This adds support for comand options to the "backtrace" command. We'll get:
(gdb) bt -
-entry-values -hide -past-main
-frame-arguments -no-filters -raw-frame-arguments
-full -past-entry
~~~~
(gdb) help backtrace
Print backtrace of all stack frames, or innermost COUNT frames.
Usage: backtrace [OPTION]... [QUALIFIER]... [COUNT | -COUNT]
Options:
-entry-values no|only|preferred|if-needed|both|compact|default
Set printing of function arguments at function entry
GDB can sometimes determine the values of function arguments at entry,
in addition to their current values. This option tells GDB whether
to print the current value, the value at entry (marked as val@entry),
or both. Note that one or both of these values may be <optimized out>.
-frame-arguments all|scalars|none
Set printing of non-scalar frame arguments
-raw-frame-arguments [on|off]
Set whether to print frame arguments in raw form.
If set, frame arguments are printed in raw form, bypassing any
pretty-printers for that value.
-past-main [on|off]
Set whether backtraces should continue past "main".
Normally the caller of "main" is not of interest, so GDB will terminate
the backtrace at "main". Set this if you need to see the rest
of the stack trace.
-past-entry [on|off]
Set whether backtraces should continue past the entry point of a program.
Normally there are no callers beyond the entry point of a program, so GDB
will terminate the backtrace there. Set this if you need to see
the rest of the stack trace.
-full
Print values of local variables.
-no-filters
Prohibit frame filters from executing on a backtrace.
-hide
Causes Python frame filter elided frames to not be printed.
For backward compatibility, the following qualifiers are supported:
full - same as -full option.
no-filters - same as -no-filters option.
hide - same as -hide.
With a negative COUNT, print outermost -COUNT frames.
~~~~
Implementation wise, this:
- Moves relevant options/settings globals to structures.
- Tweaks a number of functions to pass down references to such structures.
- Adds option_def structures describing the options/settings.
- Makes backtrace_command parse the options, with gdb::option::process_options.
- Tweaks "backtrace"'s help to describe the new options.
- Adds testcases.
Note that backtrace is a PROCESS_OPTIONS_UNKNOWN_IS_OPERAND command,
because of the "-COUNT" argument.
The COUNT/-COUNT argument is currently parsed as an expression. I
considered whether it would be prudent here to require "--", but
concluded that the risk of causing a significant breakage here is much
lower compared to "print", since printing the expression is not the
whole point of the "backtrace" command. Seems OK to me to require
typing "backtrace -past-main -- -p" if the user truly wants to refer
to the negative of a backtrace count stored in an inferior variable
called "p".
gdb/ChangeLog:
2019-06-13 Pedro Alves <palves@redhat.com>
* frame.c: Include "cli/cli-option.h.
(user_set_backtrace_options): New.
(backtrace_past_main, backtrace_past_entry, backtrace_limit):
Delete.
(get_prev_frame): Adjust.
(boolean_option_def, uinteger_option_def)
(set_backtrace_option_defs): New.
(_initialize_frame): Adjust and use
gdb::option::add_setshow_cmds_for_options to install "set
backtrace past-main" and "set backtrace past-entry".
* frame.h: Include "cli/cli-option.h".
(struct frame_print_options): Forward declare.
(print_frame_arguments_all, print_frame_arguments_scalars)
(print_frame_arguments_none): Declare.
(print_entry_values): Delete declaration.
(struct frame_print_options, user_frame_print_options): New.
(struct set_backtrace_options): New.
(set_backtrace_option_defs, user_set_backtrace_options): Declare.
* mi/mi-cmd-stack.c (mi_cmd_stack_list_frames)
(mi_cmd_stack_list_locals, mi_cmd_stack_list_args)
(mi_cmd_stack_list_variables): Pass down USER_FRAME_PRINT_OPTIONS.
(list_args_or_locals): Add frame_print_options parameter.
(mi_cmd_stack_info_frame): Pass down USER_FRAME_PRINT_OPTIONS.
* python/py-framefilter.c (enumerate_args): Pass down
USER_FRAME_PRINT_OPTIONS.
* stack.c: Include "cli/cli-option.h".
(print_frame_arguments_all, print_frame_arguments_scalars)
(print_frame_arguments_none): Declare.
(print_raw_frame_arguments, print_entry_values): Delete.
(user_frame_print_options): New.
(boolean_option_def, enum_option_def, frame_print_option_defs):
New.
(struct backtrace_cmd_options): New.
(bt_flag_option_def): New.
(backtrace_command_option_defs): New.
(print_stack_frame): Pass down USER_FRAME_PRINT_OPTIONS.
(print_frame_arg, read_frame_arg, print_frame_args)
(print_frame_info, print_frame): Add frame_print_options parameter
and use it.
(info_frame_command_core): Pass down USER_FRAME_PRINT_OPTIONS.
(backtrace_command_1): Add frame_print_options and
backtrace_cmd_options parameters and use them.
(make_backtrace_options_def_group): New.
(backtrace_command): Process command options with
gdb::option::process_options.
(backtrace_command_completer): New.
(_initialize_stack): Extend "backtrace"'s help to mention
supported options. Install completer for "backtrace".
Install some settings commands with add_setshow_cmds_for_options.
gdb/testsuite/ChangeLog:
2019-06-13 Pedro Alves <palves@redhat.com>
* gdb.base/options.exp (test-backtrace): New.
(top level): Call it.
A following patch will introduce options for the "backtrace" command,
based on some "set print" and "set backtrace" settings. There's one
setting in particular that is a bit annoying if we want to describe
the backtrace options and the settings commands using the same data
structures:
"set print raw frame-arguments"
The problem is that space between "raw" and "frame-arguments".
Calling the option
"bt -raw frame-arguments"
would be odd. So I'm calling the option
"bt -raw-frame-arguments"
instead.
And for consistency, this patch renames the set/show commands to:
"set print raw-frame-arguments"
"show print raw-frame-arguments"
I.e., dash instead of space. The old commands are left in place, but
marked deprecated.
We need to adjust a couple testcases, because the relevant tests use
gdb_test_no_output and the old commands are no longer silent:
(gdb) set print raw frame-arguments on
Warning: command 'set print raw frame-arguments' is deprecated.
Use 'set print raw-frame-arguments'.
gdb/ChangeLog:
2019-06-13 Pedro Alves <palves@redhat.com>
* NEWS (Changed commands): Mention set/show print raw-frame-arguments,
and that "set/show print raw frame-arguments" are now deprecated.
* cli/cli-decode.c (add_setshow_boolean_cmd): Now returns the
command.
* command.h (add_setshow_boolean_cmd): Return cmd_list_element *.
* stack.c (_initialize_stack): Install "set/show print
raw-frame-arguments", and deprecate "set/show print raw
frame-arguments".
* valprint.c (_initialize_valprint): Deprecate "set/show print
raw".
gdb/doc/ChangeLog:
2019-06-13 Pedro Alves <palves@redhat.com>
* gdb.texinfo (Print Settings): Document "set/show print
raw-frame-arguments" instead of "set/show print raw
frame-arguments".
gdb/testsuite/ChangeLog:
2019-06-13 Pedro Alves <palves@redhat.com>
* gdb.guile/scm-frame-args.exp: Use "set print
raw-frame-arguments" instead of "set print raw frame-arguments".
* gdb.python/py-frame-args.exp: Likewise.
As I was in the neighbourhood, I converted the other "compile"
subcommands to the new options framework too. Specifically, "compile
code" and "compile file".
The user-visible changes are:
- All abbreviations of "-raw" are accepted now, instead of just -r.
Obviously that means "-ra" is now accepted.
- Option completion now works.
- "compile file" did not have a completer yet, and now it knows to
complete on filenames.
- You couldn't use "compile file" with a file named "-something".
You can now, with "compile file -- -something".
gdb/ChangeLog:
2019-06-13 Pedro Alves <palves@redhat.com>
* compile/compile.c (struct compile_options): New.
(compile_flag_option_def, compile_command_option_defs)
(make_compile_options_def_group): New.
(compile_file_command): Handle options with
gdb::option::process_options.
(compile_file_command_completer): New function.
(compile_code_command): Handle options with
gdb::option::process_options.
(compile_code_command_completer): New function.
(_initialize_compiler): Install completers for "compile code" and
"compile file". Mention available options in "compile code" and
"compile code"'s help.
* completer.c (advance_to_completion_word): New, factored out from
...
(advance_to_expression_complete_word_point): ... this.
(advance_to_filename_complete_word_point): New.
* completer.h (advance_to_filename_complete_word_point): New
declaration.
gdb/testsuite/ChangeLog:
2019-06-13 Pedro Alves <palves@redhat.com>
* gdb.compile/compile.exp: Adjust expected output to option
processing changes.
This patch adds support for "print -option optval --", etc.
Likewise for "compile print".
We'll get:
~~~~~~
(gdb) help print
Print value of expression EXP.
Usage: print [[OPTION]... --] [/FMT] [EXP]
Options:
-address [on|off]
Set printing of addresses.
-array [on|off]
Set pretty formatting of arrays.
-array-indexes [on|off]
Set printing of array indexes.
-elements NUMBER|unlimited
Set limit on string chars or array elements to print.
"unlimited" causes there to be no limit.
-max-depth NUMBER|unlimited
Set maximum print depth for nested structures, unions and arrays.
When structures, unions, or arrays are nested beyond this depth then they
will be replaced with either '{...}' or '(...)' depending on the language.
Use "unlimited" to print the complete structure.
-null-stop [on|off]
Set printing of char arrays to stop at first null char.
-object [on|off]
Set printing of C++ virtual function tables.
-pretty [on|off]
Set pretty formatting of structures.
-repeats NUMBER|unlimited
Set threshold for repeated print elements.
"unlimited" causes all elements to be individually printed.
-static-members [on|off]
Set printing of C++ static members.
-symbol [on|off]
Set printing of symbol names when printing pointers.
-union [on|off]
Set printing of unions interior to structures.
-vtbl [on|off]
Set printing of C++ virtual function tables.
Note: because this command accepts arbitrary expressions, if you
specify any command option, you must use a double dash ("--")
to mark the end of option processing. E.g.: "print -o -- myobj".
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
I want to highlight the comment above about "--".
At first, I thought we could make the print command parse the options,
and if the option wasn't recognized, fallback to parsing as an
expression. Then, if the user wanted to disambiguate, he'd use the
"--" option delimiter. For example, if you had a variable called
"object" and you wanted to print its negative, you'd have to do:
(gdb) print -- -object
After getting that working, I saw that gdb.pascal/floats.exp
regressed, in these tests:
gdb_test "print -r" " = -1\\.2(499.*|5|500.*)"
gdb_test "print -(r)" " = -1.2(499.*|5|500.*)"
gdb_test "print -(r + s)" " = -3\\.4(499.*|5|500.*)"
It's the first one that I found most concerning. It regressed because
"-r" is the abbreviation of "-raw". I realized then that the behavior
change was a bit risker than I'd like, considering scripts, wrappers
around gdb, etc., and even user expectation. So instead, I made the
print command _require_ the "--" options delimiter if you want to
specify any option. So:
(gdb) print -r
is parsed as an expression, and
(gdb) print -r --
is parsed as an option.
I noticed that that's also what lldb's expr (the equivalent of print)
does to handle the same problem.
Going back the options themselves, note that:
- you can shorten option names, as long as unambiguous.
- For boolean options, 0/1 stand for off/on.
- For boolean options, "true" is implied.
So these are all equivalent:
(gdb) print -object on -static-members off -pretty on -- foo
(gdb) print -object -static-members off -pretty -- foo
(gdb) print -object -static-members 0 -pretty -- foo
(gdb) print -o -st 0 -p -- foo
TAB completion is fully supported:
(gdb) p -[TAB]
-address -elements -pretty -symbol
-array -null-stop -repeats -union
-array-indexes -object -static-members -vtbl
Note that the code is organized such that some of the options and the
"set/show" commands code is shared. In particular, the "print"
options and the corresponding "set print" commands are defined with
the same structures. The commands are installed with the
gdb::option::add_setshow_cmds_for_options function.
gdb/ChangeLog:
2019-06-13 Pedro Alves <palves@redhat.com>
* compile/compile.c: Include "cli/cli-option.h".
(compile_print_value): Scope data pointer is now a
value_print_options pointer; adjust.
(compile_print_command): Process options. Scope data pointer is
now a value_print_options pointer; adjust.
(_initialize_compile): Update "compile print"'s help to include
supported options. Install a completer for "compile print".
* cp-valprint.c (show_vtblprint, show_objectprint)
(show_static_field_print): Delete.
(_initialize_cp_valprint): Don't install "set print
static-members", "set print vtbl", "set print object" here.
* printcmd.c: Include "cli/cli-option.h" and
"common/gdb_optional.h".
(print_command_parse_format): Rework to fill in a
value_print_options instead of a format_data.
(print_value): Change parameter type from format_data pointer to
value_print_options reference. Adjust.
(print_command_1): Process options. Adjust to pass down a
value_print_options.
(print_command_completer): New.
(_initialize_printcmd): Install print_command_completer as
handle_brkchars completer for the "print" command. Update
"print"'s help to include supported options.
* valprint.c: Include "cli/cli-option.h".
(show_vtblprint, show_objectprint, show_static_field_print): Moved
here from cp-valprint.c.
(boolean_option_def, uinteger_option_def)
(value_print_option_defs, make_value_print_options_def_group):
New. Use gdb::option::add_setshow_cmds_for_options to install
"set print elements", "set print null-stop", "set print repeats",
"set print pretty", "set print union", "set print array", "set
print address", "set print symbol", "set print array-indexes".
* valprint.h: Include <string> and "cli/cli-option.h".
(make_value_print_options_def_group): Declare.
(print_value): Change parameter type from format_data pointer to
value_print_options reference.
(print_command_completer): Declare.
gdb/testsuite/ChangeLog:
2019-06-13 Pedro Alves <palves@redhat.com>
* gdb.base/options.exp: Build executable.
(test-print): New procedure.
(top level): Call it, once for "print" and another for "compile
print".
This commit adds a generic command options framework, that makes it
easy enough to add '-'-style options to commands in a uniform way,
instead of each command implementing option parsing in its own way.
Options are defined in arrays of option_def objects (for option
definition), and the same options definitions are used for supporting
TAB completion, and also for generating the relevant help fragment of
the "help" command. See the gdb::options::build_help function, which
returns a string with the result of replacing %OPTIONS% in a template
string with an auto-generated "help" string fragment for all the
passed-in options.
Since most options in GDB are in the form of "-OPT", with a single
dash, this is the format that the framework supports.
I like to think of gdb's "-OPT" as the equivalent to getopt's long
options format ("--OPT"), and gdb's "/" as the equivalent to getopt's
short options format. getopt's short options format allows mixing
several one-character options, like "ls -als", kind of similar to
gdb's "x /FMT" and "disassemble /MOD", etc. While with gdb's "-"
options, the option is expected to have a full name, and to be
abbreviatable. E.g., "watch -location", "break -function main", etc.
This patch only deals with "-" options. The above comment serves more
to disclose why I don't think we should support mixing several
unrelated options in a single "-" option invocation, like "thread
apply -qcs" instead of "thread apply -q -c -s".
The following patches will add uses of the infrastructure to several
key commands. Most notably, "print", "compile print", "backtrace",
"frame apply" and "thread apply". I tried to add options to several
commands in order to make sure the framework didn't leave that many
open holes open.
Options use the same type as set commands -- enum var_types. So
boolean options are var_boolean, enum options are var_enum, etc. The
idea is to share code between settings commands and command options.
The "print" options will be based on the "set print" commands, and
their names will be the same. Actually, their definitions will be the
same too. There is a function to create "set/show" commands from an
array for option definitions:
/* Install set/show commands for options defined in OPTIONS. DATA is
a pointer to the structure that holds the data associated with the
OPTIONS array. */
extern void add_setshow_cmds_for_options (command_class cmd_class, void *data,
gdb::array_view<const option_def> options,
struct cmd_list_element **set_list,
struct cmd_list_element **show_list);
That will be used by several following patches.
Other features:
- You can use the "--" delimiter to explicitly indicate end of
options. Several existing commands use this token sequence for
this effect already, so this just standardizes it.
- You can shorten option names, as long as unambiguous. Currently,
some commands allow this (e.g., break -function), while others do
not (thread apply all -ascending). As GDB allows abbreviating
command names and other things, it feels more GDB-ish to allow
abbreviating option names too, to me.
- For boolean options, 0/1 stands for off/on, just like with boolean
"set" commands.
- For boolean options, "true" is implied, just like with boolean "set
commands.
These are the option types supported, with a few examples:
- boolean options (var_boolean). The option's argument is optional.
(gdb) print -pretty on -- *obj
(gdb) print -pretty off -- *obj
(gdb) print -p -- *obj
(gdb) print -p 0 -- *obj
- flag options (like var_boolean, but no option argument (on/off))
(gdb) thread apply all -s COMMAND
- enum options (var_enum)
(gdb) bt -entry-values compact
(gdb) bt -e c
- uinteger options (var_uinteger)
(gdb) print -elements 100 -- *obj
(gdb) print -e 100 -- *obj
(gdb) print -elements unlimited -- *obj
(gdb) print -e u -- *obj
- zuinteger-unlimited options (var_zuinteger_unlimited)
(gdb) print -max-depth 100 -- obj
(gdb) print -max-depth -1 -- obj
(gdb) print -max-depth unlimited -- obj
Other var_types could be supported, of course. These were just the
types that I needed for the commands that I ported over, in the
following patches.
It was interesting (and unfortunate) to find that we need at least 3
different modes to cover the existing commands:
- Commands that require ending options with "--" if you specify any
option: "print" and "compile print".
- Commands that do not want to require "--", and want to error out if
you specify an unknown option (i.e., an unknown argument that starts
with '-'): "compile code" / "compile file".
- Commands that do not want to require "--", and want to process
unknown options themselves: "bt", because of "bt -COUNT",
"thread/frame apply", because "-" is a valid command.
The different behavior is encoded in the process_options_mode enum,
passed to process_options/complete_options.
For testing, this patch adds one representative maintenance command
for each of the process_options_mode values, that are used by the
testsuite to exercise the options framework:
(gdb) maint test-options require-delimiter
(gdb) maint test-options unknown-is-error
(gdb) maint test-options unknown-is-operand
and adds another command to help with TAB-completion testing:
(gdb) maint show test-options-completion-result
See their description at the top of the maint-test-options.c file.
Docs/NEWS are in a patch later in the series.
gdb/ChangeLog:
2019-06-13 Pedro Alves <palves@redhat.com>
* Makefile.in (SUBDIR_CLI_SRCS): Add cli/cli-option.c.
(COMMON_SFILES): Add maint-test-settings.c.
* cli/cli-decode.c (boolean_enums): New global, factored out from
...
(add_setshow_boolean_cmd): ... here.
* cli/cli-decode.h (boolean_enums): Declare.
* cli/cli-option.c: New file.
* cli/cli-option.h: New file.
* cli/cli-setshow.c (parse_cli_boolean_value(const char **)): New,
factored out from ...
(parse_cli_boolean_value(const char *)): ... this.
(is_unlimited_literal): Change parameter type to pointer to
pointer. Adjust and advance ARG pointer.
(parse_cli_var_uinteger, parse_cli_var_zuinteger_unlimited)
(parse_cli_var_enum): New, factored out from ...
(do_set_command): ... this. Adjust.
* cli/cli-setshow.h (parse_cli_boolean_value)
(parse_cli_var_uinteger, parse_cli_var_zuinteger_unlimited)
(parse_cli_var_enum): Declare.
* cli/cli-utils.c: Include "cli/cli-option.h".
(get_ulongest): New.
* cli/cli-utils.h (get_ulongest): Declare.
(check_for_argument): New overloads.
* maint-test-options.c: New file.
gdb/testsuite/ChangeLog:
2019-06-13 Pedro Alves <palves@redhat.com>
* gdb.base/options.c: New file.
* gdb.base/options.exp: New file.
While adding -OPT options to "frame apply level", I noticed that:
(gdb) frame apply level 0 -[TAB]
wasn't completing on the supported options. This commit fixes it.
We'll get instead:
(gdb) frame apply level 0 -
-c -past-entry -past-main -q -s
I added the isspace check because this case:
(gdb) frame apply level 0-
can't be an option.
Tests for this will be in a new gdb.base/options.exp file, in a
following patch. It will exercise all of:
(gdb) frame apply level 0-
(gdb) frame apply level 0 -
(gdb) frame apply level 0 --
(gdb) frame apply level 0 -- -
gdb/ChangeLog:
2019-06-13 Pedro Alves <palves@redhat.com>
* cli/cli-utils.c (number_or_range_parser::get_number): Do not
parse a range if "-" is at the end of the string.
We currently accept "o" with boolean/auto-boolean commands, taking it
to mean "on". But "o" is ambiguous, between "on" and "off". I can't
imagine why assuming the user wanted to type "on" is a good idea, it
might have been a typo.
This commit makes gdb error out. We now get:
(gdb) maint test-settings set boolean o
"on" or "off" expected.
(gdb) maint test-settings set auto-boolean o
"on", "off" or "auto" expected.
gdb/ChangeLog:
2019-06-13 Pedro Alves <palves@redhat.com>
* cli/cli-setshow.c (parse_auto_binary_operation)
(parse_cli_boolean_value): Don't allow "o".
gdb/testsuite/ChangeLog:
2019-06-13 Pedro Alves <palves@redhat.com>
* gdb.base/settings.exp (test-boolean, test-auto-boolean): Check
that "o" is ambiguous.
This commit adds new representative commands for all types of settings
commands supported by gdb (enum var_types), and then uses them to
exercise settings parsing and completion.
(gdb) maint test-settings s[TAB]
set show
(gdb) maint test-settings set [TAB]
auto-boolean integer uinteger
boolean optional-filename zinteger
enum string zuinteger
filename string-noescape zuinteger-unlimited
(gdb) maint test-settings set enum [TAB]
xxx yyy zzz
etc.
This is basically unit testing, except that it goes fully via GDB. It
must be done this way in order to exercise TAB completion properly,
which must go via readline.
gdb/ChangeLog:
2019-06-13 Pedro Alves <palves@redhat.com>
* Makefile.in (COMMON_SFILES): Add maint-test-settings.c.
* NEWS: Mention maint test-settings KIND.
* maint-test-settings.c: New file.
gdb/doc/ChangeLog:
2019-06-13 Pedro Alves <palves@redhat.com>
* gdb.texinfo (Maintenance Commands): Document "maint
test-settings" commands.
gdb/testsuite/ChangeLog:
2019-06-13 Pedro Alves <palves@redhat.com>
* gdb.base/settings.c: New file.
* gdb.base/settings.exp: New file.
Noticed this while writing the following patch. We cd to $srcdir, not $objdir.
gdb/testsuite/ChangeLog:
2019-06-13 Pedro Alves <palves@redhat.com>
* gdb.base/completion.exp: Fix comment typo.
The default command completer is symbol_completer, but it makes no
sense for a "show" command to complete on symbols, or anything else,
really.
I wonder whether we should instead make the default be no completer.
That seems like a much larger/complicated audit/change, so I'd like to
move forward with this version, as it'll be covered by tests. I
noticed this because a following patch will add a new
gdb.base/settings.exp testcase that exercises all sorts of details of
settings commands, including completing the show commands, using new
representative "maint test-settings <type or settings command>"
commands.
Also remove the completer for var_string and var_string_noescape
commands. No point in completing symbols when GDB is expecting a
string.
gdb/ChangeLog:
2019-06-13 Pedro Alves <palves@redhat.com>
* cli/cli-decode.c (add_setshow_cmd_full): Remove "show"
completer.
(add_setshow_string_cmd, add_setshow_string_noescape_cmd): Remove
"set" completers.
With enum commands, we currently fail to notice junk after the value.
Currently:
(gdb) set print entry-values compact foo
(gdb) show print entry-values foo
Printing of function arguments at function entry is "compact".
After this fix:
(gdb) set print entry-values compact foo
Junk after item "compact": foo
gdb/ChangeLog:
2019-06-13 Pedro Alves <palves@redhat.com>
* cli/cli-setshow.c (do_set_command) <var_enum>: Detect junk
after item.
Currently we can abbreviate "on/off/enable/disable/yes/no" in boolean
settings,
(gdb) set non-stop of
(gdb) set non-stop en
we can abbreviate the items of enumeration commands,
(gdb) set print frame-arguments scal
(gdb) set scheduler-locking rep
but we cannot abbreviate "unlimited" in integer commands.
(gdb) set print elements u
No symbol "u" in current context.
This commit fixes that.
Testcases will be in gdb.base/settings.exp and gdb.base/options.exp,
in following patches.
gdb/ChangeLog:
2019-06-13 Pedro Alves <palves@redhat.com>
* cli/cli-setshow.c (is_unlimited_literal): Allow abbreviations.
Basically every caller of check_for_argument needs to skip space after
the argument. This patch makes check_for_argument do it itself.
Suggested by Philippe Waroquiers.
gdb/ChangeLog:
2019-06-13 Pedro Alves <palves@redhat.com>
* ax-gdb.c (agent_command_1): Remove skip_spaces call.
* breakpoint.c (watch_maybe_just_location): Remove skip_spaces
call.
* cli/cli-cmds.c (apropos_command): Remove skip_spaces call.
* cli/cli-utils.c (extract_info_print_args): Remove skip_spaces
calls.
(check_for_argument): Skip spaces after argument.
I noticed this inconsistency in the error messages below:
(gdb) print --1
Left operand of assignment is not an lvalue.
(gdb) thread apply 1 print --1
Thread 1 (Thread 0x7ffff7fb6740 (LWP 17805)):
inverted range
The "inverted range" error happens because get_number_trailer returns
0 to indicate error, but number_or_range_parser::get_number is not
checking for that. I tried detected the error there, but that doesn't
work because number_of_range_parser is used in places that _do_ want
to legitimately handle 0. IMO we should fix get_number_trailer's
interface or use something else when we want to parse 0 too.
I've decided to fix it in a different way, similarly to how
number_or_range_parser::finished was changed in commit 529c08b25e
("Add helper functions parse_flags and parse_flags_qcs").
Seems like a good change, even if we tweaked
number_or_range_parser::get_number, as it simplifies
thread_apply_command and makes them consistent with
number_or_range_parser::finished().
We now get the same error message in both cases:
(gdb) print --1
Left operand of assignment is not an lvalue.
(gdb) thread apply 1 print --1
Thread 1 (Thread 0x7ffff7fb6740 (LWP 17805)):
Left operand of assignment is not an lvalue.
gdb/ChangeLog:
2019-06-13 Pedro Alves <palves@redhat.com>
* thread.c (thread_apply_command): Adjust TID parsing.
* tid-parse.c (tid_range_parser::finished): Ensure parsing end is
detected before end of string.
(tid_is_in_list): Error out if LIST is invalid.
gdb/testsuite/ChangeLog:
2019-06-13 Pedro Alves <palves@redhat.com>
* gdb.multi/tids.exp: Adjust expected output. Add "thread apply 1
foo --1" test.
Completion routines that use a custom word point, and that then
recurse into complete_line (e.g., if we make "thread apply" a custom
word point completer, and complete on the command passed as argument),
we stumble on this latent bug:
(gdb) thread apply all pri[TAB]
(gdb) thread apply all priprint
The problem is that there's a spot in complete_line_internal_1 that
rewinds the completion word but does not reflect that change in the
custom word point in the tracker. This patch fixes it.
gdb/ChangeLog:
2019-06-13 Pedro Alves <palves@redhat.com>
* completer.c (complete_line_internal_1): Rewind completion word
point.
(completion_tracker::advance_custom_word_point_by): Change
parameter type to int.
* completer.h (completion_tracker::advance_custom_word_point_by):
Likewise.
Without this fix, if we switch the "print" completer to custom word
point handling, we regress gdb.base/completion.exp like this:
(gdb) p "break1.c FAIL: gdb.base/completion.exp: complete 'p "break1' (timeout)
The problem is that completing an expression that starts with double
quotes, and resolves to a filename, like this:
(gdb) p "break1[TAB]
would change from this, with current master:
(gdb) p "break1.c"|
^^^^^^^^^^|
\- cursor here
to this:
(gdb) p "break1.c |
^^^^^^^^^^|
\- quote replaced by space
The issue is that completer.c:advance_to_completion_word misses
telling the completion tracker to emulate readline's handling of
completing a string when rl_find_completion_word returns a delimiter.
This commit fixes the latent bug.
gdb/ChangeLog:
2019-06-13 Pedro Alves <palves@redhat.com>
* completer.c (advance_to_completion_word): Handle delimiters.
The fpu compare symbols where not including 'sf' in the mnemonic. So
instead of `lf-sfeq` (implying set flag if operands are equal) we were
having `lf-eq`. This patch adds the 'sf'. This helps with making the
generated CGEN documentation consistent and ordered correctly.
cpu/ChangeLog:
yyyy-mm-dd Stafford Horne <shorne@gmail.com>
* or1korfpx.cpu (float-setflag-insn-base): Add 'sf' to symbol.
The 'nd' architectures did not mention what the 'nd' stands for.
Document that these mean 'no brach delay slot'.
cpu/ChangeLog:
yyyy-mm-dd Stafford Horne <shorne@gmail.com>
* or1k.cpu (or64nd, or32nd, or1200nd): Update comment.a
(l-adrp): Improve comment.
Add support for new floating point unordered comparisons. These have been
defined in OpenRISC architecture proposal 7[0] and are now included in the
architecture specification 1.3.
These new instructions provide the ability for floating point comparisons to
detect NaNs.
[0] https://openrisc.io/proposals/lfsf
cpu/ChangeLog:
yyyy-mm-dd Stafford Horne <shorne@gmail.com>
* or1korfpx.cpu (insn-opcode-float-regreg): Add SFUEQ_S, SFUNE_S,
SFUGT_S, SFUGE_S, SFULT_S, SFULE_S, SFUN_S, SFUEQ_D, SFUNE_D, SFUGT_D,
SFUGE_D, SFULT_D, SFULE_D, SFUN_D opcodes.
(float-setflag-insn-base): New pmacro based on float-setflag-insn.
(float-setflag-symantics, float-setflag-unordered-cmp-symantics,
float-setflag-unordered-symantics): New pmacro for instruction
symantics.
(float-setflag-insn): Update to use float-setflag-insn-base.
(float-setflag-unordered-insn): New pmacro for generating instructions.
This patch adds support for OpenRISC 64-bit FPU operations on 32-bit cores by
using register pairs. The functionality has been added to OpenRISC architecture
specification version 1.3 as per architecture proposal 14[0].
For supporting assembly of both 64-bit and 32-bit precision instructions we have
defined CGEN_VALIDATE_INSN_SUPPORTED. This allows cgen to use 64-bit bit
architecture assembly parsing on 64-bit toolchains and 32-bit architecture
assembly parsing on 32-bit toolchains. Without this the assembler has issues
parsing register pairs.
This patch also contains a few fixes to the symantics for existing OpenRISC
single and double precision FPU operations.
[0] https://openrisc.io/proposals/orfpx64a32
cpu/ChangeLog:
yyyy-mm-dd Andrey Bacherov <avbacherov@opencores.org>
Stafford Horne <shorne@gmail.com>
* or1k.cpu (ORFPX64A32-MACHS): New pmacro.
(ORFPX-MACHS): Removed pmacro.
* or1k.opc (or1k_cgen_insn_supported): New function.
(CGEN_VALIDATE_INSN_SUPPORTED): Define macro.
(parse_regpair, print_regpair): New functions.
* or1kcommon.cpu (h-spr, spr-shift, spr-address, h-gpr): Reorder
and add comments.
(h-fdr): Update comment to indicate or64.
(reg-pair-reg-lo, reg-pair-reg-hi): New pmacros for register pairs.
(h-fd32r): New hardware for 64-bit fpu registers.
(h-i64r): New hardware for 64-bit int registers.
* or1korbis.cpu (f-resv-8-1): New field.
* or1korfpx.cpu (rDSF, rASF, rBSF): Update attribute to ORFPX32-MACHS.
(rDDF, rADF, rBDF): Update operand comment to indicate or64.
(f-rdoff-10-1, f-raoff-9-1, f-rboff-8-1): New fields.
(h-roff1): New hardware.
(double-field-and-ops mnemonic): New pmacro to generate operations
rDD32F, rAD32F, rBD32F, rDDI and rADI.
(float-regreg-insn): Update single precision generator to MACH
ORFPX32-MACHS. Add generator for or32 64-bit instructions.
(float-setflag-insn): Update single precision generator to MACH
ORFPX32-MACHS. Fix double instructions from single to double
precision. Add generator for or32 64-bit instructions.
(float-cust-insn cust-num): Update single precision generator to MACH
ORFPX32-MACHS. Add generator for or32 64-bit instructions.
(lf-rem-s, lf-itof-s, lf-ftoi-s, lf-madd-s): Update MACH to
ORFPX32-MACHS.
(lf-rem-d): Fix operation from mod to rem.
(lf-rem-d32, lf-itof-d32, lf-ftoi-d32, lf-madd-d32): New instruction.
(lf-itof-d): Fix operands from single to double.
(lf-ftoi-d): Update operand mode from DI to WI.
PR 24643
* elf32-arm.c (arm_elf_find_function): Fail if the symol table is
absent, or the bfd is not in the ELF formart.
* elfnn-aarch64.c (aarch64_elf_find_function): Likewise.
A name for BLOCK DATA in Fortran is optional. If no name has been
assigned, GDB crashes during read-in of DWARF when BLOCK DATA is
represented via DW_TAG_module. BLOCK DATA is used for one-time
initialization of non-pointer variables in named common blocks.
As of now there is no issue when gfortran is used as DW_TAG_module is
not emitted. However, with Intel ifort the nameless DW_TAG_module is
present and has the following form:
...
<1><dd>: Abbrev Number: 7 (DW_TAG_module)
<de> DW_AT_decl_line : 46
<df> DW_AT_decl_file : 1
<e0> DW_AT_description : (indirect string, offset: 0x110): block
data
<e4> DW_AT_high_pc : 0x402bb7
<ec> DW_AT_low_pc : 0x402bb7
...
The missing name leads to a crash in add_partial_symbol, during length
calculation.
gdb/ChangeLog:
2019-06-11 Bernhard Heckel <bernhard.heckel@intel.com>
* dwarf2read.c (add_partial_symbol): Skip nameless modules.
gdb/testsuite/Changelog:
2019-06-11 Bernhard Heckel <bernhard.heckel@intel.com>
* gdb.fortran/block-data.f: New.
* gdb.fortran/block-data.exp: New.
When "common" becomes a library, linking will cause a symbol clash,
because "xmalloc" and some related symbols are defined in that
library, libiberty, and readline.
To work around this problem, this patch moves the clashing symbols to
a new file, which is then compiled separately for both gdb and
gdbserver.
gdb/ChangeLog
2019-06-11 Tom Tromey <tom@tromey.com>
* common/common-utils.c (xmalloc, xrealloc, xcalloc)
(xmalloc_failed): Move to alloc.c.
* alloc.c: New file.
* Makefile.in (COMMON_SFILES): Add alloc.c.
gdb/gdbserver/ChangeLog
2019-06-11 Tom Tromey <tom@tromey.com>
* Makefile.in (SFILES): Add alloc.c.
(OBS): Add alloc.o.
(IPA_OBJS): Add alloc-ipa.o.
(alloc-ipa.o): New target.
(%.o: ../%.c): New pattern rule.
The debugging code in linux-waitpid.c is one of the few remaining
spots that depends on the gdb/gdbserver difference.
My first thought was that this code is not extremely useful, so this
patch removes this code. (However, if it is actually useful to
someone, we could make it work by introducing a new abstraction.)
gdb/ChangeLog
2019-06-11 Tom Tromey <tom@tromey.com>
* nat/linux-waitpid.c: Don't include server.h.
(linux_debug): Remove.
(my_waitpid): Update.
I noticed recently that some command had a trailing newline in its
"help" output. So, I temporarily hacked cli-decode.c to print
something when a new command was installed that had a trailing newline
in its help message, and wrote this patch, which removes all the ones
I could find this way. (There could still be a few more in *-nat
files.)
Tested on x86-64 Fedora 29.
gdb/ChangeLog
2019-06-11 Tom Tromey <tromey@adacore.com>
* infcall.c (_initialize_infcall): Remove trailing newline from
help.
* user-regs.c (_initialize_user_regs): Remove trailing newline
from help.
* typeprint.c (_initialize_typeprint): Remove trailing newline
from help.
* reverse.c (_initialize_reverse): Remove trailing newlines from
help.
* tracepoint.c (_initialize_tracepoint): Remove trailing newlines
from help.
* language.c (add_set_language_command): Remove trailing newline
from help.
* infcmd.c (_initialize_infcmd): Remove trailing newlines from
help.
* disasm.c (_initialize_disasm): Remove trailing newline from
help.
* top.c (init_main): Remove trailing newline from help.
* interps.c (_initialize_interpreter): Remove trailing newline
from help.
* btrace.c (_initialize_btrace): Remove trailing newlines from
help.
* breakpoint.c (_initialize_breakpoint): Remove trailing newline
from help.
* python/python.c (_initialize_python): Remove trailing newline
from help.
* spu-tdep.c (_initialize_spu_tdep): Remove trailing newlines from
help.
* tui/tui-win.c (_initialize_tui_win): Remove trailing newlines
from help. Reformat some text.
* tui/tui-stack.c (_initialize_tui_stack): Remove trailing newline
from help.
* tui/tui-layout.c (_initialize_tui_layout): Remove trailing
newline from help.
We see this failure with the readnow board:
...
FAIL: gdb.multi/remove-inferiors.exp: load binary
...
When running with board readnow, an extra message "Expanding full symbols" is
emitted after the "Reading symbols" message, and the regexp corresponding to
the FAIL only allows the first message.
Fix this by allowing the extra message in the regexp.
gdb/testsuite/ChangeLog:
2019-06-11 Tom de Vries <tdevries@suse.de>
PR testsuite/24521
* gdb.multi/remove-inferiors.exp: Allow "Expanding full symbols"
message.
Building on Darwin with gcc showed that darwin-nat.c had some
shadowing variable declarations. This removes them.
gdb/ChangeLog
2019-06-11 Tom Tromey <tromey@adacore.com>
* darwin-nat.c (darwin_decode_exception_message)
(darwin_decode_message, darwin_nat_target::kill): Fix shadowing.
When running gdb.dwarf2/nonvar-access.exp with board readnow, we have:
...
FAIL: gdb.dwarf2/nonvar-access.exp: print/x def_implicit_s
...
and 12 more similar failures.
I've tracked this down to the range of main being hardcoded to
[_main, _main+0x10000) in the dwarf assembly:
...
DW_TAG_subprogram {
{name main}
{DW_AT_external 1 flag}
{low_pc [gdb_target_symbol main] DW_FORM_addr}
{high_pc [gdb_target_symbol main]+0x10000 DW_FORM_addr}
} {
...
which overlaps with the .debug_info for the elf-init.c CU (containing
__libc_csu_init and __libc_csu_fini).
Fix this by using function_range to find the actual range of main.
Tested on x86_64-linux.
gdb/testsuite/ChangeLog:
2019-06-11 Tom de Vries <tdevries@suse.de>
PR testsuite/24521
* gdb.dwarf2/nonvar-access.exp: Fix main high_pc.
When extracting an array slice we should give up if the array is
not-allocated or not-associated. For Fortran, at least in gfortran
compiled code, the upper and lower bounds are undefined if the array
is not allocated or not associated, in which case performing checks
against these bounds will result in undefined behaviour.
Better then to throw an error if we try to slice such an array. This
changes the error message that the user will receive in these
cases (if they got an error message before). Previously they may have
gotten "slice out of range" now they'll get "array not allocated" or
"array not associated".
gdb/ChangeLog:
* valops.c (value_slice): Check for not allocated or not
associated values.
gdb/testsuite/ChangeLog:
* gdb.fortran/vla-sizeof.exp: Update expected results.
When compiling gdb with '-lasan -fsanitizer=address' and running tests with:
- export ASAN_OPTIONS="detect_leaks=0:alloc_dealloc_mismatch=0",
- target board cc-with-gdb-index,
- the "[gdb/testsuite] Fix gdb.base/break-probes.exp with native-gdbserver"
commit reverted to avoid running into PR24617,
we get with gdb.arch/amd64-init-x87-values.exp:
...
==31229==ERROR: AddressSanitizer: heap-buffer-overflow on address \
0x62500098c93c at pc 0x000000bcc748 bp 0x7ffe39487660 sp 0x7ffe39487658
READ of size 1 at 0x62500098c93c thread T0
#0 0xbcc747 in cp_find_first_component_aux src/gdb/cp-support.c:999
#1 0xbcc6e9 in cp_find_first_component(char const*) \
src/gdb/cp-support.c:977
#2 0xcc2cf3 in mapped_index_base::build_name_components() \
src/gdb/dwarf2read.c:4499
#3 0xcc3322 in dw2_expand_symtabs_matching_symbol src/gdb/dwarf2read.c:4552
#4 0xcc817f in dw2_expand_symtabs_matching src/gdb/dwarf2read.c:5228
#5 0xfe8f48 in iterate_over_all_matching_symtabs src/gdb/linespec.c:1147
#6 0x1003506 in add_matching_symbols_to_info src/gdb/linespec.c:4413
#7 0xffe21b in find_function_symbols src/gdb/linespec.c:3886
#8 0xffe4a2 in find_linespec_symbols src/gdb/linespec.c:3914
#9 0xfee3ad in linespec_parse_basic src/gdb/linespec.c:1865
#10 0xff5128 in parse_linespec src/gdb/linespec.c:2655
#11 0xff8872 in event_location_to_sals src/gdb/linespec.c:3150
#12 0xff90a8 in decode_line_full(event_location const*, int, \
program_space*, symtab*, int, linespec_result*, \
char const*, char const*) src/gdb/linespec.c:3230
#13 0x9ce449 in parse_breakpoint_sals src/gdb/breakpoint.c:9057
#14 0x9ea022 in create_sals_from_location_default src/gdb/breakpoint.c:13708
#15 0x9e2c1f in bkpt_create_sals_from_location src/gdb/breakpoint.c:12514
#16 0x9cff06 in create_breakpoint(gdbarch*, event_location const*, \
char const*, int, char const*, int, int, bptype, int, \
auto_boolean, breakpoint_ops const*, int, int, int, \
unsigned int) src/gdb/breakpoint.c:9238
#17 0x9d114a in break_command_1 src/gdb/breakpoint.c:9402
#18 0x9d1b60 in break_command(char const*, int) src/gdb/breakpoint.c:9473
#19 0xac96aa in do_const_cfunc src/gdb/cli/cli-decode.c:106
#20 0xad0e5a in cmd_func(cmd_list_element*, char const*, int) \
src/gdb/cli/cli-decode.c:1892
#21 0x15226f6 in execute_command(char const*, int) src/gdb/top.c:630
#22 0xddde37 in command_handler(char const*) src/gdb/event-top.c:586
#23 0xdde7c1 in command_line_handler(std::unique_ptr<char, \
gdb::xfree_deleter<char> >&&) src/gdb/event-top.c:773
#24 0xddc9e8 in gdb_rl_callback_handler src/gdb/event-top.c:217
#25 0x16f2198 in rl_callback_read_char src/readline/callback.c:220
#26 0xddc5a1 in gdb_rl_callback_read_char_wrapper_noexcept \
src/gdb/event-top.c:175
#27 0xddc773 in gdb_rl_callback_read_char_wrapper src/gdb/event-top.c:192
#28 0xddd9f5 in stdin_event_handler(int, void*) src/gdb/event-top.c:514
#29 0xdd7d8f in handle_file_event src/gdb/event-loop.c:731
#30 0xdd8607 in gdb_wait_for_event src/gdb/event-loop.c:857
#31 0xdd629c in gdb_do_one_event() src/gdb/event-loop.c:321
#32 0xdd6344 in start_event_loop() src/gdb/event-loop.c:370
#33 0x10a7715 in captured_command_loop src/gdb/main.c:331
#34 0x10aa548 in captured_main src/gdb/main.c:1173
#35 0x10aa5d8 in gdb_main(captured_main_args*) src/gdb/main.c:1188
#36 0x87bd35 in main src/gdb/gdb.c:32
#37 0x7f16e1434f89 in __libc_start_main (/lib64/libc.so.6+0x20f89)
#38 0x87bb49 in _start (build/gdb/gdb+0x87bb49)
0x62500098c93c is located 0 bytes to the right of 8252-byte region \
[0x62500098a900,0x62500098c93c)
allocated by thread T0 here:
#0 0x7f16e359a600 in malloc (/usr/lib64/libasan.so.5+0xeb600)
#1 0x1742ddf in bfd_malloc src/bfd/libbfd.c:275
#2 0x1738824 in bfd_get_full_section_contents src/bfd/compress.c:253
#3 0xe30044 in gdb_bfd_map_section(bfd_section*, unsigned long*) \
src/gdb/gdb_bfd.c:704
#4 0xcb56bf in dwarf2_read_section(objfile*, dwarf2_section_info*) \
src/gdb/dwarf2read.c:2539
#5 0xd5bcd0 in get_gdb_index_contents_from_section<dwarf2_per_objfile> \
src/gdb/dwarf2read.c:6217
#6 0xd7fc7d in gdb::function_view<gdb::array_view<unsigned char const> \
(...) const src/gdb/common/function-view.h:284
#7 0xd7fddd in gdb::function_view<gdb::array_view<unsigned char const> \
(...) src/gdb/common/function-view.h:278
#8 0xd730cf in gdb::function_view<gdb::array_view<unsigned char const> \
(...) const src/gdb/common/function-view.h:247
#9 0xcbc7ee in dwarf2_read_gdb_index src/gdb/dwarf2read.c:3582
#10 0xcce731 in dwarf2_initialize_objfile(objfile*, dw_index_kind*) \
src/gdb/dwarf2read.c:6297
#11 0xdb88c4 in elf_symfile_read src/gdb/elfread.c:1256
#12 0x141262a in read_symbols src/gdb/symfile.c:798
#13 0x14140a7 in syms_from_objfile_1 src/gdb/symfile.c:1000
#14 0x1414393 in syms_from_objfile src/gdb/symfile.c:1017
#15 0x1414fb7 in symbol_file_add_with_addrs src/gdb/symfile.c:1124
#16 0x14159b7 in symbol_file_add_from_bfd(bfd*, char const*, \
enum_flags<symfile_add_flag>, std::vector<other_sections, \
std::allocator<other_sections> >*, \
enum_flags<objfile_flag>, objfile*) src/gdb/symfile.c:1203
#17 0x1415b6c in symbol_file_add(char const*,
enum_flags<symfile_add_flag>, std::vector<other_sections, \
std::allocator<other_sections> >*, \
enum_flags<objfile_flag>) src/gdb/symfile.c:1216
#18 0x1415f2f in symbol_file_add_main_1 src/gdb/symfile.c:1240
#19 0x1418599 in symbol_file_command(char const*, int) \
src/gdb/symfile.c:1675
#20 0xde2fa6 in file_command src/gdb/exec.c:433
#21 0xac96aa in do_const_cfunc src/gdb/cli/cli-decode.c:106
#22 0xad0e5a in cmd_func(cmd_list_element*, char const*, int) \
src/gdb/cli/cli-decode.c:1892
#23 0x15226f6 in execute_command(char const*, int) src/gdb/top.c:630
#24 0xddde37 in command_handler(char const*) src/gdb/event-top.c:586
#25 0xdde7c1 in command_line_handler(std::unique_ptr<char, \
gdb::xfree_deleter<char> >&&) src/gdb/event-top.c:773
#26 0xddc9e8 in gdb_rl_callback_handler src/gdb/event-top.c:217
#27 0x16f2198 in rl_callback_read_char src/readline/callback.c:220
#28 0xddc5a1 in gdb_rl_callback_read_char_wrapper_noexcept \
src/gdb/event-top.c:175
#29 0xddc773 in gdb_rl_callback_read_char_wrapper src/gdb/event-top.c:192
SUMMARY: AddressSanitizer: heap-buffer-overflow src/gdb/cp-support.c:999 in \
cp_find_first_component_aux
Shadow bytes around the buggy address:
0x0c4a801298d0: 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00
0x0c4a801298e0: 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00
0x0c4a801298f0: 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00
0x0c4a80129900: 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00
0x0c4a80129910: 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00
=>0x0c4a80129920: 00 00 00 00 00 00 00[04]fa fa fa fa fa fa fa fa
0x0c4a80129930: fa fa fa fa fa fa fa fa fa fa fa fa fa fa fa fa
0x0c4a80129940: fa fa fa fa fa fa fa fa fa fa fa fa fa fa fa fa
0x0c4a80129950: fa fa fa fa fa fa fa fa fa fa fa fa fa fa fa fa
0x0c4a80129960: fa fa fa fa fa fa fa fa fa fa fa fa fa fa fa fa
0x0c4a80129970: fa fa fa fa fa fa fa fa fa fa fa fa fa fa fa fa
Shadow byte legend (one shadow byte represents 8 application bytes):
Addressable: 00
Partially addressable: 01 02 03 04 05 06 07
Heap left redzone: fa
Freed heap region: fd
Stack left redzone: f1
Stack mid redzone: f2
Stack right redzone: f3
Stack after return: f5
Stack use after scope: f8
Global redzone: f9
Global init order: f6
Poisoned by user: f7
Container overflow: fc
Array cookie: ac
Intra object redzone: bb
ASan internal: fe
Left alloca redzone: ca
Right alloca redzone: cb
==31229==ABORTING
...
The problem happens as follows.
The executable amd64-init-x87-values gets an index (due to target board
cc-with-gdb-index), which looks as follows:
...
Hex dump of section '.gdb_index':
0x00000000 08000000 18000000 28000000 28000000 ........(...(...
0x00000010 3c000000 3c200000 00000000 00000000 <...< ..........
0x00000020 2e000000 00000000 d4004000 00000000 ..........@.....
0x00000030 db004000 00000000 00000000 00000000 ..@.............
0x00000040 00000000 00000000 00000000 00000000 ................
0x00000050 00000000 00000000 00000000 00000000 ................
... more zeroes ...
0x00002010 00000000 00000000 00000000 00000000 ................
0x00002020 00000000 00000000 00000000 00000000 ................
0x00002030 00000000 00000000 00000000 ............
...
The structure of this index is:
...
header : [0x0, 0x18) : size 0x18
culist : [0x18 ,0x28) : size 0x10
typesculist : [0x28, 0x28) : size 0x0
adress area : [0x28, 0x3c) : size 0x14
symbol table : [0x3c, 0x203c) : size 0x2000
constant pool: [0x203c, 0x203c): size 0x0
EOF : 0x203c
...
Note that the symbol table consists entirely of empty slots (where an empty
slot is a pair of 32-bit zeroes), and that the constant pool is empty.
The problem happens here in mapped_index_base::build_name_components:
...
auto count = this->symbol_name_count ();
for (offset_type idx = 0; idx < count; idx++)
{
if (this->symbol_name_slot_invalid (idx))
continue;
const char *name = this->symbol_name_at (idx);
...
when accessing the slot at idx == 0 in the symbol table,
symbol_name_slot_invalid returns false so we calculate name, which is
calculated using 'constant_pool + symbol_table[idx].name', which means we get
name == constant_pool. And given that the constant pool is empty, name now
points past the memory allocated for the index, and when we access name[0] for
the first time in cp_find_first_component_aux, we run into the
heap-buffer-overflow.
Fix this by fixing the definition of symbol_name_slot_invalid:
...
- return bucket.name == 0 && bucket.vec;
+ return bucket.name == 0 && bucket.vec == 0;
...
Tested on x86_64-linux.
gdb/ChangeLog:
2019-06-10 Tom de Vries <tdevries@suse.de>
PR gdb/24618
* dwarf2read.c (struct mapped_index::symbol_name_slot_invalid): Make
sure an empty slot (defined by a 32-bit zero pair) is recognized as
invalid.
When compiling gdb with '-lasan -fsanitizer=address' and running tests with
'export ASAN_OPTIONS="detect_leaks=0:alloc_dealloc_mismatch=0"', I run into:
...
ERROR: GDB process no longer exists
UNRESOLVED: gdb.linespec/cpls-abi-tag.exp: \
test_abi_tag: completion: at tag: tab complete "b test_abi_tag_function[abi:"
...
In more detail:
...
==3637==ERROR: AddressSanitizer: dynamic-stack-buffer-overflow on address \
0x7fff5952bbdd at pc 0x000000fe5c57 bp 0x7fff5952af30 sp 0x7fff5952af28
READ of size 1 at 0x7fff5952bbdd thread T0
#0 0xfe5c56 in linespec_lexer_lex_string src/gdb/linespec.c:727
#1 0xfe7473 in linespec_lexer_lex_one src/gdb/linespec.c:946
#2 0xfe799d in linespec_lexer_consume_token src/gdb/linespec.c:982
#3 0xff446d in parse_linespec src/gdb/linespec.c:2564
#4 0xff78be in linespec_complete(completion_tracker&, char const*, \
symbol_name_match_type) src/gdb/linespec.c:2961
#5 0xb9299c in complete_address_and_linespec_locations \
src/gdb/completer.c:573
#6 0xb93e90 in location_completer(cmd_list_element*, completion_tracker&, \
char const*, char const*) src/gdb/completer.c:919
#7 0xb940c5 in location_completer_handle_brkchars src/gdb/completer.c:956
#8 0xb957ec in complete_line_internal_normal_command \
src/gdb/completer.c:1208
#9 0xb96507 in complete_line_internal_1 src/gdb/completer.c:1430
#10 0xb965c2 in complete_line_internal src/gdb/completer.c:1449
#11 0xb98630 in gdb_completion_word_break_characters_throw \
src/gdb/completer.c:1862
#12 0xb98838 in gdb_completion_word_break_characters() \
src/gdb/completer.c:1897
#13 0x16c6362 in _rl_find_completion_word src/readline/complete.c:943
#14 0x16ca8d0 in rl_complete_internal src/readline/complete.c:1843
#15 0x16c460c in rl_complete src/readline/complete.c:408
#16 0x16b3368 in _rl_dispatch_subseq src/readline/readline.c:774
#17 0x16b3092 in _rl_dispatch src/readline/readline.c:724
#18 0x16b2939 in readline_internal_char src/readline/readline.c:552
#19 0x16f1fb0 in rl_callback_read_char src/readline/callback.c:201
#20 0xddc5a1 in gdb_rl_callback_read_char_wrapper_noexcept \
src/gdb/event-top.c:175
#21 0xddc773 in gdb_rl_callback_read_char_wrapper src/gdb/event-top.c:192
#22 0xddd9f5 in stdin_event_handler(int, void*) src/gdb/event-top.c:514
#23 0xdd7d8f in handle_file_event src/gdb/event-loop.c:731
#24 0xdd8607 in gdb_wait_for_event src/gdb/event-loop.c:857
#25 0xdd629c in gdb_do_one_event() src/gdb/event-loop.c:321
#26 0xdd6344 in start_event_loop() src/gdb/event-loop.c:370
#27 0x10a7715 in captured_command_loop src/gdb/main.c:331
#28 0x10aa548 in captured_main src/gdb/main.c:1173
#29 0x10aa5d8 in gdb_main(captured_main_args*) src/gdb/main.c:1188
#30 0x87bd35 in main src/gdb/gdb.c:32
#31 0x7fb0364c6f89 in __libc_start_main (/lib64/libc.so.6+0x20f89)
#32 0x87bb49 in _start (build/gdb/gdb+0x87bb49)
Address 0x7fff5952bbdd is located in stack of thread T0 at offset 557 in frame
#0 0xb93702 in location_completer(cmd_list_element*, completion_tracker&, \
char const*, char const*) src/gdb/completer.c:831
This frame has 4 object(s):
[32, 40) 'copy'
[96, 104) 'location'
[160, 168) 'text'
[224, 256) 'completion_info' <== Memory access at offset 557 overflows \
this variable
HINT: this may be a false positive if your program uses some custom stack \
unwind mechanism or swapcontext
(longjmp and C++ exceptions *are* supported)
SUMMARY: AddressSanitizer: dynamic-stack-buffer-overflow \
src/gdb/linespec.c:727 in linespec_lexer_lex_string
Shadow bytes around the buggy address:
0x10006b29d720: 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00
0x10006b29d730: 00 00 00 00 00 00 f1 f1 f1 f1 00 f2 f2 f2 f2 f2
0x10006b29d740: f2 f2 00 f2 f2 f2 f2 f2 f2 f2 00 f2 f2 f2 f2 f2
0x10006b29d750: f2 f2 00 00 00 00 f3 f3 f3 f3 00 00 00 00 00 00
0x10006b29d760: 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00
=>0x10006b29d770: 00 00 00 00 ca ca ca ca 00 00 00[05]cb cb cb cb
0x10006b29d780: 00 00 00 00 00 00 00 00 00 00 00 00 f1 f1 f1 f1
0x10006b29d790: 00 f2 f2 f2 f2 f2 f2 f2 00 f2 f2 f2 f3 f3 f3 f3
0x10006b29d7a0: 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00
0x10006b29d7b0: 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00
0x10006b29d7c0: 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00
Shadow byte legend (one shadow byte represents 8 application bytes):
Addressable: 00
Partially addressable: 01 02 03 04 05 06 07
Heap left redzone: fa
Freed heap region: fd
Stack left redzone: f1
Stack mid redzone: f2
Stack right redzone: f3
Stack after return: f5
Stack use after scope: f8
Global redzone: f9
Global init order: f6
Poisoned by user: f7
Container overflow: fc
Array cookie: ac
Intra object redzone: bb
ASan internal: fe
Left alloca redzone: ca
Right alloca redzone: cb
==3637==ABORTING
...
The problem happens in linespec_lexer_lex_string when lexing
"b test_abi_tag_function[abi:\0" (using a notation where we make the implicit
terminating \0 explicit).
We arrrive here with (PARSER_STREAM (parser)) == ":\0":
...
/* Do not tokenize ABI tags such as "[abi:cxx11]". */
else if (PARSER_STREAM (parser) - start > 4
&& startswith (PARSER_STREAM (parser) - 4, "[abi"))
++(PARSER_STREAM (parser));
...
and consume ':', after which we end up here and consume '\0':
...
/* Advance the stream. */
++(PARSER_STREAM (parser));
...
after which (PARSER_STREAM (parser)) points past the end of the string.
Fix this by removing the first "++(PARSER_STREAM (parser))", and add an assert
to the second one to detect moving past the end-of-string.
Build and tested on x86_64-linux.
gdb/ChangeLog:
2019-06-10 Tom de Vries <tdevries@suse.de>
PR gdb/24611
* linespec.c (linespec_lexer_lex_string): Remove incorrect
"++(PARSER_STREAM (parser))" for "[abi"-prefixed colon. Add assert.
The commit "[gdb/symtab] Fix language of duplicate static minimal symbol"
introduces a performance regression, when loading a cc1 executable build with
-O0 -g and gcc 7.4.0. The performance regression, measured in 'real' time is
about 175%.
The slower execution comes from the fact that the fix in symbol_set_names
makes the call to symbol_find_demangled_name unconditional.
Fix this by reverting the commit, and redoing the fix as follows.
Recapturing the original problem, the first time symbol_set_names is called
with gsymbol.language == lang_auto and linkage_name == "_ZL3foov", the name is
not present in the per_bfd->demangled_names_hash hash table, so
symbol_find_demangled_name is called to demangle the name, after which the
mangled/demangled pair is added to the hashtable. The call to
symbol_find_demangled_name also sets gsymbol.language to lang_cplus.
The second time symbol_set_names is called with gsymbol.language == lang_auto
and linkage_name == "_ZL3foov", the name is present in the hash table, so the
demangled name from the hash table is used. However, the language of the
symbol remains lang_auto.
Fix this by adding a field language in struct demangled_name_entry, and using
the field in symbol_set_names to set the language of gsymbol, if necessary.
Tested on x86_64-linux.
gdb/ChangeLog:
2019-06-10 Tom de Vries <tdevries@suse.de>
PR symtab/24545
* symtab.c (struct demangled_name_entry): Add language field.
(symbol_set_names): Revert "[gdb/symtab] Fix language of duplicate
static minimal symbol". Set and use language field.
I noticed that the "catch assert" help text erroneously claimed to
accept an argument, and while fixing this I went ahead and added
"Usage" text and made other minor updates to the commands in
ada-lang.c.
Tested on x86-64 Fedora 29.
gdb/ChangeLog
2019-06-10 Tom Tromey <tromey@adacore.com>
* ada-lang.c (_initialize_ada_language): Update help text.
ARI pointed out that a recent patch introduced a call to "warning"
with a string that ended in a newline:
https://sourceware.org/ml/gdb-patches/2019-06/msg00000.html
This is generally forbidden, I believe, because warning adds its own
newline.
This patch removes all of the trailing newlines I was able to find. I
searched for 'warning (.*\\n"' and then fixed the ones where the
newline appeared at the end of the string (some had internal
newlines).
Tested on x86-64 Fedora 29.
gdb/ChangeLog
2019-06-10 Tom Tromey <tromey@adacore.com>
* m32c-tdep.c (m32c_m16c_address_to_pointer): Don't end warning
with a newline.
* guile/guile.c (handle_boot_error): Don't end warning with a
newline.
* cli/cli-cmds.c (exit_status_set_internal_vars): Don't end
warning with a newline.
* s12z-tdep.c (s12z_skip_prologue): Don't end warning with a
newline.
(s12z_frame_cache): Likewise.
* dwarf-index-cache.c (index_cache::store): Don't end warning with
a newline.
* solib-svr4.c (disable_probes_interface): Don't end warning with
a newline.
* nat/fork-inferior.c (fork_inferior): Don't end warning with a
newline.
* python/python.c (do_finish_initialization): Don't end warning
with a newline.
gdb/gdbserver/ChangeLog
2019-06-10 Tom Tromey <tromey@adacore.com>
* remote-utils.c (look_up_one_symbol, relocate_instruction): Don't
end warning with a newline.
* linux-s390-low.c (s390_get_wordsize): Don't end warning with a
newline.
* thread-db.c (attach_thread): Don't end warning with a newline.
(thread_db_notice_clone): Likewise.
* tracepoint.c (gdb_agent_helper_thread): Don't end warning with a
newline.
* linux-x86-low.c (x86_get_min_fast_tracepoint_insn_len): Don't
end warning with a newline.
PR 24650
* elf.c (elfcore_make_auxv_note_section): New function.
(elfcore_grok_note): Use it.
(elfcore_grok_freebsd_note): Likewise.
(elfcore_grok_openbsd_note): Likewise.
(elfcore_grok_netbsd_note): Likewise. Plus add support for
NT_NETBSDCORE_AUXV notes.
A few spots in py-breakpoint.c acquire the GIL manually. However,
because these spots generate events, and because events are expected
to be arbitrary gdb-flavored Python code, it's important to use
gdbpy_enter instead, in order to ensure that the other gdb-related
Python globals are set correctly.
This patch makes this change. Tested on x86-64 Fedora 29.
gdb/ChangeLog
2019-06-10 Tom Tromey <tom@tromey.com>
* python/py-breakpoint.c (gdbpy_breakpoint_created)
(gdbpy_breakpoint_deleted, gdbpy_breakpoint_modified): Use
gdbpy_enter.
I noticed that elfread.c always allocates a dbx_symfile_info, even
though this is only ever needed in the unusual case of reading stabs
in ELF.
This patch moves the allocation into dbxread.c, and applies the same
treatment to similar code in coffread.c.
Regression tested on x86-64 Fedora 29.
gdb/ChangeLog
2019-06-10 Tom Tromey <tromey@adacore.com>
* elfread.c (elf_read_minimal_symbols): Don't set the dbx objfile
data.
(elf_new_init): Don't call stabsread_new_init.
* dbxread.c (coffstab_build_psymtabs): Set dbx objfile data.
(elfstab_build_psymtabs): Likewise. Call stabsread_new_init.
* coffread.c (coff_symfile_init): Don't set the dbx objfile data.
