Hi,

This patch fixes a regression, possibly introduced by 2a3c1174c3c0db1140180fb3fc56ac324d1c0a7c, in this part of the change:

---
@@ -2064,13 +2096,13 @@ vfprintf_unfiltered (struct ui_file *stream, const char *format, va_list args)
       fputs_unfiltered (timestamp.c_str (), stream);
     }
   else
-    fputs_unfiltered (linebuffer.c_str (), stream);
+    vfprintf_maybe_filtered (stream, format, args, false, true);
 }

 void
 vprintf_filtered (const char *format, va_list args)
---

The significance of this is that printf_unfiltered writes messages to wrap_buffer, whereas puts_unfiltered pushes them immediately to stdout, resulting in "post-" messages being printed out of order.

Not sure about how to go about testing this, looking at the testsuite, such as gdb.base/annota1.exp, everything appears to be in order.  Perhaps this is because the testsuite triggers one of these conditions in fputs_maybe_filtered() though.

---
if (stream != gdb_stdout
    || !pagination_enabled
    || pagination_disabled_for_command
    || batch_flag
    || (lines_per_page == UINT_MAX && chars_per_line == UINT_MAX)
    || top_level_interpreter () == NULL
    || top_level_interpreter ()->interp_ui_out ()->is_mi_like_p ())
---


However, the actual observed behaviour in gdb is:
---
Reading symbols from a.out...
(gdb) set annotate 2

\032\032pre-prompt
(gdb)
\032\032prompt
start

prompt\032\032post-
Temporary breakpoint 1 at 0x13716: file test.c, line 3.
---

With this patch applied, instead "\032\032post-prompt" is printed.

I think this can be applied as obvious, but wanted to have someone else have a quick check, just in case it would be preferred to change fputs_maybe_filtered instead to flush the buffer on scope exit for unfiltered messages.

--
Iain

---