Re: [PATCH v11] Add pretty printers for the NPTL lock types

[Torvald Riegel <triegel at redhat dot com>]

On Thu, 2016-12-22 at 20:39 -0300, Martin Galvan wrote:
> 2016-12-22 19:50 GMT-03:00 Torvald Riegel <triegel@redhat.com>:
> > I first thought it would pick the system gdb and not the one from $PATH.
> 
> What would be the difference between the "system gdb" and the one from
> $PATH? Unless you altered your $PATH to make it point to the newly
> built gdb, I don't understand what you mean.

I thought it would not pick up the gdb from PATH.  But it did, and the
problem was elsewhere (see below).

> >  It seems that the new gdb is used, but the test script
> > still complaints that the gdb doesn't support python ('gdb must have
> > python support to test the pretty printers.').  pexpect should be
> > installed for the python that's used.  What am I doing wrong?
> 
> How did you come to the conclusion that the new gdb is being used?

system gdb is too old and fails earlier tests.

> In any case, I think it would be helpful to see the value of
> gdb_python_error before the test script exits.

After *lots* of poking, it seems that something in the pexpect/gdb
combination in my case was adding terminal escape sequences to the gdb
output.  That made the error string look empty when printed, but it was
not a zero-sized string and thus does not evaluate to false.
I do not know what caused this, and whether the pretty printers script
should have filtered that out or pexpect/gdb shouldn't have produced the
escape sequence in the first place.  The combination I used was:
pexpect from pip on system python 2.7.5, gdb 7.8.5 built wiht
--with-python, and I had ncurses-devel-5.9-13.20130511.el7.x86_64
installed (picked up in the gdb build).
Switching to gdb 7.12 does not require ncurses or a termcap library
anymore, and works with the existing pretty printers script.

I suppose raising the minimally required gdb version would be helpful.

It would also be helpful if scripts/test_printers_common.py would not
discard the error when it cannot load the pretty printer modules for
some reason.  I've changed
    for printer_file in printer_files:
        test('source {0}'.format(printer_file))
to the following, but don't know whether that's the proper way to do it:
    for printer_file in printer_files:
        test('source {0}'.format(printer_file), '')

And nptl/test-mutex-printers fails in my setup, BTW.