On Windows, debug builds insert stack probes, and recursive functions
tend to exhaust the stack faster.
This patch reduces the marshal maximum depth from 2000 to 1500 for debug
builds only. Optimized builds are not affected.
This allows test_marshal to pass with debug builds.

2. Fix some debug prints in RemoteDebugger.py - use py3k syntax.

test_float crashes on Windows, because the %zd format is used in a call
to PyOS_snprintf().
The attached patch properly uses PY_FORMAT_SIZE_T.

1/ getargs.diff adds the 'Z' and 'Z#' format specifiers for
PyArg_ParseTuple. They mimic z and z# for unicode strings, by accepting
a Unicode or None (in which case the Py_UNICODE* pointer is set to
NULL). With doc and tests.

2/ subprocess.diff converts file PC/_subprocess.c to unicode. We use the
Unicode version of the win32 api (and Z conversion from previous patch)

3/ stdout.diff: sys.stdout must not convert the line endings, Windows
already does it.
Without this patch, when redirecting the output of python, the file
contains \r\r\n for each line. (test_subprocess did catch this)

However, I (GvR) removed the change to _fileio.c (included in the
patches) that prevents closing file descripors < 3 from being closed;
I think that needs to be solved in a different way.

so that even on Windows, after s.write("x\n"), s.getvalue() == "x\n".

instance.  This was effectively already the case, but the error
reporting was lousy.

Reflow long lines.
Remove whitespace inside parents.
Move open braces off lines by themselves.

This breaks a few things that I'll patch up in a minute.

has an extra (optional) output parameter through which it returns the size.
Use this in a few places where I used PyUnicode_AsString() + strlen(),
and in one new place (which fixes test_pep263).

previous version of PEP 3101.  It's currently not compiled in, but I want to
leave it because it might be useful in the future and it makes
calc_number_widths() clearer.  It justifies NumberFieldWidths.rsign
and .n__rsign.

From now on, trying to write str to a binary stream
is an error (I'm still working on the reverse).
There are still (at least) two failing tests:
- test_asynchat
- test_urllib2_localnet
but I'm sure these will be fixed by someone.

This fixes test_doctest with strict bytes/str.

not buffer("blob").

(PyUnicode these days) to a char* + length.  The fix consists of calling
PyUnicode_AsString() and strlen().  This is not ideal, but AsCharBuffer()
is definitely not the API to use.

a tuple (literal, field_name, format_spec, conversion).

literal will always be a string, but might be of zero length.
field_name will be None if there is no markup text
format_spec will be a (possibly zero length) string if
  field_name is non-None
conversion will be a one character string, or None

This makes the Formatter class, and especially it's parse()
method, easier to understand.

Suggestion was by Jim Jewett, inspired by the "tail" of an
elementtree node.

Also, fixed a reference leak in fieldnameiter_next.

(One change to httplib.py, but not enough for test_httplib.)