1 /* C preprocessor macro expansion for GDB.
2 Copyright (C) 2002, 2007, 2008, 2009, 2010 Free Software Foundation, Inc.
3 Contributed by Red Hat, Inc.
5 This file is part of GDB.
7 This program is free software; you can redistribute it and/or modify
8 it under the terms of the GNU General Public License as published by
9 the Free Software Foundation; either version 3 of the License, or
10 (at your option) any later version.
12 This program is distributed in the hope that it will be useful,
13 but WITHOUT ANY WARRANTY; without even the implied warranty of
14 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
15 GNU General Public License for more details.
17 You should have received a copy of the GNU General Public License
18 along with this program. If not, see <http://www.gnu.org/licenses/>. */
21 #include "gdb_obstack.h"
25 #include "gdb_assert.h"
30 /* A resizeable, substringable string type. */
33 /* A string type that we can resize, quickly append to, and use to
34 refer to substrings of other strings. */
37 /* An array of characters. The first LEN bytes are the real text,
38 but there are SIZE bytes allocated to the array. If SIZE is
39 zero, then this doesn't point to a malloc'ed block. If SHARED is
40 non-zero, then this buffer is actually a pointer into some larger
41 string, and we shouldn't append characters to it, etc. Because
42 of sharing, we can't assume in general that the text is
46 /* The number of characters in the string. */
49 /* The number of characters allocated to the string. If SHARED is
50 non-zero, this is meaningless; in this case, we set it to zero so
51 that any "do we have room to append something?" tests will fail,
52 so we don't always have to check SHARED before using this field. */
55 /* Zero if TEXT can be safely realloc'ed (i.e., it's its own malloc
56 block). Non-zero if TEXT is actually pointing into the middle of
57 some other block, and we shouldn't reallocate it. */
60 /* For detecting token splicing.
62 This is the index in TEXT of the first character of the token
63 that abuts the end of TEXT. If TEXT contains no tokens, then we
64 set this equal to LEN. If TEXT ends in whitespace, then there is
65 no token abutting the end of TEXT (it's just whitespace), and
66 again, we set this equal to LEN. We set this to -1 if we don't
67 know the nature of TEXT. */
70 /* If this buffer is holding the result from get_token, then this
71 is non-zero if it is an identifier token, zero otherwise. */
76 /* Set the macro buffer *B to the empty string, guessing that its
77 final contents will fit in N bytes. (It'll get resized if it
78 doesn't, so the guess doesn't have to be right.) Allocate the
79 initial storage with xmalloc. */
81 init_buffer (struct macro_buffer *b, int n)
85 b->text = (char *) xmalloc (n);
94 /* Set the macro buffer *BUF to refer to the LEN bytes at ADDR, as a
97 init_shared_buffer (struct macro_buffer *buf, char *addr, int len)
103 buf->last_token = -1;
107 /* Free the text of the buffer B. Raise an error if B is shared. */
109 free_buffer (struct macro_buffer *b)
111 gdb_assert (! b->shared);
117 /* A cleanup function for macro buffers. */
119 cleanup_macro_buffer (void *untyped_buf)
121 free_buffer ((struct macro_buffer *) untyped_buf);
125 /* Resize the buffer B to be at least N bytes long. Raise an error if
126 B shouldn't be resized. */
128 resize_buffer (struct macro_buffer *b, int n)
130 /* We shouldn't be trying to resize shared strings. */
131 gdb_assert (! b->shared);
139 b->text = xrealloc (b->text, b->size);
143 /* Append the character C to the buffer B. */
145 appendc (struct macro_buffer *b, int c)
147 int new_len = b->len + 1;
149 if (new_len > b->size)
150 resize_buffer (b, new_len);
157 /* Append the LEN bytes at ADDR to the buffer B. */
159 appendmem (struct macro_buffer *b, char *addr, int len)
161 int new_len = b->len + len;
163 if (new_len > b->size)
164 resize_buffer (b, new_len);
166 memcpy (b->text + b->len, addr, len);
172 /* Recognizing preprocessor tokens. */
176 macro_is_whitespace (int c)
187 macro_is_digit (int c)
189 return ('0' <= c && c <= '9');
194 macro_is_identifier_nondigit (int c)
197 || ('a' <= c && c <= 'z')
198 || ('A' <= c && c <= 'Z'));
203 set_token (struct macro_buffer *tok, char *start, char *end)
205 init_shared_buffer (tok, start, end - start);
208 /* Presumed; get_identifier may overwrite this. */
209 tok->is_identifier = 0;
214 get_comment (struct macro_buffer *tok, char *p, char *end)
231 set_token (tok, tok_start, p);
235 error (_("Unterminated comment in macro expansion."));
247 set_token (tok, tok_start, p);
256 get_identifier (struct macro_buffer *tok, char *p, char *end)
259 && macro_is_identifier_nondigit (*p))
264 && (macro_is_identifier_nondigit (*p)
265 || macro_is_digit (*p)))
268 set_token (tok, tok_start, p);
269 tok->is_identifier = 1;
278 get_pp_number (struct macro_buffer *tok, char *p, char *end)
281 && (macro_is_digit (*p)
284 && macro_is_digit (p[1]))))
291 && strchr ("eEpP", *p)
292 && (p[1] == '+' || p[1] == '-'))
294 else if (macro_is_digit (*p)
295 || macro_is_identifier_nondigit (*p)
302 set_token (tok, tok_start, p);
311 /* If the text starting at P going up to (but not including) END
312 starts with a character constant, set *TOK to point to that
313 character constant, and return 1. Otherwise, return zero.
314 Signal an error if it contains a malformed or incomplete character
317 get_character_constant (struct macro_buffer *tok, char *p, char *end)
319 /* ISO/IEC 9899:1999 (E) Section 6.4.4.4 paragraph 1
320 But of course, what really matters is that we handle it the same
321 way GDB's C/C++ lexer does. So we call parse_escape in utils.c
322 to handle escape sequences. */
323 if ((p + 1 <= end && *p == '\'')
325 && (p[0] == 'L' || p[0] == 'u' || p[0] == 'U')
334 else if (*p == 'L' || *p == 'u' || *p == 'U')
343 error (_("Unmatched single quote."));
347 error (_("A character constant must contain at least one "
355 char_count += c_parse_escape (&p, NULL);
364 set_token (tok, tok_start, p);
372 /* If the text starting at P going up to (but not including) END
373 starts with a string literal, set *TOK to point to that string
374 literal, and return 1. Otherwise, return zero. Signal an error if
375 it contains a malformed or incomplete string literal. */
377 get_string_literal (struct macro_buffer *tok, char *p, char *end)
382 && (p[0] == 'L' || p[0] == 'u' || p[0] == 'U')
389 else if (*p == 'L' || *p == 'u' || *p == 'U')
397 error (_("Unterminated string in expression."));
404 error (_("Newline characters may not appear in string "
409 c_parse_escape (&p, NULL);
415 set_token (tok, tok_start, p);
424 get_punctuator (struct macro_buffer *tok, char *p, char *end)
426 /* Here, speed is much less important than correctness and clarity. */
428 /* ISO/IEC 9899:1999 (E) Section 6.4.6 Paragraph 1.
429 Note that this table is ordered in a special way. A punctuator
430 which is a prefix of another punctuator must appear after its
431 "extension". Otherwise, the wrong token will be returned. */
432 static const char * const punctuators[] = {
433 "[", "]", "(", ")", "{", "}", "?", ";", ",", "~",
435 "->", "--", "-=", "-",
441 "%>", "%:%:", "%:", "%=", "%",
446 "<<=", "<<", "<=", "<:", "<%", "<",
447 ">>=", ">>", ">=", ">",
456 for (i = 0; punctuators[i]; i++)
458 const char *punctuator = punctuators[i];
460 if (p[0] == punctuator[0])
462 int len = strlen (punctuator);
465 && ! memcmp (p, punctuator, len))
467 set_token (tok, p, p + len);
478 /* Peel the next preprocessor token off of SRC, and put it in TOK.
479 Mutate TOK to refer to the first token in SRC, and mutate SRC to
480 refer to the text after that token. SRC must be a shared buffer;
481 the resulting TOK will be shared, pointing into the same string SRC
482 does. Initialize TOK's last_token field. Return non-zero if we
483 succeed, or 0 if we didn't find any more tokens in SRC. */
485 get_token (struct macro_buffer *tok,
486 struct macro_buffer *src)
489 char *end = p + src->len;
491 gdb_assert (src->shared);
493 /* From the ISO C standard, ISO/IEC 9899:1999 (E), section 6.4:
502 each non-white-space character that cannot be one of the above
504 We don't have to deal with header-name tokens, since those can
505 only occur after a #include, which we will never see. */
508 if (macro_is_whitespace (*p))
510 else if (get_comment (tok, p, end))
512 else if (get_pp_number (tok, p, end)
513 || get_character_constant (tok, p, end)
514 || get_string_literal (tok, p, end)
515 /* Note: the grammar in the standard seems to be
516 ambiguous: L'x' can be either a wide character
517 constant, or an identifier followed by a normal
518 character constant. By trying `get_identifier' after
519 we try get_character_constant and get_string_literal,
520 we give the wide character syntax precedence. Now,
521 since GDB doesn't handle wide character constants
522 anyway, is this the right thing to do? */
523 || get_identifier (tok, p, end)
524 || get_punctuator (tok, p, end))
526 /* How many characters did we consume, including whitespace? */
527 int consumed = p - src->text + tok->len;
528 src->text += consumed;
529 src->len -= consumed;
534 /* We have found a "non-whitespace character that cannot be
535 one of the above." Make a token out of it. */
538 set_token (tok, p, p + 1);
539 consumed = p - src->text + tok->len;
540 src->text += consumed;
541 src->len -= consumed;
550 /* Appending token strings, with and without splicing */
553 /* Append the macro buffer SRC to the end of DEST, and ensure that
554 doing so doesn't splice the token at the end of SRC with the token
555 at the beginning of DEST. SRC and DEST must have their last_token
556 fields set. Upon return, DEST's last_token field is set correctly.
560 If DEST is "(" and SRC is "y", then we can return with
561 DEST set to "(y" --- we've simply appended the two buffers.
563 However, if DEST is "x" and SRC is "y", then we must not return
564 with DEST set to "xy" --- that would splice the two tokens "x" and
565 "y" together to make a single token "xy". However, it would be
566 fine to return with DEST set to "x y". Similarly, "<" and "<" must
567 yield "< <", not "<<", etc. */
569 append_tokens_without_splicing (struct macro_buffer *dest,
570 struct macro_buffer *src)
572 int original_dest_len = dest->len;
573 struct macro_buffer dest_tail, new_token;
575 gdb_assert (src->last_token != -1);
576 gdb_assert (dest->last_token != -1);
578 /* First, just try appending the two, and call get_token to see if
580 appendmem (dest, src->text, src->len);
582 /* If DEST originally had no token abutting its end, then we can't
583 have spliced anything, so we're done. */
584 if (dest->last_token == original_dest_len)
586 dest->last_token = original_dest_len + src->last_token;
590 /* Set DEST_TAIL to point to the last token in DEST, followed by
591 all the stuff we just appended. */
592 init_shared_buffer (&dest_tail,
593 dest->text + dest->last_token,
594 dest->len - dest->last_token);
596 /* Re-parse DEST's last token. We know that DEST used to contain
597 at least one token, so if it doesn't contain any after the
598 append, then we must have spliced "/" and "*" or "/" and "/" to
599 make a comment start. (Just for the record, I got this right
600 the first time. This is not a bug fix.) */
601 if (get_token (&new_token, &dest_tail)
602 && (new_token.text + new_token.len
603 == dest->text + original_dest_len))
605 /* No splice, so we're done. */
606 dest->last_token = original_dest_len + src->last_token;
610 /* Okay, a simple append caused a splice. Let's chop dest back to
611 its original length and try again, but separate the texts with a
613 dest->len = original_dest_len;
615 appendmem (dest, src->text, src->len);
617 init_shared_buffer (&dest_tail,
618 dest->text + dest->last_token,
619 dest->len - dest->last_token);
621 /* Try to re-parse DEST's last token, as above. */
622 if (get_token (&new_token, &dest_tail)
623 && (new_token.text + new_token.len
624 == dest->text + original_dest_len))
626 /* No splice, so we're done. */
627 dest->last_token = original_dest_len + 1 + src->last_token;
631 /* As far as I know, there's no case where inserting a space isn't
632 enough to prevent a splice. */
633 internal_error (__FILE__, __LINE__,
634 _("unable to avoid splicing tokens during macro expansion"));
637 /* Stringify an argument, and insert it into DEST. ARG is the text to
638 stringify; it is LEN bytes long. */
641 stringify (struct macro_buffer *dest, char *arg, int len)
643 /* Trim initial whitespace from ARG. */
644 while (len > 0 && macro_is_whitespace (*arg))
650 /* Trim trailing whitespace from ARG. */
651 while (len > 0 && macro_is_whitespace (arg[len - 1]))
654 /* Insert the string. */
658 /* We could try to handle strange cases here, like control
659 characters, but there doesn't seem to be much point. */
660 if (macro_is_whitespace (*arg))
662 /* Replace a sequence of whitespace with a single space. */
664 while (len > 1 && macro_is_whitespace (arg[1]))
670 else if (*arg == '\\' || *arg == '"')
672 appendc (dest, '\\');
673 appendc (dest, *arg);
676 appendc (dest, *arg);
681 dest->last_token = dest->len;
685 /* Expanding macros! */
688 /* A singly-linked list of the names of the macros we are currently
689 expanding --- for detecting expansion loops. */
690 struct macro_name_list {
692 struct macro_name_list *next;
696 /* Return non-zero if we are currently expanding the macro named NAME,
697 according to LIST; otherwise, return zero.
699 You know, it would be possible to get rid of all the NO_LOOP
700 arguments to these functions by simply generating a new lookup
701 function and baton which refuses to find the definition for a
702 particular macro, and otherwise delegates the decision to another
703 function/baton pair. But that makes the linked list of excluded
704 macros chained through untyped baton pointers, which will make it
705 harder to debug. :( */
707 currently_rescanning (struct macro_name_list *list, const char *name)
709 for (; list; list = list->next)
710 if (strcmp (name, list->name) == 0)
717 /* Gather the arguments to a macro expansion.
719 NAME is the name of the macro being invoked. (It's only used for
720 printing error messages.)
722 Assume that SRC is the text of the macro invocation immediately
723 following the macro name. For example, if we're processing the
724 text foo(bar, baz), then NAME would be foo and SRC will be (bar,
727 If SRC doesn't start with an open paren ( token at all, return
728 zero, leave SRC unchanged, and don't set *ARGC_P to anything.
730 If SRC doesn't contain a properly terminated argument list, then
733 For a variadic macro, NARGS holds the number of formal arguments to
734 the macro. For a GNU-style variadic macro, this should be the
735 number of named arguments. For a non-variadic macro, NARGS should
738 Otherwise, return a pointer to the first element of an array of
739 macro buffers referring to the argument texts, and set *ARGC_P to
740 the number of arguments we found --- the number of elements in the
741 array. The macro buffers share their text with SRC, and their
742 last_token fields are initialized. The array is allocated with
743 xmalloc, and the caller is responsible for freeing it.
745 NOTE WELL: if SRC starts with a open paren ( token followed
746 immediately by a close paren ) token (e.g., the invocation looks
747 like "foo()"), we treat that as one argument, which happens to be
748 the empty list of tokens. The caller should keep in mind that such
749 a sequence of tokens is a valid way to invoke one-parameter
750 function-like macros, but also a valid way to invoke zero-parameter
751 function-like macros. Eeew.
753 Consume the tokens from SRC; after this call, SRC contains the text
754 following the invocation. */
756 static struct macro_buffer *
757 gather_arguments (const char *name, struct macro_buffer *src,
758 int nargs, int *argc_p)
760 struct macro_buffer tok;
761 int args_len, args_size;
762 struct macro_buffer *args = NULL;
763 struct cleanup *back_to = make_cleanup (free_current_contents, &args);
765 /* Does SRC start with an opening paren token? Read from a copy of
766 SRC, so SRC itself is unaffected if we don't find an opening
769 struct macro_buffer temp;
770 init_shared_buffer (&temp, src->text, src->len);
772 if (! get_token (&tok, &temp)
774 || tok.text[0] != '(')
776 discard_cleanups (back_to);
781 /* Consume SRC's opening paren. */
782 get_token (&tok, src);
786 args = (struct macro_buffer *) xmalloc (sizeof (*args) * args_size);
790 struct macro_buffer *arg;
793 /* Make sure we have room for the next argument. */
794 if (args_len >= args_size)
797 args = xrealloc (args, sizeof (*args) * args_size);
800 /* Initialize the next argument. */
801 arg = &args[args_len++];
802 set_token (arg, src->text, src->text);
804 /* Gather the argument's tokens. */
808 if (! get_token (&tok, src))
809 error (_("Malformed argument list for macro `%s'."), name);
811 /* Is tok an opening paren? */
812 if (tok.len == 1 && tok.text[0] == '(')
815 /* Is tok is a closing paren? */
816 else if (tok.len == 1 && tok.text[0] == ')')
818 /* If it's a closing paren at the top level, then that's
819 the end of the argument list. */
822 /* In the varargs case, the last argument may be
823 missing. Add an empty argument in this case. */
824 if (nargs != -1 && args_len == nargs - 1)
826 /* Make sure we have room for the argument. */
827 if (args_len >= args_size)
830 args = xrealloc (args, sizeof (*args) * args_size);
832 arg = &args[args_len++];
833 set_token (arg, src->text, src->text);
836 discard_cleanups (back_to);
844 /* If tok is a comma at top level, then that's the end of
845 the current argument. However, if we are handling a
846 variadic macro and we are computing the last argument, we
847 want to include the comma and remaining tokens. */
848 else if (tok.len == 1 && tok.text[0] == ',' && depth == 0
849 && (nargs == -1 || args_len < nargs))
852 /* Extend the current argument to enclose this token. If
853 this is the current argument's first token, leave out any
854 leading whitespace, just for aesthetics. */
857 arg->text = tok.text;
863 arg->len = (tok.text + tok.len) - arg->text;
864 arg->last_token = tok.text - arg->text;
871 /* The `expand' and `substitute_args' functions both invoke `scan'
872 recursively, so we need a forward declaration somewhere. */
873 static void scan (struct macro_buffer *dest,
874 struct macro_buffer *src,
875 struct macro_name_list *no_loop,
876 macro_lookup_ftype *lookup_func,
880 /* A helper function for substitute_args.
882 ARGV is a vector of all the arguments; ARGC is the number of
883 arguments. IS_VARARGS is true if the macro being substituted is a
884 varargs macro; in this case VA_ARG_NAME is the name of the
885 "variable" argument. VA_ARG_NAME is ignored if IS_VARARGS is
888 If the token TOK is the name of a parameter, return the parameter's
889 index. If TOK is not an argument, return -1. */
892 find_parameter (const struct macro_buffer *tok,
893 int is_varargs, const struct macro_buffer *va_arg_name,
894 int argc, const char * const *argv)
898 if (! tok->is_identifier)
901 for (i = 0; i < argc; ++i)
902 if (tok->len == strlen (argv[i]) && ! memcmp (tok->text, argv[i], tok->len))
905 if (is_varargs && tok->len == va_arg_name->len
906 && ! memcmp (tok->text, va_arg_name->text, tok->len))
912 /* Given the macro definition DEF, being invoked with the actual
913 arguments given by ARGC and ARGV, substitute the arguments into the
914 replacement list, and store the result in DEST.
916 IS_VARARGS should be true if DEF is a varargs macro. In this case,
917 VA_ARG_NAME should be the name of the "variable" argument -- either
918 __VA_ARGS__ for c99-style varargs, or the final argument name, for
919 GNU-style varargs. If IS_VARARGS is false, this parameter is
922 If it is necessary to expand macro invocations in one of the
923 arguments, use LOOKUP_FUNC and LOOKUP_BATON to find the macro
924 definitions, and don't expand invocations of the macros listed in
928 substitute_args (struct macro_buffer *dest,
929 struct macro_definition *def,
930 int is_varargs, const struct macro_buffer *va_arg_name,
931 int argc, struct macro_buffer *argv,
932 struct macro_name_list *no_loop,
933 macro_lookup_ftype *lookup_func,
936 /* A macro buffer for the macro's replacement list. */
937 struct macro_buffer replacement_list;
938 /* The token we are currently considering. */
939 struct macro_buffer tok;
940 /* The replacement list's pointer from just before TOK was lexed. */
941 char *original_rl_start;
942 /* We have a single lookahead token to handle token splicing. */
943 struct macro_buffer lookahead;
944 /* The lookahead token might not be valid. */
946 /* The replacement list's pointer from just before LOOKAHEAD was
948 char *lookahead_rl_start;
950 init_shared_buffer (&replacement_list, (char *) def->replacement,
951 strlen (def->replacement));
953 gdb_assert (dest->len == 0);
954 dest->last_token = 0;
956 original_rl_start = replacement_list.text;
957 if (! get_token (&tok, &replacement_list))
959 lookahead_rl_start = replacement_list.text;
960 lookahead_valid = get_token (&lookahead, &replacement_list);
964 /* Just for aesthetics. If we skipped some whitespace, copy
966 if (tok.text > original_rl_start)
968 appendmem (dest, original_rl_start, tok.text - original_rl_start);
969 dest->last_token = dest->len;
972 /* Is this token the stringification operator? */
974 && tok.text[0] == '#')
978 if (!lookahead_valid)
979 error (_("Stringification operator requires an argument."));
981 arg = find_parameter (&lookahead, is_varargs, va_arg_name,
982 def->argc, def->argv);
984 error (_("Argument to stringification operator must name "
985 "a macro parameter."));
987 stringify (dest, argv[arg].text, argv[arg].len);
989 /* Read one token and let the loop iteration code handle the
991 lookahead_rl_start = replacement_list.text;
992 lookahead_valid = get_token (&lookahead, &replacement_list);
994 /* Is this token the splicing operator? */
995 else if (tok.len == 2
996 && tok.text[0] == '#'
997 && tok.text[1] == '#')
998 error (_("Stray splicing operator"));
999 /* Is the next token the splicing operator? */
1000 else if (lookahead_valid
1001 && lookahead.len == 2
1002 && lookahead.text[0] == '#'
1003 && lookahead.text[1] == '#')
1006 int prev_was_comma = 0;
1008 /* Note that GCC warns if the result of splicing is not a
1009 token. In the debugger there doesn't seem to be much
1010 benefit from doing this. */
1012 /* Insert the first token. */
1013 if (tok.len == 1 && tok.text[0] == ',')
1017 int arg = find_parameter (&tok, is_varargs, va_arg_name,
1018 def->argc, def->argv);
1020 appendmem (dest, argv[arg].text, argv[arg].len);
1022 appendmem (dest, tok.text, tok.len);
1025 /* Apply a possible sequence of ## operators. */
1028 if (! get_token (&tok, &replacement_list))
1029 error (_("Splicing operator at end of macro"));
1031 /* Handle a comma before a ##. If we are handling
1032 varargs, and the token on the right hand side is the
1033 varargs marker, and the final argument is empty or
1034 missing, then drop the comma. This is a GNU
1035 extension. There is one ambiguous case here,
1036 involving pedantic behavior with an empty argument,
1037 but we settle that in favor of GNU-style (GCC uses an
1038 option). If we aren't dealing with varargs, we
1039 simply insert the comma. */
1043 && tok.len == va_arg_name->len
1044 && !memcmp (tok.text, va_arg_name->text, tok.len)
1045 && argv[argc - 1].len == 0))
1046 appendmem (dest, ",", 1);
1050 /* Insert the token. If it is a parameter, insert the
1051 argument. If it is a comma, treat it specially. */
1052 if (tok.len == 1 && tok.text[0] == ',')
1056 int arg = find_parameter (&tok, is_varargs, va_arg_name,
1057 def->argc, def->argv);
1059 appendmem (dest, argv[arg].text, argv[arg].len);
1061 appendmem (dest, tok.text, tok.len);
1064 /* Now read another token. If it is another splice, we
1066 original_rl_start = replacement_list.text;
1067 if (! get_token (&tok, &replacement_list))
1074 && tok.text[0] == '#'
1075 && tok.text[1] == '#'))
1081 /* We saw a comma. Insert it now. */
1082 appendmem (dest, ",", 1);
1085 dest->last_token = dest->len;
1087 lookahead_valid = 0;
1090 /* Set up for the loop iterator. */
1092 lookahead_rl_start = original_rl_start;
1093 lookahead_valid = 1;
1098 /* Is this token an identifier? */
1099 int substituted = 0;
1100 int arg = find_parameter (&tok, is_varargs, va_arg_name,
1101 def->argc, def->argv);
1105 struct macro_buffer arg_src;
1107 /* Expand any macro invocations in the argument text,
1108 and append the result to dest. Remember that scan
1109 mutates its source, so we need to scan a new buffer
1110 referring to the argument's text, not the argument
1112 init_shared_buffer (&arg_src, argv[arg].text, argv[arg].len);
1113 scan (dest, &arg_src, no_loop, lookup_func, lookup_baton);
1117 /* If it wasn't a parameter, then just copy it across. */
1119 append_tokens_without_splicing (dest, &tok);
1122 if (! lookahead_valid)
1126 original_rl_start = lookahead_rl_start;
1128 lookahead_rl_start = replacement_list.text;
1129 lookahead_valid = get_token (&lookahead, &replacement_list);
1134 /* Expand a call to a macro named ID, whose definition is DEF. Append
1135 its expansion to DEST. SRC is the input text following the ID
1136 token. We are currently rescanning the expansions of the macros
1137 named in NO_LOOP; don't re-expand them. Use LOOKUP_FUNC and
1138 LOOKUP_BATON to find definitions for any nested macro references.
1140 Return 1 if we decided to expand it, zero otherwise. (If it's a
1141 function-like macro name that isn't followed by an argument list,
1142 we don't expand it.) If we return zero, leave SRC unchanged. */
1144 expand (const char *id,
1145 struct macro_definition *def,
1146 struct macro_buffer *dest,
1147 struct macro_buffer *src,
1148 struct macro_name_list *no_loop,
1149 macro_lookup_ftype *lookup_func,
1152 struct macro_name_list new_no_loop;
1154 /* Create a new node to be added to the front of the no-expand list.
1155 This list is appropriate for re-scanning replacement lists, but
1156 it is *not* appropriate for scanning macro arguments; invocations
1157 of the macro whose arguments we are gathering *do* get expanded
1159 new_no_loop.name = id;
1160 new_no_loop.next = no_loop;
1162 /* What kind of macro are we expanding? */
1163 if (def->kind == macro_object_like)
1165 struct macro_buffer replacement_list;
1167 init_shared_buffer (&replacement_list, (char *) def->replacement,
1168 strlen (def->replacement));
1170 scan (dest, &replacement_list, &new_no_loop, lookup_func, lookup_baton);
1173 else if (def->kind == macro_function_like)
1175 struct cleanup *back_to = make_cleanup (null_cleanup, 0);
1177 struct macro_buffer *argv = NULL;
1178 struct macro_buffer substituted;
1179 struct macro_buffer substituted_src;
1180 struct macro_buffer va_arg_name;
1185 if (strcmp (def->argv[def->argc - 1], "...") == 0)
1187 /* In C99-style varargs, substitution is done using
1189 init_shared_buffer (&va_arg_name, "__VA_ARGS__",
1190 strlen ("__VA_ARGS__"));
1195 int len = strlen (def->argv[def->argc - 1]);
1197 && strcmp (def->argv[def->argc - 1] + len - 3, "...") == 0)
1199 /* In GNU-style varargs, the name of the
1200 substitution parameter is the name of the formal
1201 argument without the "...". */
1202 init_shared_buffer (&va_arg_name,
1203 (char *) def->argv[def->argc - 1],
1210 make_cleanup (free_current_contents, &argv);
1211 argv = gather_arguments (id, src, is_varargs ? def->argc : -1,
1214 /* If we couldn't find any argument list, then we don't expand
1218 do_cleanups (back_to);
1222 /* Check that we're passing an acceptable number of arguments for
1224 if (argc != def->argc)
1226 if (is_varargs && argc >= def->argc - 1)
1230 /* Remember that a sequence of tokens like "foo()" is a
1231 valid invocation of a macro expecting either zero or one
1233 else if (! (argc == 1
1236 error (_("Wrong number of arguments to macro `%s' "
1237 "(expected %d, got %d)."),
1238 id, def->argc, argc);
1241 /* Note that we don't expand macro invocations in the arguments
1242 yet --- we let subst_args take care of that. Parameters that
1243 appear as operands of the stringifying operator "#" or the
1244 splicing operator "##" don't get macro references expanded,
1245 so we can't really tell whether it's appropriate to macro-
1246 expand an argument until we see how it's being used. */
1247 init_buffer (&substituted, 0);
1248 make_cleanup (cleanup_macro_buffer, &substituted);
1249 substitute_args (&substituted, def, is_varargs, &va_arg_name,
1250 argc, argv, no_loop, lookup_func, lookup_baton);
1252 /* Now `substituted' is the macro's replacement list, with all
1253 argument values substituted into it properly. Re-scan it for
1254 macro references, but don't expand invocations of this macro.
1256 We create a new buffer, `substituted_src', which points into
1257 `substituted', and scan that. We can't scan `substituted'
1258 itself, since the tokenization process moves the buffer's
1259 text pointer around, and we still need to be able to find
1260 `substituted's original text buffer after scanning it so we
1262 init_shared_buffer (&substituted_src, substituted.text, substituted.len);
1263 scan (dest, &substituted_src, &new_no_loop, lookup_func, lookup_baton);
1265 do_cleanups (back_to);
1270 internal_error (__FILE__, __LINE__, _("bad macro definition kind"));
1274 /* If the single token in SRC_FIRST followed by the tokens in SRC_REST
1275 constitute a macro invokation not forbidden in NO_LOOP, append its
1276 expansion to DEST and return non-zero. Otherwise, return zero, and
1277 leave DEST unchanged.
1279 SRC_FIRST and SRC_REST must be shared buffers; DEST must not be one.
1280 SRC_FIRST must be a string built by get_token. */
1282 maybe_expand (struct macro_buffer *dest,
1283 struct macro_buffer *src_first,
1284 struct macro_buffer *src_rest,
1285 struct macro_name_list *no_loop,
1286 macro_lookup_ftype *lookup_func,
1289 gdb_assert (src_first->shared);
1290 gdb_assert (src_rest->shared);
1291 gdb_assert (! dest->shared);
1293 /* Is this token an identifier? */
1294 if (src_first->is_identifier)
1296 /* Make a null-terminated copy of it, since that's what our
1297 lookup function expects. */
1298 char *id = xmalloc (src_first->len + 1);
1299 struct cleanup *back_to = make_cleanup (xfree, id);
1300 memcpy (id, src_first->text, src_first->len);
1301 id[src_first->len] = 0;
1303 /* If we're currently re-scanning the result of expanding
1304 this macro, don't expand it again. */
1305 if (! currently_rescanning (no_loop, id))
1307 /* Does this identifier have a macro definition in scope? */
1308 struct macro_definition *def = lookup_func (id, lookup_baton);
1310 if (def && expand (id, def, dest, src_rest, no_loop,
1311 lookup_func, lookup_baton))
1313 do_cleanups (back_to);
1318 do_cleanups (back_to);
1325 /* Expand macro references in SRC, appending the results to DEST.
1326 Assume we are re-scanning the result of expanding the macros named
1327 in NO_LOOP, and don't try to re-expand references to them.
1329 SRC must be a shared buffer; DEST must not be one. */
1331 scan (struct macro_buffer *dest,
1332 struct macro_buffer *src,
1333 struct macro_name_list *no_loop,
1334 macro_lookup_ftype *lookup_func,
1337 gdb_assert (src->shared);
1338 gdb_assert (! dest->shared);
1342 struct macro_buffer tok;
1343 char *original_src_start = src->text;
1345 /* Find the next token in SRC. */
1346 if (! get_token (&tok, src))
1349 /* Just for aesthetics. If we skipped some whitespace, copy
1351 if (tok.text > original_src_start)
1353 appendmem (dest, original_src_start, tok.text - original_src_start);
1354 dest->last_token = dest->len;
1357 if (! maybe_expand (dest, &tok, src, no_loop, lookup_func, lookup_baton))
1358 /* We didn't end up expanding tok as a macro reference, so
1359 simply append it to dest. */
1360 append_tokens_without_splicing (dest, &tok);
1363 /* Just for aesthetics. If there was any trailing whitespace in
1364 src, copy it to dest. */
1367 appendmem (dest, src->text, src->len);
1368 dest->last_token = dest->len;
1374 macro_expand (const char *source,
1375 macro_lookup_ftype *lookup_func,
1376 void *lookup_func_baton)
1378 struct macro_buffer src, dest;
1379 struct cleanup *back_to;
1381 init_shared_buffer (&src, (char *) source, strlen (source));
1383 init_buffer (&dest, 0);
1384 dest.last_token = 0;
1385 back_to = make_cleanup (cleanup_macro_buffer, &dest);
1387 scan (&dest, &src, 0, lookup_func, lookup_func_baton);
1389 appendc (&dest, '\0');
1391 discard_cleanups (back_to);
1397 macro_expand_once (const char *source,
1398 macro_lookup_ftype *lookup_func,
1399 void *lookup_func_baton)
1401 error (_("Expand-once not implemented yet."));
1406 macro_expand_next (char **lexptr,
1407 macro_lookup_ftype *lookup_func,
1410 struct macro_buffer src, dest, tok;
1411 struct cleanup *back_to;
1413 /* Set up SRC to refer to the input text, pointed to by *lexptr. */
1414 init_shared_buffer (&src, *lexptr, strlen (*lexptr));
1416 /* Set up DEST to receive the expansion, if there is one. */
1417 init_buffer (&dest, 0);
1418 dest.last_token = 0;
1419 back_to = make_cleanup (cleanup_macro_buffer, &dest);
1421 /* Get the text's first preprocessing token. */
1422 if (! get_token (&tok, &src))
1424 do_cleanups (back_to);
1428 /* If it's a macro invocation, expand it. */
1429 if (maybe_expand (&dest, &tok, &src, 0, lookup_func, lookup_baton))
1431 /* It was a macro invocation! Package up the expansion as a
1432 null-terminated string and return it. Set *lexptr to the
1433 start of the next token in the input. */
1434 appendc (&dest, '\0');
1435 discard_cleanups (back_to);
1441 /* It wasn't a macro invocation. */
1442 do_cleanups (back_to);