a boolean value is required here."),
120 atom_type_name (actual_type));
125 return finish_expression (expr_optimize (n, e), e);
128 /* Parses a numeric expression that is intended to be assigned to newly created
129 variable NEW_VAR_NAME. (This allows for a better error message if the
130 expression is not numeric.) Otherwise similar to expr_parse(). */
132 expr_parse_new_variable (struct lexer *lexer, struct dataset *ds,
133 const char *new_var_name)
135 struct expression *e = expr_create (ds);
136 struct expr_node *n = parse_expr (lexer, e);
143 atom_type actual_type = expr_node_returns (n);
144 if (actual_type != OP_number && actual_type != OP_boolean)
146 msg (SE, _("This command tries to create a new variable %s by assigning a "
147 "string value to it, but this is not supported. Use "
148 "the STRING command to create the new variable with the "
149 "correct width before assigning to it, e.g. STRING %s(A20)."),
150 new_var_name, new_var_name);
155 return finish_expression (expr_optimize (n, e), e);
158 /* Free expression E. */
160 expr_free (struct expression *e)
163 pool_destroy (e->expr_pool);
167 expr_parse_any (struct lexer *lexer, struct dataset *ds, bool optimize)
170 struct expression *e;
172 e = expr_create (ds);
173 n = parse_expr (lexer, e);
181 n = expr_optimize (n, e);
182 return finish_expression (n, e);
185 /* Finishing up expression building. */
187 /* Height of an expression's stacks. */
190 int number_height; /* Height of number stack. */
191 int string_height; /* Height of string stack. */
194 /* Stack heights used by different kinds of arguments. */
195 static const struct stack_heights on_number_stack = {1, 0};
196 static const struct stack_heights on_string_stack = {0, 1};
197 static const struct stack_heights not_on_stack = {0, 0};
199 /* Returns the stack heights used by an atom of the given
201 static const struct stack_heights *
202 atom_type_stack (atom_type type)
204 assert (is_atom (type));
210 case OP_num_vec_elem:
211 return &on_number_stack;
214 return &on_string_stack;
225 return ¬_on_stack;
232 /* Measures the stack height needed for node N, supposing that
233 the stack height is initially *HEIGHT and updating *HEIGHT to
234 the final stack height. Updates *MAX, if necessary, to
235 reflect the maximum intermediate or final height. */
237 measure_stack (const struct expr_node *n,
238 struct stack_heights *height, struct stack_heights *max)
240 const struct stack_heights *return_height;
242 if (is_composite (n->type))
244 struct stack_heights args;
248 for (i = 0; i < n->n_args; i++)
249 measure_stack (n->args[i], &args, max);
251 return_height = atom_type_stack (operations[n->type].returns);
254 return_height = atom_type_stack (n->type);
256 height->number_height += return_height->number_height;
257 height->string_height += return_height->string_height;
259 if (height->number_height > max->number_height)
260 max->number_height = height->number_height;
261 if (height->string_height > max->string_height)
262 max->string_height = height->string_height;
265 /* Allocates stacks within E sufficient for evaluating node N. */
267 allocate_stacks (struct expr_node *n, struct expression *e)
269 struct stack_heights initial = {0, 0};
270 struct stack_heights max = {0, 0};
272 measure_stack (n, &initial, &max);
273 e->number_stack = pool_alloc (e->expr_pool,
274 sizeof *e->number_stack * max.number_height);
275 e->string_stack = pool_alloc (e->expr_pool,
276 sizeof *e->string_stack * max.string_height);
279 /* Finalizes expression E for evaluating node N. */
280 static struct expression *
281 finish_expression (struct expr_node *n, struct expression *e)
283 /* Allocate stacks. */
284 allocate_stacks (n, e);
286 /* Output postfix representation. */
289 /* The eval_pool might have been used for allocating strings
290 during optimization. We need to keep those strings around
291 for all subsequent evaluations, so start a new eval_pool. */
292 e->eval_pool = pool_create_subpool (e->expr_pool);
297 /* Verifies that expression E, whose root node is *N, can be
298 converted to type EXPECTED_TYPE, inserting a conversion at *N
299 if necessary. Returns true if successful, false on failure. */
301 type_check (const struct expression *e, const struct expr_node *n,
302 enum val_type expected_type)
304 atom_type actual_type = expr_node_returns (n);
306 switch (expected_type)
309 if (actual_type != OP_number && actual_type != OP_boolean)
311 msg_at (SE, expr_location (e, n),
312 _("Type mismatch: expression has type '%s', "
313 "but a numeric value is required."),
314 atom_type_name (actual_type));
320 if (actual_type != OP_string)
322 msg_at (SE, expr_location (e, n),
323 _("Type mismatch: expression has type '%s', "
324 "but a string value is required."),
325 atom_type_name (actual_type));
337 /* Recursive-descent expression parser. */
340 free_msg_location (void *loc_)
342 struct msg_location *loc = loc_;
343 msg_location_destroy (loc);
347 expr_location__ (struct expression *e,
348 const struct expr_node *node,
349 const struct msg_location **minp,
350 const struct msg_location **maxp)
352 struct msg_location *loc = node->location;
355 const struct msg_location *min = *minp;
358 || loc->start.line < min->start.line
359 || (loc->start.line == min->start.line
360 && loc->start.column < min->start.column)))
363 const struct msg_location *max = *maxp;
366 || loc->end.line > max->end.line
367 || (loc->end.line == max->end.line
368 && loc->end.column > max->end.column)))
374 if (is_composite (node->type))
375 for (size_t i = 0; i < node->n_args; i++)
376 expr_location__ (e, node->args[i], minp, maxp);
379 /* Returns the source code location corresponding to expression NODE, computing
380 it lazily if needed. */
381 const struct msg_location *
382 expr_location (const struct expression *e_, const struct expr_node *node_)
384 struct expr_node *node = CONST_CAST (struct expr_node *, node_);
390 struct expression *e = CONST_CAST (struct expression *, e_);
391 const struct msg_location *min = NULL;
392 const struct msg_location *max = NULL;
393 expr_location__ (e, node, &min, &max);
396 node->location = msg_location_dup (min);
397 node->location->end = max->end;
398 pool_register (e->expr_pool, free_msg_location, node->location);
401 return node->location;
404 /* Sets e->location to the tokens in S's lexer from offset START_OFS to the
405 token before the current one. Has no effect if E already has a location or
408 expr_add_location (struct lexer *lexer, struct expression *e,
409 int start_ofs, struct expr_node *node)
411 if (node && !node->location)
413 node->location = lex_ofs_location (lexer, start_ofs, lex_ofs (lexer) - 1);
414 pool_register (e->expr_pool, free_msg_location, node->location);
419 type_coercion__ (struct expression *e, struct expr_node *node, size_t arg_idx,
422 assert (!!do_coercion == (e != NULL));
427 struct expr_node **argp = &node->args[arg_idx];
428 struct expr_node *arg = *argp;
432 const struct operation *op = &operations[node->type];
433 atom_type required_type = op->args[MIN (arg_idx, op->n_args - 1)];
434 atom_type actual_type = expr_node_returns (arg);
435 if (actual_type == required_type)
441 switch (required_type)
444 if (actual_type == OP_boolean)
446 /* To enforce strict typing rules, insert Boolean to
447 numeric "conversion". This conversion is a no-op,
448 so it will be removed later. */
450 *argp = expr_allocate_unary (e, OP_BOOLEAN_TO_NUM, arg);
453 else if (actual_type == OP_num_vec_elem)
456 arg->type = OP_VEC_ELEM_NUM;
462 /* No coercion to string. */
466 if (actual_type == OP_number)
468 /* Convert numeric to boolean. */
470 *argp = expr_allocate_binary (e, OP_OPERAND_TO_BOOLEAN, arg,
471 expr_allocate_expr_node (e, node));
477 if (actual_type == OP_number)
479 /* Convert number to integer. */
481 *argp = expr_allocate_unary (e, OP_NUM_TO_INTEGER, arg);
487 /* We never coerce to OP_format, only to OP_ni_format or OP_no_format. */
492 if (arg->type == OP_format
493 && fmt_check_input (&arg->format)
494 && fmt_check_type_compat (&arg->format, VAL_NUMERIC))
498 arg->type = OP_ni_format;
506 if (arg->type == OP_format
507 && fmt_check_output (&arg->format)
508 && fmt_check_type_compat (&arg->format, VAL_NUMERIC))
512 arg->type = OP_no_format;
519 if (arg->type == OP_NUM_VAR)
522 *argp = arg->args[0];
528 if (arg->type == OP_STR_VAR)
531 *argp = arg->args[0];
537 if (arg->type == OP_NUM_VAR || arg->type == OP_STR_VAR)
540 *argp = arg->args[0];
546 if (arg->type == OP_number
547 && floor (arg->number) == arg->number
548 && arg->number > 0 && arg->number < INT_MAX)
551 *argp = expr_allocate_pos_int (e, arg->number);
563 type_coercion (struct expression *e, struct expr_node *node, size_t arg_idx)
565 return type_coercion__ (e, node, arg_idx, true);
569 is_coercible (const struct expr_node *node_, size_t arg_idx)
571 struct expr_node *node = CONST_CAST (struct expr_node *, node_);
572 return type_coercion__ (NULL, node, arg_idx, false);
575 /* How to parse an operator.
577 Some operators support both numeric and string operators. For those,
578 'num_op' and 'str_op' are both nonzero. Otherwise, only one 'num_op' is
579 nonzero. (PSPP doesn't have any string-only operators.) */
582 enum token_type token; /* Operator token. */
583 operation_type num_op; /* Operation for numeric operands (or 0). */
584 operation_type str_op; /* Operation for string operands (or 0). */
587 static operation_type
588 match_operator (struct lexer *lexer, const struct operator ops[], size_t n_ops,
589 const struct expr_node *lhs)
591 bool lhs_is_numeric = operations[lhs->type].returns != OP_string;
592 for (const struct operator *op = ops; op < ops + n_ops; op++)
593 if (lex_token (lexer) == op->token)
595 if (op->token != T_NEG_NUM)
598 return op->str_op && !lhs_is_numeric ? op->str_op : op->num_op;
604 operator_name (enum token_type token)
606 return token == T_NEG_NUM ? "-" : token_type_to_string (token);
609 static struct expr_node *
610 parse_binary_operators__ (struct lexer *lexer, struct expression *e,
611 const struct operator ops[], size_t n_ops,
612 parse_recursively_func *parse_next_level,
613 const char *chain_warning, struct expr_node *lhs)
615 for (int op_count = 0; ; op_count++)
617 enum token_type token = lex_token (lexer);
618 operation_type optype = match_operator (lexer, ops, n_ops, lhs);
621 if (op_count > 1 && chain_warning)
622 msg_at (SW, expr_location (e, lhs), "%s", chain_warning);
627 struct expr_node *rhs = parse_next_level (lexer, e);
631 struct expr_node *node = expr_allocate_binary (e, optype, lhs, rhs);
632 if (!is_coercible (node, 0) || !is_coercible (node, 1))
635 for (size_t i = 0; i < n_ops; i++)
636 if (ops[i].token == token)
637 both = ops[i].num_op && ops[i].str_op;
639 const char *name = operator_name (token);
641 msg_at (SE, expr_location (e, node),
642 _("Both operands of %s must have the same type."), name);
643 else if (operations[node->type].args[0] != OP_string)
644 msg_at (SE, expr_location (e, node),
645 _("Both operands of %s must be numeric."), name);
647 msg_at (SE, expr_location (e, node),
648 _("Both operands of %s must be strings."), name);
650 msg_at (SN, expr_location (e, node->args[0]),
651 _("This operand has type '%s'."),
652 atom_type_name (expr_node_returns (node->args[0])));
653 msg_at (SN, expr_location (e, node->args[1]),
654 _("This operand has type '%s'."),
655 atom_type_name (expr_node_returns (node->args[1])));
660 if (!type_coercion (e, node, 0) || !type_coercion (e, node, 1))
667 static struct expr_node *
668 parse_binary_operators (struct lexer *lexer, struct expression *e,
669 const struct operator ops[], size_t n_ops,
670 parse_recursively_func *parse_next_level,
671 const char *chain_warning)
673 struct expr_node *lhs = parse_next_level (lexer, e);
677 return parse_binary_operators__ (lexer, e, ops, n_ops, parse_next_level,
681 static struct expr_node *
682 parse_inverting_unary_operator (struct lexer *lexer, struct expression *e,
683 const struct operator *op,
684 parse_recursively_func *parse_next_level)
686 int start_ofs = lex_ofs (lexer);
687 unsigned int op_count = 0;
688 while (lex_match (lexer, op->token))
691 struct expr_node *inner = parse_next_level (lexer, e);
692 if (!inner || !op_count)
695 struct expr_node *outer = expr_allocate_unary (e, op->num_op, inner);
696 expr_add_location (lexer, e, start_ofs, outer);
698 if (!type_coercion (e, outer, 0))
700 assert (operations[outer->type].args[0] != OP_string);
702 const char *name = operator_name (op->token);
703 msg_at (SE, expr_location (e, outer),
704 _("The unary %s operator requires a numeric operand."), name);
706 msg_at (SN, expr_location (e, outer->args[0]),
707 _("The operand of %s has type '%s'."),
708 name, atom_type_name (expr_node_returns (outer->args[0])));
713 return op_count % 2 ? outer : outer->args[0];
716 /* Parses the OR level. */
717 static struct expr_node *
718 parse_or (struct lexer *lexer, struct expression *e)
720 static const struct operator op = { .token = T_OR, .num_op = OP_OR };
721 return parse_binary_operators (lexer, e, &op, 1, parse_and, NULL);
724 /* Parses the AND level. */
725 static struct expr_node *
726 parse_and (struct lexer *lexer, struct expression *e)
728 static const struct operator op = { .token = T_AND, .num_op = OP_AND };
730 return parse_binary_operators (lexer, e, &op, 1, parse_not, NULL);
733 /* Parses the NOT level. */
734 static struct expr_node *
735 parse_not (struct lexer *lexer, struct expression *e)
737 static const struct operator op = { .token = T_NOT, .num_op = OP_NOT };
738 return parse_inverting_unary_operator (lexer, e, &op, parse_rel);
741 /* Parse relational operators. */
742 static struct expr_node *
743 parse_rel (struct lexer *lexer, struct expression *e)
745 const char *chain_warning =
746 _("Chaining relational operators (e.g. `a < b < c') will "
747 "not produce the mathematically expected result. "
748 "Use the AND logical operator to fix the problem "
749 "(e.g. `a < b AND b < c'). "
750 "To disable this warning, insert parentheses.");
752 static const struct operator ops[] =
754 { .token = T_EQUALS, .num_op = OP_EQ, .str_op = OP_EQ_STRING },
755 { .token = T_EQ, .num_op = OP_EQ, .str_op = OP_EQ_STRING },
756 { .token = T_GE, .num_op = OP_GE, .str_op = OP_GE_STRING },
757 { .token = T_GT, .num_op = OP_GT, .str_op = OP_GT_STRING },
758 { .token = T_LE, .num_op = OP_LE, .str_op = OP_LE_STRING },
759 { .token = T_LT, .num_op = OP_LT, .str_op = OP_LT_STRING },
760 { .token = T_NE, .num_op = OP_NE, .str_op = OP_NE_STRING },
763 return parse_binary_operators (lexer, e, ops, sizeof ops / sizeof *ops,
764 parse_add, chain_warning);
767 /* Parses the addition and subtraction level. */
768 static struct expr_node *
769 parse_add (struct lexer *lexer, struct expression *e)
771 static const struct operator ops[] =
773 { .token = T_PLUS, .num_op = OP_ADD },
774 { .token = T_DASH, .num_op = OP_SUB },
775 { .token = T_NEG_NUM, .num_op = OP_ADD },
778 return parse_binary_operators (lexer, e, ops, sizeof ops / sizeof *ops,
782 /* Parses the multiplication and division level. */
783 static struct expr_node *
784 parse_mul (struct lexer *lexer, struct expression *e)
786 static const struct operator ops[] =
788 { .token = T_ASTERISK, .num_op = OP_MUL },
789 { .token = T_SLASH, .num_op = OP_DIV },
792 return parse_binary_operators (lexer, e, ops, sizeof ops / sizeof *ops,
796 /* Parses the unary minus level. */
797 static struct expr_node *
798 parse_neg (struct lexer *lexer, struct expression *e)
800 static const struct operator op = { .token = T_DASH, .num_op = OP_NEG };
801 return parse_inverting_unary_operator (lexer, e, &op, parse_exp);
804 static struct expr_node *
805 parse_exp (struct lexer *lexer, struct expression *e)
807 static const struct operator op = { .token = T_EXP, .num_op = OP_POW };
809 const char *chain_warning =
810 _("The exponentiation operator (`**') is left-associative: "
811 "`a**b**c' equals `(a**b)**c', not `a**(b**c)'. "
812 "To disable this warning, insert parentheses.");
814 if (lex_token (lexer) != T_NEG_NUM || lex_next_token (lexer, 1) != T_EXP)
815 return parse_binary_operators (lexer, e, &op, 1,
816 parse_primary, chain_warning);
818 /* Special case for situations like "-5**6", which must be parsed as
821 int start_ofs = lex_ofs (lexer);
822 struct expr_node *lhs = expr_allocate_number (e, -lex_tokval (lexer));
824 expr_add_location (lexer, e, start_ofs, lhs);
826 struct expr_node *node = parse_binary_operators__ (
827 lexer, e, &op, 1, parse_primary, chain_warning, lhs);
831 node = expr_allocate_unary (e, OP_NEG, node);
832 expr_add_location (lexer, e, start_ofs, node);
837 ymd_to_offset (int y, int m, int d)
840 double retval = calendar_gregorian_to_offset (
841 y, m, d, settings_get_fmt_settings (), &error);
844 msg (SE, "%s", error);
850 static struct expr_node *
851 expr_date (struct expression *e, int year_digits)
853 static const char *months[12] =
855 "JAN", "FEB", "MAR", "APR", "MAY", "JUN",
856 "JUL", "AUG", "SEP", "OCT", "NOV", "DEC",
859 time_t last_proc_time = time_of_last_procedure (e->ds);
860 struct tm *time = localtime (&last_proc_time);
862 char *tmp = (year_digits == 2
863 ? xasprintf ("%02d-%s-%02d", time->tm_mday, months[time->tm_mon],
865 : xasprintf ("%02d-%s-%04d", time->tm_mday, months[time->tm_mon],
866 time->tm_year + 1900));
869 ss_alloc_substring_pool (&s, ss_cstr (tmp), e->expr_pool);
873 return expr_allocate_string (e, s);
876 /* Parses system variables. */
877 static struct expr_node *
878 parse_sysvar (struct lexer *lexer, struct expression *e)
880 if (lex_match_id (lexer, "$CASENUM"))
881 return expr_allocate_nullary (e, OP_CASENUM);
882 else if (lex_match_id (lexer, "$DATE"))
883 return expr_date (e, 2);
884 else if (lex_match_id (lexer, "$DATE11"))
885 return expr_date (e, 4);
886 else if (lex_match_id (lexer, "$TRUE"))
887 return expr_allocate_boolean (e, 1.0);
888 else if (lex_match_id (lexer, "$FALSE"))
889 return expr_allocate_boolean (e, 0.0);
890 else if (lex_match_id (lexer, "$SYSMIS"))
891 return expr_allocate_number (e, SYSMIS);
892 else if (lex_match_id (lexer, "$JDATE"))
894 time_t time = time_of_last_procedure (e->ds);
895 struct tm *tm = localtime (&time);
896 return expr_allocate_number (e, ymd_to_offset (tm->tm_year + 1900,
900 else if (lex_match_id (lexer, "$TIME"))
902 time_t time = time_of_last_procedure (e->ds);
903 struct tm *tm = localtime (&time);
904 return expr_allocate_number (e, ymd_to_offset (tm->tm_year + 1900,
907 + tm->tm_hour * 60 * 60.
911 else if (lex_match_id (lexer, "$LENGTH"))
912 return expr_allocate_number (e, settings_get_viewlength ());
913 else if (lex_match_id (lexer, "$WIDTH"))
914 return expr_allocate_number (e, settings_get_viewwidth ());
917 msg (SE, _("Unknown system variable %s."), lex_tokcstr (lexer));
922 /* Parses numbers, varnames, etc. */
923 static struct expr_node *
924 parse_primary__ (struct lexer *lexer, struct expression *e)
926 switch (lex_token (lexer))
929 if (lex_next_token (lexer, 1) == T_LPAREN)
931 /* An identifier followed by a left parenthesis may be
932 a vector element reference. If not, it's a function
934 if (e->ds != NULL && dict_lookup_vector (dataset_dict (e->ds), lex_tokcstr (lexer)) != NULL)
935 return parse_vector_element (lexer, e);
937 return parse_function (lexer, e);
939 else if (lex_tokcstr (lexer)[0] == '$')
941 /* $ at the beginning indicates a system variable. */
942 return parse_sysvar (lexer, e);
944 else if (e->ds != NULL && dict_lookup_var (dataset_dict (e->ds), lex_tokcstr (lexer)))
946 /* It looks like a user variable.
947 (It could be a format specifier, but we'll assume
948 it's a variable unless proven otherwise. */
949 return allocate_unary_variable (e, parse_variable (lexer, dataset_dict (e->ds)));
953 /* Try to parse it as a format specifier. */
958 ok = parse_format_specifier (lexer, &fmt);
962 return expr_allocate_format (e, &fmt);
964 /* All attempts failed. */
965 msg (SE, _("Unknown identifier %s."), lex_tokcstr (lexer));
973 struct expr_node *node = expr_allocate_number (e, lex_tokval (lexer));
980 const char *dict_encoding;
981 struct expr_node *node;
984 dict_encoding = (e->ds != NULL
985 ? dict_get_encoding (dataset_dict (e->ds))
987 s = recode_string_pool (dict_encoding, "UTF-8", lex_tokcstr (lexer),
988 ss_length (lex_tokss (lexer)), e->expr_pool);
989 node = expr_allocate_string (e, ss_cstr (s));
998 struct expr_node *node = parse_or (lexer, e);
999 return !node || !lex_force_match (lexer, T_RPAREN) ? NULL : node;
1003 lex_error (lexer, NULL);
1008 static struct expr_node *
1009 parse_primary (struct lexer *lexer, struct expression *e)
1011 int start_ofs = lex_ofs (lexer);
1012 struct expr_node *node = parse_primary__ (lexer, e);
1013 expr_add_location (lexer, e, start_ofs, node);
1017 static struct expr_node *
1018 parse_vector_element (struct lexer *lexer, struct expression *e)
1020 int vector_start_ofs = lex_ofs (lexer);
1022 /* Find vector, skip token.
1023 The caller must already have verified that the current token
1024 is the name of a vector. */
1025 const struct vector *vector = dict_lookup_vector (dataset_dict (e->ds),
1026 lex_tokcstr (lexer));
1027 assert (vector != NULL);
1030 /* Skip left parenthesis token.
1031 The caller must have verified that the lookahead is a left
1033 assert (lex_token (lexer) == T_LPAREN);
1036 int element_start_ofs = lex_ofs (lexer);
1037 struct expr_node *element = parse_or (lexer, e);
1040 expr_add_location (lexer, e, element_start_ofs, element);
1042 if (!lex_match (lexer, T_RPAREN))
1045 operation_type type = (vector_get_type (vector) == VAL_NUMERIC
1046 ? OP_VEC_ELEM_NUM_RAW : OP_VEC_ELEM_STR);
1047 struct expr_node *node = expr_allocate_binary (
1048 e, type, element, expr_allocate_vector (e, vector));
1049 expr_add_location (lexer, e, vector_start_ofs, node);
1051 if (!type_coercion (e, node, 0))
1053 msg_at (SE, expr_location (e, node),
1054 _("A vector index must be numeric."));
1056 msg_at (SN, expr_location (e, node->args[0]),
1057 _("This vector index has type '%s'."),
1058 atom_type_name (expr_node_returns (node->args[0])));
1066 /* Individual function parsing. */
1068 const struct operation operations[OP_first + n_OP] = {
1069 #include "parse.inc"
1073 word_matches (const char **test, const char **name)
1075 size_t test_len = strcspn (*test, ".");
1076 size_t name_len = strcspn (*name, ".");
1077 if (test_len == name_len)
1079 if (buf_compare_case (*test, *name, test_len))
1082 else if (test_len < 3 || test_len > name_len)
1086 if (buf_compare_case (*test, *name, test_len))
1092 if (**test != **name)
1103 /* Returns 0 if TOKEN and FUNC do not match,
1104 1 if TOKEN is an acceptable abbreviation for FUNC,
1105 2 if TOKEN equals FUNC. */
1107 compare_function_names (const char *token_, const char *func_)
1109 const char *token = token_;
1110 const char *func = func_;
1111 while (*token || *func)
1112 if (!word_matches (&token, &func))
1114 return !c_strcasecmp (token_, func_) ? 2 : 1;
1118 lookup_function (const char *token,
1119 const struct operation **first,
1120 const struct operation **last)
1122 *first = *last = NULL;
1123 const struct operation *best = NULL;
1125 for (const struct operation *f = operations + OP_function_first;
1126 f <= operations + OP_function_last; f++)
1128 int score = compare_function_names (token, f->name);
1134 else if (score == 1 && !(f->flags & OPF_NO_ABBREV) && !best)
1143 const struct operation *f = best;
1144 while (f <= operations + OP_function_last
1145 && !c_strcasecmp (f->name, best->name))
1153 extract_min_valid (const char *s)
1155 char *p = strrchr (s, '.');
1157 || p[1] < '0' || p[1] > '9'
1158 || strspn (p + 1, "0123456789") != strlen (p + 1))
1161 return atoi (p + 1);
1165 match_function__ (struct expr_node *node, const struct operation *f)
1167 if (node->n_args < f->n_args
1168 || (node->n_args > f->n_args && (f->flags & OPF_ARRAY_OPERAND) == 0)
1169 || node->n_args - (f->n_args - 1) < f->array_min_elems)
1172 node->type = f - operations;
1173 for (size_t i = 0; i < node->n_args; i++)
1174 if (!is_coercible (node, i))
1180 static const struct operation *
1181 match_function (struct expr_node *node,
1182 const struct operation *first, const struct operation *last)
1184 for (const struct operation *f = first; f < last; f++)
1185 if (match_function__ (node, f))
1191 validate_function_args (const struct expression *e, const struct expr_node *n,
1192 const struct operation *f, int n_args, int min_valid)
1194 /* Count the function arguments that go into the trailing array (if any). We
1195 know that there must be at least the minimum number because
1196 match_function() already checked. */
1197 int array_n_args = n_args - (f->n_args - 1);
1198 assert (array_n_args >= f->array_min_elems);
1200 if ((f->flags & OPF_ARRAY_OPERAND)
1201 && array_n_args % f->array_granularity != 0)
1203 /* RANGE is the only case we have so far. It has paired arguments with
1204 one initial argument, and that's the only special case we deal with
1206 assert (f->array_granularity == 2);
1207 assert (n_args % 2 == 0);
1208 msg_at (SE, expr_location (e, n),
1209 _("%s must have an odd number of arguments."), f->prototype);
1213 if (min_valid != -1)
1215 if (f->array_min_elems == 0)
1217 assert ((f->flags & OPF_MIN_VALID) == 0);
1218 msg_at (SE, expr_location (e, n),
1219 _("%s function cannot accept suffix .%d to specify the "
1220 "minimum number of valid arguments."),
1221 f->prototype, min_valid);
1226 assert (f->flags & OPF_MIN_VALID);
1227 if (min_valid > array_n_args)
1229 msg_at (SE, expr_location (e, n),
1230 _("For %s with %d arguments, at most %d (not %d) may be "
1231 "required to be valid."),
1232 f->prototype, n_args, array_n_args, min_valid);
1242 add_arg (struct expr_node ***args, size_t *n_args, size_t *allocated_args,
1243 struct expr_node *arg,
1244 struct expression *e, struct lexer *lexer, int arg_start_ofs)
1246 if (*n_args >= *allocated_args)
1247 *args = x2nrealloc (*args, allocated_args, sizeof **args);
1249 expr_add_location (lexer, e, arg_start_ofs, arg);
1250 (*args)[(*n_args)++] = arg;
1254 put_invocation (struct string *s,
1255 const char *func_name, struct expr_node *node)
1259 ds_put_format (s, "%s(", func_name);
1260 for (i = 0; i < node->n_args; i++)
1263 ds_put_cstr (s, ", ");
1264 ds_put_cstr (s, operations[expr_node_returns (node->args[i])].prototype);
1266 ds_put_byte (s, ')');
1270 no_match (struct expression *e, const char *func_name, struct expr_node *node,
1271 const struct operation *ops, size_t n)
1279 ds_put_format (&s, _("Type mismatch invoking %s as "), ops->prototype);
1280 put_invocation (&s, func_name, node);
1284 ds_put_cstr (&s, _("Function invocation "));
1285 put_invocation (&s, func_name, node);
1286 ds_put_cstr (&s, _(" does not match any known function. Candidates are:"));
1288 for (size_t i = 0; i < n; i++)
1289 ds_put_format (&s, "\n%s", ops[i].prototype);
1291 ds_put_byte (&s, '.');
1293 msg_at (SE, expr_location (e, node), "%s", ds_cstr (&s));
1295 if (n == 1 && ops->n_args == node->n_args)
1297 for (size_t i = 0; i < ops->n_args; i++)
1298 if ((ops->args[i] == OP_ni_format
1299 || ops->args[i] == OP_no_format)
1300 && expr_node_returns (node->args[i]) == OP_format)
1302 const struct fmt_spec *f = &node->args[i]->format;
1303 char *error = fmt_check__ (f, (ops->args[i] == OP_ni_format
1304 ? FMT_FOR_INPUT : FMT_FOR_OUTPUT));
1306 error = fmt_check_type_compat__ (f, VAL_NUMERIC);
1309 msg_at (SN, expr_location (e, node->args[i]), "%s", error);
1318 static struct expr_node *
1319 parse_function (struct lexer *lexer, struct expression *e)
1321 struct string func_name;
1322 ds_init_substring (&func_name, lex_tokss (lexer));
1324 int min_valid = extract_min_valid (lex_tokcstr (lexer));
1326 const struct operation *first, *last;
1327 if (!lookup_function (lex_tokcstr (lexer), &first, &last))
1329 msg (SE, _("No function or vector named %s."), lex_tokcstr (lexer));
1330 ds_destroy (&func_name);
1334 int func_start_ofs = lex_ofs (lexer);
1336 if (!lex_force_match (lexer, T_LPAREN))
1338 ds_destroy (&func_name);
1342 struct expr_node **args = NULL;
1344 size_t allocated_args = 0;
1345 if (lex_token (lexer) != T_RPAREN)
1348 int arg_start_ofs = lex_ofs (lexer);
1349 if (lex_token (lexer) == T_ID
1350 && lex_next_token (lexer, 1) == T_TO)
1352 const struct variable **vars;
1355 if (!parse_variables_const (lexer, dataset_dict (e->ds),
1356 &vars, &n_vars, PV_SINGLE))
1358 for (size_t i = 0; i < n_vars; i++)
1359 add_arg (&args, &n_args, &allocated_args,
1360 allocate_unary_variable (e, vars[i]),
1361 e, lexer, arg_start_ofs);
1366 struct expr_node *arg = parse_or (lexer, e);
1370 add_arg (&args, &n_args, &allocated_args, arg,
1371 e, lexer, arg_start_ofs);
1373 if (lex_match (lexer, T_RPAREN))
1375 else if (!lex_match (lexer, T_COMMA))
1377 lex_error_expecting (lexer, "`,'", "`)'");
1382 struct expr_node *n = expr_allocate_composite (e, first - operations,
1384 expr_add_location (lexer, e, func_start_ofs, n);
1385 const struct operation *f = match_function (n, first, last);
1388 no_match (e, ds_cstr (&func_name), n, first, last - first);
1391 n->type = f - operations;
1392 n->min_valid = min_valid != -1 ? min_valid : f->array_min_elems;
1394 for (size_t i = 0; i < n_args; i++)
1395 if (!type_coercion (e, n, i))
1397 /* Unreachable because match_function already checked that the
1398 arguments were coercible. */
1401 if (!validate_function_args (e, n, f, n_args, min_valid))
1404 if ((f->flags & OPF_EXTENSION) && settings_get_syntax () == COMPATIBLE)
1405 msg_at (SW, expr_location (e, n),
1406 _("%s is a PSPP extension."), f->prototype);
1407 if (f->flags & OPF_UNIMPLEMENTED)
1409 msg_at (SE, expr_location (e, n),
1410 _("%s is not available in this version of PSPP."), f->prototype);
1413 if ((f->flags & OPF_PERM_ONLY) &&
1414 proc_in_temporary_transformations (e->ds))
1416 msg_at (SE, expr_location (e, n),
1417 _("%s may not appear after %s."), f->prototype, "TEMPORARY");
1421 if (n->type == OP_LAG_Vn || n->type == OP_LAG_Vs)
1422 dataset_need_lag (e->ds, 1);
1423 else if (n->type == OP_LAG_Vnn || n->type == OP_LAG_Vsn)
1425 assert (n->n_args == 2);
1426 assert (n->args[1]->type == OP_pos_int);
1427 dataset_need_lag (e->ds, n->args[1]->integer);
1431 ds_destroy (&func_name);
1436 ds_destroy (&func_name);
1440 /* Utility functions. */
1442 static struct expression *
1443 expr_create (struct dataset *ds)
1445 struct pool *pool = pool_create ();
1446 struct expression *e = pool_alloc (pool, sizeof *e);
1447 *e = (struct expression) {
1450 .eval_pool = pool_create_subpool (pool),
1456 expr_node_returns (const struct expr_node *n)
1459 assert (is_operation (n->type));
1460 if (is_atom (n->type))
1462 else if (is_composite (n->type))
1463 return operations[n->type].returns;
1469 atom_type_name (atom_type type)
1471 assert (is_atom (type));
1473 /* The Boolean type is purely an internal concept that the documentation
1474 doesn't mention, so it might confuse users if we talked about them in
1476 return type == OP_boolean ? "number" : operations[type].name;
1480 expr_allocate_nullary (struct expression *e, operation_type op)
1482 return expr_allocate_composite (e, op, NULL, 0);
1486 expr_allocate_unary (struct expression *e, operation_type op,
1487 struct expr_node *arg0)
1489 return expr_allocate_composite (e, op, &arg0, 1);
1493 expr_allocate_binary (struct expression *e, operation_type op,
1494 struct expr_node *arg0, struct expr_node *arg1)
1496 struct expr_node *args[2];
1499 return expr_allocate_composite (e, op, args, 2);
1503 expr_allocate_composite (struct expression *e, operation_type op,
1504 struct expr_node **args, size_t n_args)
1506 for (size_t i = 0; i < n_args; i++)
1510 struct expr_node *n = pool_alloc (e->expr_pool, sizeof *n);
1511 *n = (struct expr_node) {
1514 .args = pool_clone (e->expr_pool, args, sizeof *n->args * n_args),
1520 expr_allocate_number (struct expression *e, double d)
1522 struct expr_node *n = pool_alloc (e->expr_pool, sizeof *n);
1523 *n = (struct expr_node) { .type = OP_number, .number = d };
1528 expr_allocate_boolean (struct expression *e, double b)
1530 assert (b == 0.0 || b == 1.0 || b == SYSMIS);
1532 struct expr_node *n = pool_alloc (e->expr_pool, sizeof *n);
1533 *n = (struct expr_node) { .type = OP_boolean, .number = b };
1538 expr_allocate_integer (struct expression *e, int i)
1540 struct expr_node *n = pool_alloc (e->expr_pool, sizeof *n);
1541 *n = (struct expr_node) { .type = OP_integer, .integer = i };
1546 expr_allocate_pos_int (struct expression *e, int i)
1550 struct expr_node *n = pool_alloc (e->expr_pool, sizeof *n);
1551 *n = (struct expr_node) { .type = OP_pos_int, .integer = i };
1556 expr_allocate_vector (struct expression *e, const struct vector *vector)
1558 struct expr_node *n = pool_alloc (e->expr_pool, sizeof *n);
1559 *n = (struct expr_node) { .type = OP_vector, .vector = vector };
1564 expr_allocate_string (struct expression *e, struct substring s)
1566 struct expr_node *n = pool_alloc (e->expr_pool, sizeof *n);
1567 *n = (struct expr_node) { .type = OP_string, .string = s };
1572 expr_allocate_variable (struct expression *e, const struct variable *v)
1574 struct expr_node *n = pool_alloc (e->expr_pool, sizeof *n);
1575 *n = (struct expr_node) {
1576 .type = var_is_numeric (v) ? OP_num_var : OP_str_var,
1583 expr_allocate_format (struct expression *e, const struct fmt_spec *format)
1585 struct expr_node *n = pool_alloc (e->expr_pool, sizeof *n);
1586 *n = (struct expr_node) { .type = OP_format, .format = *format };
1591 expr_allocate_expr_node (struct expression *e,
1592 const struct expr_node *expr_node)
1594 struct expr_node *n = pool_alloc (e->expr_pool, sizeof *n);
1595 *n = (struct expr_node) { .type = OP_expr_node, .expr_node = expr_node };
1599 /* Allocates a unary composite node that represents the value of
1600 variable V in expression E. */
1601 static struct expr_node *
1602 allocate_unary_variable (struct expression *e, const struct variable *v)
1605 return expr_allocate_unary (e, var_is_numeric (v) ? OP_NUM_VAR : OP_STR_VAR,
1606 expr_allocate_variable (e, v));
1609 /* Export function details to other modules. */
1611 /* Returns the operation structure for the function with the
1613 const struct operation *
1614 expr_get_function (size_t idx)
1616 assert (idx < n_OP_function);
1617 return &operations[OP_function_first + idx];
1620 /* Returns the number of expression functions. */
1622 expr_get_n_functions (void)
1624 return n_OP_function;
1627 /* Returns the name of operation OP. */
1629 expr_operation_get_name (const struct operation *op)
1634 /* Returns the human-readable prototype for operation OP. */
1636 expr_operation_get_prototype (const struct operation *op)
1638 return op->prototype;
1641 /* Returns the number of arguments for operation OP. */
1643 expr_operation_get_n_args (const struct operation *op)