Re: Output of printf command

[Kai Ruottu <kai dot ruottu at luukku dot com>]

Keith Seitz wrote:
> 
> On Thu, 7 Feb 2002, Keith Seitz wrote:
> 
> > On Thu, 7 Feb 2002, David Mc Kenna wrote:
> >
> > > This GDB was configured as "--host=i686-pc-cygwin --target=arm-elf".
> >
> > Ok, that's starting to make some sense. Cygwin hosts obviously don't have
> > xterms (even those with XFree86, as far as Insight is concerned).
> >
> > I will check into this and get back to you.
> 
> Ok, I've checked into this, and it is a simulator problem. The ARM
> simulator (sim/arm) doesn't use the gdb/sim callback interface for os-like
> functions. It just does printf/fprintf and ignores that there might be a
> ui on the other side.
> 
> I've modified src/sim/arm/armos.c to use the callback interface, and it
> works for me for my simple testcase.
> 
> I've contacted some sim folks, and I'll send a note when/if this is fixed.
> I don't know how this has gone unnoticed for so long.

 All the other sims could also be checked, the japanese targets used to output
to the GDB-console ("Console Window") as default, while some like the PowerPC
'psim' use 'write(1,...)' directly, this forces the output to the 'system console',
to the xterm or to the Win32-console. This has been compatible with the native
Insights (this far). Outputting to the GDB-console has been seen uncompatible
by me, but it has been the only possibility on systems without a 'system console'
like Win32s.  Here is a clip taken from my 'sim/ppc/emul-unix.c':

----------------------- clip ----------------------
static void
do_unix_write(os_emul_data *emul,
	      unsigned call,
	      const int arg0,
	      cpu *processor,
	      unsigned_word cia)
{
  void *scratch_buffer = NULL;
  int d = (int)cpu_registers(processor)->gpr[arg0];
  unsigned_word buf = cpu_registers(processor)->gpr[arg0+1];
  int nbytes = cpu_registers(processor)->gpr[arg0+2];
  int status;

  if (WITH_TRACE && ppc_trace[trace_os_emul])
    printf_filtered ("%d, 0x%lx, %d", d, (long)buf, nbytes);

  /* get a tempoary bufer */
  scratch_buffer = zalloc(nbytes); /* FIXME - nbytes == 0 */

  /* copy in */
  emul_read_buffer(scratch_buffer, buf, nbytes,
		   processor, cia);

  /* write */
#ifdef __WIN32S__
  status = ppc_callback->write(ppc_callback, d, scratch_buffer, nbytes);
#else
  status = write(d, scratch_buffer, nbytes);
#endif
  emul_write_status(processor, status, errno);
  zfree(scratch_buffer);

  flush_stdoutput();
}
----------------------- clip ----------------------

 The GDB-console has been the only available place for the 'callback-writes'
AFAIK... Is there something else available now?
 
 This far ('...5.0') the direct 'write()'s have however worked on Unix and
Win32, but is there now some change in '5.1.x' ?  The last sources I have
tried are from the sad day '20010911' and a little later and then the output
to the system console still worked... But I may have some local patches added,
to continue to use the 'system console' just as earlier, which I have now
forgotten. Anyway with the Autumn-2001 snapshots the following happens on
Win32 when I click a '<prefix>-gdb' icon:

 1. The black Win32 console, the 'system console' for the 'stdout' from the
    debugged program opens

 2. After a while the empty Source Window, ie. the Insight-GUI opens

 If the 1. doesn't happen, there is some essential difference between my
(patched) and the plain vanilla Insight sources... Or the user is still
using Win3.1x/Win32s ;-)

 I use Mingw as the Win32-host, not Cygwin... Neither I build the Win32-
binaries under Windoze, but under Linux (using Linux-x-Mingw tools), but
this should make any difference (than making the builds much easier and
quicker...)

Cheers, Kai