1/* Line completion stuff for GDB, the GNU debugger.
2   Copyright (C) 2000, 2001, 2007 Free Software Foundation, Inc.
3
4   This file is part of GDB.
5
6   This program is free software; you can redistribute it and/or modify
7   it under the terms of the GNU General Public License as published by
8   the Free Software Foundation; either version 3 of the License, or
9   (at your option) any later version.
10
11   This program is distributed in the hope that it will be useful,
12   but WITHOUT ANY WARRANTY; without even the implied warranty of
13   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
14   GNU General Public License for more details.
15
16   You should have received a copy of the GNU General Public License
17   along with this program.  If not, see <http://www.gnu.org/licenses/>.  */
18
19#include "defs.h"
20#include "symtab.h"
21#include "gdbtypes.h"
22#include "expression.h"
23#include "filenames.h"		/* For DOSish file names.  */
24#include "language.h"
25
26#include "cli/cli-decode.h"
27
28/* FIXME: This is needed because of lookup_cmd_1 ().  We should be
29   calling a hook instead so we eliminate the CLI dependency.  */
30#include "gdbcmd.h"
31
32/* Needed for rl_completer_word_break_characters() and for
33   rl_filename_completion_function.  */
34#include "readline/readline.h"
35
36/* readline defines this.  */
37#undef savestring
38
39#include "completer.h"
40
41/* Prototypes for local functions.  */
42static
43char *line_completion_function (const char *text, int matches,
44				char *line_buffer,
45				int point);
46
47/* readline uses the word breaks for two things:
48   (1) In figuring out where to point the TEXT parameter to the
49   rl_completion_entry_function.  Since we don't use TEXT for much,
50   it doesn't matter a lot what the word breaks are for this purpose, but
51   it does affect how much stuff M-? lists.
52   (2) If one of the matches contains a word break character, readline
53   will quote it.  That's why we switch between
54   current_language->la_word_break_characters() and
55   gdb_completer_command_word_break_characters.  I'm not sure when
56   we need this behavior (perhaps for funky characters in C++ symbols?).  */
57
58/* Variables which are necessary for fancy command line editing.  */
59
60/* When completing on command names, we remove '-' from the list of
61   word break characters, since we use it in command names.  If the
62   readline library sees one in any of the current completion strings,
63   it thinks that the string needs to be quoted and automatically supplies
64   a leading quote.  */
65static char *gdb_completer_command_word_break_characters =
66" \t\n!@#$%^&*()+=|~`}{[]\"';:?/>.<,";
67
68/* When completing on file names, we remove from the list of word
69   break characters any characters that are commonly used in file
70   names, such as '-', '+', '~', etc.  Otherwise, readline displays
71   incorrect completion candidates.  */
72#ifdef HAVE_DOS_BASED_FILE_SYSTEM
73/* MS-DOS and MS-Windows use colon as part of the drive spec, and most
74   programs support @foo style response files.  */
75static char *gdb_completer_file_name_break_characters = " \t\n*|\"';?><@";
76#else
77static char *gdb_completer_file_name_break_characters = " \t\n*|\"';:?><";
78#endif
79
80/* These are used when completing on locations, which can mix file
81   names and symbol names separated by a colon.  */
82static char *gdb_completer_loc_break_characters = " \t\n*|\"';:?><,";
83
84/* Characters that can be used to quote completion strings.  Note that we
85   can't include '"' because the gdb C parser treats such quoted sequences
86   as strings.  */
87static char *gdb_completer_quote_characters = "'";
88
89/* Accessor for some completer data that may interest other files.  */
90
91char *
92get_gdb_completer_quote_characters (void)
93{
94  return gdb_completer_quote_characters;
95}
96
97/* Line completion interface function for readline.  */
98
99char *
100readline_line_completion_function (const char *text, int matches)
101{
102  return line_completion_function (text, matches, rl_line_buffer, rl_point);
103}
104
105/* This can be used for functions which don't want to complete on symbols
106   but don't want to complete on anything else either.  */
107char **
108noop_completer (char *text, char *prefix)
109{
110  return NULL;
111}
112
113/* Complete on filenames.  */
114char **
115filename_completer (char *text, char *word)
116{
117  int subsequent_name;
118  char **return_val;
119  int return_val_used;
120  int return_val_alloced;
121
122  return_val_used = 0;
123  /* Small for testing.  */
124  return_val_alloced = 1;
125  return_val = (char **) xmalloc (return_val_alloced * sizeof (char *));
126
127  subsequent_name = 0;
128  while (1)
129    {
130      char *p, *q;
131      p = rl_filename_completion_function (text, subsequent_name);
132      if (return_val_used >= return_val_alloced)
133	{
134	  return_val_alloced *= 2;
135	  return_val =
136	    (char **) xrealloc (return_val,
137				return_val_alloced * sizeof (char *));
138	}
139      if (p == NULL)
140	{
141	  return_val[return_val_used++] = p;
142	  break;
143	}
144      /* We need to set subsequent_name to a non-zero value before the
145	 continue line below, because otherwise, if the first file seen
146	 by GDB is a backup file whose name ends in a `~', we will loop
147	 indefinitely.  */
148      subsequent_name = 1;
149      /* Like emacs, don't complete on old versions.  Especially useful
150         in the "source" command.  */
151      if (p[strlen (p) - 1] == '~')
152	{
153	  xfree (p);
154	  continue;
155	}
156
157      if (word == text)
158	/* Return exactly p.  */
159	return_val[return_val_used++] = p;
160      else if (word > text)
161	{
162	  /* Return some portion of p.  */
163	  q = xmalloc (strlen (p) + 5);
164	  strcpy (q, p + (word - text));
165	  return_val[return_val_used++] = q;
166	  xfree (p);
167	}
168      else
169	{
170	  /* Return some of TEXT plus p.  */
171	  q = xmalloc (strlen (p) + (text - word) + 5);
172	  strncpy (q, word, text - word);
173	  q[text - word] = '\0';
174	  strcat (q, p);
175	  return_val[return_val_used++] = q;
176	  xfree (p);
177	}
178    }
179#if 0
180  /* There is no way to do this just long enough to affect quote inserting
181     without also affecting the next completion.  This should be fixed in
182     readline.  FIXME.  */
183  /* Ensure that readline does the right thing
184     with respect to inserting quotes.  */
185  rl_completer_word_break_characters = "";
186#endif
187  return return_val;
188}
189
190/* Complete on locations, which might be of two possible forms:
191
192       file:line
193   or
194       symbol+offset
195
196   This is intended to be used in commands that set breakpoints etc.  */
197char **
198location_completer (char *text, char *word)
199{
200  int n_syms = 0, n_files = 0;
201  char ** fn_list = NULL;
202  char ** list = NULL;
203  char *p;
204  int quote_found = 0;
205  int quoted = *text == '\'' || *text == '"';
206  int quote_char = '\0';
207  char *colon = NULL;
208  char *file_to_match = NULL;
209  char *symbol_start = text;
210  char *orig_text = text;
211  size_t text_len;
212
213  /* Do we have an unquoted colon, as in "break foo.c::bar"?  */
214  for (p = text; *p != '\0'; ++p)
215    {
216      if (*p == '\\' && p[1] == '\'')
217	p++;
218      else if (*p == '\'' || *p == '"')
219	{
220	  quote_found = *p;
221	  quote_char = *p++;
222	  while (*p != '\0' && *p != quote_found)
223	    {
224	      if (*p == '\\' && p[1] == quote_found)
225		p++;
226	      p++;
227	    }
228
229	  if (*p == quote_found)
230	    quote_found = 0;
231	  else
232	    break;		/* Hit the end of text.  */
233	}
234#if HAVE_DOS_BASED_FILE_SYSTEM
235      /* If we have a DOS-style absolute file name at the beginning of
236	 TEXT, and the colon after the drive letter is the only colon
237	 we found, pretend the colon is not there.  */
238      else if (p < text + 3 && *p == ':' && p == text + 1 + quoted)
239	;
240#endif
241      else if (*p == ':' && !colon)
242	{
243	  colon = p;
244	  symbol_start = p + 1;
245	}
246      else if (strchr (current_language->la_word_break_characters(), *p))
247	symbol_start = p + 1;
248    }
249
250  if (quoted)
251    text++;
252  text_len = strlen (text);
253
254  /* Where is the file name?  */
255  if (colon)
256    {
257      char *s;
258
259      file_to_match = (char *) xmalloc (colon - text + 1);
260      strncpy (file_to_match, text, colon - text + 1);
261      /* Remove trailing colons and quotes from the file name.  */
262      for (s = file_to_match + (colon - text);
263	   s > file_to_match;
264	   s--)
265	if (*s == ':' || *s == quote_char)
266	  *s = '\0';
267    }
268  /* If the text includes a colon, they want completion only on a
269     symbol name after the colon.  Otherwise, we need to complete on
270     symbols as well as on files.  */
271  if (colon)
272    {
273      list = make_file_symbol_completion_list (symbol_start, word,
274					       file_to_match);
275      xfree (file_to_match);
276    }
277  else
278    {
279      list = make_symbol_completion_list (symbol_start, word);
280      /* If text includes characters which cannot appear in a file
281	 name, they cannot be asking for completion on files.  */
282      if (strcspn (text,
283		   gdb_completer_file_name_break_characters) == text_len)
284	fn_list = make_source_files_completion_list (text, text);
285    }
286
287  /* How many completions do we have in both lists?  */
288  if (fn_list)
289    for ( ; fn_list[n_files]; n_files++)
290      ;
291  if (list)
292    for ( ; list[n_syms]; n_syms++)
293      ;
294
295  /* Make list[] large enough to hold both lists, then catenate
296     fn_list[] onto the end of list[].  */
297  if (n_syms && n_files)
298    {
299      list = xrealloc (list, (n_syms + n_files + 1) * sizeof (char *));
300      memcpy (list + n_syms, fn_list, (n_files + 1) * sizeof (char *));
301      xfree (fn_list);
302    }
303  else if (n_files)
304    {
305      /* If we only have file names as possible completion, we should
306	 bring them in sync with what rl_complete expects.  The
307	 problem is that if the user types "break /foo/b TAB", and the
308	 possible completions are "/foo/bar" and "/foo/baz"
309	 rl_complete expects us to return "bar" and "baz", without the
310	 leading directories, as possible completions, because `word'
311	 starts at the "b".  But we ignore the value of `word' when we
312	 call make_source_files_completion_list above (because that
313	 would not DTRT when the completion results in both symbols
314	 and file names), so make_source_files_completion_list returns
315	 the full "/foo/bar" and "/foo/baz" strings.  This produces
316	 wrong results when, e.g., there's only one possible
317	 completion, because rl_complete will prepend "/foo/" to each
318	 candidate completion.  The loop below removes that leading
319	 part.  */
320      for (n_files = 0; fn_list[n_files]; n_files++)
321	{
322	  memmove (fn_list[n_files], fn_list[n_files] + (word - text),
323		   strlen (fn_list[n_files]) + 1 - (word - text));
324	}
325      /* Return just the file-name list as the result.  */
326      list = fn_list;
327    }
328  else if (!n_syms)
329    {
330      /* No completions at all.  As the final resort, try completing
331	 on the entire text as a symbol.  */
332      list = make_symbol_completion_list (orig_text, word);
333      xfree (fn_list);
334    }
335  else
336    xfree (fn_list);
337
338  return list;
339}
340
341/* Complete on command names.  Used by "help".  */
342char **
343command_completer (char *text, char *word)
344{
345  return complete_on_cmdlist (cmdlist, text, word);
346}
347
348
349/* Here are some useful test cases for completion.  FIXME: These should
350   be put in the test suite.  They should be tested with both M-? and TAB.
351
352   "show output-" "radix"
353   "show output" "-radix"
354   "p" ambiguous (commands starting with p--path, print, printf, etc.)
355   "p "  ambiguous (all symbols)
356   "info t foo" no completions
357   "info t " no completions
358   "info t" ambiguous ("info target", "info terminal", etc.)
359   "info ajksdlfk" no completions
360   "info ajksdlfk " no completions
361   "info" " "
362   "info " ambiguous (all info commands)
363   "p \"a" no completions (string constant)
364   "p 'a" ambiguous (all symbols starting with a)
365   "p b-a" ambiguous (all symbols starting with a)
366   "p b-" ambiguous (all symbols)
367   "file Make" "file" (word break hard to screw up here)
368   "file ../gdb.stabs/we" "ird" (needs to not break word at slash)
369 */
370
371/* Generate completions all at once.  Returns a NULL-terminated array
372   of strings.  Both the array and each element are allocated with
373   xmalloc.  It can also return NULL if there are no completions.
374
375   TEXT is the caller's idea of the "word" we are looking at.
376
377   LINE_BUFFER is available to be looked at; it contains the entire text
378   of the line.  POINT is the offset in that line of the cursor.  You
379   should pretend that the line ends at POINT.  */
380
381char **
382complete_line (const char *text, char *line_buffer, int point)
383{
384  char **list = NULL;
385  char *tmp_command, *p;
386  /* Pointer within tmp_command which corresponds to text.  */
387  char *word;
388  struct cmd_list_element *c, *result_list;
389
390  /* Choose the default set of word break characters to break completions.
391     If we later find out that we are doing completions on command strings
392     (as opposed to strings supplied by the individual command completer
393     functions, which can be any string) then we will switch to the
394     special word break set for command strings, which leaves out the
395     '-' character used in some commands.  */
396
397  rl_completer_word_break_characters =
398    current_language->la_word_break_characters();
399
400  /* Decide whether to complete on a list of gdb commands or on symbols. */
401  tmp_command = (char *) alloca (point + 1);
402  p = tmp_command;
403
404  strncpy (tmp_command, line_buffer, point);
405  tmp_command[point] = '\0';
406  /* Since text always contains some number of characters leading up
407     to point, we can find the equivalent position in tmp_command
408     by subtracting that many characters from the end of tmp_command.  */
409  word = tmp_command + point - strlen (text);
410
411  if (point == 0)
412    {
413      /* An empty line we want to consider ambiguous; that is, it
414	 could be any command.  */
415      c = (struct cmd_list_element *) -1;
416      result_list = 0;
417    }
418  else
419    {
420      c = lookup_cmd_1 (&p, cmdlist, &result_list, 1);
421    }
422
423  /* Move p up to the next interesting thing.  */
424  while (*p == ' ' || *p == '\t')
425    {
426      p++;
427    }
428
429  if (!c)
430    {
431      /* It is an unrecognized command.  So there are no
432	 possible completions.  */
433      list = NULL;
434    }
435  else if (c == (struct cmd_list_element *) -1)
436    {
437      char *q;
438
439      /* lookup_cmd_1 advances p up to the first ambiguous thing, but
440	 doesn't advance over that thing itself.  Do so now.  */
441      q = p;
442      while (*q && (isalnum (*q) || *q == '-' || *q == '_'))
443	++q;
444      if (q != tmp_command + point)
445	{
446	  /* There is something beyond the ambiguous
447	     command, so there are no possible completions.  For
448	     example, "info t " or "info t foo" does not complete
449	     to anything, because "info t" can be "info target" or
450	     "info terminal".  */
451	  list = NULL;
452	}
453      else
454	{
455	  /* We're trying to complete on the command which was ambiguous.
456	     This we can deal with.  */
457	  if (result_list)
458	    {
459	      list = complete_on_cmdlist (*result_list->prefixlist, p,
460					  word);
461	    }
462	  else
463	    {
464	      list = complete_on_cmdlist (cmdlist, p, word);
465	    }
466	  /* Ensure that readline does the right thing with respect to
467	     inserting quotes.  */
468	  rl_completer_word_break_characters =
469	    gdb_completer_command_word_break_characters;
470	}
471    }
472  else
473    {
474      /* We've recognized a full command.  */
475
476      if (p == tmp_command + point)
477	{
478	  /* There is no non-whitespace in the line beyond the command.  */
479
480	  if (p[-1] == ' ' || p[-1] == '\t')
481	    {
482	      /* The command is followed by whitespace; we need to complete
483		 on whatever comes after command.  */
484	      if (c->prefixlist)
485		{
486		  /* It is a prefix command; what comes after it is
487		     a subcommand (e.g. "info ").  */
488		  list = complete_on_cmdlist (*c->prefixlist, p, word);
489
490		  /* Ensure that readline does the right thing
491		     with respect to inserting quotes.  */
492		  rl_completer_word_break_characters =
493		    gdb_completer_command_word_break_characters;
494		}
495	      else if (c->enums)
496		{
497		  list = complete_on_enum (c->enums, p, word);
498		  rl_completer_word_break_characters =
499		    gdb_completer_command_word_break_characters;
500		}
501	      else
502		{
503		  /* It is a normal command; what comes after it is
504		     completed by the command's completer function.  */
505		  if (c->completer == filename_completer)
506		    {
507		      /* Many commands which want to complete on
508			 file names accept several file names, as
509			 in "run foo bar >>baz".  So we don't want
510			 to complete the entire text after the
511			 command, just the last word.  To this
512			 end, we need to find the beginning of the
513			 file name by starting at `word' and going
514			 backwards.  */
515		      for (p = word;
516			   p > tmp_command
517			     && strchr (gdb_completer_file_name_break_characters, p[-1]) == NULL;
518			   p--)
519			;
520		      rl_completer_word_break_characters =
521			gdb_completer_file_name_break_characters;
522		    }
523		  else if (c->completer == location_completer)
524		    {
525		      /* Commands which complete on locations want to
526			 see the entire argument.  */
527		      for (p = word;
528			   p > tmp_command
529			     && p[-1] != ' ' && p[-1] != '\t';
530			   p--)
531			;
532		    }
533		  list = (*c->completer) (p, word);
534		}
535	    }
536	  else
537	    {
538	      /* The command is not followed by whitespace; we need to
539		 complete on the command itself.  e.g. "p" which is a
540		 command itself but also can complete to "print", "ptype"
541		 etc.  */
542	      char *q;
543
544	      /* Find the command we are completing on.  */
545	      q = p;
546	      while (q > tmp_command)
547		{
548		  if (isalnum (q[-1]) || q[-1] == '-' || q[-1] == '_')
549		    --q;
550		  else
551		    break;
552		}
553
554	      list = complete_on_cmdlist (result_list, q, word);
555
556	      /* Ensure that readline does the right thing
557		 with respect to inserting quotes.  */
558	      rl_completer_word_break_characters =
559		gdb_completer_command_word_break_characters;
560	    }
561	}
562      else
563	{
564	  /* There is non-whitespace beyond the command.  */
565
566	  if (c->prefixlist && !c->allow_unknown)
567	    {
568	      /* It is an unrecognized subcommand of a prefix command,
569		 e.g. "info adsfkdj".  */
570	      list = NULL;
571	    }
572	  else if (c->enums)
573	    {
574	      list = complete_on_enum (c->enums, p, word);
575	    }
576	  else
577	    {
578	      /* It is a normal command.  */
579	      if (c->completer == filename_completer)
580		{
581		  /* See the commentary above about the specifics
582		     of file-name completion.  */
583		  for (p = word;
584		       p > tmp_command
585			 && strchr (gdb_completer_file_name_break_characters, p[-1]) == NULL;
586		       p--)
587		    ;
588		  rl_completer_word_break_characters =
589		    gdb_completer_file_name_break_characters;
590		}
591	      else if (c->completer == location_completer)
592		{
593		  for (p = word;
594		       p > tmp_command
595			 && p[-1] != ' ' && p[-1] != '\t';
596		       p--)
597		    ;
598		}
599	      list = (*c->completer) (p, word);
600	    }
601	}
602    }
603
604  return list;
605}
606
607/* Generate completions one by one for the completer.  Each time we are
608   called return another potential completion to the caller.
609   line_completion just completes on commands or passes the buck to the
610   command's completer function, the stuff specific to symbol completion
611   is in make_symbol_completion_list.
612
613   TEXT is the caller's idea of the "word" we are looking at.
614
615   MATCHES is the number of matches that have currently been collected from
616   calling this completion function.  When zero, then we need to initialize,
617   otherwise the initialization has already taken place and we can just
618   return the next potential completion string.
619
620   LINE_BUFFER is available to be looked at; it contains the entire text
621   of the line.  POINT is the offset in that line of the cursor.  You
622   should pretend that the line ends at POINT.
623
624   Returns NULL if there are no more completions, else a pointer to a string
625   which is a possible completion, it is the caller's responsibility to
626   free the string.  */
627
628static char *
629line_completion_function (const char *text, int matches,
630			  char *line_buffer, int point)
631{
632  static char **list = (char **) NULL;	/* Cache of completions.  */
633  static int index;			/* Next cached completion.  */
634  char *output = NULL;
635
636  if (matches == 0)
637    {
638      /* The caller is beginning to accumulate a new set of completions, so
639         we need to find all of them now, and cache them for returning one at
640         a time on future calls.  */
641
642      if (list)
643	{
644	  /* Free the storage used by LIST, but not by the strings inside.
645	     This is because rl_complete_internal () frees the strings.  */
646	  xfree (list);
647	}
648      index = 0;
649      list = complete_line (text, line_buffer, point);
650    }
651
652  /* If we found a list of potential completions during initialization then
653     dole them out one at a time.  The vector of completions is NULL
654     terminated, so after returning the last one, return NULL (and continue
655     to do so) each time we are called after that, until a new list is
656     available.  */
657
658  if (list)
659    {
660      output = list[index];
661      if (output)
662	{
663	  index++;
664	}
665    }
666
667#if 0
668  /* Can't do this because readline hasn't yet checked the word breaks
669     for figuring out whether to insert a quote.  */
670  if (output == NULL)
671    /* Make sure the word break characters are set back to normal for the
672       next time that readline tries to complete something.  */
673    rl_completer_word_break_characters =
674      current_language->la_word_break_characters();
675#endif
676
677  return (output);
678}
679
680/* Skip over the possibly quoted word STR (as defined by the quote
681   characters QUOTECHARS and the the word break characters
682   BREAKCHARS).  Returns pointer to the location after the "word".  If
683   either QUOTECHARS or BREAKCHARS is NULL, use the same values used
684   by the completer.  */
685
686char *
687skip_quoted_chars (char *str, char *quotechars, char *breakchars)
688{
689  char quote_char = '\0';
690  char *scan;
691
692  if (quotechars == NULL)
693    quotechars = gdb_completer_quote_characters;
694
695  if (breakchars == NULL)
696    breakchars = current_language->la_word_break_characters();
697
698  for (scan = str; *scan != '\0'; scan++)
699    {
700      if (quote_char != '\0')
701	{
702	  /* Ignore everything until the matching close quote char.  */
703	  if (*scan == quote_char)
704	    {
705	      /* Found matching close quote.  */
706	      scan++;
707	      break;
708	    }
709	}
710      else if (strchr (quotechars, *scan))
711	{
712	  /* Found start of a quoted string. */
713	  quote_char = *scan;
714	}
715      else if (strchr (breakchars, *scan))
716	{
717	  break;
718	}
719    }
720
721  return (scan);
722}
723
724/* Skip over the possibly quoted word STR (as defined by the quote
725   characters and word break characters used by the completer).
726   Returns pointer to the location after the "word".  */
727
728char *
729skip_quoted (char *str)
730{
731  return skip_quoted_chars (str, NULL, NULL);
732}
733