1.0.2 (2015-03-31)
------------------

- Released to fix an issue where the universal wheel was tagged on PyPI
  as Py Version "2.7" instead of "py2.py3".

1.0.1 (2015-03-31)
------------------

- Fixed a bug introduced in 1.0.0 where unicode strings could not be used
  as attribute values.  Thanks to Terrence Brannon for reporting the issue
  and supplying the initial patch.

- A ``setup.cfg`` file has been added with a ``[bdist_wheel]`` section to
  mark this package as a universal wheel.

1.0.0 (2014-04-10)
------------------

- Dropped support for Python 2.3 / 2.4.  Users of those Python versions
  should pin to earlier releases, (e.g., "meld3<1.0.0dev").

- Added compatibility with Python 3.2, 3.3, and 3.4.  Thanks to Scott
  Maxwell and Tres Seaver for contributing the patches.

- Installation now requires setuptools.

- The optional C extension has been removed.  It was only used internally
  by meld3, was not enabled by default, and did not support all meld3
  features.  The pure Python version has very acceptable performance.

- The license has been changed from ZPL 2.1 to the BSD-derived Repoze
  license, which is the same license used by Supervisor.

- Profiling scripts that had been used early in meld3 development have
  been removed.

- The example scripts example.py and melddiff.py have been removed.  See
  README.rst for the same example that was in example.py and the unit tests
  for examples of the ``diffmeld`` function.

- Support for using meld3 directly from the command line with
  ``python meld3.py <file> <mutator>`` has been removed.

0.6.10 (2012-11-27)
-------------------

- Fixed a bug where an exception could be raised when escaping certain
  attribute or cdata values.  This was caused by meld3 trying to use
  _encode_entity from xml.etree without importing it first.  Thanks to
  Jorge Puente Sarrin for contributing the patch.

0.6.9 (2012-09-12)
------------------

- Fixed a test failure that only occurred on some builds of
  Python 2.7 where parsing an unknown entity could raise
  an expat.error instead of SyntaxError.

0.6.8 (2012-01-12)
------------------

- Added the C extension source (cmeld3.c) to the release package
  by including it in MANIFEST.in.  Thanks to Soren Hansen for
  noticing it was missing from prior releases.

- Running setup.py will now halt on error if a compatible version
  of Python is not detected.

0.6.7 (2010-08-04)
------------------

- Make compatible with Python 2.7 (patch kindly contributed by
  Jonathan Riboux).

0.6.6 (2009-09-30)
------------------

- Change download location.  This really should be a setuptools
  package so we can upload it to PyPI.  No functionality changes.

0.6.5 (2008-07-21)
------------------

- Apply patch for Python 2.5 compatibility from both
  Toshio and Anders.

- Create distro tarball via:

   rm MANIFEST
   USE_MELD3_EXTENSION_MODULES=1 python setup.py sdist

  .. in order to get the cmeld3.c file in the distribution.

0.6.4 (2008-01-17)
------------------

- Make the default build use the Python-based meld "helper" instead of
  the C-based one.  Since the primary consumer of meld3 (as far as I
  know) is "supervisor", and since the typical supervisor consumer is
  likely not to have a C compiler and Python development libraries on
  his system, it makes more sense for the default build not to compile
  the C extensions.  However, if the environment variable
  "USE_MELD3_EXTENSION_MODULES" is set when "setup.py install" is
  invoked, the C extensions will be built.  meld3 is much slower
  without the C extensions, so using "USE_MELD3_EXTENSION_MODULES" is
  usually a good idea for performance-sensitive systems.

  As a result of this change, the "NO_MELD3_EXTENSION_MODULES"
  environment variable introduced in 0.6.1 now has no effect.

0.6.3 (2007-08-25)
------------------

- Fixed two more memory leaks (one in bfclonehandler, the other in
  findmeldhandler) in the c helper module.

0.6.2 (2007-08-23)
------------------

- Fixed a number of memory leaks in the C implementation of the helper
  module.  Any use of "findmeld", "clone", "getiterator", or "content"
  previously leaked references.

0.6.1 (2007-08-21)
------------------

- Allow people to install meld3 without building extension modules.
  If the environment variable "NO_MELD3_EXTENSION_MODULES" is set, the
  meld3 setup.py file will not build any extension modules.  meld3
  will still work, only more slowly than if the extension modules
  existed.

0.6 (2006-09-18)
----------------

- Fixed crashbug when repeating trees with "structure" nodes in them.
  Symptom: bus error.  Thanks to Terrence Brannon for the report.

0.5 (2006-02-19)
----------------

- Add 'fillmeldhtmlform' method to nodes.

- Fix obscure parsing bug that arose when attempting to change the
  URI used as a "meld id" (do not lowercase).

- Added three methods to element nodes:

    write_htmlstring
    write_xmlstring
    write_xhtmlstring

  These methods have the same respective argument lists as their
  "write_foo" cousins, except they don't accept a "file" argument.
  Instead of writing to a file, they return a string containing the
  renderering of the element.

- You can now use the meld3.py module to interactively try out
  different renderings.  To do this, invoke meld3.py with a filename
  and a dotted-python-path name to a mutator function that accepts a
  single argument (the root element), e.g.:

    python meld3.py sample.html meld3.sample_mutator

  The rendering will be sent to stdout

- Remove unused parse method.

- Add __setslice__, __delslice__, remove methods that correctly update
  parent pointers.

- Don't call into superclass to do append, insert, __setitem__, etc.

- Internal speedups:

    - Avoid function call overhead in various places by inlining code.

    - _write_html/_write_xml now accept a callable "write" argument
      instead of a file argument (avoid file.write call overhead).

    - HTML serializer only calls _escape_cdata/_escape_attrib if necessary
      instead of calling it without regard to its need.

    - Use string.encode(encoding) instead of calling an _encode function.

    - Perform special-case rendering for English-centric encodings
      during write_html.

    - Ignore things that might be "QNames" during rendering.

    - "getiterator", "content", "clone", and "findmeld" now
      implemented in a C module.  Experimental.  Disable via changing
      "import cmeld3 as helper" in meld3.py to something like "import
      wontbehere".

- Fix text and attribute serialization to only quote ampersands that
  aren't already part of entities.

- Do the minimal possible thing to escape text and attribute values during
  rendering.  For text, this is escaping ampersands that aren't parts of
  entities and the less-than (<) character.  For attribute values, this is
  amps and less-than as well as the quote (") character.  This was done
  partly because I think the spec allows it but it's also what Kid does,
  shrug.

- Change 'diffmeld()' to return a dictionary of dictionaries.  The
  dictionary returned by diffmeld has the keys 'reduced' and
  'unreduced'.  the values of both 'reduced' and 'unreduced' is
  another dictionary.  The leafmost dictionary has the keys 'added',
  'removed', and 'moved.  In the 'unreduced' dictionary, *all* meld
  tags that have been added, removed, or moved are present (this was
  the behavior of diffmeld previously).  In the 'reduced' dictionary,
  the added, removed, and moved values are reduced to the smallest
  number of tags which share a common lineage.

- Add a meldprofile.py script.

0.4 (2006-01-01)
----------------

- The clone() method of elements copied neither the text nor the tail
  of the element (this is what caused markup created by "repeats" to
  fall all on one line).

- Add diffing capability to meld nodes.  The file 'melddiff.py' shows an
  example of using the diff API.

- Add a 'meldid()' method to elements.  This returns None if the element
  has no meld id, otherwise it returns the id.

- Add a 'fillmeld(**kw)' method to elements.  This does the same thing
  as '__mod__' but returns meld ids (the keys of **kw) that cannot be
  found anywhere in the tree.

- Make source distro into a distutils-installable package.

0.3 (2005-12-26)
----------------

- Fix broken example.py file.

- Add ZPT-alike methods on elements: 'content', 'replace', and
  'attributes'.  'content' replaces the node's content; 'replace'
  replaces the node itself with a text value, and 'attributes' sets
  the attributes of the node.  Using the ElementTree API to do the
  same things usually causes the code to run faster, but these
  functions are more convenient and more easily grokked by ZPT people.

- Override __delitem__ on meld elements in order to relieve deleted
  items of their parent pointers.

- Strip all xhtml namespace identifiers out of XHTML output.  Browsers
  just can't deal with this.

- Undocumented element API method 'remove' renamed to 'deparent' (it
  was shadowing an ElementTree API method).

- Documentation improvements and change examples to use ZPT-alike
  methods.

- Add support for HTML input files.  Input files don't need to be
  strictly well-formed XML anymore.

- Remove the 'parse' top-level function in favor of explicit separate
  xml parsing and html parsing functions.

- Add 'parse_xml' and 'parse_html' top-level parsing functions.

- Add 'parse_xmlstring' and 'parse_htmlstring' module-scope functions
  which calls their respective 'parse_xxx' function with a StringIO
  containing the passed text.

0.2 (2005-12-24)
----------------

- Use a method on elements to do writing rather than requiring a user
  call a "write" function.  The equivalent is now a method of the
  element named "write_xml".  element.write_xml(file) performs a write
  of XML into the file.  element.write_xml(...) includes an XML
  declaration in its serialization (but no doctype, at least by
  default).

- Various non-XML serialization methods have been added.  The default
  arguments of these serialization methods are what I'm guessing are
  the most common cases desired for various kinds of output::

     element.write_html(...).  This serializes the node and its
     children to HTML.  This feature was inspired by and based on code
     Ian Bicking.  By default, the serialization will include a
     'loose' HTML DTD doctype (this can be overridden with the
     doctype= argument).  "Empty" shortcut elements such as "<div/>"
     will be converted to a balanced pair of tags e.g. "<div></div>".
     But some HTML tags (defined as per the HTML 4 spec as area, base,
     basefont, br, col, frame, hr, img, input, isindex, link, meta,
     param) will not be followed with a balanced ending tag; only the
     beginning tag will be output.  Additionally, "boolean" tag
     attributes will not be followed with any value.  The "boolean"
     tags are selected, checked, compact, declare, defer, disabled,
     ismap, multiple, nohref, noresize, noshade, and nowrap.  So the
     XML input "<input type="checkbox" checked="checked"/>" will be
     turned into "<input type="checkbox" checked>".  Additionally,
     'script' and 'style' tags will not have their contents escaped
     (e.g. so "&" will not be turned into &amp; when it's iside the
     textual content of a script or style tag.)

     element.write_xhtml(...).  This serializes the node and its
     children to XHTML.  By default, the serialization will include a
     'loose' XHTML doctype (this can be overridden with the doctype=
     argument).  No XML declaration is included in the serialization by
     default.  If you want to serialize an XML declaration, pass
     'declaration=True'.

- All serialization methods have a number of optional
  arguments::

     fragment: If this is true, serialize an element as a "fragment".
     When an element is serialized as a fragment, it will not include
     either a declaration nor a doctype (the declaration= and doctype=
     arguments will be ignored).

     doctype: Output a custom doctype during the writing of XML and
     HTML (see write, write_xml, write_xhtml, and write_html).  Use
     the constants in meld3.doctype (xhtml, xhtml_strict, html, and
     html_strict) to avoid passing a literal 3-tuple of (name, pubid,
     system) as the doctype parameter.  If fragment=True is specified
     for serialization, this argument has no effect.

     encoding: Specify a character encoding to be used during writing
     (see write, write_xml write_html, and write_xhtml).  The encoding
     must be a valid Python codec name (e.g. 'utf-8').  If this is
     provided for write_xml and write_xhtml, and the XML declaration
     is serialized, the declaration will include the encoding.  If an
     encoding is passed to write_html, no explicit encoding is
     included in the declaration but the serialization will be done
     with utf-8.

- XML serializations (write_xml and write_xhtml) have the
  aforementioned arguments but expose two additional optional
  arguments::

     declaration: If this is true, an xml declaration header is output
     during the writing of XML (see write, write_xml, and
     write_xhtml).  If the encoding is specified, and the
     serialization is meant to include an XML declaration (via
     declaration=), the declaration will include the encoding.  If
     'fragment=True' is specified for serialization, this argument has
     no effect.  It doesn't matter if your input document had a
     declaration header; this option must be used to control
     declaration output.

     pipeline: If this is true, allow meld identifiers to be preserved
     during the writing of XML and XHTML (see write, write_xml and
     write_xhtml).  meld identifiers cannot be preserved on HTML
     serializations because HTML doesn't understand namespaces.

- HTML entities can now be parsed properly (magically) when a DOCTYPE
  is not supplied in the source of the XML passed to 'parse'.  If your
  source document does not contain a DOCTYPE declaration, the DOCTYPE
  is set to 'loose' XHTML 'by magic'.  If your source document does
  contain a DOCTYPE declaration, the existing DOCTYPE is used (and
  HTML entities thus may or may not work as a result, depending on the
  DOCTYPE).  To prevent this behavior, pass a false value to the
  xhtml= parameter of the 'parse' function.  This in no way effects
  output, which is independent of parsing.  This does not imply that
  any *non*-HTML entity can be parsed in the input stream under any
  circumstance without having it defined it in your source document.

- Comments are now preserved in output.  They are also present in the
  ElementTree node tree (as Comment elements), so beware. Processing
  instructions (e.g. <?xml version="1.0">) are completely thrown away
  at parse time and do not exist anywhere in the element tree.

- Avoid use of deepcopy in the clone() method of elements (much
  speedier to explicitly recurse).

- The "meld helper" namespace (e.g. element.meld) is no longer present
  or supported.  Instead of using element.meld['foo'] to find an
  element with the meld:id "foo", use element.findmeld('foo').  This
  returns None if the node cannot be found.  Instead of using
  element.meld.get('foo', 'somedefault'), use element.findmeld('foo',
  'somedefault').  Instead of using element.meld.repeat(...), use
  element.repeat(...).

- Elements now support a __mod__ which can accept a dictionarylike
  operand and which causes the text of elements with meld ids which
  match the keys in the dictionary to be set to the key's value in the
  dictionary.  For example, if an element contains subelements with
  the meld ids "foo" and "bar", you can replace those nodes' text
  values with the following::

    element % {'foo':'foo text', 'bar':'bar text'}

- __mod__ will not accept a non-dictionary-like object (such as a list
  or tuple).  __mod__ will never raise an error unless you pass it a
  non-dictionary-like object; if it can't find a node corresponding to
  a key in the dictionary, it moves on to the next key.  Only the text
  values of the nodes which are found during this process are
  replaced.

- Using duplicate meld identifiers on separate elements in the document
  now causes a ValueError to be raised at parse time.

0.1 (2005-12-18)
----------------

- Initial release.