Rewrite/enhance explicit locations completer, parse left->right
authorPedro Alves <palves@redhat.com>
Mon, 17 Jul 2017 19:21:33 +0000 (20:21 +0100)
committerPedro Alves <palves@redhat.com>
Mon, 17 Jul 2017 19:21:33 +0000 (20:21 +0100)
commitc6756f62e04846d68c24ee922ddb0377d4bd17f2
treee46b28cd0cb1ac766d8c086aef72d08aca2709e3
parentc5bb8910e80c6cd80c63541f86471c18375c8198
Rewrite/enhance explicit locations completer, parse left->right

One of the most annoying (to me) things about GDB's completion is 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 -f func[TAB]
 (gdb) b -f function(       # ok, gdb completed as much as possible.
 (gdb) b -f function([TAB]  # show me the overloads, please.
 <_all_ symbols in the program are shown...>

E.g., when debugging GDB, that'd be:

 (gdb) b -f function([TAB]
 (anonymous namespace)::get_global()::global  pt_insn_get_offset@plt                       scm_new_port_table_entry
 asprintf                                     pt_pkt_alloc_decoder                         scm_new_port_table_entry@plt
 asprintf@plt                                 pt_pkt_alloc_decoder@plt                     scm_out_of_range
 bt_ctf_get_char_array                        pt_pkt_sync_forward                          scm_out_of_range@plt
 bt_ctf_get_char_array@plt                    pt_pkt_sync_forward@plt                      scm_putc
 bt_ctf_get_uint64                            pwrite                                       scm_putc@plt
 bt_ctf_get_uint64@plt                        pwrite@plt                                   scm_reverse_x
 bt_ctf_iter_read_event                       PyErr_Restore                                scm_reverse_x@plt
 bt_ctf_iter_read_event@plt                   PyErr_Restore@plt                            scm_set_port_filename_x
 <snip...>

Now that's a load of completely useless completions.

The reason GDB offers those is that the completer relies on readline
figuring out the completion word point in the input line based on the
language's word break characters, which include "(".  So readline
tells the completer to complete on "", the string that is after '('.
Likewise, if you type "function(i[TAB]" to try to complete to "int",
you're out of luck.  GDB shows you all the symbols in the program that
start with "i"...  This makes sense for the expression completer, as
what you'd want to type is e.g., a global variable, say:

(gdb) print function(i[TAB]

but, it makes no sense when specifying a function name for a
breakpoint location.

To get around that limitation, users need to quote the function name,
like:

 (gdb) b -f 'function([TAB]
 function(int)      function(long)
 (gdb) b 'function(i[TAB]
 (gdb) b 'function(int)' # now completes correctly!

Note that the quoting is only necessary for completion.  Creating the
breakpoint does not require the quoting:

 (gdb) b -f function(int) [RET]
 Breakpoint 1 at ....

This patch removes this limitation.

(
Actually, it's a necessary patch, though not sufficient.  That'll
start working correctly by the end of the series.  With this patch, if try it,
you'll see:

 (gdb) b -f function(i[TAB]
 (gdb) b -f function

i.e., gdb strips everything after the "(".  That's caused by some code
in symtab.c that'll be eliminated further down the series.  These
patches are all unfortunately interrelated, which is also the reason
new tests only appear much later in the series.
But let's ignore that reality for the remainder of the description.
)

So... this patch gets rid of the need for quoting.

It does that by adding a way for a completer to control the exact
completion word point that readline should start the completion
request for, instead of letting readline try to figure it out using
the current language's word break chars array, and often failing.

In the case above, we want the completer to figure out that it's
completing a function name that starts with "function(i".  It now
does.

It took me a while to figure out a way to ask readline to "use this
exact word point", and for a while I feared that it'd be impossible
with current readline (and having to rely on master readline for core
functionality is something I'd like to avoid very much).  Eventually,
after several different attempts, I came up with what is described in
the comment above gdb_custom_word_point_brkchars in the patch.

With this patch, the handle_brkchars phase of the explicit location
completer advances the expected word point as it parses the input line
left to right, until it figures out exactly what we're completing,
instead of expecting readline to break the string using the word break
characters, and then having the completer heuristically fix up a bad
decision by parsing the input string backwards.  This allows correctly
knowning that we're completing a symbol name after -function, complete
functions without quoting, etc.

Later, we'll make use of this same mechanims to implement a proper
linespec completer that avoids need for quoting too.

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

* ada-lang.c (ada_collect_symbol_completion_matches): Add
complete_symbol_mode parameter.
* cli/cli-cmds.c (complete_command): Get the completion result out
of the handle_brkchars tracker if used a custom word point.
* completer.c: Include "linespec.h".
(enum explicit_location_match_type) <MATCH_LINE>: New enumerator.
(advance_to_expression_complete_word_point): New.
(completion_tracker::completes_to_completion_word): New.
(complete_files_symbols): Pass down
complete_symbol_mode::EXPRESSION.
(explicit_options, probe_options): New.
(collect_explicit_location_matches): Complete on the
explictit_loc->foo instead of word.  Use
linespec_complete_function.  Handle MATCH_LINE.  Handle offering
keyword and options completions.
(backup_text_ptr): Delete.
(skip_keyword): New.
(complete_explicit_location): Remove 'word' parameter.  Add
language, quoted_arg_start and quoted_arg_end parameters.
Rewrite, parsing left to right.
(location_completer): Rewrite.
(location_completer_handle_brkchars): New function.
(symbol_completer): Pass down complete_symbol_mode::EXPRESSION.
(enum complete_line_internal_reason): Adjust comments.
(completion_tracker::discard_completions): New.
(completer_handle_brkchars_func_for_completer): Handle
location_completer.
(gdb_custom_word_point_brkchars)
(gdb_org_rl_basic_quote_characters): New.
(gdb_completion_word_break_characters_throw)
(completion_find_completion_word): Handle trackers that use a
custom word point.
(completion_tracker::advance_custom_word_point_by): New.
(completion_tracker::build_completion_result): Don't rely on
readline appending the quote char.
(gdb_rl_attempted_completion_function_throw): Handle trackers that
use a custom word point.
(gdb_rl_attempted_completion_function): Restore
rl_basic_quote_characters.
* completer.h (class completion_tracker): Extend intro comment.
(completion_tracker::set_quote_char)
(completion_tracker::quote_char)
(completion_tracker::set_use_custom_word_point)
(completion_tracker::use_custom_word_point)
(completion_tracker::custom_word_point)
(completion_tracker::set_custom_word_point)
(completion_tracker::advance_custom_word_point_by)
(completion_tracker::completes_to_completion_word)
(completion_tracker::discard_completions): New methods.
(completion_tracker::m_quote_char)
(completion_tracker::m_use_custom_word_point)
(completion_tracker::m_custom_word_point): New fields.
(advance_to_expression_complete_word_point): Declare.
* f-lang.c (f_collect_symbol_completion_matches): Add
complete_symbol_mode parameter.
* language.h (struct language_defn)
<la_collect_symbol_completion_matches>: Add complete_symbol_mode
parameter.
* linespec.c (linespec_keywords): Add NULL terminator.  Make extern.
(linespec_complete_function): New function.
(linespec_lexer_lex_keyword): Adjust.
* linespec.h (linespec_keywords, linespec_complete_function): New
declarations.
* location.c (find_end_quote): New function.
(explicit_location_lex_one): Add explicit_completion_info
parameter.  Save quoting info.  Don't throw if being called for
completion.  Don't handle Ada operators here.
(is_cp_operator, skip_op_false_positives, first_of)
(explicit_location_lex_one_function): New function.
(string_to_explicit_location): Replace 'dont_throw' parameter with
an explicit_completion_info pointer parameter.  Handle it.  Don't
use explicit_location_lex_one to lex function names.  Use
explicit_location_lex_one_function instead.
* location.h (struct explicit_completion_info): New.
(string_to_explicit_location): Replace 'dont_throw' parameter with
an explicit_completion_info pointer parameter.
* symtab.c (default_collect_symbol_completion_matches_break_on):
Add complete_symbol_mode parameter.  Handle LINESPEC mode.
(default_collect_symbol_completion_matches)
(collect_symbol_completion_matches): Add complete_symbol_mode
parameter.
(collect_symbol_completion_matches_type): Pass down
complete_symbol_mode::EXPRESSION.
(collect_file_symbol_completion_matches): Add complete_symbol_mode
parameter.  Handle LINESPEC mode.
* symtab.h (complete_symbol_mode): New.
(default_collect_symbol_completion_matches_break_on)
(default_collect_symbol_completion_matches)
(collect_symbol_completion_matches)
(collect_file_symbol_completion_matches): Add complete_symbol_mode
parameter.

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

* gdb.linespec/ls-errs.exp (do_test): Adjust expected output.
15 files changed:
gdb/ChangeLog
gdb/ada-lang.c
gdb/cli/cli-cmds.c
gdb/completer.c
gdb/completer.h
gdb/f-lang.c
gdb/language.h
gdb/linespec.c
gdb/linespec.h
gdb/location.c
gdb/location.h
gdb/symtab.c
gdb/symtab.h
gdb/testsuite/ChangeLog
gdb/testsuite/gdb.linespec/ls-errs.exp