Mwparserfromhell

- Added support for Python 3.6.
- Fixed parsing bugs involving:
- wikitables nested in templates;
- wikitable error recovery when unable to recurse;
- templates nested in template parameters before other parameters.
- Fixed parsing file-like objects.
- Made builds deterministic.
- Documented caveats.

- Added Windows binaries for Python 3.5.
- Fixed edge cases involving wikilinks inside of external links and vice versa.
- Fixed a C tokenizer crash when a keyboard interrupt happens while parsing.

- Fixed setup script not including header files in releases.
- Fixed Windows binary uploads.

- The process for building Windows binaries has been fixed, and these should be
distributed along with new releases. Windows users can now take advantage of
C speedups without having a compiler of their own.
- Added support for Python 3.5.
- '<' and '>' are now disallowed in wikilink titles and template names. This
includes when denoting tags, but not comments.
- Fixed the behavior of preserve_spacing in Template.add() and keep_field in
Template.remove() on parameters with hidden keys.
- Removed _ListProxy.detach(). SmartLists now use weak references and their
children are garbage-collected properly.
- Fixed parser bugs involving:
- templates with completely blank names;
- templates with newlines and comments.
- Heavy refactoring and fixes to the C tokenizer, including:
- corrected a design flaw in text handling, allowing for substantial speed
improvements when parsing long strings of plain text;
- implemented new Python 3.3 PEP 393 Unicode APIs.
- Fixed various bugs in SmartList, including one that was causing memory issues
on 64-bit builds of Python 2 on Windows.
- Fixed some bugs in the release scripts.

- The parser now falls back on pure Python mode if C extensions cannot be
built. This fixes an issue that prevented some Windows users from installing
the parser.
- Added support for parsing wikicode tables (patches by David Winegar).
- Added a script to test for memory leaks in scripts/memtest.py.
- Added a script to do releases in scripts/release.sh.
- skip_style_tags can now be passed to mwparserfromhell.parse() (previously,
only Parser().parse() allowed it).
- The 'recursive' argument to Wikicode's filter methods now accepts a third
option, RECURSE_OTHERS, which recurses over all children except instances of
'forcetype' (for example, `code.filter_templates(code.RECURSE_OTHERS)`
returns all un-nested templates).
- The parser now understands HTML tag attributes quoted with single quotes.
When setting a tag attribute's value, quotes will be added if necessary. As
part of this, Attribute's 'quoted' attribute has been changed to 'quotes',
and is now either a string or None.
- Calling Template.remove() with a Parameter object that is not part of the
template now raises ValueError instead of doing nothing.
- Parameters with non-integer keys can no longer be created with
'showkey=False', nor have the value of this attribute be set to False later.
- _ListProxy.destroy() has been changed to _ListProxy.detach(), and now works
in a more useful way.
- If something goes wrong while parsing, ParserError will now be raised.
Previously, the parser would produce an unclear BadRoute exception or allow
an incorrect node tree to be build.
- Fixed parser bugs involving:
- nested tags;
- comments in template names;
- tags inside of <nowiki> tags.
- Added tests to ensure that parsed trees convert back to wikicode without
unintentional modifications.
- Added support for a NOWEB environment variable, which disables a unit test
that makes a web call.
- Test coverage has been improved, and some minor related bugs have been fixed.
- Updated and fixed some documentation.

- Added support for Python 2.6 and 3.4.
- Template.has() is now passed 'ignore_empty=False' by default instead of True.
This fixes a bug when adding parameters to templates with empty fields, and
is a breaking change if you rely on the default behavior.
- The 'matches' argument of Wikicode's filter methods now accepts a function
(taking one argument, a Node, and returning a bool) in addition to a regex.
- Re-added 'flat' argument to Wikicode.get_sections(), fixed the order in which
it returns sections, and made it faster.
- Wikicode.matches() now accepts a tuple or list of strings/Wikicode objects
instead of just a single string or Wikicode.
- Given the frequency of issues with the (admittedly insufficient) tag parser,
there's a temporary skip_style_tags argument to parse() that ignores '' and
''' until these issues are corrected.
- Fixed a parser bug involving nested wikilinks and external links.
- C code cleanup and speed improvements.