Re: MI testsuite to use PTY for inferior

[Bob Rossi <bob at brasko dot net>]

On Tue, Aug 02, 2005 at 04:47:50PM -0400, Daniel Jacobowitz wrote:
> On Tue, Aug 02, 2005 at 11:45:20PM +0300, Eli Zaretskii wrote:
> > > Date: Mon, 1 Aug 2005 23:50:20 -0400
> > > From: Daniel Jacobowitz <drow@false.org>
> > > Cc: gdb-patches@sources.redhat.com
> > > 
> > > Yes, but the problem's that the buffering is in the program being
> > > debugged.  It's not the buffering of the GDB MI stream that's the
> > > problem, but the fact that the stdout of the program you're debugging
> > > would suddenly become block buffered instead of line buffered, if you
> > > fed it to a pipe when it expects a TTY.
> > 
> > This started as a discussion how to separate MI output from the
> > debuggee's.  For that, I suggested to redirect MI's output (_not_ the
> > debuggee's) to a different file handle.  That should not change the
> > buffering of the debuggee's stdout in any way.  While the problem you
> > mention is real, it has nothing to do with separation of these two
> > streams: if sending the program's output to a pipe changes its
> > buffering, we already have that annoyance on Windows.
> 
> Ah!  Thank you, I misunderstood your solution.  This is a very
> interesting idea indeed.

This is a good solution. I just had yet another idea, that might be
worth while. It would be a simple way to allow the FE to separate the MI
output command and the inferior output data.

Currently, GDB works like this,

(gdb) 
-interpreter-exec console "b test.c:49"
~"Breakpoint 1 at 0x8048525: file test.c, line 49.\n"
^done
(gdb) 
-interpreter-exec console "r"
~"Starting program: /home/bob/cvs/cgdb/cgdb.readline/objdir/test \n"
OUTPUT FROM INFERIOR
~"\n"
~"Breakpoint 1, main (argc=1, argv=0xbffffb54) at test.c:49\n"
~"49\t   for ( i = 0; i < 100; i++)\n"
^done
(gdb) 

If we let GDB do what it does know, except change the MI output command
to allow something like,

(gdb) 
-interpreter-exec console "b test.c:49"
~"Breakpoint 1 at 0x8048525: file test.c, line 49.\n"
^done
(gdb) 
-interpreter-exec console "r"
~"Starting program: /home/bob/cvs/cgdb/cgdb.readline/objdir/test \n"
=inferior-starting unique-key
OUTPUT FROM INFERIOR
=inferior-stopped unique-key
~"\n"
~"Breakpoint 1, main (argc=1, argv=0xbffffb54) at test.c:49\n"
~"49\t   for ( i = 0; i < 100; i++)\n"
^done
(gdb) 

In this way, the FE can still break the output into lines, and it can
strcmp the line to determine if the output is from GDB or the inferior.
The "unique-key" takes care of the case where you are actually debugging
GDB, because once the FE get's the '=inferior-starting unique-key' line,
it doesn't care how many =inferior-started or =inferior-stopped lines
get sent out until the key matches.

This sounds like a very simple solution. What does everyone think?

Thanks,
Bob Rossi