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 expr_node *, enum val_type expected_type);
64 static struct expr_node *allocate_unary_variable (struct expression *,
65 const struct variable *);
67 /* Public functions. */
69 /* Parses an expression of the given TYPE. If DS is nonnull then variables and
70 vectors within it may be referenced within the expression; otherwise, the
71 expression must not reference any variables or vectors. Returns the new
72 expression if successful or a null pointer otherwise. */
74 expr_parse (struct lexer *lexer, struct dataset *ds, enum val_type type)
76 assert (val_type_is_valid (type));
78 struct expression *e = expr_create (ds);
79 struct expr_node *n = parse_or (lexer, e);
80 if (!n || !type_check (n, type))
86 return finish_expression (expr_optimize (n, e), e);
89 /* Parses a boolean expression, otherwise similar to expr_parse(). */
91 expr_parse_bool (struct lexer *lexer, struct dataset *ds)
93 struct expression *e = expr_create (ds);
94 struct expr_node *n = parse_or (lexer, e);
101 atom_type actual_type = expr_node_returns (n);
102 if (actual_type == OP_number)
103 n = expr_allocate_unary (e, OP_NUM_TO_BOOLEAN, n);
104 else if (actual_type != OP_boolean)
106 msg (SE, _("Type mismatch: expression has %s type, "
107 "but a boolean value is required here."),
108 atom_type_name (actual_type));
113 return finish_expression (expr_optimize (n, e), e);
116 /* Parses a numeric expression that is intended to be assigned to newly created
117 variable NEW_VAR_NAME. (This allows for a better error message if the
118 expression is not numeric.) Otherwise similar to expr_parse(). */
120 expr_parse_new_variable (struct lexer *lexer, struct dataset *ds,
121 const char *new_var_name)
123 struct expression *e = expr_create (ds);
124 struct expr_node *n = parse_or (lexer, e);
131 atom_type actual_type = expr_node_returns (n);
132 if (actual_type != OP_number && actual_type != OP_boolean)
134 msg (SE, _("This command tries to create a new variable %s by assigning a "
135 "string value to it, but this is not supported. Use "
136 "the STRING command to create the new variable with the "
137 "correct width before assigning to it, e.g. STRING %s(A20)."),
138 new_var_name, new_var_name);
143 return finish_expression (expr_optimize (n, e), e);
146 /* Free expression E. */
148 expr_free (struct expression *e)
151 pool_destroy (e->expr_pool);
155 expr_parse_any (struct lexer *lexer, struct dataset *ds, bool optimize)
158 struct expression *e;
160 e = expr_create (ds);
161 n = parse_or (lexer, e);
169 n = expr_optimize (n, e);
170 return finish_expression (n, e);
173 /* Finishing up expression building. */
175 /* Height of an expression's stacks. */
178 int number_height; /* Height of number stack. */
179 int string_height; /* Height of string stack. */
182 /* Stack heights used by different kinds of arguments. */
183 static const struct stack_heights on_number_stack = {1, 0};
184 static const struct stack_heights on_string_stack = {0, 1};
185 static const struct stack_heights not_on_stack = {0, 0};
187 /* Returns the stack heights used by an atom of the given
189 static const struct stack_heights *
190 atom_type_stack (atom_type type)
192 assert (is_atom (type));
198 return &on_number_stack;
201 return &on_string_stack;
211 return ¬_on_stack;
218 /* Measures the stack height needed for node N, supposing that
219 the stack height is initially *HEIGHT and updating *HEIGHT to
220 the final stack height. Updates *MAX, if necessary, to
221 reflect the maximum intermediate or final height. */
223 measure_stack (const struct expr_node *n,
224 struct stack_heights *height, struct stack_heights *max)
226 const struct stack_heights *return_height;
228 if (is_composite (n->type))
230 struct stack_heights args;
234 for (i = 0; i < n->n_args; i++)
235 measure_stack (n->args[i], &args, max);
237 return_height = atom_type_stack (operations[n->type].returns);
240 return_height = atom_type_stack (n->type);
242 height->number_height += return_height->number_height;
243 height->string_height += return_height->string_height;
245 if (height->number_height > max->number_height)
246 max->number_height = height->number_height;
247 if (height->string_height > max->string_height)
248 max->string_height = height->string_height;
251 /* Allocates stacks within E sufficient for evaluating node N. */
253 allocate_stacks (struct expr_node *n, struct expression *e)
255 struct stack_heights initial = {0, 0};
256 struct stack_heights max = {0, 0};
258 measure_stack (n, &initial, &max);
259 e->number_stack = pool_alloc (e->expr_pool,
260 sizeof *e->number_stack * max.number_height);
261 e->string_stack = pool_alloc (e->expr_pool,
262 sizeof *e->string_stack * max.string_height);
265 /* Finalizes expression E for evaluating node N. */
266 static struct expression *
267 finish_expression (struct expr_node *n, struct expression *e)
269 /* Allocate stacks. */
270 allocate_stacks (n, e);
272 /* Output postfix representation. */
275 /* The eval_pool might have been used for allocating strings
276 during optimization. We need to keep those strings around
277 for all subsequent evaluations, so start a new eval_pool. */
278 e->eval_pool = pool_create_subpool (e->expr_pool);
283 /* Verifies that expression E, whose root node is *N, can be
284 converted to type EXPECTED_TYPE, inserting a conversion at *N
285 if necessary. Returns true if successful, false on failure. */
287 type_check (const struct expr_node *n, enum val_type expected_type)
289 atom_type actual_type = expr_node_returns (n);
291 switch (expected_type)
294 if (actual_type != OP_number && actual_type != OP_boolean)
296 msg (SE, _("Type mismatch: expression has %s type, "
297 "but a numeric value is required here."),
298 atom_type_name (actual_type));
304 if (actual_type != OP_string)
306 msg (SE, _("Type mismatch: expression has %s type, "
307 "but a string value is required here."),
308 atom_type_name (actual_type));
320 /* Recursive-descent expression parser. */
323 free_msg_location (void *loc_)
325 struct msg_location *loc = loc_;
326 msg_location_destroy (loc);
330 expr_location__ (struct expression *e,
331 const struct expr_node *node,
332 const struct msg_location **minp,
333 const struct msg_location **maxp)
335 struct msg_location *loc = node->location;
338 const struct msg_location *min = *minp;
341 || loc->start.line < min->start.line
342 || (loc->start.line == min->start.line
343 && loc->start.column < min->start.column)))
346 const struct msg_location *max = *maxp;
349 || loc->end.line > max->end.line
350 || (loc->end.line == max->end.line
351 && loc->end.column > max->end.column)))
357 if (is_composite (node->type))
358 for (size_t i = 0; i < node->n_args; i++)
359 expr_location__ (e, node->args[i], minp, maxp);
362 /* Returns the source code location corresponding to expression NODE, computing
363 it lazily if needed. */
364 static const struct msg_location *
365 expr_location (struct expression *e, const struct expr_node *node_)
367 struct expr_node *node = CONST_CAST (struct expr_node *, node_);
373 const struct msg_location *min = NULL;
374 const struct msg_location *max = NULL;
375 expr_location__ (e, node, &min, &max);
378 node->location = msg_location_dup (min);
379 node->location->end = max->end;
380 pool_register (e->expr_pool, free_msg_location, node->location);
383 return node->location;
386 /* Sets e->location to the tokens in S's lexer from offset START_OFS to the
387 token before the current one. Has no effect if E already has a location or
390 expr_add_location (struct lexer *lexer, struct expression *e,
391 int start_ofs, struct expr_node *node)
393 if (node && !node->location)
395 node->location = lex_ofs_location (lexer, start_ofs, lex_ofs (lexer) - 1);
396 pool_register (e->expr_pool, free_msg_location, node->location);
401 type_coercion__ (struct expression *e, struct expr_node *node, size_t arg_idx,
404 assert (!!do_coercion == (e != NULL));
409 struct expr_node **argp = &node->args[arg_idx];
410 struct expr_node *arg = *argp;
414 const struct operation *op = &operations[node->type];
415 atom_type required_type = op->args[MIN (arg_idx, op->n_args - 1)];
416 atom_type actual_type = expr_node_returns (arg);
417 if (actual_type == required_type)
423 switch (required_type)
426 if (actual_type == OP_boolean)
428 /* To enforce strict typing rules, insert Boolean to
429 numeric "conversion". This conversion is a no-op,
430 so it will be removed later. */
432 *argp = expr_allocate_unary (e, OP_BOOLEAN_TO_NUM, arg);
438 /* No coercion to string. */
442 if (actual_type == OP_number)
444 /* Convert numeric to boolean. */
446 *argp = expr_allocate_unary (e, OP_NUM_TO_BOOLEAN, arg);
456 if (arg->type == OP_format
457 && fmt_check_input (&arg->format)
458 && fmt_check_type_compat (&arg->format, VAL_NUMERIC))
462 arg->type = OP_ni_format;
470 if (arg->type == OP_format
471 && fmt_check_output (&arg->format)
472 && fmt_check_type_compat (&arg->format, VAL_NUMERIC))
476 arg->type = OP_no_format;
483 if (arg->type == OP_NUM_VAR)
486 *argp = arg->args[0];
492 if (arg->type == OP_STR_VAR)
495 *argp = arg->args[0];
501 if (arg->type == OP_NUM_VAR || arg->type == OP_STR_VAR)
504 *argp = arg->args[0];
510 if (arg->type == OP_number
511 && floor (arg->number) == arg->number
512 && arg->number > 0 && arg->number < INT_MAX)
515 *argp = expr_allocate_pos_int (e, arg->number);
527 type_coercion (struct expression *e, struct expr_node *node, size_t arg_idx)
529 return type_coercion__ (e, node, arg_idx, true);
533 is_coercible (const struct expr_node *node_, size_t arg_idx)
535 struct expr_node *node = CONST_CAST (struct expr_node *, node_);
536 return type_coercion__ (NULL, node, arg_idx, false);
539 /* How to parse an operator.
541 Some operators support both numeric and string operators. For those,
542 'num_op' and 'str_op' are both nonzero. Otherwise, only one 'num_op' is
543 nonzero. (PSPP doesn't have any string-only operators.) */
546 enum token_type token; /* Operator token. */
547 operation_type num_op; /* Operation for numeric operands (or 0). */
548 operation_type str_op; /* Operation for string operands (or 0). */
551 static operation_type
552 match_operator (struct lexer *lexer, const struct operator ops[], size_t n_ops,
553 const struct expr_node *lhs)
555 bool lhs_is_numeric = operations[lhs->type].returns != OP_string;
556 for (const struct operator *op = ops; op < ops + n_ops; op++)
557 if (lex_token (lexer) == op->token)
559 if (op->token != T_NEG_NUM)
562 return op->str_op && !lhs_is_numeric ? op->str_op : op->num_op;
568 operator_name (enum token_type token)
570 return token == T_NEG_NUM ? "-" : token_type_to_string (token);
573 static struct expr_node *
574 parse_binary_operators__ (struct lexer *lexer, struct expression *e,
575 const struct operator ops[], size_t n_ops,
576 parse_recursively_func *parse_next_level,
577 const char *chain_warning, struct expr_node *lhs)
579 for (int op_count = 0; ; op_count++)
581 enum token_type token = lex_token (lexer);
582 operation_type optype = match_operator (lexer, ops, n_ops, lhs);
585 if (op_count > 1 && chain_warning)
586 msg_at (SW, expr_location (e, lhs), "%s", chain_warning);
591 struct expr_node *rhs = parse_next_level (lexer, e);
595 struct expr_node *node = expr_allocate_binary (e, optype, lhs, rhs);
596 if (!is_coercible (node, 0) || !is_coercible (node, 1))
599 for (size_t i = 0; i < n_ops; i++)
600 if (ops[i].token == token)
601 both = ops[i].num_op && ops[i].str_op;
603 const char *name = operator_name (token);
605 msg_at (SE, expr_location (e, node),
606 _("Both operands of %s must have the same type."), name);
607 else if (operations[node->type].args[0] != OP_string)
608 msg_at (SE, expr_location (e, node),
609 _("Both operands of %s must be numeric."), name);
611 msg_at (SE, expr_location (e, node),
612 _("Both operands of %s must be strings."), name);
614 msg_at (SN, expr_location (e, node->args[0]),
615 _("This operand has type '%s'."),
616 atom_type_name (expr_node_returns (node->args[0])));
617 msg_at (SN, expr_location (e, node->args[1]),
618 _("This operand has type '%s'."),
619 atom_type_name (expr_node_returns (node->args[1])));
624 if (!type_coercion (e, node, 0) || !type_coercion (e, node, 1))
631 static struct expr_node *
632 parse_binary_operators (struct lexer *lexer, struct expression *e,
633 const struct operator ops[], size_t n_ops,
634 parse_recursively_func *parse_next_level,
635 const char *chain_warning)
637 struct expr_node *lhs = parse_next_level (lexer, e);
641 return parse_binary_operators__ (lexer, e, ops, n_ops, parse_next_level,
645 static struct expr_node *
646 parse_inverting_unary_operator (struct lexer *lexer, struct expression *e,
647 const struct operator *op,
648 parse_recursively_func *parse_next_level)
650 int start_ofs = lex_ofs (lexer);
651 unsigned int op_count = 0;
652 while (lex_match (lexer, op->token))
655 struct expr_node *inner = parse_next_level (lexer, e);
656 if (!inner || !op_count)
659 struct expr_node *outer = expr_allocate_unary (e, op->num_op, inner);
660 expr_add_location (lexer, e, start_ofs, outer);
662 if (!type_coercion (e, outer, 0))
664 assert (operations[outer->type].args[0] != OP_string);
666 const char *name = operator_name (op->token);
667 msg_at (SE, expr_location (e, outer),
668 _("The unary %s operator requires a numeric operand."), name);
670 msg_at (SN, expr_location (e, outer->args[0]),
671 _("The operand of %s has type '%s'."),
672 name, atom_type_name (expr_node_returns (outer->args[0])));
677 return op_count % 2 ? outer : outer->args[0];
680 /* Parses the OR level. */
681 static struct expr_node *
682 parse_or (struct lexer *lexer, struct expression *e)
684 static const struct operator op = { .token = T_OR, .num_op = OP_OR };
685 return parse_binary_operators (lexer, e, &op, 1, parse_and, NULL);
688 /* Parses the AND level. */
689 static struct expr_node *
690 parse_and (struct lexer *lexer, struct expression *e)
692 static const struct operator op = { .token = T_AND, .num_op = OP_AND };
694 return parse_binary_operators (lexer, e, &op, 1, parse_not, NULL);
697 /* Parses the NOT level. */
698 static struct expr_node *
699 parse_not (struct lexer *lexer, struct expression *e)
701 static const struct operator op = { .token = T_NOT, .num_op = OP_NOT };
702 return parse_inverting_unary_operator (lexer, e, &op, parse_rel);
705 /* Parse relational operators. */
706 static struct expr_node *
707 parse_rel (struct lexer *lexer, struct expression *e)
709 const char *chain_warning =
710 _("Chaining relational operators (e.g. `a < b < c') will "
711 "not produce the mathematically expected result. "
712 "Use the AND logical operator to fix the problem "
713 "(e.g. `a < b AND b < c'). "
714 "If chaining is really intended, parentheses will disable "
715 "this warning (e.g. `(a < b) < c'.)");
717 static const struct operator ops[] =
719 { .token = T_EQUALS, .num_op = OP_EQ, .str_op = OP_EQ_STRING },
720 { .token = T_EQ, .num_op = OP_EQ, .str_op = OP_EQ_STRING },
721 { .token = T_GE, .num_op = OP_GE, .str_op = OP_GE_STRING },
722 { .token = T_GT, .num_op = OP_GT, .str_op = OP_GT_STRING },
723 { .token = T_LE, .num_op = OP_LE, .str_op = OP_LE_STRING },
724 { .token = T_LT, .num_op = OP_LT, .str_op = OP_LT_STRING },
725 { .token = T_NE, .num_op = OP_NE, .str_op = OP_NE_STRING },
728 return parse_binary_operators (lexer, e, ops, sizeof ops / sizeof *ops,
729 parse_add, chain_warning);
732 /* Parses the addition and subtraction level. */
733 static struct expr_node *
734 parse_add (struct lexer *lexer, struct expression *e)
736 static const struct operator ops[] =
738 { .token = T_PLUS, .num_op = OP_ADD },
739 { .token = T_DASH, .num_op = OP_SUB },
740 { .token = T_NEG_NUM, .num_op = OP_ADD },
743 return parse_binary_operators (lexer, e, ops, sizeof ops / sizeof *ops,
747 /* Parses the multiplication and division level. */
748 static struct expr_node *
749 parse_mul (struct lexer *lexer, struct expression *e)
751 static const struct operator ops[] =
753 { .token = T_ASTERISK, .num_op = OP_MUL },
754 { .token = T_SLASH, .num_op = OP_DIV },
757 return parse_binary_operators (lexer, e, ops, sizeof ops / sizeof *ops,
761 /* Parses the unary minus level. */
762 static struct expr_node *
763 parse_neg (struct lexer *lexer, struct expression *e)
765 static const struct operator op = { .token = T_DASH, .num_op = OP_NEG };
766 return parse_inverting_unary_operator (lexer, e, &op, parse_exp);
769 static struct expr_node *
770 parse_exp (struct lexer *lexer, struct expression *e)
772 static const struct operator op = { .token = T_EXP, .num_op = OP_POW };
774 const char *chain_warning =
775 _("The exponentiation operator (`**') is left-associative, "
776 "even though right-associative semantics are more useful. "
777 "That is, `a**b**c' equals `(a**b)**c', not as `a**(b**c)'. "
778 "To disable this warning, insert parentheses.");
780 if (lex_token (lexer) != T_NEG_NUM || lex_next_token (lexer, 1) != T_EXP)
781 return parse_binary_operators (lexer, e, &op, 1,
782 parse_primary, chain_warning);
784 /* Special case for situations like "-5**6", which must be parsed as
787 int start_ofs = lex_ofs (lexer);
788 struct expr_node *lhs = expr_allocate_number (e, -lex_tokval (lexer));
791 struct expr_node *node = parse_binary_operators__ (
792 lexer, e, &op, 1, parse_primary, chain_warning, lhs);
796 node = expr_allocate_unary (e, OP_NEG, node);
797 expr_add_location (lexer, e, start_ofs, node);
801 /* Parses system variables. */
802 static struct expr_node *
803 parse_sysvar (struct lexer *lexer, struct expression *e)
805 if (lex_match_id (lexer, "$CASENUM"))
806 return expr_allocate_nullary (e, OP_CASENUM);
807 else if (lex_match_id (lexer, "$DATE"))
809 static const char *months[12] =
811 "JAN", "FEB", "MAR", "APR", "MAY", "JUN",
812 "JUL", "AUG", "SEP", "OCT", "NOV", "DEC",
815 time_t last_proc_time = time_of_last_procedure (e->ds);
820 time = localtime (&last_proc_time);
821 sprintf (temp_buf, "%02d %s %02d", abs (time->tm_mday) % 100,
822 months[abs (time->tm_mon) % 12], abs (time->tm_year) % 100);
824 ss_alloc_substring (&s, ss_cstr (temp_buf));
825 return expr_allocate_string (e, s);
827 else if (lex_match_id (lexer, "$TRUE"))
828 return expr_allocate_boolean (e, 1.0);
829 else if (lex_match_id (lexer, "$FALSE"))
830 return expr_allocate_boolean (e, 0.0);
831 else if (lex_match_id (lexer, "$SYSMIS"))
832 return expr_allocate_number (e, SYSMIS);
833 else if (lex_match_id (lexer, "$JDATE"))
835 time_t time = time_of_last_procedure (e->ds);
836 struct tm *tm = localtime (&time);
837 return expr_allocate_number (e, expr_ymd_to_ofs (tm->tm_year + 1900,
841 else if (lex_match_id (lexer, "$TIME"))
843 time_t time = time_of_last_procedure (e->ds);
844 struct tm *tm = localtime (&time);
845 return expr_allocate_number (e,
846 expr_ymd_to_date (tm->tm_year + 1900,
849 + tm->tm_hour * 60 * 60.
853 else if (lex_match_id (lexer, "$LENGTH"))
854 return expr_allocate_number (e, settings_get_viewlength ());
855 else if (lex_match_id (lexer, "$WIDTH"))
856 return expr_allocate_number (e, settings_get_viewwidth ());
859 msg (SE, _("Unknown system variable %s."), lex_tokcstr (lexer));
864 /* Parses numbers, varnames, etc. */
865 static struct expr_node *
866 parse_primary__ (struct lexer *lexer, struct expression *e)
868 switch (lex_token (lexer))
871 if (lex_next_token (lexer, 1) == T_LPAREN)
873 /* An identifier followed by a left parenthesis may be
874 a vector element reference. If not, it's a function
876 if (e->ds != NULL && dict_lookup_vector (dataset_dict (e->ds), lex_tokcstr (lexer)) != NULL)
877 return parse_vector_element (lexer, e);
879 return parse_function (lexer, e);
881 else if (lex_tokcstr (lexer)[0] == '$')
883 /* $ at the beginning indicates a system variable. */
884 return parse_sysvar (lexer, e);
886 else if (e->ds != NULL && dict_lookup_var (dataset_dict (e->ds), lex_tokcstr (lexer)))
888 /* It looks like a user variable.
889 (It could be a format specifier, but we'll assume
890 it's a variable unless proven otherwise. */
891 return allocate_unary_variable (e, parse_variable (lexer, dataset_dict (e->ds)));
895 /* Try to parse it as a format specifier. */
900 ok = parse_format_specifier (lexer, &fmt);
904 return expr_allocate_format (e, &fmt);
906 /* All attempts failed. */
907 msg (SE, _("Unknown identifier %s."), lex_tokcstr (lexer));
915 struct expr_node *node = expr_allocate_number (e, lex_tokval (lexer));
922 const char *dict_encoding;
923 struct expr_node *node;
926 dict_encoding = (e->ds != NULL
927 ? dict_get_encoding (dataset_dict (e->ds))
929 s = recode_string_pool (dict_encoding, "UTF-8", lex_tokcstr (lexer),
930 ss_length (lex_tokss (lexer)), e->expr_pool);
931 node = expr_allocate_string (e, ss_cstr (s));
939 /* Count number of left parentheses so that we can match them against
940 an equal number of right parentheses. This defeats trivial attempts
941 to exhaust the stack with a lot of left parentheses. (More
942 sophisticated attacks will still succeed.) */
944 while (lex_match (lexer, T_LPAREN))
947 struct expr_node *node = parse_or (lexer, e);
951 for (size_t i = 0; i < n; i++)
952 if (!lex_force_match (lexer, T_RPAREN))
959 lex_error (lexer, NULL);
964 static struct expr_node *
965 parse_primary (struct lexer *lexer, struct expression *e)
967 int start_ofs = lex_ofs (lexer);
968 struct expr_node *node = parse_primary__ (lexer, e);
969 expr_add_location (lexer, e, start_ofs, node);
973 static struct expr_node *
974 parse_vector_element (struct lexer *lexer, struct expression *e)
976 int vector_start_ofs = lex_ofs (lexer);
978 /* Find vector, skip token.
979 The caller must already have verified that the current token
980 is the name of a vector. */
981 const struct vector *vector = dict_lookup_vector (dataset_dict (e->ds),
982 lex_tokcstr (lexer));
983 assert (vector != NULL);
986 /* Skip left parenthesis token.
987 The caller must have verified that the lookahead is a left
989 assert (lex_token (lexer) == T_LPAREN);
992 int element_start_ofs = lex_ofs (lexer);
993 struct expr_node *element = parse_or (lexer, e);
996 expr_add_location (lexer, e, element_start_ofs, element);
998 if (!lex_match (lexer, T_RPAREN))
1001 operation_type type = (vector_get_type (vector) == VAL_NUMERIC
1002 ? OP_VEC_ELEM_NUM : OP_VEC_ELEM_STR);
1003 struct expr_node *node = expr_allocate_binary (
1004 e, type, element, expr_allocate_vector (e, vector));
1005 expr_add_location (lexer, e, vector_start_ofs, node);
1007 if (!type_coercion (e, node, 1))
1009 msg_at (SE, expr_location (e, node),
1010 _("A vector index must be numeric."));
1012 msg_at (SN, expr_location (e, node->args[0]),
1013 _("This vector index has type '%s'."),
1014 atom_type_name (expr_node_returns (node->args[0])));
1022 /* Individual function parsing. */
1024 const struct operation operations[OP_first + n_OP] = {
1025 #include "parse.inc"
1029 word_matches (const char **test, const char **name)
1031 size_t test_len = strcspn (*test, ".");
1032 size_t name_len = strcspn (*name, ".");
1033 if (test_len == name_len)
1035 if (buf_compare_case (*test, *name, test_len))
1038 else if (test_len < 3 || test_len > name_len)
1042 if (buf_compare_case (*test, *name, test_len))
1048 if (**test != **name)
1059 /* Returns 0 if TOKEN and FUNC do not match,
1060 1 if TOKEN is an acceptable abbreviation for FUNC,
1061 2 if TOKEN equals FUNC. */
1063 compare_function_names (const char *token_, const char *func_)
1065 const char *token = token_;
1066 const char *func = func_;
1067 while (*token || *func)
1068 if (!word_matches (&token, &func))
1070 return !c_strcasecmp (token_, func_) ? 2 : 1;
1074 lookup_function (const char *token,
1075 const struct operation **first,
1076 const struct operation **last)
1078 *first = *last = NULL;
1079 const struct operation *best = NULL;
1081 for (const struct operation *f = operations + OP_function_first;
1082 f <= operations + OP_function_last; f++)
1084 int score = compare_function_names (token, f->name);
1090 else if (score == 1 && !(f->flags & OPF_NO_ABBREV) && !best)
1099 const struct operation *f = best;
1100 while (f <= operations + OP_function_last
1101 && !c_strcasecmp (f->name, best->name))
1109 extract_min_valid (const char *s)
1111 char *p = strrchr (s, '.');
1113 || p[1] < '0' || p[1] > '9'
1114 || strspn (p + 1, "0123456789") != strlen (p + 1))
1117 return atoi (p + 1);
1121 match_function__ (struct expr_node *node, const struct operation *f)
1123 if (node->n_args < f->n_args
1124 || (node->n_args > f->n_args && (f->flags & OPF_ARRAY_OPERAND) == 0)
1125 || node->n_args - (f->n_args - 1) < f->array_min_elems)
1128 node->type = f - operations;
1129 for (size_t i = 0; i < node->n_args; i++)
1130 if (!is_coercible (node, i))
1136 static const struct operation *
1137 match_function (struct expr_node *node,
1138 const struct operation *first, const struct operation *last)
1140 for (const struct operation *f = first; f < last; f++)
1141 if (match_function__ (node, f))
1147 validate_function_args (const struct operation *f, int n_args, int min_valid)
1149 /* Count the function arguments that go into the trailing array (if any). We
1150 know that there must be at least the minimum number because
1151 match_function() already checked. */
1152 int array_n_args = n_args - (f->n_args - 1);
1153 assert (array_n_args >= f->array_min_elems);
1155 if ((f->flags & OPF_ARRAY_OPERAND)
1156 && array_n_args % f->array_granularity != 0)
1158 /* RANGE is the only case we have so far. It has paired arguments with
1159 one initial argument, and that's the only special case we deal with
1161 assert (f->array_granularity == 2);
1162 assert (n_args % 2 == 0);
1163 msg (SE, _("%s must have an odd number of arguments."), f->prototype);
1167 if (min_valid != -1)
1169 if (f->array_min_elems == 0)
1171 assert ((f->flags & OPF_MIN_VALID) == 0);
1172 msg (SE, _("%s function cannot accept suffix .%d to specify the "
1173 "minimum number of valid arguments."),
1174 f->prototype, min_valid);
1179 assert (f->flags & OPF_MIN_VALID);
1180 if (min_valid > array_n_args)
1182 msg (SE, _("For %s with %d arguments, at most %d (not %d) may be "
1183 "required to be valid."),
1184 f->prototype, n_args, array_n_args, min_valid);
1194 add_arg (struct expr_node ***args, size_t *n_args, size_t *allocated_args,
1195 struct expr_node *arg,
1196 struct expression *e, struct lexer *lexer, int arg_start_ofs)
1198 if (*n_args >= *allocated_args)
1199 *args = x2nrealloc (*args, allocated_args, sizeof **args);
1201 expr_add_location (lexer, e, arg_start_ofs, arg);
1202 (*args)[(*n_args)++] = arg;
1206 put_invocation (struct string *s,
1207 const char *func_name, struct expr_node *node)
1211 ds_put_format (s, "%s(", func_name);
1212 for (i = 0; i < node->n_args; i++)
1215 ds_put_cstr (s, ", ");
1216 ds_put_cstr (s, operations[expr_node_returns (node->args[i])].prototype);
1218 ds_put_byte (s, ')');
1222 no_match (struct expression *e, const char *func_name, struct expr_node *node,
1223 const struct operation *first, const struct operation *last)
1226 const struct operation *f;
1230 if (last - first == 1)
1232 ds_put_format (&s, _("Type mismatch invoking %s as "), first->prototype);
1233 put_invocation (&s, func_name, node);
1237 ds_put_cstr (&s, _("Function invocation "));
1238 put_invocation (&s, func_name, node);
1239 ds_put_cstr (&s, _(" does not match any known function. Candidates are:"));
1241 for (f = first; f < last; f++)
1242 ds_put_format (&s, "\n%s", f->prototype);
1244 ds_put_byte (&s, '.');
1246 msg_at (SE, expr_location (e, node), "%s", ds_cstr (&s));
1251 static struct expr_node *
1252 parse_function (struct lexer *lexer, struct expression *e)
1254 struct string func_name;
1255 ds_init_substring (&func_name, lex_tokss (lexer));
1257 int min_valid = extract_min_valid (lex_tokcstr (lexer));
1259 const struct operation *first, *last;
1260 if (!lookup_function (lex_tokcstr (lexer), &first, &last))
1262 msg (SE, _("No function or vector named %s."), lex_tokcstr (lexer));
1263 ds_destroy (&func_name);
1267 int func_start_ofs = lex_ofs (lexer);
1269 if (!lex_force_match (lexer, T_LPAREN))
1271 ds_destroy (&func_name);
1275 struct expr_node **args = NULL;
1277 size_t allocated_args = 0;
1278 if (lex_token (lexer) != T_RPAREN)
1281 int arg_start_ofs = lex_ofs (lexer);
1282 if (lex_token (lexer) == T_ID
1283 && lex_next_token (lexer, 1) == T_TO)
1285 const struct variable **vars;
1288 if (!parse_variables_const (lexer, dataset_dict (e->ds),
1289 &vars, &n_vars, PV_SINGLE))
1291 for (size_t i = 0; i < n_vars; i++)
1292 add_arg (&args, &n_args, &allocated_args,
1293 allocate_unary_variable (e, vars[i]),
1294 e, lexer, arg_start_ofs);
1299 struct expr_node *arg = parse_or (lexer, e);
1303 add_arg (&args, &n_args, &allocated_args, arg,
1304 e, lexer, arg_start_ofs);
1306 if (lex_match (lexer, T_RPAREN))
1308 else if (!lex_match (lexer, T_COMMA))
1310 lex_error_expecting (lexer, "`,'", "`)'");
1315 struct expr_node *n = expr_allocate_composite (e, first - operations,
1317 expr_add_location (lexer, e, func_start_ofs, n);
1318 const struct operation *f = match_function (n, first, last);
1321 no_match (e, ds_cstr (&func_name), n, first, last);
1324 n->type = f - operations;
1325 n->min_valid = min_valid != -1 ? min_valid : f->array_min_elems;
1327 for (size_t i = 0; i < n_args; i++)
1328 if (!type_coercion (e, n, i))
1330 /* Unreachable because match_function already checked that the
1331 arguments were coercible. */
1334 if (!validate_function_args (f, n_args, min_valid))
1337 if ((f->flags & OPF_EXTENSION) && settings_get_syntax () == COMPATIBLE)
1338 msg (SW, _("%s is a PSPP extension."), f->prototype);
1339 if (f->flags & OPF_UNIMPLEMENTED)
1341 msg (SE, _("%s is not available in this version of PSPP."),
1345 if ((f->flags & OPF_PERM_ONLY) &&
1346 proc_in_temporary_transformations (e->ds))
1348 msg (SE, _("%s may not appear after %s."), f->prototype, "TEMPORARY");
1352 if (n->type == OP_LAG_Vn || n->type == OP_LAG_Vs)
1353 dataset_need_lag (e->ds, 1);
1354 else if (n->type == OP_LAG_Vnn || n->type == OP_LAG_Vsn)
1356 assert (n->n_args == 2);
1357 assert (n->args[1]->type == OP_pos_int);
1358 dataset_need_lag (e->ds, n->args[1]->integer);
1362 ds_destroy (&func_name);
1367 ds_destroy (&func_name);
1371 /* Utility functions. */
1373 static struct expression *
1374 expr_create (struct dataset *ds)
1376 struct pool *pool = pool_create ();
1377 struct expression *e = pool_alloc (pool, sizeof *e);
1378 *e = (struct expression) {
1381 .eval_pool = pool_create_subpool (pool),
1387 expr_node_returns (const struct expr_node *n)
1390 assert (is_operation (n->type));
1391 if (is_atom (n->type))
1393 else if (is_composite (n->type))
1394 return operations[n->type].returns;
1400 atom_type_name (atom_type type)
1402 assert (is_atom (type));
1404 /* The Boolean type is purely an internal concept that the documentation
1405 doesn't mention, so it might confuse users if we talked about them in
1407 return type == OP_boolean ? "number" : operations[type].name;
1411 expr_allocate_nullary (struct expression *e, operation_type op)
1413 return expr_allocate_composite (e, op, NULL, 0);
1417 expr_allocate_unary (struct expression *e, operation_type op,
1418 struct expr_node *arg0)
1420 return expr_allocate_composite (e, op, &arg0, 1);
1424 expr_allocate_binary (struct expression *e, operation_type op,
1425 struct expr_node *arg0, struct expr_node *arg1)
1427 struct expr_node *args[2];
1430 return expr_allocate_composite (e, op, args, 2);
1434 expr_allocate_composite (struct expression *e, operation_type op,
1435 struct expr_node **args, size_t n_args)
1437 for (size_t i = 0; i < n_args; i++)
1441 struct expr_node *n = pool_alloc (e->expr_pool, sizeof *n);
1442 *n = (struct expr_node) {
1445 .args = pool_clone (e->expr_pool, args, sizeof *n->args * n_args),
1451 expr_allocate_number (struct expression *e, double d)
1453 struct expr_node *n = pool_alloc (e->expr_pool, sizeof *n);
1454 *n = (struct expr_node) { .type = OP_number, .number = d };
1459 expr_allocate_boolean (struct expression *e, double b)
1461 assert (b == 0.0 || b == 1.0 || b == SYSMIS);
1463 struct expr_node *n = pool_alloc (e->expr_pool, sizeof *n);
1464 *n = (struct expr_node) { .type = OP_boolean, .number = b };
1469 expr_allocate_integer (struct expression *e, int i)
1471 struct expr_node *n = pool_alloc (e->expr_pool, sizeof *n);
1472 *n = (struct expr_node) { .type = OP_integer, .integer = i };
1477 expr_allocate_pos_int (struct expression *e, int i)
1481 struct expr_node *n = pool_alloc (e->expr_pool, sizeof *n);
1482 *n = (struct expr_node) { .type = OP_pos_int, .integer = i };
1487 expr_allocate_vector (struct expression *e, const struct vector *vector)
1489 struct expr_node *n = pool_alloc (e->expr_pool, sizeof *n);
1490 *n = (struct expr_node) { .type = OP_vector, .vector = vector };
1495 expr_allocate_string (struct expression *e, struct substring s)
1497 struct expr_node *n = pool_alloc (e->expr_pool, sizeof *n);
1498 *n = (struct expr_node) { .type = OP_string, .string = s };
1503 expr_allocate_variable (struct expression *e, const struct variable *v)
1505 struct expr_node *n = pool_alloc (e->expr_pool, sizeof *n);
1506 *n = (struct expr_node) {
1507 .type = var_is_numeric (v) ? OP_num_var : OP_str_var,
1514 expr_allocate_format (struct expression *e, const struct fmt_spec *format)
1516 struct expr_node *n = pool_alloc (e->expr_pool, sizeof *n);
1517 *n = (struct expr_node) { .type = OP_format, .format = *format };
1521 /* Allocates a unary composite node that represents the value of
1522 variable V in expression E. */
1523 static struct expr_node *
1524 allocate_unary_variable (struct expression *e, const struct variable *v)
1527 return expr_allocate_unary (e, var_is_numeric (v) ? OP_NUM_VAR : OP_STR_VAR,
1528 expr_allocate_variable (e, v));
1531 /* Export function details to other modules. */
1533 /* Returns the operation structure for the function with the
1535 const struct operation *
1536 expr_get_function (size_t idx)
1538 assert (idx < n_OP_function);
1539 return &operations[OP_function_first + idx];
1542 /* Returns the number of expression functions. */
1544 expr_get_n_functions (void)
1546 return n_OP_function;
1549 /* Returns the name of operation OP. */
1551 expr_operation_get_name (const struct operation *op)
1556 /* Returns the human-readable prototype for operation OP. */
1558 expr_operation_get_prototype (const struct operation *op)
1560 return op->prototype;
1563 /* Returns the number of arguments for operation OP. */
1565 expr_operation_get_n_args (const struct operation *op)