A smarter linespec completer

Continuing the theme of the explicit locations patch, this patch gets
rid of the need for quoting function names in linespec TAB completion.
To recap, when you have overloads in your program, and you want to set
a breakpoint in one of them:

 void function(int);  // set breakpoint here.
 void function(long);

 (gdb) b function(i[TAB]
 <all the symbols in the program that start with "i" are uselessly shown...>

This patch gets rid of the need for quoting by switching the linespec
completer to use the custom completion word point mechanism added in
the previous explicit location patch (extending it as needed), to
correctly determine the right completion word point.  In the case
above, we want the completer to figure out that it's completing a
function name that starts with "function(i", and it now does.

We also want the completer to know when it's potentially completing a
source file name, for:

(gdb) break source.[TAB] -> source.c:
(gdb) break source.c:  # Type line number or function name now

And we want it to know to complete label names, which it doesn't today:

(gdb) break function:lab[TAB]

etc., etc.

So what we want is for completion to grok the input string as closely
to how the linespec parser groks it.

With that in mind, the solution suggests itself - make the linespec
completer use the same parsing code as normal linespec parsing.

That's what the patch does.  The old completer is replaced by one that
reuses the actual linespec parser as much as possible.  This (ideally)
eliminate differences between what completion understands and actually
setting breakpoints understands by design.

The completer now offers sensible completion candidates depending on
which component of the linespec is being completed, source filename,
function, line number, expression, and (a new addition), labels.  For
example, when completing the function part, we now show the full name
of the method as completion candidates, instead of showing whatever
comes after what readline considered the word break character:

 (gdb) break klass::method[TAB]
 klass:method1(int)
 klass:method2()

If input is past the function, then we now offer keyword condidates:

  (gdb) b function(int) [TAB]
  if      task    thread

If input is past a keyword, we offer expression completion, which is
different from linespec completion:

  (gdb) b main if 1 + glo[TAB]
  global

(e.g., completes on types, struct data fields, etc.)

As mentioned, this teaches the linespec completer about completing
label symbols too:

  (gdb) b source.c:function:lab[TAB]

A nice convenience is that when completion uniquely matches a source
name, gdb adds the ":" automatically for you:

  (gdb) b filenam[TAB]
  (gdb) b filename.c:  # ':' auto-added, cursor right after it.

It's the little details.  :-)

I worked on this patch in parallel with writing the (big) testcase
added closer to the end of the series, which exercises many many
tricky cases around quoting and whitespace insertion placement.  In
general, I think it now all Just Works.

gdb/ChangeLog:
2017-07-17  Pedro Alves  <palves@redhat.com>

	* completer.c (complete_source_filenames): New function.
	(complete_address_and_linespec_locations): New function.
	(location_completer): Use complete_address_and_linespec_locations.
	(completion_tracker::build_completion_result): Honor the tracker's
	request to suppress append.
	* completer.h (completion_tracker::suppress_append_ws)
	(completion_tracker::set_suppress_append_ws): New methods.
	(completion_tracker::m_suppress_append_ws): New field.
	(complete_source_filenames): New declaration.
	* linespec.c (linespec_complete_what): New.
	(struct ls_parser) <complete_what, completion_word,
	completion_quote_char, completion_quote_end, completion_tracker>:
	New fields.
	(string_find_incomplete_keyword_at_end): New.
	(linespec_lexer_lex_string): Record quote char.  If in completion
	mode, don't throw.
	(linespec_lexer_consume_token): Advance the completion word point.
	(linespec_lexer_peek_token): Save/restore completion info.
	(save_stream_and_consume_token): New.
	(set_completion_after_number): New.
	(linespec_parse_basic): Set what to complete next depending on
	token.  Handle function and label completions specially.
	(parse_linespec): Disable objc shortcut in completion mode.  Set
	what to complete next depending on token type.  Skip keyword if in
	completion mode.
	(complete_linespec_component, linespec_complete): New.
	* linespec.h (linespec_complete): Declare.

gdb/testsuite/ChangeLog:
2017-07-17  Pedro Alves  <palves@redhat.com>

	* gdb.base/completion.exp: Adjust expected output.
	* gdb.linespec/ls-errs.exp: Don't send tab characters, now that
	the completer works.

8 files changed, 722 insertions, 65 deletions

diff --git a/gdb/ChangeLog b/gdb/ChangeLog
index 8147cc6..20344f7 100644
--- a/gdb/ChangeLog
+++ b/gdb/ChangeLog
@@ -1,5 +1,35 @@
 2017-07-17  Pedro Alves  <palves@redhat.com>
 
+	* completer.c (complete_source_filenames): New function.
+	(complete_address_and_linespec_locations): New function.
+	(location_completer): Use complete_address_and_linespec_locations.
+	(completion_tracker::build_completion_result): Honor the tracker's
+	request to suppress append.
+	* completer.h (completion_tracker::suppress_append_ws)
+	(completion_tracker::set_suppress_append_ws): New methods.
+	(completion_tracker::m_suppress_append_ws): New field.
+	(complete_source_filenames): New declaration.
+	* linespec.c (linespec_complete_what): New.
+	(struct ls_parser) <complete_what, completion_word,
+	completion_quote_char, completion_quote_end, completion_tracker>:
+	New fields.
+	(string_find_incomplete_keyword_at_end): New.
+	(linespec_lexer_lex_string): Record quote char.  If in completion
+	mode, don't throw.
+	(linespec_lexer_consume_token): Advance the completion word point.
+	(linespec_lexer_peek_token): Save/restore completion info.
+	(save_stream_and_consume_token): New.
+	(set_completion_after_number): New.
+	(linespec_parse_basic): Set what to complete next depending on
+	token.  Handle function and label completions specially.
+	(parse_linespec): Disable objc shortcut in completion mode.  Set
+	what to complete next depending on token type.  Skip keyword if in
+	completion mode.
+	(complete_linespec_component, linespec_complete): New.
+	* linespec.h (linespec_complete): Declare.
+
+2017-07-17  Pedro Alves  <palves@redhat.com>
+
 	* linespec.c (linespec_lexer_lex_string, find_toplevel_char):
 	Handle 'operator<' / 'operator<<'.
 
diff --git a/gdb/completer.c b/gdb/completer.c
index c2bb4ee..ba2e860 100644
--- a/gdb/completer.c
+++ b/gdb/completer.c
@@ -45,9 +45,6 @@
 
 #include "completer.h"
 
-static void complete_expression (completion_tracker &tracker,
-				 const char *text, const char *word);
-
 /* Misc state that needs to be tracked across several different
    readline completer entry point calls, all related to a single
    completion invocation.  */
@@ -558,8 +555,46 @@ complete_files_symbols (completion_tracker &tracker,
     }
 }
 
+/* See completer.h.  */
+
+completion_list
+complete_source_filenames (const char *text)
+{
+  size_t text_len = strlen (text);
+
+  /* If text includes characters which cannot appear in a file name,
+     the user cannot be asking for completion on files.  */
+  if (strcspn (text,
+	       gdb_completer_file_name_break_characters)
+      == text_len)
+    return make_source_files_completion_list (text, text);
+
+  return {};
+}
+
+/* Complete address and linespec locations.  */
+
+static void
+complete_address_and_linespec_locations (completion_tracker &tracker,
+					 const char *text)
+{
+  if (*text == '*')
+    {
+      tracker.advance_custom_word_point_by (1);
+      text++;
+      const char *word
+	= advance_to_expression_complete_word_point (tracker, text);
+      complete_expression (tracker, text, word);
+    }
+  else
+    {
+      linespec_complete (tracker, text);
+    }
+}
+
 /* The explicit location options.  Note that indexes into this array
    must match the explicit_location_match_type enumerators.  */
+
 static const char *const explicit_options[] =
   {
     "-source",
@@ -801,7 +836,7 @@ complete_explicit_location (completion_tracker &tracker,
 void
 location_completer (struct cmd_list_element *ignore,
 		    completion_tracker &tracker,
-		    const char *text, const char *word_entry)
+		    const char *text, const char * /* word */)
 {
   int found_probe_option = -1;
 
@@ -872,27 +907,7 @@ location_completer (struct cmd_list_element *ignore,
   else
     {
       /* This is an address or linespec location.  */
-      if (*text == '*')
-	{
-	  tracker.advance_custom_word_point_by (1);
-	  text++;
-	  const char *word
-	    = advance_to_expression_complete_word_point (tracker, text);
-	  complete_expression (tracker, text, word);
-	}
-      else
-	{
-	  /* Fall back to the old linespec completer, for now.  */
-
-	  if (word_entry == NULL)
-	    {
-	     /* We're in the handle_brkchars phase.  */
-	      tracker.set_use_custom_word_point (false);
-	      return;
-	    }
-
-	  complete_files_symbols (tracker, text, word_entry);
-	}
+      complete_address_and_linespec_locations (tracker, text);
     }
 
   /* Add matches for option names, if either:
@@ -984,11 +999,9 @@ add_struct_fields (struct type *type, completion_list &output,
     }
 }
 
-/* Complete on expressions.  Often this means completing on symbol
-   names, but some language parsers also have support for completing
-   field names.  */
+/* See completer.h.  */
 
-static void
+void
 complete_expression (completion_tracker &tracker,
 		     const char *text, const char *word)
 {
@@ -1944,10 +1957,12 @@ completion_tracker::build_completion_result (const char *text,
 				buf, (char *) NULL);
       match_list[1] = NULL;
 
-      /* If we already have a space at the end of the match, tell
-	 readline to skip appending another.  */
+      /* If the tracker wants to, or we already have a space at the
+	 end of the match, tell readline to skip appending
+	 another.  */
       bool completion_suppress_append
-	= (match_list[0][strlen (match_list[0]) - 1] == ' ');
+	= (suppress_append_ws ()
+	   || match_list[0][strlen (match_list[0]) - 1] == ' ');
 
       return completion_result (match_list, 1, completion_suppress_append);
     }
diff --git a/gdb/completer.h b/gdb/completer.h
index 40d2f58..f68c6dc 100644
--- a/gdb/completer.h
+++ b/gdb/completer.h
@@ -192,6 +192,16 @@ public:
   /* Advance the custom word point by LEN.  */
   void advance_custom_word_point_by (size_t len);
 
+  /* Whether to tell readline to skip appending a whitespace after the
+     completion.  See m_suppress_append_ws.  */
+  bool suppress_append_ws () const
+  { return m_suppress_append_ws; }
+
+  /* Set whether to tell readline to skip appending a whitespace after
+     the completion.  See m_suppress_append_ws.  */
+  void set_suppress_append_ws (bool suppress)
+  { m_suppress_append_ws = suppress; }
+
   /* Return true if we only have one completion, and it matches
      exactly the completion word.  I.e., completing results in what we
      already have.  */
@@ -255,6 +265,14 @@ private:
      completable words.  */
   int m_custom_word_point = 0;
 
+  /* If true, tell readline to skip appending a whitespace after the
+     completion.  Automatically set if we have a unique completion
+     that already has a space at the end.  A completer may also
+     explicitly set this.  E.g., the linespec completer sets this when
+     the completion ends with the ":" separator between filename and
+     function name.  */
+  bool m_suppress_append_ws = false;
+
   /* Our idea of lowest common denominator to hand over to readline.
      See intro.  */
   char *m_lowest_common_denominator = NULL;
@@ -353,6 +371,16 @@ extern completer_handle_brkchars_ftype *
 
 /* Exported to linespec.c */
 
+/* Return a list of all source files whose names begin with matching
+   TEXT.  */
+extern completion_list complete_source_filenames (const char *text);
+
+/* Complete on expressions.  Often this means completing on symbol
+   names, but some language parsers also have support for completing
+   field names.  */
+extern void complete_expression (completion_tracker &tracker,
+				 const char *text, const char *word);
+
 extern const char *skip_quoted_chars (const char *, const char *,
 				      const char *);
 
diff --git a/gdb/linespec.c b/gdb/linespec.c
index 26baad0..136cb65 100644
--- a/gdb/linespec.c
+++ b/gdb/linespec.c
@@ -46,6 +46,35 @@
 #include "location.h"
 #include "common/function-view.h"
 
+/* An enumeration of the various things a user might attempt to
+   complete for a linespec location.  */
+
+enum class linespec_complete_what
+{
+  /* Nothing, no possible completion.  */
+  NOTHING,
+
+  /* A function/method name.  Due to ambiguity between
+
+       (gdb) b source[TAB]
+       source_file.c
+       source_function
+
+     this can also indicate a source filename, iff we haven't seen a
+     separate source filename component, as in "b source.c:function".  */
+  FUNCTION,
+
+  /* A label symbol.  E.g., break file.c:function:LABEL.  */
+  LABEL,
+
+  /* An expression.  E.g., "break foo if EXPR", or "break *EXPR".  */
+  EXPRESSION,
+
+  /* A linespec keyword ("if"/"thread"/"task").
+     E.g., "break func threa<tab>".  */
+  KEYWORD,
+};
+
 typedef struct symbol *symbolp;
 DEF_VEC_P (symbolp);
 
@@ -271,6 +300,29 @@ struct ls_parser
   /* The result of the parse.  */
   struct linespec result;
 #define PARSER_RESULT(PPTR) (&(PPTR)->result)
+
+  /* What the parser believes the current word point should complete
+     to.  */
+  linespec_complete_what complete_what;
+
+  /* The completion word point.  The parser advances this as it skips
+     tokens.  At some point the input string will end or parsing will
+     fail, and then we attempt completion at the captured completion
+     word point, interpreting the string at completion_word as
+     COMPLETE_WHAT.  */
+  const char *completion_word;
+
+  /* If the current token was a quoted string, then this is the
+     quoting character (either " or ').  */
+  int completion_quote_char;
+
+  /* If the current token was a quoted string, then this points at the
+     end of the quoted string.  */
+  const char *completion_quote_end;
+
+  /* If parsing for completion, then this points at the completion
+     tracker.  Otherwise, this is NULL.  */
+  struct completion_tracker *completion_tracker;
 };
 typedef struct ls_parser linespec_parser;
 
@@ -543,6 +595,30 @@ find_parameter_list_end (const char *input)
   return p;
 }
 
+/* If the [STRING, STRING_LEN) string ends with what looks like a
+   keyword, return the keyword start offset in STRING.  Return -1
+   otherwise.  */
+
+static size_t
+string_find_incomplete_keyword_at_end (const char * const *keywords,
+				       const char *string, size_t string_len)
+{
+  const char *end = string + string_len;
+  const char *p = end;
+
+  while (p > string && *p != ' ')
+    --p;
+  if (p > string)
+    {
+      p++;
+      size_t len = end - p;
+      for (size_t i = 0; keywords[i] != NULL; ++i)
+	if (strncmp (keywords[i], p, len) == 0)
+	  return p - string;
+    }
+
+  return -1;
+}
 
 /* Lex a string from the input in PARSER.  */
 
@@ -590,13 +666,31 @@ linespec_lexer_lex_string (linespec_parser *parser)
       /* Skip to the ending quote.  */
       end = skip_quote_char (PARSER_STREAM (parser), quote_char);
 
-      /* Error if the input did not terminate properly.  */
-      if (end == NULL)
-	error (_("unmatched quote"));
+      /* This helps the completer mode decide whether we have a
+	 complete string.  */
+      parser->completion_quote_char = quote_char;
+      parser->completion_quote_end = end;
 
-      /* Skip over the ending quote and mark the length of the string.  */
-      PARSER_STREAM (parser) = (char *) ++end;
-      LS_TOKEN_STOKEN (token).length = PARSER_STREAM (parser) - 2 - start;
+      /* Error if the input did not terminate properly, unless in
+	 completion mode.  */
+      if (end == NULL)
+	{
+	  if (parser->completion_tracker == NULL)
+	    error (_("unmatched quote"));
+
+	  /* In completion mode, we'll try to complete the incomplete
+	     token.  */
+	  token.type = LSTOKEN_STRING;
+	  while (*PARSER_STREAM (parser) != '\0')
+	    PARSER_STREAM (parser)++;
+	  LS_TOKEN_STOKEN (token).length = PARSER_STREAM (parser) - 1 - start;
+	}
+      else
+	{
+	  /* Skip over the ending quote and mark the length of the string.  */
+	  PARSER_STREAM (parser) = (char *) ++end;
+	  LS_TOKEN_STOKEN (token).length = PARSER_STREAM (parser) - 2 - start;
+	}
     }
   else
     {
@@ -835,13 +929,48 @@ linespec_lexer_lex_one (linespec_parser *parser)
 }
 
 /* Consume the current token and return the next token in PARSER's
-   input stream.  */
+   input stream.  Also advance the completion word for completion
+   mode.  */
 
 static linespec_token
 linespec_lexer_consume_token (linespec_parser *parser)
 {
+  gdb_assert (parser->lexer.current.type != LSTOKEN_EOI);
+
+  bool advance_word = (parser->lexer.current.type != LSTOKEN_STRING
+		       || *PARSER_STREAM (parser) != '\0');
+
+  /* If we're moving past a string to some other token, it must be the
+     quote was terminated.  */
+  if (parser->completion_quote_char)
+    {
+      gdb_assert (parser->lexer.current.type == LSTOKEN_STRING);
+
+      /* If the string was the last (non-EOI) token, we're past the
+	 quote, but remember that for later.  */
+      if (*PARSER_STREAM (parser) != '\0')
+	{
+	  parser->completion_quote_char = '\0';
+	  parser->completion_quote_end = NULL;;
+	}
+    }
+
   parser->lexer.current.type = LSTOKEN_CONSUMED;
-  return linespec_lexer_lex_one (parser);
+  linespec_lexer_lex_one (parser);
+
+  if (parser->lexer.current.type == LSTOKEN_STRING)
+    {
+      /* Advance the completion word past a potential initial
+	 quote-char.  */
+      parser->completion_word = LS_TOKEN_STOKEN (parser->lexer.current).ptr;
+    }
+  else if (advance_word)
+    {
+      /* Advance the completion word past any whitespace.  */
+      parser->completion_word = PARSER_STREAM (parser);
+    }
+
+  return parser->lexer.current;
 }
 
 /* Return the next token without consuming the current token.  */
@@ -852,10 +981,16 @@ linespec_lexer_peek_token (linespec_parser *parser)
   linespec_token next;
   const char *saved_stream = PARSER_STREAM (parser);
   linespec_token saved_token = parser->lexer.current;
+  int saved_completion_quote_char = parser->completion_quote_char;
+  const char *saved_completion_quote_end = parser->completion_quote_end;
+  const char *saved_completion_word = parser->completion_word;
 
   next = linespec_lexer_consume_token (parser);
   PARSER_STREAM (parser) = saved_stream;
   parser->lexer.current = saved_token;
+  parser->completion_quote_char = saved_completion_quote_char;
+  parser->completion_quote_end = saved_completion_quote_end;
+  parser->completion_word = saved_completion_word;
   return next;
 }
 
@@ -1599,6 +1734,17 @@ source_file_not_found_error (const char *name)
   throw_error (NOT_FOUND_ERROR, _("No source file named %s."), name);
 }
 
+/* Unless at EIO, save the current stream position as completion word
+   point, and consume the next token.  */
+
+static linespec_token
+save_stream_and_consume_token (linespec_parser *parser)
+{
+  if (linespec_lexer_peek_token (parser).type != LSTOKEN_EOI)
+    parser->completion_word = PARSER_STREAM (parser);
+  return linespec_lexer_consume_token (parser);
+}
+
 /* See description in linespec.h.  */
 
 struct line_offset
@@ -1626,6 +1772,26 @@ linespec_parse_line_offset (const char *string)
   return line_offset;
 }
 
+/* In completion mode, if the user is still typing the number, there's
+   no possible completion to offer.  But if there's already input past
+   the number, setup to expect NEXT.  */
+
+static void
+set_completion_after_number (linespec_parser *parser,
+			     linespec_complete_what next)
+{
+  if (*PARSER_STREAM (parser) == ' ')
+    {
+      parser->completion_word = skip_spaces_const (PARSER_STREAM (parser) + 1);
+      parser->complete_what = next;
+    }
+  else
+    {
+      parser->completion_word = PARSER_STREAM (parser);
+      parser->complete_what = linespec_complete_what::NOTHING;
+    }
+}
+
 /* Parse the basic_spec in PARSER's input.  */
 
 static void
@@ -1641,11 +1807,20 @@ linespec_parse_basic (linespec_parser *parser)
   token = linespec_lexer_lex_one (parser);
 
   /* If it is EOI or KEYWORD, issue an error.  */
-  if (token.type == LSTOKEN_KEYWORD || token.type == LSTOKEN_EOI)
-    unexpected_linespec_error (parser);
+  if (token.type == LSTOKEN_KEYWORD)
+    {
+      parser->complete_what = linespec_complete_what::NOTHING;
+      unexpected_linespec_error (parser);
+    }
+  else if (token.type == LSTOKEN_EOI)
+    {
+      unexpected_linespec_error (parser);
+    }
   /* If it is a LSTOKEN_NUMBER, we have an offset.  */
   else if (token.type == LSTOKEN_NUMBER)
     {
+      set_completion_after_number (parser, linespec_complete_what::KEYWORD);
+
       /* Record the line offset and get the next token.  */
       name = copy_token_string (token);
       cleanup = make_cleanup (xfree, name);
@@ -1657,7 +1832,10 @@ linespec_parse_basic (linespec_parser *parser)
 
       /* If the next token is a comma, stop parsing and return.  */
       if (token.type == LSTOKEN_COMMA)
-	return;
+	{
+	  parser->complete_what = linespec_complete_what::NOTHING;
+	  return;
+	}
 
       /* If the next token is anything but EOI or KEYWORD, issue
 	 an error.  */
@@ -1670,12 +1848,58 @@ linespec_parse_basic (linespec_parser *parser)
 
   /* Next token must be LSTOKEN_STRING.  */
   if (token.type != LSTOKEN_STRING)
-    unexpected_linespec_error (parser);
+    {
+      parser->complete_what = linespec_complete_what::NOTHING;
+      unexpected_linespec_error (parser);
+    }
 
   /* The current token will contain the name of a function, method,
      or label.  */
-  name  = copy_token_string (token);
-  cleanup = make_cleanup (xfree, name);
+  name = copy_token_string (token);
+  cleanup = make_cleanup (free_current_contents, &name);
+
+  if (parser->completion_tracker != NULL)
+    {
+      /* If the function name ends with a ":", then this may be an
+	 incomplete "::" scope operator instead of a label separator.
+	 E.g.,
+	   "b klass:<tab>"
+	 which should expand to:
+	   "b klass::method()"
+
+	 Do a tentative completion assuming the later.  If we find
+	 completions, advance the stream past the colon token and make
+	 it part of the function name/token.  */
+
+      if (!parser->completion_quote_char
+	  && strcmp (PARSER_STREAM (parser), ":") == 0)
+	{
+	  completion_tracker tmp_tracker;
+	  const char *source_filename
+	    = PARSER_EXPLICIT (parser)->source_filename;
+
+	  linespec_complete_function (tmp_tracker,
+				      parser->completion_word,
+				      source_filename);
+
+	  if (tmp_tracker.have_completions ())
+	    {
+	      PARSER_STREAM (parser)++;
+	      LS_TOKEN_STOKEN (token).length++;
+
+	      xfree (name);
+	      name = savestring (parser->completion_word,
+				 (PARSER_STREAM (parser)
+				  - parser->completion_word));
+	    }
+	}
+
+      PARSER_EXPLICIT (parser)->function_name = name;
+      discard_cleanups (cleanup);
+    }
+  else
+    {
+      /* XXX Reindent before pushing.  */
 
   /* Try looking it up as a function/method.  */
   find_linespec_symbols (PARSER_STATE (parser),
@@ -1735,11 +1959,19 @@ linespec_parse_basic (linespec_parser *parser)
 	  return;
 	}
     }
+    }
+
+  int previous_qc = parser->completion_quote_char;
 
   /* Get the next token.  */
   token = linespec_lexer_consume_token (parser);
 
-  if (token.type == LSTOKEN_COLON)
+  if (token.type == LSTOKEN_EOI)
+    {
+      if (previous_qc && !parser->completion_quote_char)
+	parser->complete_what = linespec_complete_what::KEYWORD;
+    }
+  else if (token.type == LSTOKEN_COLON)
     {
       /* User specified a label or a lineno.  */
       token = linespec_lexer_consume_token (parser);
@@ -1748,17 +1980,56 @@ linespec_parse_basic (linespec_parser *parser)
 	{
 	  /* User specified an offset.  Record the line offset and
 	     get the next token.  */
+	  set_completion_after_number (parser, linespec_complete_what::KEYWORD);
+
 	  name = copy_token_string (token);
 	  cleanup = make_cleanup (xfree, name);
 	  PARSER_EXPLICIT (parser)->line_offset
 	    = linespec_parse_line_offset (name);
 	  do_cleanups (cleanup);
 
-	  /* Ge the next token.  */
+	  /* Get the next token.  */
 	  token = linespec_lexer_consume_token (parser);
 	}
+      else if (token.type == LSTOKEN_EOI && parser->completion_tracker != NULL)
+	{
+	  parser->complete_what = linespec_complete_what::LABEL;
+	}
       else if (token.type == LSTOKEN_STRING)
 	{
+	  parser->complete_what = linespec_complete_what::LABEL;
+
+	  /* If we have text after the label separated by whitespace
+	     (e.g., "b func():lab i<tab>"), don't consider it part of
+	     the label.  In completion mode that should complete to
+	     "if", in normal mode, the 'i' should be treated as
+	     garbage.  */
+	  if (parser->completion_quote_char == '\0')
+	    {
+	      const char *ptr = LS_TOKEN_STOKEN (token).ptr;
+	      for (size_t i = 0; i < LS_TOKEN_STOKEN (token).length; i++)
+		{
+		  if (ptr[i] == ' ')
+		    {
+		      LS_TOKEN_STOKEN (token).length = i;
+		      PARSER_STREAM (parser) = skip_spaces_const (ptr + i + 1);
+		      break;
+		    }
+		}
+	    }
+
+	  if (parser->completion_tracker != NULL)
+	    {
+	      if (PARSER_STREAM (parser)[-1] == ' ')
+		{
+		  parser->completion_word = PARSER_STREAM (parser);
+		  parser->complete_what = linespec_complete_what::KEYWORD;
+		}
+	    }
+	  else
+	    {
+	      /* XXX Reindent before pushing.  */
+
 	  /* Grab a copy of the label's name and look it up.  */
 	  name = copy_token_string (token);
 	  cleanup = make_cleanup (xfree, name);
@@ -1781,8 +2052,10 @@ linespec_parse_basic (linespec_parser *parser)
 				     name);
 	    }
 
+	    }
+
 	  /* Check for a line offset.  */
-	  token = linespec_lexer_consume_token (parser);
+	  token = save_stream_and_consume_token (parser);
 	  if (token.type == LSTOKEN_COLON)
 	    {
 	      /* Get the next token.  */
@@ -2272,11 +2545,15 @@ parse_linespec (linespec_parser *parser, const char *arg)
   struct gdb_exception file_exception = exception_none;
   struct cleanup *cleanup;
 
+  values.nelts = 0;
+  values.sals = NULL;
+
   /* A special case to start.  It has become quite popular for
      IDEs to work around bugs in the previous parser by quoting
      the entire linespec, so we attempt to deal with this nicely.  */
   parser->is_quote_enclosed = 0;
-  if (!is_ada_operator (arg)
+  if (parser->completion_tracker == NULL
+      && !is_ada_operator (arg)
       && strchr (linespec_quote_characters, *arg) != NULL)
     {
       const char *end;
@@ -2293,20 +2570,34 @@ parse_linespec (linespec_parser *parser, const char *arg)
 
   parser->lexer.saved_arg = arg;
   parser->lexer.stream = arg;
+  parser->completion_word = arg;
+  parser->complete_what = linespec_complete_what::FUNCTION;
 
   /* Initialize the default symtab and line offset.  */
   initialize_defaults (&PARSER_STATE (parser)->default_symtab,
 		       &PARSER_STATE (parser)->default_line);
 
   /* Objective-C shortcut.  */
-  values = decode_objc (PARSER_STATE (parser), PARSER_RESULT (parser), arg);
-  if (values.sals != NULL)
-    return values;
+  if (parser->completion_tracker == NULL)
+    {
+      values = decode_objc (PARSER_STATE (parser), PARSER_RESULT (parser), arg);
+      if (values.sals != NULL)
+	return values;
+    }
+  else
+    {
+      /* "-"/"+" is either an objc selector, or a number.  There's
+	 nothing to complete the latter to, so just let the caller
+	 complete on functions, which finds objc selectors, if there's
+	 any.  */
+      if ((arg[0] == '-' || arg[0] == '+') && arg[1] == '\0')
+	return {};
+    }
 
   /* Start parsing.  */
 
   /* Get the first token.  */
-  token = linespec_lexer_lex_one (parser);
+  token = linespec_lexer_consume_token (parser);
 
   /* It must be either LSTOKEN_STRING or LSTOKEN_NUMBER.  */
   if (token.type == LSTOKEN_STRING && *LS_TOKEN_STOKEN (token).ptr == '$')
@@ -2314,7 +2605,8 @@ parse_linespec (linespec_parser *parser, const char *arg)
       char *var;
 
       /* A NULL entry means to use GLOBAL_DEFAULT_SYMTAB.  */
-      VEC_safe_push (symtab_ptr, PARSER_RESULT (parser)->file_symtabs, NULL);
+      if (parser->completion_tracker == NULL)
+	VEC_safe_push (symtab_ptr, PARSER_RESULT (parser)->file_symtabs, NULL);
 
       /* User specified a convenience variable or history value.  */
       var = copy_token_string (token);
@@ -2333,8 +2625,16 @@ parse_linespec (linespec_parser *parser, const char *arg)
 	  goto convert_to_sals;
 	}
     }
+  else if (token.type == LSTOKEN_EOI && parser->completion_tracker != NULL)
+    {
+      /* Let the default linespec_complete_what::FUNCTION kick in.  */
+      unexpected_linespec_error (parser);
+    }
   else if (token.type != LSTOKEN_STRING && token.type != LSTOKEN_NUMBER)
-    unexpected_linespec_error (parser);
+    {
+      parser->complete_what = linespec_complete_what::NOTHING;
+      unexpected_linespec_error (parser);
+    }
 
   /* Shortcut: If the next token is not LSTOKEN_COLON, we know that
      this token cannot represent a filename.  */
@@ -2382,8 +2682,9 @@ parse_linespec (linespec_parser *parser, const char *arg)
 	}
     }
   /* If the next token is not EOI, KEYWORD, or COMMA, issue an error.  */
-  else if (token.type != LSTOKEN_EOI && token.type != LSTOKEN_KEYWORD
-	   && token.type != LSTOKEN_COMMA)
+  else if (parser->completion_tracker == NULL
+	   && (token.type != LSTOKEN_EOI && token.type != LSTOKEN_KEYWORD
+	       && token.type != LSTOKEN_COMMA))
     {
       /* TOKEN is the _next_ token, not the one currently in the parser.
 	 Consuming the token will give the correct error message.  */
@@ -2399,7 +2700,8 @@ parse_linespec (linespec_parser *parser, const char *arg)
   /* Parse the rest of the linespec.  */
   linespec_parse_basic (parser);
 
-  if (PARSER_RESULT (parser)->function_symbols == NULL
+  if (parser->completion_tracker == NULL
+      && PARSER_RESULT (parser)->function_symbols == NULL
       && PARSER_RESULT (parser)->labels.label_symbols == NULL
       && PARSER_EXPLICIT (parser)->line_offset.sign == LINE_OFFSET_UNKNOWN
       && PARSER_RESULT (parser)->minimal_symbols == NULL)
@@ -2420,11 +2722,21 @@ parse_linespec (linespec_parser *parser, const char *arg)
      if necessary.  */
   token = linespec_lexer_lex_one (parser);
   if (token.type != LSTOKEN_EOI && token.type != LSTOKEN_KEYWORD)
-    PARSER_STREAM (parser) = LS_TOKEN_STOKEN (token).ptr;
+    unexpected_linespec_error (parser);
+  else if (token.type == LSTOKEN_KEYWORD)
+    {
+      /* Setup the completion word past the keyword.  Lexing never
+	 advances past a keyword automatically, so skip it
+	 manually.  */
+      parser->completion_word
+	= skip_spaces_const (skip_to_space_const (PARSER_STREAM (parser)));
+      parser->complete_what = linespec_complete_what::EXPRESSION;
+    }
 
   /* Convert the data in PARSER_RESULT to SALs.  */
-  values = convert_linespec_to_sals (PARSER_STATE (parser),
-				     PARSER_RESULT (parser));
+  if (parser->completion_tracker == NULL)
+    values = convert_linespec_to_sals (PARSER_STATE (parser),
+				       PARSER_RESULT (parser));
 
   return values;
 }
@@ -2562,6 +2874,67 @@ linespec_complete_function (completion_tracker &tracker,
     collect_symbol_completion_matches (tracker, mode, function, function);
 }
 
+/* Helper for complete_linespec to simplify it.  SOURCE_FILENAME is
+   only meaningful if COMPONENT is FUNCTION.  */
+
+static void
+complete_linespec_component (linespec_parser *parser,
+			     completion_tracker &tracker,
+			     const char *text,
+			     linespec_complete_what component,
+			     const char *source_filename)
+{
+  if (component == linespec_complete_what::KEYWORD)
+    {
+      complete_on_enum (tracker, linespec_keywords, text, text);
+    }
+  else if (component == linespec_complete_what::EXPRESSION)
+    {
+      const char *word
+	= advance_to_expression_complete_word_point (tracker, text);
+      complete_expression (tracker, text, word);
+    }
+  else if (component == linespec_complete_what::FUNCTION)
+    {
+      completion_list fn_list;
+
+      linespec_complete_function (tracker, text, source_filename);
+      if (source_filename == NULL)
+	{
+	  /* Haven't seen a source component, like in "b
+	     file.c:function[TAB]".  Maybe this wasn't a function, but
+	     a filename instead, like "b file.[TAB]".  */
+	  fn_list = complete_source_filenames (text);
+	}
+
+      /* If we only have a single filename completion, append a ':' for
+	 the user, since that's the only thing that can usefully follow
+	 the filename.  */
+      if (fn_list.size () == 1 && !tracker.have_completions ())
+	{
+	  char *fn = fn_list[0].release ();
+
+	  /* If we also need to append a quote char, it needs to be
+	     appended before the ':'.  Append it now, and make ':' the
+	     new "quote" char.  */
+	  if (tracker.quote_char ())
+	    {
+	      char quote_char_str[2] = { tracker.quote_char () };
+
+	      fn = reconcat (fn, fn, quote_char_str, (char *) NULL);
+	      tracker.set_quote_char (':');
+	    }
+	  else
+	    fn = reconcat (fn, fn, ":", (char *) NULL);
+	  fn_list[0].reset (fn);
+
+	  /* Tell readline to skip appending a space.  */
+	  tracker.set_suppress_append_ws (true);
+	}
+      tracker.add_completions (std::move (fn_list));
+    }
+}
+
 /* Helper for linespec_complete_label.  Find labels that match
    LABEL_NAME in the function symbols listed in the PARSER, and add
    them to the tracker.  */
@@ -2625,6 +2998,201 @@ linespec_complete_label (completion_tracker &tracker,
   do_cleanups (cleanup);
 }
 
+/* See description in linespec.h.  */
+
+void
+linespec_complete (completion_tracker &tracker, const char *text)
+{
+  linespec_parser parser;
+  struct cleanup *cleanup;
+  const char *orig = text;
+
+  linespec_parser_new (&parser, 0, current_language, NULL, NULL, 0, NULL);
+  cleanup = make_cleanup (linespec_parser_delete, &parser);
+  parser.lexer.saved_arg = text;
+  PARSER_STREAM (&parser) = text;
+
+  parser.completion_tracker = &tracker;
+  PARSER_STATE (&parser)->is_linespec = 1;
+
+  /* Parse as much as possible.  parser.completion_word will hold
+     furthest completion point we managed to parse to.  */
+  TRY
+    {
+      parse_linespec (&parser, text);
+    }
+  CATCH (except, RETURN_MASK_ERROR)
+    {
+    }
+  END_CATCH
+
+  if (parser.completion_quote_char != '\0'
+      && parser.completion_quote_end != NULL
+      && parser.completion_quote_end[1] == '\0')
+    {
+      /* If completing a quoted string with the cursor right at
+	 terminating quote char, complete the completion word without
+	 interpretation, so that readline advances the cursor one
+	 whitespace past the quote, even if there's no match.  This
+	 makes these cases behave the same:
+
+	   before: "b function()"
+	   after:  "b function() "
+
+	   before: "b 'function()'"
+	   after:  "b 'function()' "
+
+	 and trusts the user in this case:
+
+	   before: "b 'not_loaded_function_yet()'"
+	   after:  "b 'not_loaded_function_yet()' "
+      */
+      parser.complete_what = linespec_complete_what::NOTHING;
+      parser.completion_quote_char = '\0';
+
+      gdb::unique_xmalloc_ptr<char> text_copy
+	(xstrdup (parser.completion_word));
+      tracker.add_completion (std::move (text_copy));
+    }
+
+  tracker.set_quote_char (parser.completion_quote_char);
+
+  if (parser.complete_what == linespec_complete_what::LABEL)
+    {
+      parser.complete_what = linespec_complete_what::NOTHING;
+
+      const char *func_name = PARSER_EXPLICIT (&parser)->function_name;
+
+      VEC (symbolp) *function_symbols;
+      VEC (bound_minimal_symbol_d) *minimal_symbols;
+      find_linespec_symbols (PARSER_STATE (&parser),
+			     PARSER_RESULT (&parser)->file_symtabs,
+			     func_name,
+			     &function_symbols, &minimal_symbols);
+
+      PARSER_RESULT (&parser)->function_symbols = function_symbols;
+      PARSER_RESULT (&parser)->minimal_symbols = minimal_symbols;
+
+      complete_label (tracker, &parser, parser.completion_word);
+    }
+  else if (parser.complete_what == linespec_complete_what::FUNCTION)
+    {
+      /* While parsing/lexing, we didn't know whether the completion
+	 word completes to a unique function/source name already or
+	 not.
+
+	 E.g.:
+	   "b function() <tab>"
+	 may need to complete either to:
+	   "b function() const"
+	 or to:
+	   "b function() if/thread/task"
+
+	 Or, this:
+	   "b foo t"
+	 may need to complete either to:
+	   "b foo template_fun<T>()"
+	 with "foo" being the template function's return type, or to:
+	   "b foo thread/task"
+
+	 Or, this:
+	   "b file<TAB>"
+	 may need to complete either to a source file name:
+	   "b file.c"
+	 or this, also a filename, but a unique completion:
+	   "b file.c:"
+	 or to a function name:
+	   "b file_function"
+
+	 Address that by completing assuming source or function, and
+	 seeing if we find a completion that matches exactly the
+	 completion word.  If so, then it must be a function (see note
+	 below) and we advance the completion word to the end of input
+	 and switch to KEYWORD completion mode.
+
+	 Note: if we find a unique completion for a source filename,
+	 then it won't match the completion word, because the LCD will
+	 contain a trailing ':'.  And if we're completing at or after
+	 the ':', then complete_linespec_component won't try to
+	 complete on source filenames.  */
+
+      const char *text = parser.completion_word;
+      const char *word = parser.completion_word;
+
+      complete_linespec_component (&parser, tracker,
+				   parser.completion_word,
+				   linespec_complete_what::FUNCTION,
+				   PARSER_EXPLICIT (&parser)->source_filename);
+
+      parser.complete_what = linespec_complete_what::NOTHING;
+
+      if (tracker.quote_char ())
+	{
+	  /* The function/file name was not close-quoted, so this
+	     can't be a keyword.  Note: complete_linespec_component
+	     may have swapped the original quote char for ':' when we
+	     get here, but that still indicates the same.  */
+	}
+      else if (!tracker.have_completions ())
+	{
+	  size_t key_start;
+	  size_t wordlen = strlen (parser.completion_word);
+
+	  key_start
+	    = string_find_incomplete_keyword_at_end (linespec_keywords,
+						     parser.completion_word,
+						     wordlen);
+
+	  if (key_start != -1
+	      || (wordlen > 0
+		  && parser.completion_word[wordlen - 1] == ' '))
+	    {
+	      parser.completion_word += key_start;
+	      parser.complete_what = linespec_complete_what::KEYWORD;
+	    }
+	}
+      else if (tracker.completes_to_completion_word (word))
+	{
+	  /* Skip the function and complete on keywords.  */
+	  parser.completion_word += strlen (word);
+	  parser.complete_what = linespec_complete_what::KEYWORD;
+	  tracker.discard_completions ();
+	}
+    }
+
+  tracker.advance_custom_word_point_by (parser.completion_word - orig);
+
+  complete_linespec_component (&parser, tracker,
+			       parser.completion_word,
+			       parser.complete_what,
+			       PARSER_EXPLICIT (&parser)->source_filename);
+
+  /* If we're past the "filename:function:label:offset" linespec, and
+     didn't find any match, then assume the user might want to create
+     a pending breakpoint anyway and offer the keyword
+     completions.  */
+  if (!parser.completion_quote_char
+      && (parser.complete_what == linespec_complete_what::FUNCTION
+	  || parser.complete_what == linespec_complete_what::LABEL
+	  || parser.complete_what == linespec_complete_what::NOTHING)
+      && !tracker.have_completions ())
+    {
+      const char *end
+	= parser.completion_word + strlen (parser.completion_word);
+
+      if (end > orig && end[-1] == ' ')
+	{
+	  tracker.advance_custom_word_point_by (end - parser.completion_word);
+
+	  complete_linespec_component (&parser, tracker, end,
+				       linespec_complete_what::KEYWORD,
+				       NULL);
+	}
+    }
+
+  do_cleanups (cleanup);
+}
+
 /* A helper function for decode_line_full and decode_line_1 to
    turn LOCATION into symtabs_and_lines.  */
 
diff --git a/gdb/linespec.h b/gdb/linespec.h
index f16bb81..27d237a 100644
--- a/gdb/linespec.h
+++ b/gdb/linespec.h
@@ -186,6 +186,11 @@ extern void linespec_lex_to_end (char **stringp);
 
 extern const char * const linespec_keywords[];
 
+/* Complete a linespec.  */
+
+extern void linespec_complete (completion_tracker &tracker,
+			       const char *text);
+
 /* Complete a function symbol, in linespec mode.  If SOURCE_FILENAME
    is non-NULL, limits completion to the list of functions defined in
    source files that match SOURCE_FILENAME.  */
diff --git a/gdb/testsuite/ChangeLog b/gdb/testsuite/ChangeLog
index f1f1ecb..3f0a13e 100644
--- a/gdb/testsuite/ChangeLog
+++ b/gdb/testsuite/ChangeLog
@@ -1,5 +1,11 @@
 2017-07-17  Pedro Alves  <palves@redhat.com>
 
+	* gdb.base/completion.exp: Adjust expected output.
+	* gdb.linespec/ls-errs.exp: Don't send tab characters, now that
+	the completer works.
+
+2017-07-17  Pedro Alves  <palves@redhat.com>
+
 	* gdb.linespec/ls-errs.exp (do_test): Adjust expected output.
 
 2017-07-15  Andrew Burgess  <andrew.burgess@embecosm.com>
diff --git a/gdb/testsuite/gdb.base/completion.exp b/gdb/testsuite/gdb.base/completion.exp
index 6597ea7..f03bfc3 100644
--- a/gdb/testsuite/gdb.base/completion.exp
+++ b/gdb/testsuite/gdb.base/completion.exp
@@ -790,7 +790,7 @@ gdb_test_multiple "" $test {
     -re "break\.c.*break1\.c.*$gdb_prompt " {
 	send_gdb "1\t\n"
 	gdb_test_multiple "" $test {
-	    -re ".*Function \"$srcfile2\" not defined\..*$gdb_prompt " {
+	    -re "malformed linespec error: unexpected end of input\r\n$gdb_prompt " {
 		pass $test
 	    }
 	    -re "$gdb_prompt p$" {
diff --git a/gdb/testsuite/gdb.linespec/ls-errs.exp b/gdb/testsuite/gdb.linespec/ls-errs.exp
index 7942f1f..f55cd0e 100644
--- a/gdb/testsuite/gdb.linespec/ls-errs.exp
+++ b/gdb/testsuite/gdb.linespec/ls-errs.exp
@@ -96,8 +96,13 @@ proc do_test {lang} {
     }
 
     # Some commonly used whitespace tests around ':'.
-    set spaces [list ":" ": " " :" " : " "\t:  " "  :\t" "\t:\t" \
-		     " \t:\t " "\t  \t:\t  \t  \t"]
+    set spaces [list \
+		    ":" \
+		    ": " \
+		    " :" \
+		    " : " \
+		    "  :  " \
+		   ]
 
     # A list of invalid offsets.
     set invalid_offsets [list -100 +500 1000]