overflow in parsing "es#" and "et#" format units.  SystemError is now raised
instead of TypeError on programmical error in parsing format string.

Fixed the documented about buffer overflow error for "es#" and "et#" format
units.

* Various sections were pointing to the section on the string.Formatter
  class, when the section on the common format string syntax is probably more
  appropriate
* Fix references to various format() functions and methods
* Nested replacement fields may contain conversions and format specifiers,
  and this is tested; see Issue #19729 for instance

Patch by Thomas Kluyver.

Fix issue 26287: While handling FORMAT_VALUE opcode, the top of stack was being corrupted if an error occurred in PyObject_Format().

Based on patches by Brian Brazil and Daniel Rocco.

Reported by Denis Akhiyarov on docs@

Reported by Jon Tetlak on docs@

Fixed a crash when unpickle the functools.partial object with wrong state.
Fixed a leak in failed functools.partial constructor.
"args" and "keywords" attributes of functools.partial have now always types
tuple and dict correspondingly.

Based on patch by Aviv Palivoda.

Fixed a crash when unpickle the functools.partial object with wrong state.
Fixed a leak in failed functools.partial constructor.
"args" and "keywords" attributes of functools.partial have now always types
tuple and dict correspondingly.

Patch written by Carlo Beccarini <hackdiablo.cb@gmail.com>.

* test_contains() did not override anything
* test_expandtabs/upper/lower() in FixedStringTest were masking usable tests
  in string_tests. These tests now get run for bytearray() and bytes().
* test_expandtabs/upper/lower() in buffer_tests were only run on bytearray()
  and are redundant with string_tests

Test test_wrong_cert() runs a server that rejects the client's certificate,
so ECONNRESET is reasonable in addition to SSLError. On the other hand, the
other three tests don't even need to run a server because they are just
testing the parsing of invalid certificate files.

Also fix a ResourceWarning by closing the wrapped socket.

Accept time delta up to 0.12 second, instead of 0.11, for the "AMD64 FreeBSD
9.x" buildbot slave.

Issue #26248, patch written by Ben Hoyt:

1) Clarify that the return values of is_dir()/is_file()/etc are cached
   separately for follow_symlinks True and False.
2) Be more specific about when the functions require a system call, and how it
   relates to caching and follow_symlinks.
3) DRY up common stuff between is_dir and is_file by saying "Caching, system
   calls made, and exceptions raised are as per is_dir" in is_file.
4) Tweak to the first paragraph of docs for is_dir/is_file to simplify: assume
   the follow_symlinks=True default, then note the follow_symlinks=False
   non-default case after.