Nowadays, GDB propagates C++ exceptions across readline using
setjmp/longjmp 89525768cd ("Propagate GDB/C++ exceptions across
readline using sj/lj-based TRY/CATCH") because DWARF-based unwinding
can't cross C functions compiled without -fexceptions (see details
from the commit above).
Unfortunately, toolchains that use SjLj-based C++ exceptions got
broken with that fix, because _Unwind_SjLj_Unregister, which is put at
the exit of a function, is not executed due to the longjmp added by
that commit.
(gdb) [New Thread 2936.0xb80]
kill
Thread 1 received signal SIGSEGV, Segmentation fault.
0x03ff662b in ?? ()
top?bt 15
#0 0x03ff662b in ?? ()
#1 0x00526b92 in stdin_event_handler (error=0, client_data=0x172ed8)
at ../../binutils-gdb/gdb/event-top.c:555
#2 0x00525a94 in handle_file_event (ready_mask=<optimized out>,
file_ptr=0x3ff5cb8) at ../../binutils-gdb/gdb/event-loop.c:733
#3 gdb_wait_for_event (block=block@entry=1)
at ../../binutils-gdb/gdb/event-loop.c:884
#4 0x00525bfb in gdb_do_one_event ()
at ../../binutils-gdb/gdb/event-loop.c:347
#5 0x00525ce5 in start_event_loop ()
at ../../binutils-gdb/gdb/event-loop.c:371
#6 0x0051fada in captured_command_loop (data=0x0)
at ../../binutils-gdb/gdb/main.c:324
#7 0x0051cf5d in catch_errors (
func=func@entry=0x51fab0 <captured_command_loop(void*)>,
func_args=func_args@entry=0x0,
errstring=errstring@entry=0x7922bf <VEC_interp_factory_p_quick_push(VEC_inte rp_factory_p*, interp_factory*, char const*, unsigned int)::__PRETTY_FUNCTION__+351> "", mask=mask@entry=RETURN_MASK_ALL)
at ../../binutils-gdb/gdb/exceptions.c:236
#8 0x00520f0c in captured_main (data=0x328feb4)
at ../../binutils-gdb/gdb/main.c:1149
#9 gdb_main (args=args@entry=0x328feb4) at ../../binutils-gdb/gdb/main.c:1159
#10 0x0071e400 in main (argc=1, argv=0x171220)
at ../../binutils-gdb/gdb/gdb.c:32
Fix this by making the functions involved in setjmp/longjmp as
noexcept, so that the compiler knows it doesn't need to emit the
_Unwind_SjLj_Register / _Unwind_SjLj_Unregister calls for C++
exceptions.
Tested on x86_64 Fedora 23 with:
- GCC 5.3.1 w/ DWARF-based exceptions.
- GCC 7 built with --enable-sjlj-exceptions.
gdb/ChangeLog:
2016-12-20 Pedro Alves <palves@redhat.com>
Yao Qi <yao.qi@linaro.org>
PR gdb/20977
* event-top.c (gdb_rl_callback_read_char_wrapper_noexcept): New
noexcept function, factored out from ...
(gdb_rl_callback_read_char_wrapper): ... this.
(gdb_rl_callback_handler): Mark noexcept.
Since GDB has switched to C++ but the file names are still .c emacs does
not load the proper mode when opening files in the gdb directory.
This patch fixes that by enabling c++ mode.
This patch also fixes indentation tweaks as discussed in this thread:
https://sourceware.org/ml/gdb-patches/2016-12/msg00074.html
Indent with gdb-code-style.el included and the .dir-locals.el is as such:
namespace TestNameSpace {
class test
{
public:
test test() {}
int m_a;
};
struct teststruct
{
int a;
}
}
gdb/ChangeLog:
* .dir-locals.el: Set c++ mode for the directory and set indent
properly.
* gdb-code-style.el: Set c-set-offset 'innamespace as a safe value
to be used in .dir-locals.el.
Due to changes introduced by
commit 4d01a485d2
('struct expression *' -> gdb::unique_xmalloc_ptr<expression>)
compilation is broken on Darwin.
../gdb/darwin-nat-info.c:733:8: error: assigning to 'struct expression *'
from incompatible type
'expression_up' (aka 'std::__1::unique_ptr<expression, gdb::xfree_deleter<expression> >')
expr = parse_expression (exp);
Beside compilation, memory leak was solved as 'make_clean_up' was not called in previous
version.
2016-12-16 Bernhard Heckel <bernhard.heckel@intel.com>
gdb/Changelog:
* darwin-nat-info.c (info_mach_region_command): Use expression_up.
cris_delayed_get_disassembler has an assert that exec_bfd can't be
NULL, but this assert can be triggered like this,
(gdb) set architecture cris
The target architecture is assumed to be cris
(gdb) disassemble 0x0,+4
Dump of assembler code from 0x0 to 0x4:
0x00000000:
../../binutils-gdb/gdb/cris-tdep.c:3798: internal-error: int cris_delayed_get_disassembler(bfd_vma, disassemble_info*): Assertion `exec_bfd != NULL' failed.
A problem internal to GDB has been detected,
further debugging may prove unreliable.
however, cris_get_disassembler does have code to handle the case that
bfd is NULL,
/* If there's no bfd in sight, we return what is valid as input in all
contexts if fed back to the assembler: disassembly *with* register
prefix. Unfortunately this will be totally wrong for v32. */
if (abfd == NULL)
return print_insn_cris_with_register_prefix;
This patch is to remove this assert.
gdb:
2016-12-12 Yao Qi <yao.qi@linaro.org>
PR tdep/20955
* cris-tdep.c (cris_delayed_get_disassembler): Remove the
assert.
This patch change aarch prologue analyzer using code cache, in order
to improve the performance of remote debugging.
gdb.perf/skip-prologue.exp (measured by wall-time) is improved when
the program is compiled without debug information.
Original Patched Original Patched
without dbg without dbg with dbg with dbg
/ 11.1635239124 9.99472999573 9.65339517593 9.66648793221
-fstack-protector-all 11.2560930252 9.338118 9.63896489143 9.59474396706
gdb:
2016-12-9 Yao Qi <yao.qi@linaro.org>
* aarch64-tdep.c (instruction_reader::read): Call
read_code_unsigned_integer instead of
read_memory_unsigned_integer.
This patch change arm prologue analyzer using code cache, in order
to improve the performance of remote debugging.
gdb.perf/skip-prologue.exp (measured by wall-time) is improved a lot,
Original Patched Original Patched
without dbg without dbg with dbg with dbg
-marm 14.166741848 9.32852292061 11.4908499718 9.16302204132
-marm 14.6705040932 9.34849786758 18.2788009644 9.14823913574
\-fstack-protector-all
-mthumb 34.4391930103 10.6062178612 13.7886838913 10.3094120026
-mthumb
\-fstack-protector-all 34.9310460091 10.6413481236 25.3875930309 10.6294929981
gdb:
2016-12-09 Yao Qi <yao.qi@linaro.org>
* arm-tdep.c (skip_prologue_function): Call
read_code_unsigned_integer instead of
read_memory_unsigned_integer.
(thumb_analyze_prologue): Likewise.
(arm_analyze_load_stack_chk_guard): Likewise.
(arm_skip_stack_protector): Likewise.
(arm_analyze_prologue):Likewise.
(extend_buffer_earlier): Call target_read_code instead
of target_read_memory.
(arm_adjust_breakpoint_address): Likewise.
For "info threads", we currently run into:
$ gdb/gdb -q -nw -nx --batch -ex start -ex info\ threads bfd/doc/chew
Temporary breakpoint 1 at 0x80486e0: file ../../../W._C._Handy/bfd/doc/chew.c, line 1535.
[New Thread 10656.5]
Thread 4 hit Temporary breakpoint 1, main (ac=1, av=0x102cd84) at ../../../W._C._Handy/bfd/doc/chew.c:1535
1535 {
Id Target Id Frame
1 bogus thread id 1 Can't fetch registers from thread bogus thread id 1: No such thread
Before commit e8032dde10,
gdb/thread.c:update_thread_list used to call prune_threads, after that change
it doesn't anymore, and we don't implement the to_update_thread_list target
method where the prune_threads call got moved. For now, apply a fix, related
to commit c82f56d9d7 "Hurd: Adjust to
startup-with-shell changes", which restores the previous behavior:
Id Target Id Frame
* 4 Thread 10688.4 main (ac=1, av=0x102cd84) at ../../../W._C._Handy/bfd/doc/chew.c:1535
5 Thread 10688.5 0x0106096c in ?? () from /lib/i386-gnu/libc.so.0.3
Not perfect, but at least better.
gdb/
* gnu-nat.c (gnu_create_inferior): After startup_inferior, call
prune_threads.
Follow-up to commit 14f6890677.
global_thread_id_to_ptid expects global thread numbers, which are nowadays only
used in MI, never presented to the user in the CLI. Since this is a CLI
command, it should accept the inferior-qualified format instead.
gdb/
* gnu-nat.c (set_sig_thread_cmd): Use parse_thread_id instead of
global_thread_id_to_ptid.
..., so handle these in "C" mode still:
gdb/
* config/i386/i386gnu.mh (%_S.o %_U.o): Add "-x c" to
"COMPILE.post".
* gnu-nat.c: #include Mach/Hurd headers before all others. Wrap
Mach/Hurd headers and MIG stubs' prototypes in 'extern "C"'.
* i386-gnu-nat.c: Likewise.
GNU/Hurd uses its own "typedef enum __error_t_codes error_t;"
([glibc]/sysdeps/mach/hurd/bits/errno.h), contrary to the default
"typedef int error_t;" ([glibc]/stdlib/errno.h).
The Mach/Hurd RPCs return kern_return_t values, for which, upon assigning them
to an error_t variable, GCC in C++ mode tells us "error: invalid conversion
from 'kern_return_t {aka int}' to 'error_t {aka __error_t_codes}'". Instead of
casting all these RPC return values to "error_t", just use "kern_return_t"
variables:
gdb/
* gnu-nat.c (proc_get_exception_port, proc_set_exception_port)
(INF_RESUME_MSGPORT_RPC, proc_get_state, _proc_get_exc_port)
(proc_steal_exc_port, proc_restore_exc_port, make_proc)
(inf_startup, inf_set_pid, inf_validate_procinfo)
(inf_validate_task_sc, inf_set_traced, inf_validate_procs)
(inf_signal, inf_continue, gnu_wait, S_exception_raise_request)
(do_mach_notify_dead_name, S_proc_wait_reply)
(S_msg_sig_post_untraced_reply, S_msg_sig_post_reply)
(port_msgs_queued, gnu_read_inferior, gnu_write_inferior)
(gnu_find_memory_regions, steal_exc_port, thread_takeover_sc_cmd)
(flush_inferior_icache): Instead of "error_t" use "kern_return_t".
* i386-gnu-nat.c (fetch_fpregs, store_fpregs, i386_gnu_dr_get)
(i386_gnu_dr_set): Likewise.
... by a bit of code refactoring:
gdb/
* gnu-nat.c (set_task_pause_cmd, set_signals_cmd)
(set_exceptions_cmd): Add variants taking an "int arg" instead of
a "char *". Make the "char *" variants use the former.
(set_noninvasive_cmd): Also use the "int arg" variants.
C++ doesn't do implicit type conversions from "void *", so we have to...
gdb/
* i386-gnu-nat.c (i386_gnu_dr_set_control_one)
(i386_gnu_dr_set_addr_one): Explicitly cast "void *".
In the case where we switch to a non-running inferior, we do a
"find_inferior_id (num)", although we did the same call right before.
gdb/ChangeLog:
* inferior.c (inferior_command): Remove duplicate
find_inferior_id call.
This patch adds asserts where the value's lval must be lval_register.
This triggers an error in frame_register_unwind because VALUE_REGNUM
is used but value's lval is not lval_register.
This also reveals a design issue in frame_register_unwind, that is
arguments addrp and realnump are mutually exclusive, we either use
addrp (for lval_memory), or use realnump (for lval_register). This
can be done in a separate patch.
gdb:
2016-12-06 Yao Qi <yao.qi@linaro.org>
* frame.c (frame_register_unwind): Set *realnump if *lvalp is
lval_register.
* value.c (deprecated_value_next_frame_id_hack): Assert
value->lval is lval_register.
(deprecated_value_regnum_hack): Likewise.
New in v2:
- A few adjustments / simplifications were possible now that we
require C++11:
. Use std::unique_ptr to make the user_args_stack std::vector own
its elements:
static std::vector<std::unique_ptr<user_args>> user_args_stack;
. use vector::emplace_back to construct elements directly in the
corresponding vectors.
. use std::to_string instead of adding a gdb::to_string
replacement.
- Now includes a test.
Docs/NEWS are unchanged from v1 and have already been approved.
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
I recently wrote a user-defined command that could benefit from
supporting an unlimited number of arguments:
http://palves.net/list-active-signal-handlers-with-gdb/
E.g., 'info signal-dispositions 1 2 3 4 5 6 7 8 9 10 11'
However, we currently only support up to 10 arguments passed to
user-defined commands ($arg0..$arg9).
I can't find a good reason for that, other than "old code with hard
coded limits". This patch removes that limit and modernizes the code
along the way:
- Makes the user_args struct a real C++ class that uses std::vector
for storage.
- Removes the "next" pointer from within user_args and uses a
std::vector to maintain a stack instead.
- Adds a new RAII-based scoped_user_args_level class to help
push/pop user args in the stack instead of using a cleanup.
gdb/ChangeLog:
2016-12-02 Pedro Alves <palves@redhat.com>
* NEWS: Mention that user commands now accept an unlimited number
of arguments.
* cli/cli-script.c: Include <vector>.
(struct string_view): New type.
(MAXUSERARGS): Delete.
(struct user_args): Now a C++ class.
(user_args_stack): New.
(struct scoped_user_args_level): New type.
(execute_user_command): Use scoped_user_args_level.
(arg_cleanup): Delete.
(setup_user_args): Deleted, and refactored as ...
(user_args::user_args): ... this new constructor. Limit of number
of arguments removed.
(insert_user_defined_cmd_args): Defer to user_args_stack.
(user_args::insert_args): New, bits based on old
insert_user_defined_cmd_args with limit of number of arguments
eliminated.
gdb/doc/ChangeLog:
2016-12-02 Pedro Alves <palves@redhat.com>
* gdb.texinfo (User-defined Commands): Limit on number of
arguments passed to user-defined commands removed; update.
gdb/testsuite/ChangeLog:
2016-12-02 Pedro Alves <palves@redhat.com>
* gdb.base/commands.exp (user_defined_command_manyargs_test): New
procedure.
(top level): Call it.
It'd be handy to be able to iterate over command arguments in
user-defined commands, in order to support optional arguments
($arg0..$argN).
I thought I could make it work with "eval", but alas, it doesn't work
currently. E.g., with:
define test
set $i = 0
while $i < $argc
eval "print $arg%d", $i
set $i = $i + 1
end
end
we get:
(gdb) test 1
$1 = void
(gdb) test 1 2 3
$2 = void
$3 = void
$4 = void
(gdb)
The problem is that "eval" doesn't do user-defined command arguments
substitution after expanding its own argument. This patch fixes that,
which makes the example above work:
(gdb) test 1
$1 = 1
(gdb) test 1 2 3
$2 = 1
$3 = 2
$4 = 3
(gdb)
New test included, similar the above, but also exercises expanding
$argc.
I think this is likely to simplify many scripts out there, so I'm
adding an example to the manual and mentioning it in NEWS as well.
gdb/ChangeLog:
2016-12-02 Pedro Alves <palves@redhat.com>
PR cli/20559
* NEWS: Mention "eval" expands user-defined command arguments.
* cli/cli-script.c (execute_control_command): Adjust to rename.
(insert_args): Rename to ...
(insert_user_defined_cmd_args): ... this, and make extern.
* cli/cli-script.h (insert_user_defined_cmd_args): New
declaration.
* printcmd.c: Include "cli/cli-script.h".
(eval_command): Call insert_user_defined_cmd_args.
gdb/doc/ChangeLog:
2016-12-02 Pedro Alves <palves@redhat.com>
PR cli/20559
* gdb.texinfo (Define): Add example of using "eval" to process a
variable number of arguments.
(Output) <eval>: Add anchor.
gdb/testsuite/ChangeLog:
2016-12-02 Pedro Alves <palves@redhat.com>
PR cli/20559
* gdb.base/commands.exp (user_defined_command_args_eval): New
procedure.
(top level): Call it.
This patch adds support for DW_AT_main_subprogram.
This is PR symtab/16264.
DW_AT_main_subprogram is used to mark a program's entry point. GCC
can emit this, and I hope to change the Rust compiler to emit it as
well.
GDB already supports an older, pre-DWARF 4 convention adopted by
FORTRAN compilers, namely to emit DW_AT_calling_convention for the
"main" function. However, I think this support in GDB had a small
bug, in that it seems to rely on the DW_AT_name being read before
DW_AT_calling_convention. This patch fixes this as well.
Built and regtested on x86-64 Fedora 24 and the buildbot. New test
case included.
2016-12-02 Tom Tromey <tom@tromey.com>
PR symtab/16264:
* dwarf2read.c (struct partial_die_info) <main_subprogram>: New
member.
(add_partial_symbol): Call set_objfile_main_name.
(read_partial_die): Handle DW_AT_main_subprogram.
<DW_AT_calling_convention>: don't call set_objfile_main_name, but
set main_subprogram flag.
2016-12-02 Tom Tromey <tom@tromey.com>
* gdb.dwarf2/main-subprogram.c: New file.
* gdb.dwarf2/main-subprogram.exp: New file.
This patch renames a few trace-related functions, so that they adhere to
the de facto standard of naming command entry point functions
<command>_command. I like the ease of looking up a command entry point
if they all follow that rule.
An enum label "tstop_command" conflicts with a new function name, so I
renamed this one trace_stop_command.
In v2:
- Rename functions of the trace_find family, as well as
trace_dump_command.
gdb/ChangeLog:
* tracefile-tfile.c (tfile_write_status): Adjust to renames.
* tracefile.c (trace_save_command): Rename to...
(tsave_command): ...this.
(_initialize_tracefile): Adjust to renames.
* tracepoint.c (trace_actions_command): Rename to...
(actions_command): ...this.
(trace_start_command): Rename to...
(tstart_command): ...this, and adjust to renames..
(trace_stop_command): Rename to...
(tstop_command): ...this.
(trace_status_command): Rename to...
(tstatus_command): ...this, and adjust to renames.
(trace_find_command): Rename to...
(tfind_command): ...this.
(trace_find_pc_command): Rename to...
(tfind_pc_command): ...this.
(trace_find_tracepoint_command): Rename to...
(tfind_tracepoint_command): ...this.
(trace_find_line_command): Rename to...
(tfind_line_command): ...this.
(trace_find_range_command): Rename to...
(tfind_range_command): ...this.
(trace_find_outside_command): Rename to...
(tfind_outside_command): ...this.
(trace_dump_command): Rename to...
(tdump_command): ...this.
(tfind_1): Adjust to renames.
(trace_find_end_command): Rename to...
(tfind_end_command): ...this, and adjust to renames..
(trace_status_mi): Adjust to renames.
(parse_trace_status): Adjust to renames.
(_initialize_tracepoint): Adjust to renames.
* tracepoint.h (enum trace_stop_reason) <tstop_command>: Rename
to...
<trace_stop_command>: ...this.
The suppress_output field of the mi_ui_out_data structure is never actually
set to 1/true. We can therefore remove it, and remove all the
if (suppress_output)
checks.
gdb/ChangeLog:
* mi/mi-out.c (mi_ui_out_data) <suppress_output>: Remove.
(mi_table_body): Remove suppress_output check.
(mi_table_end): Likewise.
(mi_table_header): Likewise.
(mi_begin): Likewise.
(mi_end): Likewise.
(mi_field_int): Likewise.
(mi_field_string): Likewise.
(mi_field_fmt): Likewise.
(mi_out_data_ctor): Likewise.
This patch teaches GDB AArch64 backend to recognize STR instructions
in prologue, like 'str x19, [sp, #-48]!' or 'str w0, [sp, #44]'.
The unit test is added too.
gdb:
2016-12-02 Yao Qi <yao.qi@linaro.org>
Pedro Alves <palves@redhat.com>
* aarch64-tdep.c (aarch64_analyze_prologue): Recognize STR
instruction.
(aarch64_analyze_prologue_test): More tests.
We don't have an effective way to test prologue analyzer which is
highly dependent on instruction patterns in prologue generated by
compiler. GDB prologue analyzer may not handle the new sequences
generated by new compiler, or may still handle some sequences that
generated by very old compilers which are no longer used. The
former is a functionality issue, while the latter is a maintenance
issue.
The input and output of prologue analyzer is quite clear, so it
fits for unit test. The input is series of instructions, and the
output are 1) where prologue end, 2) where registers are saved.
In aarch64, they are represented in 'struct aarch64_prologue_cache'.
This patch refactors aarch64_analyze_prologue so it can read
instructions from either real target or test harness. In unit
test aarch64_analyze_prologue_test, aarch64_analyze_prologue gets
instructions we prepared in the test, as the input of prologue
analyzer. Then, we checked various fields in
'struct aarch64_prologue_cache'.
gdb:
2016-12-02 Yao Qi <yao.qi@linaro.org>
Pedro Alves <palves@redhat.com>
* aarch64-tdep.c: Include "selftest.h".
(abstract_instruction_reader): New class.
(instruction_reader): New class.
(aarch64_analyze_prologue): Add new parameter reader. Call
reader.read instead of read_memory_unsigned_integer.
[GDB_SELF_TEST] (instruction_reader_test): New class.
(aarch64_analyze_prologue_test): New function.
(_initialize_aarch64_tdep) [GDB_SELF_TEST]: Register
selftests::aarch64_analyze_prologue_test.
* trad-frame.c (trad_frame_cache_zalloc):
(trad_frame_alloc_saved_regs): Add a new function.
* trad-frame.h (trad_frame_alloc_saved_regs): Declare.
This patch makes a class out of the ui_out_table structure, the
structure responsible for managing the generation of an UI table.
To simplify the ui_out_table object, I changed it so that it can only be
used for generating a single object. Instead of clearing the header
list when starting a new table, we an ui_out_table when starting a
table and delete it when we're done. Therefore, the checks:
if (uiout->table->flag)
if (!uiout->table->flag)
are respectively replaced with
if (uiout->table != nullptr)
if (uiout->table == nullptr)
Note: I removed the check at the beginning of ui_out_begin, because
there is an equivalent check at the beginning of verify_field.
New in v2:
- use "enum class" for ui_out_table::state and update references.
gdb/ChangeLog:
* ui-out.c (enum ui_out_table_state): Move to class
ui_out_table as ui_out_table::state.
(struct ui_out_table): Change to ...
(class ui_out_table): ... this.
<flag>: Remove.
<entry_level>: Rename to ...
<m_entry_level>: ... this.
<columns>: Rename to ...
<m_nr_cols>: ... this.
<id>: Rename to ...
<m_id>: ... this.
<headers>: Rename to ...
<m_headers>: ... this.
<headers_iterator>: Rename to ...
<m_headers_iterator>: ... this.
<start_body, append_header, start_row, get_next_header,
query_field, current_state, entry_level>: New methods.
(struct ui_out) <table>: Change type to unique_ptr to
ui_out_table.
(append_header_to_list, get_next_header, clear_header_list,
clear_table): Remove.
(ui_out_table_begin): Instantiate ui_out_table object. Update
table check.
(ui_out_table_body): Update table check, replace code with call
to ui_out_table::start_body.
(ui_out_table_end): Update table check, replace manual cleanup
with assignment of uiout->table unique_ptr to nullptr.
(ui_out_table_header): Update table check, replace call to
append_header_to_list with call to append_header method.
(ui_out_begin): Remove one table state check, update another.
Replace code with call to start_row method.
(verify_field): Update table checks.
(ui_out_query_field): Update table check, replace code with call
to query_field method.
(ui_out_new): Remove table initialization code.
This patch is just a little cleanup, it replaces the body_flag field of
ui_out_table with an enum. It expresses more explicitly the
intent of the field (check that state == TABLE_STATE_HEADERS conveys
more what we want to do than checking for !body_flag).
New in v2:
- Remove unnecessary ui_out_table_state::.
gdb/ChangeLog:
* ui-out.c (enum ui_out_table_state): New enum.
(struct ui_out_table) <body_flag>: Remove field.
<state>: New field.
(ui_out_table_begin): Replace usages of body_flag with state.
(ui_out_table_body): Likewise.
(ui_out_table_end): Likewise.
(ui_out_table_header): Likewise.
(ui_out_begin): Likewise.
(verify_field): Likewise.
(ui_out_new): Likewise.
Now that we use a vector to store the levels, we don't have to keep a
separate level field in ui_out to keep track of the current level. We
can efficiently derive it from the vector size. That causes a little
change in the meaning of the level, as in they are now 1-based instead
of 0-based (the initial level has the "id" 1 now), but it shouldn't
change anything in the behavior.
Additionally, push_level and pop_level don't really need to return the
new level, making them return void simplifies the code a bit.
Finally, the ui_out_begin/ui_out_end callbacks in the ui_out_impl
interface don't need to be passed the level, it's never actually used.
New in v2:
- Remove or update stale comments.
gdb/ChangeLog:
* ui-out.h (ui_out_begin_ftype): Remove level parameter.
(ui_out_end_ftype): Likewise.
* ui-out.c (struct ui_out) <level>: Replace field with a method
that dynamically computes the result.
(current_level): Get vector's back item instead of using
uiout->level.
(push_level): Make return type void.
(pop_level): Make return type void and update access to
ui_out::level.
(uo_begin): Remove level parameter.
(uo_end): Likewise.
(ui_out_table_begin): Update access to uiout::level.
(ui_out_begin): Don't read return value from push_level, call
uiout->level() instead, update call to uo_begin.
(ui_out_end): Don't read return value from pop_level, update
call to uo_end.
(verify_field): Update access to uiout->level.
(ui_out_new): Don't initialize ui_out::level, call push_level
to push the initial level instead of doing it by hand.
* cli-out.c (cli_begin): Remove level parameter.
(cli_end): Likewise.
* mi/mi-out.c (mi_begin): Likewise.
(mi_end): Likewise.
This patch changes struct ui_out_level to be a real C++ class. No
behavioral changes.
gdb/ChangeLog:
* ui-out.c (struct ui_out_level): Replace with ...
(class ui_out_level): ... this.
(current_level): Update.
(push_level): Update.
(pop_level): Update.
(verify_field): Update.
(ui_out_new): Update.
This patch makes ui_out_hdr (the object that represents an ui-out table
header) a proper C++ class. No behavior changes, it's all about
encapsulation.
gdb/ChangeLog:
* ui-out.c (struct ui_out_hdr): Replace with ...
(class ui_out_hdr): ... this.
(append_header_to_list): Update.
(get_next_header): Update.
(ui_out_query_field): Update.
Instead of keeping pointers to first, last and current ui_out_hdr in
ui_out_table, we can use an std::vector and an iterator. Direct random
access of to vector helps make get_next_header a bit nicer by avoiding
iterating on all the headers. append_header_to_list is also a bit
simpler.
Also, using unique_ptr inside the vector allows expressing the ownership
of the ui_out_hdr objects by the ui_out_table object, and it simplifies
the destruction.
gdb/ChangeLog:
* ui-out.c (struct ui_out_hdr) <next>: Remove.
(struct ui_out_table) <header_first, header_last, header_next>: Remove.
<headers, headers_iterator>: New fields.
(ui_out_table_body): Update for the new data structure.
(ui_out_begin): Likewise.
(clear_header_list): Likewise.
(append_header_to_list): Likewise.
(get_next_header): Likewise.
(ui_out_query_field): Likewise.
(ui_out_new): Likewise.
Use std::string for the id field of the ui_out_table object.
I found that all users of ui_out_table_begin passed a non-NULL value to
the tblid parameter, so we don't have to worry about the NULL case. I
changed the tblid parameter to be a std::string while at it.
gdb/ChangeLog:
* ui-out.c (struct ui_out_table) <id>: Change type to
std::string.
(ui_out_table_begin): Change tblid parameter type to
std::string, adapt code.
update following type change.
(clear_table): Update.
(ui_out_new): Update.
Use a standard vector instead of the home-made version. I used a vector
of plain pointers, because the cli_ui_out_data object doesn't own the
streams objects (i.e. they shouldn't be deleted when the vector is
deleted).
gdb/ChangeLog:
* cli-out.h (cli_ui_out_data) <streams>: Change type to
std::vector.
* cli-out.c: Remove vec.h include.
(cli_uiout_dtor): Update.
(cli_field_fmt): Update.
(cli_spaces): Update.
(cli_text): Update.
(cli_message): Update.
(cli_flush): Update.
(cli_redirect): Update.
(out_field_fmt): Update.
(field_separator): Update.
(cli_out_data_ctor): Update.
(cli_out_new): Update.
(cli_out_set_stream): Update.
Use a standard vector instead of the home-made version. I used a vector
of plain pointers, because the mi_ui_out_data object doesn't own the
streams objects (i.e. they shouldn't be deleted when the vector is
deleted).
gdb/ChangeLog:
* mi/mi-out.c: Remove vec.h include.
(mi_ui_out_data) <streams>: Change type to std::vector.
(mi_field_string): Update.
(mi_field_fmt): Update.
(mi_flush): Update.
(mi_redirect): Update.
(field_separator): Update.
(mi_open): Update.
(mi_close): Update.
(mi_out_buffered): Update.
(mi_out_rewind): Update.
(mi_out_put): Update.
(mi_out_data_ctor): Update.
(mi_out_data_dtor): Don't free streams.
Convert the levels field of struct ui_out to be a vector of unique_ptr
to ui_out_level. This way, the ownership of the ui_out_level objects by
the ui_out instance is clear.
gdb/ChangeLog:
* ui-out.c (ui_out_level_p): Remove typedef.
(DEF_VEC_P (ui_out_level_p)): Remove definition.
(struct ui_out) <levels>: Change type to vector of unique_ptr of
ui_out_level.
(current_level): Update.
(push_level): Update.
(pop_level): Update, don't manually delete the ui_out_level
instance.
(ui_out_new): Update.
The following patches introduce C++ vectors and strings as fields of the
various ui_out structures. We therefore need to use new/delete so that
their contructor/destructor is called. I find it simpler to change all
the allocations in a separate preliminary patch, rather than in each
individual patch.
gdb/ChangeLog:
* cli-out.c (cli_uiout_dtor): Use delete instead of xfree.
(cli_out_new): Use new instead of XNEW.
* mi/mi-out.c (mi_out_data_dtor): Use delete instead of xfree.
(mi_out_new): Use new instead of XNEW.
* tui/tui-out.c (tui_out_new): Likewise.
* ui-out.c (push_level): Likewise.
(pop_level): Use delete instead of xfree.
(clear_header_list): Use delete instead of xfree.
(append_header_to_list): Use new instead of XNEW.
(ui_out_new): Likewise.
Since we don't use suffix rules nor implicit rules in gdb, we can
disable them. The advantage is a slightly faster make [1].
Here are some numbers about the speedup. I ran this on my trusty old
Intel Q6600, so the time numbers are probably higher than what you'd get
on any recent hardware. I ran "make" in the gdb/ directory of an
already built repository (configured with --enable-targets=all). I
recorded the time of execution (average of 5). I then ran "make -d" and
recorded the number of printed lines, which gives a rough idea of the
number of operations done.
I compared the following configurations, to see the impact of both the
empty .SUFFIXES target and the empty pattern rules, as well as running
"make -r", which can be considered the "ideal" case.
A - baseline
B - baseline + .SUFFIXES
C - baseline + pattern rules
D - baseline + .SUFFIXES + pattern rules
E - baseline + make -r
config | time (s) | "make -d"
-----------------------------
A | 5.74 | 2396643
B | 1.19 | 298469
C | 2.81 | 1266573
D | 1.13 | 245489
E | 1.01 | 163914
We can see that the empty .SUFFIXES target has a bigger impact than the
empty pattern rules, but still it doesn't hurt to disable the implicit
pattern rules as well.
There are still some mentions of implicit rules I can't get rid of in
the "make -d" output. For example, it's trying to build .c files from
.w files:
Looking for an implicit rule for '/home/simark/src/binutils-gdb/gdb/infrun.c'.
Trying pattern rule with stem 'infrun'.
Trying implicit prerequisite '/home/simark/src/binutils-gdb/gdb/infrun.w'.
and trying to build Makefile.in from a bunch of extensions:
Looking for an implicit rule for 'Makefile.in'.
Trying pattern rule with stem 'Makefile.in'.
Trying implicit prerequisite 'Makefile.in.o'.
Trying pattern rule with stem 'Makefile.in'.
Trying implicit prerequisite 'Makefile.in.c'.
Trying pattern rule with stem 'Makefile.in'.
Trying implicit prerequisite 'Makefile.in.cc'.
... many more ...
If somebody knows how to disable them, we can do it, but at this point
the returns are minimal, so it is not that important.
I verified that both in-tree and out-of-tree builds work.
[1] Switching from explicit rules to pattern rules for files in
subdirectories actually made it slower, so this is kind of a way to
redeem myself. But it the end it's faster than it was previously,
so it was all worth it. :)
gdb/ChangeLog:
* disable-implicit-rules.mk: New file.
* Makefile.in: Include disable-implicit-rules.mk.
* data-directory/Makefile.in: Likewise.
* gnulib/Makefile.in: Likewise.
gdb/doc/ChangeLog:
* Makefile.in: Likewise.
gdb/gdbserver/ChangeLog:
* Makefile.in: Include disable-implicit-rules.mk.
gdb/testsuite/ChangeLog:
* Makefile.in: Include disable-implicit-rules.mk.
When GDB read inferior memory as an address or an instruction,
it should be unsigned.
gdb:
2016-11-30 Yao Qi <yao.qi@linaro.org>
* arm-tdep.c (arm_scan_prologue): Read memory as unsigned integer.
(arm_exidx_unwind_sniffer): Likewise.
Constify the data path between ui_out_wrap_hint and the wrap_indent
global, because we can. It's clearer that the argument passed to
wrap_hint is not intended to be modified by the ui_out implementation.
gdb/ChangeLog:
* mi/mi-out.c (mi_wrap_hint): Constify argument.
* cli-out.c (cli_wrap_hint): Likewise.
* ui-out.c (ui_out_wrap_hint, uo_wrap_hint): Likewise.
* ui-out.h (ui_out_wrap_hint, wrap_hint_ftype): Likewise.
* utils.c (wrap_here): Likewise.
(wrap_indent): Constify.
* utils.h (wrap_here): Constify argument.
The wrapper uo_redirect seems like it should return the return value
from of implementation function, since callers rely on it, but it
doesn't.
gdb/ChangeLog:
* ui-out.c (uo_redirect): Return the return value from the
implementation function.
It's not actually used, and removing it simplifies the upcoming patches
a bit. After the whole series, destroying an ui_out object will be
simply "delete uiout", which will call the default destructor.
gdb/ChangeLog:
* ui-out.c (ui_out_destroy, uo_data_destroy): Remove.
* ui-out.h (ui_out_destroy): Remove.
Just a little cleanup, so the name is more consistent with the naming of
the equivalent structures of cli and tui. It goes away in subsequent
patches anyway, but it might help follow the changes in those patches...
gdb/ChangeLog:
* mi/mi-out.c (ui_out_data): Rename to ...
(mi_ui_out_data): ... this.
