This is the mail archive of the
gdb-patches@sourceware.org
mailing list for the GDB project.
Re: [PATCH] Speed up "gdb -tui" output
- From: Doug Evans <dje at google dot com>
- To: Eli Zaretskii <eliz at gnu dot org>
- Cc: gdb-patches <gdb-patches at sourceware dot org>
- Date: Wed, 7 Jan 2015 12:45:15 -0800
- Subject: Re: [PATCH] Speed up "gdb -tui" output
- Authentication-results: sourceware.org; auth=none
- References: <83zj9v7urq dot fsf at gnu dot org> <CADPb22Q7oD3K-dYkngEPDBbV++mLCKifTEmvJczQ=0h2FX0yXA at mail dot gmail dot com> <83sifn7mpt dot fsf at gnu dot org> <CADPb22QeiSrzD81kMJeqpnDBrx1DybpbGmHc1r0koGDtqAu8tw at mail dot gmail dot com> <83h9w278a9 dot fsf at gnu dot org> <CADPb22R_=xwonUGJMxJRtAbbK8OcSmU5ojarB=T9wjg9YEtE6Q at mail dot gmail dot com> <83bnma75yt dot fsf at gnu dot org> <CADPb22SvEa8-+y009ET7PQDuYixGEK6LbrJr_pyQ98SfVy_iJg at mail dot gmail dot com> <837fwy74ny dot fsf at gnu dot org>
On Wed, Jan 7, 2015 at 11:48 AM, Eli Zaretskii <eliz@gnu.org> wrote:
>> Date: Wed, 7 Jan 2015 11:30:37 -0800
>> From: Doug Evans <dje@google.com>
>> Cc: gdb-patches <gdb-patches@sourceware.org>
>>
>> I can't promise to have a patch ready before 7.9, but I'll put it
>> on my todo list.
>
> Thanks.
>
>> >> Note that while we do explicitly call *_unfiltered with single characters,
>> >> unfiltered output is not in itself broken up into single characters.
>> >
>> > Not sure what you mean by that. fputs_maybe_filtered, which is the
>> > workhorse of most of the output functions, explicitly writes out the
>> > stuff it gets one character at a time, by calling fputc_unfiltered.
>> > How's that not breaking output?
>>
>> fputs_maybe_filtered is the workhorse for filtered output, and
>> it early exists for a number of things, like stream != gdb_stdout.
>
> It is used both for gdb_stdout and for other streams.
static void
fputs_maybe_filtered (const char *linebuffer, struct ui_file *stream,
int filter)
{
const char *lineptr;
if (linebuffer == 0)
return;
/* Don't do any filtering if it is disabled. */
if (stream != gdb_stdout
|| !pagination_enabled
|| batch_flag
|| (lines_per_page == UINT_MAX && chars_per_line == UINT_MAX)
|| top_level_interpreter () == NULL
|| ui_out_is_mi_like_p (interp_ui_out (top_level_interpreter ())))
{
fputs_unfiltered (linebuffer, stream);
return;
}
...
}
It may be called for other streams, but it early exits
with a direct call to fputs_unfiltered if stream != gdb_stdout.
I don't know how to say it any plainer than that.
>
>> Most unfiltered output goes straight to fputs_unfiltered (which is the
>> wrapper around the ui-file to_fputs method).
>
> Once again, I fail to see how this fact, which I certainly noticed
> already, helps to solve the problem. When fputs_maybe_filtered is
> called, we have no idea whether the text it gets is complete enough to
> flush the stream after we are done with it. That information is at
> higher levels. So we don't know whether to flush the stream after the
> direct call to fputs_unfiltered.
My theory is we can punt for streams != gdb_stdout,
and for them just always refresh inside calls to tui_file_fputs.
And for gdb_stdout rely on calls to gdb_flush to do
the refresh.
Seems like not that many lines of code to give it a try.
>
>> I see there is fputstr{,n}_unfiltered which does things a character
>> at a time but it looks like it is only used by MI.
>
> Grep for fputs_unfiltered callers, and you will see that there are
> much more calls with only 1 or 2 characters.
I already did, and pointed that out.
We don't know yet if they're a problem.
Seems like the main source of the problem would be gdb_stdout,
so let's try to fix that first and go from there.