diff options
author | Pedro Alves <palves@redhat.com> | 2017-02-02 22:00:43 +0000 |
---|---|---|
committer | Pedro Alves <palves@redhat.com> | 2017-02-02 22:00:43 +0000 |
commit | 616268b639780e0819b51053c794037bcde3de16 (patch) | |
tree | 8d12ec88f6482a508d9410f9554cfc099b2d0ac6 /gdb/cli/cli-interp.h | |
parent | 5be5dbf0ce3d2eb20003fea3691eaa6bfa5710d2 (diff) | |
download | binutils-616268b639780e0819b51053c794037bcde3de16.zip binutils-616268b639780e0819b51053c794037bcde3de16.tar.gz binutils-616268b639780e0819b51053c794037bcde3de16.tar.bz2 |
Move "tee" building down to interpreter::set_logging_proc
This patch gets rid of this hack in mi_set_logging:
/* The tee created already is based on gdb_stdout, which for MI
is a console and so we end up in an infinite loop of console
writing to ui_file writing to console etc. So discard the
existing tee (it hasn't been used yet, and MI won't ever use
it), and create one based on raw_stdout instead. */
By pushing down responsibility for the tee creation to the
interpreter. I.e., pushing the CLI bits out of handle_redirections
down to the CLI interpreter's set_logging_proc method.
This fixes a few leaks that I spotted, and then confirmed with
"valgrind --leak-check=full":
[...]
==21429== 56 (32 direct, 24 indirect) bytes in 1 blocks are definitely lost in loss record 30,243 of 34,980
==21429== at 0x4C29216: operator new(unsigned long) (vg_replace_malloc.c:334)
==21429== by 0x62D9A9: mi_set_logging(interp*, int, ui_file*, ui_file*) (mi-interp.c:1395)
==21429== by 0x810B8A: current_interp_set_logging(int, ui_file*, ui_file*) (interps.c:360)
==21429== by 0x61C537: handle_redirections(int) (cli-logging.c:162)
==21429== by 0x61C6EC: set_logging_on(char*, int) (cli-logging.c:190)
==21429== by 0x6163BE: do_cfunc(cmd_list_element*, char*, int) (cli-decode.c:105)
==21429== by 0x6193C1: cmd_func(cmd_list_element*, char*, int) (cli-decode.c:1913)
==21429== by 0x8DB790: execute_command(char*, int) (top.c:674)
==21429== by 0x632AE6: mi_execute_cli_command(char const*, int, char const*) (mi-main.c:2343)
==21429== by 0x6329BA: mi_cmd_execute(mi_parse*) (mi-main.c:2306)
==21429== by 0x631E19: captured_mi_execute_command(ui_out*, mi_parse*) (mi-main.c:1998)
==21429== by 0x632389: mi_execute_command(char const*, int) (mi-main.c:2163)
==21429==
[...]
==26635== 24 bytes in 1 blocks are definitely lost in loss record 20,740 of 34,995
==26635== at 0x4C29216: operator new(unsigned long) (vg_replace_malloc.c:334)
==26635== by 0x61C355: handle_redirections(int) (cli-logging.c:131)
==26635== by 0x61C6EC: set_logging_on(char*, int) (cli-logging.c:190)
==26635== by 0x6163BE: do_cfunc(cmd_list_element*, char*, int) (cli-decode.c:105)
==26635== by 0x6193C1: cmd_func(cmd_list_element*, char*, int) (cli-decode.c:1913)
==26635== by 0x8DB7BC: execute_command(char*, int) (top.c:674)
==26635== by 0x7B9132: command_handler(char*) (event-top.c:590)
==26635== by 0x7B94F7: command_line_handler(char*) (event-top.c:780)
==26635== by 0x7B8ABB: gdb_rl_callback_handler(char*) (event-top.c:213)
==26635== by 0x933CE9: rl_callback_read_char (callback.c:220)
==26635== by 0x7B89ED: gdb_rl_callback_read_char_wrapper_noexcept() (event-top.c:175)
==26635== by 0x7B8A49: gdb_rl_callback_read_char_wrapper(void*) (event-top.c:192)
One is fixed by transfering ownership of the log file to the tee. In
pseudo-patch, since the code was moved at the same time:
- out = new tee_file (curr_output, false, logfile.get (), false);
+ out = new tee_file (curr_output, false, logfile.get (), true);
The other is this bit in mi_set_logging:
else
{
+ delete mi->raw_stdout;
I tried to split the leak fixes to a smaller preparatory patch, but
that was difficult exactly because of the tee hack in
handle_redirections -> mi_set_logging.
gdb/ChangeLog:
2017-02-02 Pedro Alves <palves@redhat.com>
* cli/cli-interp.c (struct saved_output_files, saved_output):
Moved from cli/cli-logging.c.
(cli_set_logging): New function.
(cli_interp_procs): Install cli_set_logging.
* cli/cli-interp.h (make_logging_output, cli_set_logging):
Declare.
* cli/cli-logging.c (struct saved_output_files, saved_output):
Moved to cli/cli-interp.c.
(pop_output_files): Don't save outputs here.
(make_logging_output): New function.
(handle_redirections): Don't build tee nor save previous outputs
here.
* interps.c (current_interp_set_logging): Change prototype.
Assume there's always a set_logging_proc method installed.
* interps.h (interp_set_logging_ftype): Change prototype.
(current_interp_set_logging): Change prototype and adjust comment.
* mi/mi-interp.c (mi_set_logging): Change protototype. Adjust to
use make_logging_output.
* tui/tui-interp.c (tui_interp_procs): Install cli_set_logging.
Diffstat (limited to 'gdb/cli/cli-interp.h')
-rw-r--r-- | gdb/cli/cli-interp.h | 18 |
1 files changed, 18 insertions, 0 deletions
diff --git a/gdb/cli/cli-interp.h b/gdb/cli/cli-interp.h index ef86372..f93548c 100644 --- a/gdb/cli/cli-interp.h +++ b/gdb/cli/cli-interp.h @@ -20,6 +20,24 @@ struct interp; +/* Make the output ui_file to use when logging is enabled. + CURR_OUTPUT is the stream where output is currently being sent to + (e.g., gdb_stdout for the CLI, raw output stream for the MI). + LOGFILE is the log file already opened by the caller. + LOGGING_REDIRECT is the value of the "set logging redirect" + setting. If true, the resulting output is the logfile. If false, + the output stream is a tee, with the log file as one of the + outputs. Ownership of LOGFILE is transferred to the returned + output file, which is an owning pointer. */ +extern ui_file *make_logging_output (ui_file *curr_output, + ui_file_up logfile, + bool logging_redirect); + +/* The CLI interpreter's set_logging_proc method. Exported so other + interpreters can reuse it. */ +extern void cli_set_logging (struct interp *interp, + ui_file_up logfile, bool logging_redirect); + extern int cli_interpreter_supports_command_editing (struct interp *interp); extern void cli_interpreter_pre_command_loop (struct interp *self); |