Re: [PATCH] gdb: Don't flush line wrap buffer before adding newline

[Andrew Burgess <andrew.burgess@embecosm.com>]

* Tom Tromey <tom@tromey.com> [2019-02-05 08:27:29 -0700]:

> >>>>> "Andrew" == Andrew Burgess <andrew.burgess@embecosm.com> writes:
> 
> Andrew> 	* utils.c (fputs_maybe_filtered): Don't flush the line wrap buffer
> Andrew> 	before inserting a newline for line-wrapping.
> 
> Thanks, this is ok.

Really, I was just teasing - this patch is rubbish :-/

(Not really, I spotted my mistake immediately after hitting send.)

The code used to look like this:

    if (wrap_column)
    {
      emit_style_escape (ui_file_style ());
      flush_wrap_buffer (stream);
      fputc_unfiltered ('\n', stream);
    }

The problem is that 'emit_style_escape' sends the escape string to the
wrap_buffer, while the newline is sent directly to the console.

This means that GDB is still styled when the newline is sent, which I
think means that if we hit a pager prompt this could be styled.

I did consider having a second version of 'emit_style_escape' that
by-passed the wrap_buffer, but I worried that this might be too open
to incorrect use.  Instead I thought a better approach would be backup
and clear the wrap_buffer before calling 'emit_style_escape' in the
above code, the 'flush_wrap_buffer' can then be used safely.

The patch below shows a new proposal for utils.c.

Do you agree with the issues I see above?

What do you think of the new approach?

Thanks,
Andrew


---

diff --git a/gdb/utils.c b/gdb/utils.c
index 6fb5736abb5..153de6cd0f8 100644
--- a/gdb/utils.c
+++ b/gdb/utils.c
@@ -1787,9 +1787,12 @@ fputs_maybe_filtered (const char *linebuffer, struct ui_file *stream,
 	         anyway; if it's wrong, let us keep going.  */
 	      if (wrap_column)
 		{
+		  std::string tmp = wrap_buffer;
+		  wrap_buffer.clear ();
 		  emit_style_escape (ui_file_style ());
 		  flush_wrap_buffer (stream);
 		  fputc_unfiltered ('\n', stream);
+		  wrap_buffer = tmp;
 		}
 
 	      /* Possible new page.  Note that