Pin lxml to latest version 5.2.1

[pyup-bot]

This PR pins lxml to the latest release 5.2.1.

Changelog

5.2.1

==================

Bugs fixed
----------

* LP2059910: The minimum CPU architecture for the Linux x86 binary wheels was set back to
"core2", but with SSE 4.2 enabled.

* LP2059977: ``Element.iterfind("//absolute_path")`` failed with a ``SyntaxError``
where it should have issued a warning.

* GH416: The documentation build was using the non-standard ``which`` command.
Patch by Michał Górny.

5.2.0

==================

Other changes
-------------

* LP1958539: The ``lxml.html.clean`` implementation suffered from several (only if used)
security issues in the past and was now extracted into a separate library:

https://github.com/fedora-python/lxml_html_clean

Projects that use lxml without "lxml.html.clean" will not notice any difference,
except that they won't have potentially vulnerable code installed.
The module is available as an "extra" setuptools dependency "lxml[html_clean]",
so that Projects that need "lxml.html.clean" will need to switch their requirements
from "lxml" to "lxml[html_clean]", or install the new library themselves.

* The minimum CPU architecture for the Linux x86 binary wheels was upgraded to
"sandybridge" (launched 2011), and glibc 2.28 / gcc 12 (manylinux_2_28) wheels were added.

* Built with Cython 3.0.10.

5.1.2

==================

Bugs fixed
----------

* LP2059977: ``Element.iterfind("//absolute_path")`` failed with a ``SyntaxError``
where it should have issued a warning.

5.1.1

==================

Bugs fixed
----------

* LP2048920: ``iterlinks()`` in ``lxml.html`` rejected ``bytes`` input in 5.1.0.

* High source line numbers from the parser are no longer truncated
(up to a C ``long``) when using libxml2 2.11 or later.

Other changes
-------------

* GH407: A compatibility test was adapted to recent expat versions.
Patch by Miro Hrončok.

* Binary wheels use the library versions libxml2 2.12.6 and libxslt 1.1.39.

* Windows binary wheels use the library versions libxml2 2.11.7 and libxslt 1.1.39.

* Built with Cython 3.0.9.

5.1.0

==================

Features added
--------------

* Parsing ASCII strings is slightly faster.

Bugs fixed
----------

* GH349: The HTML ``Cleaner()`` interpreted an accidentally provided string parameter
for the ``host_whitelist`` as list of characters and silently failed to reject any hosts.
Passing a non-collection is now rejected.

Other changes
-------------

* Support for Python 2.7 and Python versions < 3.6 was removed.

* The wheel build was migrated to use ``cibuildwheel``.
Patch by Primož Godec.

5.0.2

==================

Other changes
-------------

* GH407: A compatibility test was adapted to recent expat versions.
Patch by Miro Hrončok.

* Binary wheels use the library versions libxml2 2.12.6 and libxslt 1.1.39.

* Built with Cython 3.0.9.

5.0.1

==================

Bugs fixed
----------

* LP2046208: Parsing non-BMP Python Unicode strings could fail on macOS.

* LP2044225: When incrementally parsing broken HTML, reporting start events on
missing structural tags failed and could lead to subsequent exceptions.

* LP2045435: Some (not all) issues with stricter C compilers were resolved.

* The binary wheels in the 5.0.0 release did not validate cleanly (but installed ok).


.. _latest_release:

5.0.0

==================

Features added
--------------

* Character escaping in ``C14N2`` serialisation now uses a single pass over the text
instead of searching for each unescaped character separately.

* Early support for Python 3.13a2 was added.

Bugs fixed
----------

* LP1976304: The ``Element.addnext()`` method previously inserted the new element
before existing tail text.  The tail text of both sibling elements now stays on
the respective elements.

* LP1980767, GH379: ``TreeBuilder.close()`` could fail with a ``TypeError`` after
parsing incorrect input.  Original patch by Enrico Minack.

* ``Element.itertext(with_tail=False)`` returned the tail text of comments and
processing instructions, despite the explicit option.

* GH370: A crash with recent libxml2 2.11.x versions was resolved.
Patch by Michael Schlenker.

* A compile problem with recent libxml2 2.12.x versions was resolved.

* The internal exception handling in C callbacks was improved for Cython 3.0.

* The exception declarations of ``xmlInputReadCallback``, ``xmlInputCloseCallback``,
``xmlOutputWriteCallback`` and ``xmlOutputCloseCallback`` in ``tree.pxd`` were
corrected to prevent running Python code or calling into the C-API with a live
exception set.

* GH385: The long deprecated ``unittest.m̀akeSuite()`` function is no longer used.
Patch by Miro Hrončok.

* LP1522052: A file-system specific test is now optional and should no longer fail
on systems that don't support it.

* GH392: Some tests were adapted for libxml2 2.13.
Patch by Nick Wellnhofer.

* Contains all fixes from lxml 4.9.4.

Other changes
-------------

* LP1742885: lxml no longer expands external entities (XXE) by default to prevent
the security risk of loading arbitrary files and URLs.  If this feature is needed,
it can be enabled in a backwards compatible way by using a parser with the option
``resolve_entities=True``.  The new default is ``resolve_entities='internal'``.

* With libxml2 2.10.4 and later (as provided by the lxml 5.0 binary wheels),
parsing HTML tags with "prefixes" no longer builds a namespace dictionary
in ``nsmap`` but considers the ``prefix:name`` string the actual tag name.
With older libxml2 versions, since 2.9.11, the prefix was removed.  Before
that, the prefix was parsed as XML prefix.

lxml 5.0 does not try to hide this difference but now changes the ElementPath
implementation to let ``element.find("part1:part2")`` search for the tag
``part1:part2`` in documents parsed as HTML, instead of looking only for ``part2``.

* LP2024343: The validation of the schema file itself is now optional in the
ISO-Schematron implementation.  This was done because some lxml distributions
discard the RNG validation schema file due to licensing issues.  The validation
can now always be disabled with ``Schematron(..., validate_schema=False)``.
It is enabled by default if available and disabled otherwise.  The module
constant ``lxml.isoschematron.schematron_schema_valid_supported`` can be used
to detect whether schema file validation is available.

* Some redundant and long deprecated methods were removed:
``parser.setElementClassLookup()``,
``xslt_transform.apply()``,
``xpath.evaluate()``.

* Some incorrect declarations were removed from ``python.pxd``. In general, this file
should not be used by external Cython code. Use the C-API declarations provided by
Cython itself instead.

* Binary wheels use the library versions libxml2 2.12.3 and libxslt 1.1.39.

* Built with Cython 3.0.7, updated to follow recent changes in Cython 3.1-dev.

4.9.4

==================

Bugs fixed
----------

* LP2046398: Inserting/replacing an ancestor into a node's children could loop indefinitely.

* LP1980767, GH379: ``TreeBuilder.close()`` could fail with a ``TypeError`` after
parsing incorrect input.  Original patch by Enrico Minack.

* LP1522052: A file-system specific test is now optional and should no longer fail
on systems that don't support it.

Other changes
-------------

* Wheels include zlib 1.3, libxml2 2.10.3 and libxslt 1.1.39
(zlib 1.2.12, libxml2 2.10.3 and libxslt 1.1.37 on Windows).

* Built with Cython 0.29.37.

4.9.3

==================

Bugs fixed
----------

* LP2008911: ``lxml.objectify`` accepted non-decimal numbers like ``²²²`` as integers.

* A memory leak in ``lxml.html.clean`` was resolved by switching to Cython 0.29.34+.

* GH348: URL checking in the HTML cleaner was improved.
Patch by Tim McCormack.

* GH371, GH373: Some regex strings were changed to raw strings to fix Python warnings.
Patches by Jakub Wilk and Anthony Sottile.

Other changes
-------------

* Wheels include zlib 1.2.13, libxml2 2.10.3 and libxslt 1.1.38
(zlib 1.2.12, libxml2 2.10.3 and libxslt 1.1.37 on Windows).

* Built with Cython 0.29.36 to adapt to changes in Python 3.12.

4.9.2

==================

Bugs fixed
----------

* CVE-2022-2309: A Bug in libxml2 2.9.1[0-4] could let namespace declarations
from a failed parser run leak into later parser runs.  This bug was worked around
in lxml and resolved in libxml2 2.10.0.
https://gitlab.gnome.org/GNOME/libxml2/-/issues/378

Other changes
-------------

* LP1981760: ``Element.attrib`` now registers as ``collections.abc.MutableMapping``.

* lxml now has a static build setup for macOS on ARM64 machines (not used for building wheels).
Patch by Quentin Leffray.

4.9.1

==================

Bugs fixed
----------

* A crash was resolved when using ``iterwalk()`` (or ``canonicalize()``)
after parsing certain incorrect input.  Note that ``iterwalk()`` can crash
on *valid* input parsed with the same parser *after* failing to parse the
incorrect input.

4.9.0

==================

Bugs fixed
----------

* GH341: The mixin inheritance order in ``lxml.html`` was corrected.
Patch by xmo-odoo.

Other changes
-------------

* Built with Cython 0.29.30 to adapt to changes in Python 3.11 and 3.12.

* Wheels include zlib 1.2.12, libxml2 2.9.14 and libxslt 1.1.35
(libxml2 2.9.12+ and libxslt 1.1.34 on Windows).

* GH343: Windows-AArch64 build support in Visual Studio.
Patch by Steve Dower.

4.8.0

==================

Features added
--------------

* GH337: Path-like objects are now supported throughout the API instead of just strings.
Patch by Henning Janssen.

* The ``ElementMaker`` now supports ``QName`` values as tags, which always override
the default namespace of the factory.

Bugs fixed
----------

* GH338: In lxml.objectify, the XSI float annotation "nan" and "inf" were spelled in
lower case, whereas XML Schema datatypes define them as "NaN" and "INF" respectively.
Patch by Tobias Deiminger.

Other changes
-------------

* Built with Cython 0.29.28.

4.7.1

==================

Features added
--------------

* Chunked Unicode string parsing via ``parser.feed()`` now encodes the input data
to the native UTF-8 encoding directly, instead of going through ``Py_UNICODE`` /
``wchar_t`` encoding first, which previously required duplicate recoding in most cases.

Bugs fixed
----------

* The standard namespace prefixes were mishandled during "C14N2" serialisation on Python 3.
See https://mail.python.org/archives/list/lxmlpython.org/thread/6ZFBHFOVHOS5GFDOAMPCT6HM5HZPWQ4Q/

* ``lxml.objectify`` previously accepted non-XML numbers with underscores (like "1_000")
as integers or float values in Python 3.6 and later. It now adheres to the number
format of the XML spec again.

* LP1939031: Static wheels of lxml now contain the header files of zlib and libiconv
(in addition to the already provided headers of libxml2/libxslt/libexslt).

Other changes
-------------

* Wheels include libxml2 2.9.12+ and libxslt 1.1.34 (also on Windows).

4.7.0

==================

* Release retracted due to missing files in lxml/includes/.

4.6.5

==================

Bugs fixed
----------

* A vulnerability (GHSL-2021-1038) in the HTML cleaner allowed sneaking script
content through SVG images (CVE-2021-43818).

* A vulnerability (GHSL-2021-1037) in the HTML cleaner allowed sneaking script
content through CSS imports and other crafted constructs (CVE-2021-43818).

4.6.4

==================

Features added
--------------

* GH317: A new property ``system_url`` was added to DTD entities.
Patch by Thirdegree.

* GH314: The ``STATIC_*`` variables in ``setup.py`` can now be passed via env vars.
Patch by Isaac Jurado.

4.6.3

==================

Bugs fixed
----------

* A vulnerability (CVE-2021-28957) was discovered in the HTML Cleaner by Kevin Chung,
which allowed JavaScript to pass through.  The cleaner now removes the HTML5
``formaction`` attribute.

4.6.2

==================

Bugs fixed
----------

* A vulnerability (CVE-2020-27783) was discovered in the HTML Cleaner by Yaniv Nizry,
which allowed JavaScript to pass through.  The cleaner now removes more sneaky
"style" content.

4.6.1

==================

Bugs fixed
----------

* A vulnerability was discovered in the HTML Cleaner by Yaniv Nizry, which allowed
JavaScript to pass through.  The cleaner now removes more sneaky "style" content.

4.6.0

==================

Features added
--------------

* GH310: ``lxml.html.InputGetter`` supports ``__len__()`` to count the number of input fields.
Patch by Aidan Woolley.

* ``lxml.html.InputGetter`` has a new ``.items()`` method to ease processing all input fields.

* ``lxml.html.InputGetter.keys()`` now returns the field names in document order.

* GH-309: The API documentation is now generated using ``sphinx-apidoc``.
Patch by Chris Mayo.

Bugs fixed
----------

* LP1869455: C14N 2.0 serialisation failed for unprefixed attributes
when a default namespace was defined.

* ``TreeBuilder.close()`` raised ``AssertionError`` in some error cases where it
should have raised ``XMLSyntaxError``.  It now raises a combined exception to
keep up backwards compatibility, while switching to ``XMLSyntaxError`` as an
interface.

4.5.2

==================

Bugs fixed
----------

* ``Cleaner()`` now validates that only known configuration options can be set.

* LP1882606: ``Cleaner.clean_html()`` discarded comments and PIs regardless of the
corresponding configuration option, if ``remove_unknown_tags`` was set.

* LP1880251: Instead of globally overwriting the document loader in libxml2, lxml now
sets it per parser run, which improves the interoperability with other users of libxml2
such as libxmlsec.

* LP1881960: Fix build in CPython 3.10 by using Cython 0.29.21.

* The setup options "--with-xml2-config" and "--with-xslt-config" were accidentally renamed
to "--xml2-config" and "--xslt-config" in 4.5.1 and are now available again.

4.5.1

==================

Bugs fixed
----------

* LP1570388: Fix failures when serialising documents larger than 2GB in some cases.

* LP1865141, GH298: ``QName`` values were not accepted by the ``el.iter()`` method.
Patch by xmo-odoo.

* LP1863413, GH297: The build failed to detect libraries on Linux that are only
configured via pkg-config.
Patch by Hugh McMaster.

4.5.0

==================

Features added
--------------

* A new function ``indent()`` was added to insert tail whitespace for pretty-printing
an XML tree.

Bugs fixed
----------

* LP1857794: Tail text of nodes that get removed from a document using item
deletion disappeared silently instead of sticking with the node that was removed.

Other changes
-------------

* MacOS builds are 64-bit-only by default.
Set CFLAGS and LDFLAGS explicitly to override it.

* Linux/MacOS Binary wheels now use libxml2 2.9.10 and libxslt 1.1.34.

* LP1840234: The package version number is now available as ``lxml.__version__``.

4.4.3

==================

Bugs fixed
----------

* LP1844674: ``itertext()`` was missing tail text of comments and PIs since 4.4.0.

4.4.2

==================

Bugs fixed
----------

* LP1835708: ``ElementInclude`` incorrectly rejected repeated non-recursive
includes as recursive.
Patch by Rainer Hausdorf.

4.4.1

==================

Bugs fixed
----------

* LP1838252: The order of an OrderedDict was lost in 4.4.0 when passing it as
attrib mapping during element creation.

* LP1838521: The package metadata now lists the supported Python versions.

4.4.0

==================

Features added
--------------

* ``Element.clear()`` accepts a new keyword argument ``keep_tail=True`` to clear
everything but the tail text.  This is helpful in some document-style use cases
and for clearing the current element in ``iterparse()`` and pull parsing.

* When creating attributes or namespaces from a dict in Python 3.6+, lxml now
preserves the original insertion order of that dict, instead of always sorting
the items by name.  A similar change was made for ElementTree in CPython 3.8.
See https://bugs.python.org/issue34160

* Integer elements in ``lxml.objectify`` implement the ``__index__()`` special method.

* GH269: Read-only elements in XSLT were missing the ``nsmap`` property.
Original patch by Jan Pazdziora.

* ElementInclude can now restrict the maximum inclusion depth via a ``max_depth``
argument to prevent content explosion.  It is limited to 6 by default.

* The ``target`` object of the XMLParser can have ``start_ns()`` and ``end_ns()``
callback methods to listen to namespace declarations.

* The ``TreeBuilder`` has new arguments ``comment_factory`` and ``pi_factory`` to
pass factories for creating comments and processing instructions, as well as
flag arguments ``insert_comments`` and ``insert_pis`` to discard them from the
tree when set to false.

* A `C14N 2.0 <https://www.w3.org/TR/xml-c14n2/>`_ implementation was added as
``etree.canonicalize()``, a corresponding ``C14NWriterTarget`` class, and
a ``c14n2`` serialisation method.

Bugs fixed
----------

* When writing to file paths that contain the URL escape character '%', the file
path could wrongly be mangled by URL unescaping and thus write to a different
file or directory.  Code that writes to file paths that are provided by untrusted
sources, but that must work with previous versions of lxml, should best either
reject paths that contain '%' characters, or otherwise make sure that the path
does not contain maliciously injected '%XX' URL hex escapes for paths like '../'.

* Assigning to Element child slices with negative step could insert the slice at
the wrong position, starting too far on the left.

* Assigning to Element child slices with overly large step size could take very
long, regardless of the length of the actual slice.

* Assigning to Element child slices of the wrong size could sometimes fail to
raise a ValueError (like a list assignment would) and instead assign outside
of the original slice bounds or leave parts of it unreplaced.

* The ``comment`` and ``pi`` events in ``iterwalk()`` were never triggered, and
instead, comments and processing instructions in the tree were reported as
``start`` elements.  Also, when walking an ElementTree (as opposed to its root
element), comments and PIs outside of the root element are now reported.

* LP1827833: The RelaxNG compact syntax support was broken with recent versions
of ``rnc2rng``.

* LP1758553: The HTML elements ``source`` and ``track`` were added to the list
of empty tags in ``lxml.html.defs``.

* Registering a prefix other than "xml" for the XML namespace is now rejected.

* Failing to write XSLT output to a file could raise a misleading exception.
It now raises ``IOError``.

Other changes
-------------

* Support for Python 3.4 was removed.

* When using ``Element.find*()`` with prefix-namespace mappings, the empty string
is now accepted to define a default namespace, in addition to the previously
supported ``None`` prefix.  Empty strings are more convenient since they keep
all prefix keys in a namespace dict strings, which simplifies sorting etc.

* The ``ElementTree.write_c14n()`` method has been deprecated in favour of the
long preferred ``ElementTree.write(f, method="c14n")``.  It will be removed
in a future release.

4.3.5

==================

* Rebuilt with Cython 0.29.13 to support Python 3.8.

4.3.4

==================

* Rebuilt with Cython 0.29.10 to support Python 3.8.

4.3.3

==================

Bugs fixed
----------

* Fix leak of output buffer and unclosed files in ``_XSLTResultTree.write_output()``.

4.3.2

==================

Bugs fixed
----------

* Crash in 4.3.1 when appending a child subtree with certain text nodes.

Other changes
-------------

* Built with Cython 0.29.6.

4.3.1

==================

Bugs fixed
----------

* LP1814522: Crash when appending a child subtree that contains unsubstituted
entity references.

Other changes
-------------

* Built with Cython 0.29.5.

4.3.0

==================

Features added
--------------

* The module ``lxml.sax`` is compiled using Cython in order to speed it up.

* GH267: ``lxml.sax.ElementTreeProducer`` now preserves the namespace prefixes.
If two prefixes point to the same URI, the first prefix in alphabetical order
is used.  Patch by Lennart Regebro.

* Updated ISO-Schematron implementation to 2013 version (now MIT licensed)
and the corresponding schema to the 2016 version (with optional "properties").

Other changes
-------------

* GH270, GH271: Support for Python 2.6 and 3.3 was removed.
Patch by hugovk.

* The minimum dependency versions were raised to libxml2 2.9.2 and libxslt 1.1.27,
which were released in 2014 and 2012 respectively.

* Built with Cython 0.29.2.

4.2.6

==================

Bugs fixed
----------

* LP1799755: Fix a DeprecationWarning in Py3.7+.

* Import warnings in Python 3.6+ were resolved.

4.2.5

==================

Bugs fixed
----------

* Javascript URLs that used URL escaping were not removed by the HTML cleaner.
Security problem found by Omar Eissa.  (CVE-2018-19787)

4.2.4

==================

Features added
--------------

* GH259: Allow using ``pkg-config`` for build configuration.
Patch by Patrick Griffis.

Bugs fixed
----------

* LP1773749, GH268: Crash when moving an element to another document with
``Element.insert()``.
Patch by Alexander Weggerle.

4.2.3

==================

Bugs fixed
----------

* Reverted GH265: lxml links against zlib as a shared library again.

4.2.2

==================

Bugs fixed
----------

* GH266: Fix sporadic crash during GC when parse-time schema validation is used
and the parser participates in a reference cycle.
Original patch by Julien Greard.

* GH265: lxml no longer links against zlib as a shared library, only on static builds.
Patch by Nehal J Wani.

4.2.1

==================

Bugs fixed
----------

* LP1755825: ``iterwalk()`` failed to return the 'start' event for the initial
element if a tag selector is used.

* LP1756314: Failure to import 4.2.0 into PyPy due to a missing library symbol.

* LP1727864, GH258: Add "-isysroot" linker option on MacOS as needed by XCode 9.

4.2.0

==================

Features added
--------------

* GH255: ``SelectElement.value`` returns more standard-compliant and
browser-like defaults for non-multi-selects.  If no option is selected, the
value of the first option is returned (instead of None).  If multiple options
are selected, the value of the last one is returned (instead of that of the
first one).  If no options are present (not standard-compliant)
``SelectElement.value`` still returns ``None``.

* GH261: The ``HTMLParser()`` now supports the ``huge_tree`` option.
Patch by stranac.

Bugs fixed
----------

* LP1551797: Some XSLT messages were not captured by the transform error log.

* LP1737825: Crash at shutdown after an interrupted iterparse run with XMLSchema
validation.

Other changes
-------------

4.1.1

==================

* Rebuild with Cython 0.27.3 to improve support for Py3.7.

4.1.0

==================

Features added
--------------

* ElementPath supports text predicates for current node, like "[.='text']".

* ElementPath allows spaces in predicates.

* Custom Element classes and XPath functions can now be registered with a
decorator rather than explicit dict assignments.

* Static Linux wheels are now built with link time optimisation (LTO) enabled.
This should have a beneficial impact on the overall performance by providing
a tighter compiler integration between lxml and libxml2/libxslt.

Bugs fixed
----------

* LP1722776: Requesting non-Element objects like comments from a document with
``PythonElementClassLookup`` could fail with a TypeError.

4.0.0

==================

Features added
--------------

* The ElementPath implementation is now compiled using Cython,
which speeds up the ``.find*()`` methods quite significantly.

* The modules ``lxml.builder``, ``lxml.html.diff`` and ``lxml.html.clean``
are also compiled using Cython in order to speed them up.

* ``xmlfile()`` supports async coroutines using ``async with`` and ``await``.

* ``iterwalk()`` has a new method ``skip_subtree()`` that prevents walking into
the descendants of the current element.

* ``RelaxNG.from_rnc_string()`` accepts a ``base_url`` argument to
allow relative resource lookups.

* The XSLT result object has a new method ``.write_output(file)`` that serialises
output data into a file according to the ``<xsl:output>`` configuration.

Bugs fixed
----------

* GH251: HTML comments were handled incorrectly by the soupparser.
Patch by mozbugbox.

* LP1654544: The html5parser no longer passes the ``useChardet`` option
if the input is a Unicode string, unless explicitly requested.  When parsing
files, the default is to enable it when a URL or file path is passed (because
the file is then opened in binary mode), and to disable it when reading from
a file(-like) object.

Note: This is a backwards incompatible change of the default configuration.
If your code parses byte strings/streams and depends on character detection,
please pass the option ``guess_charset=True`` explicitly, which already worked
in older lxml versions.

* LP1703810: ``etree.fromstring()`` failed to parse UTF-32 data with BOM.

* LP1526522: Some RelaxNG errors were not reported in the error log.

* LP1567526: Empty and plain text input raised a TypeError in soupparser.

* LP1710429: Uninitialised variable usage in HTML diff.

* LP1415643: The closing tags context manager in ``xmlfile()`` could continue
to output end tags even after writing failed with an exception.

* LP1465357: ``xmlfile.write()`` now accepts and ignores None as input argument.

* Compilation under Py3.7-pre failed due to a modified function signature.

Other changes
-------------

* The main module source files were renamed from ``lxml.*.pyx`` to plain
``*.pyx`` (e.g. ``etree.pyx``) to simplify their handling in the build
process.  Care was taken to keep the old header files as fallbacks for
code that compiles against the public C-API of lxml, but it might still
be worth validating that third-party code does not notice this change.

3.8.0

==================

Features added
--------------

* ``ElementTree.write()`` has a new option ``doctype`` that writes out a
doctype string before the serialisation, in the same way as ``tostring()``.

* GH220: ``xmlfile`` allows switching output methods at an element level.
Patch by Burak Arslan.

* LP1595781, GH240: added a PyCapsule Python API and C-level API for
passing externally generated libxml2 documents into lxml.

* GH244: error log entries have a new property ``path`` with an XPath
expression (if known, None otherwise) that points to the tree element
responsible for the error. Patch by Bob Kline.

* The namespace prefix mapping that can be used in ElementPath now injects
a default namespace when passing a None prefix.

Bugs fixed
----------

* GH238: Character escapes were not hex-encoded in the ``xmlfile`` serialiser.
Patch by matejcik.

* GH229: fix for externally created XML documents.  Patch by Theodore Dubois.

* LP1665241, GH228: Form data handling in lxml.html no longer strips the
option values specified in form attributes but only the text values.
Patch by Ashish Kulkarni.

* LP1551797: revert previous fix for XSLT error logging as it breaks
multi-threaded XSLT processing.

* LP1673355, GH233: ``fromstring()`` html5parser failed to parse byte strings.

Other changes
-------------

* The previously undocumented ``docstring`` option in ``ElementTree.write()``
produces a deprecation warning and will eventually be removed.

3.7.4

==================

Bugs fixed
----------

* LP1551797: revert previous fix for XSLT error logging as it breaks
multi-threaded XSLT processing.

* LP1673355, GH233: ``fromstring()`` html5parser failed to parse byte strings.

3.7.3

==================

Bugs fixed
----------

* GH218 was ineffective in Python 3.

* GH222: ``lxml.html.submit_form()`` failed in Python 3.
Patch by Jakub Wilk.

3.7.2

==================

* GH220: ``xmlfile`` allows switching output methods at an element level.
Patch by Burak Arslan.

Bugs fixed
----------

* Work around installation problems in recent Python 2.7 versions
due to FTP download failures.

* GH219: ``xmlfile.element()`` was not properly quoting attribute values.
Patch by Burak Arslan.

* GH218: ``xmlfile.element()`` was not properly escaping text content of
script/style tags.  Patch by Burak Arslan.

3.7.1

==================

* No source changes, issued only to solve problems with the
binary packages released for 3.7.0.

3.7.0

==================

Features added
--------------

* GH217: ``XMLSyntaxError`` now behaves more like its ``SyntaxError``
baseclass.  Patch by Philipp A.

* GH216: ``HTMLParser()`` now supports the same ``collect_ids`` parameter
as ``XMLParser()``.  Patch by Burak Arslan.

* GH210: Allow specifying a serialisation method in ``xmlfile.write()``.
Patch by Burak Arslan.

* GH203: New option ``default_doctype`` in ``HTMLParser`` that allows
disabling the automatic doctype creation.  Patch by Shadab Zafar.

* GH201: Calling the method ``.set('attrname')`` without value argument
(or ``None``) on HTML elements creates an attribute without value that
serialises like ``<div attrname></div>``.  Patch by Daniel Holth.

* GH197: Ignore form input fields in ``form_values()`` when they are
marked as ``disabled`` in HTML.  Patch by Kristian Klemon.

Bugs fixed
----------

* GH206: File name and line number were missing from XSLT error messages.
Patch by Marcus Brinkmann.

Other changes
-------------

* Log entries no longer allow anything but plain string objects as message text
and file name.

* ``zlib`` is included in the list of statically built libraries.

3.6.4

==================

* GH204, LP1614693: build fix for MacOS-X.

3.6.3

==================

* LP1614603: change linker flags to build multi-linux wheels

3.6.2

==================

* LP1614603: release without source changes to provide cleanly built Linux wheels

3.6.1

==================

Features added
--------------

* GH180: Separate option ``inline_style`` for Cleaner that only removes ``style``
attributes instead of all styles.  Patch by Christian Pedersen.

* GH196: Windows build support for Python 3.5.  Contribution by Maximilian Hils.

Bugs fixed
----------

* GH199: Exclude ``file`` fields from ``FormElement.form_values`` (as browsers do).
Patch by Tomas Divis.

* GH198, LP1568167: Try to provide base URL from ``Resolver.resolve_string()``.
Patch by Michael van Tellingen.

* GH191: More accurate float serialisation in ``objectify.FloatElement``.
Patch by Holger Joukl.

* LP1551797: Repair XSLT error logging. Patch by Marcus Brinkmann.

3.6.0

==================

Features added
--------------

* GH187: Now supports (only) version 5.x and later of PyPy.
Patch by Armin Rigo.

* GH181: Direct support for ``.rnc`` files in `RelaxNG()` if ``rnc2rng``
is installed.  Patch by Dirkjan Ochtman.

Bugs fixed
----------

* GH189: Static builds honour FTP proxy configurations when downloading
the external libs.  Patch by Youhei Sakurai.

* GH186: Soupparser failed to process entities in Python 3.x.
Patch by Duncan Morris.

* GH185: Rare encoding related ``TypeError`` on import was fixed.
Patch by Petr Demin.

3.5.0

==================

Bugs fixed
----------

* Unicode string results failed XPath queries in PyPy.

* LP1497051: HTML target parser failed to terminate on exceptions
and continued parsing instead.

* Deprecated API usage in doctestcompare.

3.5.0b1

====================

Features added
--------------

* ``cleanup_namespaces()`` accepts a new argument ``keep_ns_prefixes``
that does not remove definitions of the provided prefix-namespace
mapping from the tree.

* ``cleanup_namespaces()`` accepts a new argument ``top_nsmap`` that
moves definitions of the provided prefix-namespace mapping to the
top of the tree.

* LP1490451: ``Element`` objects gained a ``cssselect()`` method as
known from ``lxml.html``.  Patch by Simon Sapin.

* API functions and methods behave and look more like Python functions,
which allows introspection on them etc.  One side effect to be aware of
is that the functions now bind as methods when assigned to a class
variable.  A quick fix is to wrap them in ``staticmethod()`` (as for
normal Python functions).

* ISO-Schematron support gained an option ``error_finder`` that allows
passing a filter function for picking validation errors from reports.

* LP1243600: Elements in ``lxml.html`` gained a ``classes`` property
that provides a set-like interface to the ``class`` attribute.
Original patch by masklinn.

* LP1341964: The soupparser now handles DOCTYPE declarations, comments
and processing instructions outside of the root element.
Patch by Olli Pottonen.

* LP1421512: The ``docinfo`` of a tree was made editable to allow
setting and removing the public ID and system ID of the DOCTYPE.
Patch by Olli Pottonen.

* LP1442427: More work-arounds for quirks and bugs in pypy and pypy3.

* ``lxml.html.soupparser`` now uses BeautifulSoup version 4 instead
of version 3 if available.

Bugs fixed
----------

* Memory errors that occur during tree adaptations (e.g. moving subtrees
to foreign documents) could leave the tree in a crash prone state.

* Calling ``process_children()`` in an XSLT extension element without
an ``output_parent`` argument failed with a ``TypeError``.
Fix by Jens Tröger.

* GH162: Image data in HTML ``data`` URLs is considered safe and
no longer removed by ``lxml.html.clean`` JavaScript cleaner.

* GH166: Static build could link libraries in wrong order.

* GH172: Rely a bit more on libxml2 for encoding detection rather than
rolling our own in some cases.  Patch by Olli Pottonen.

* GH159: Validity checks for names and string content were tightened
to detect the use of illegal characters early.  Patch by Olli Pottonen.

* LP1421921: Comments/PIs before the DOCTYPE declaration were not
serialised.  Patch by Olli Pottonen.

* LP659367: Some HTML DOCTYPE declarations were not serialised.
Patch by Olli Pottonen.

* LP1238503: lxml.doctestcompare is now consistent with stdlib's doctest
in how it uses ``+`` and ``-`` to refer to unexpected and missing output.

* Empty prefixes are explicitly rejected when a namespace mapping is used
with ElementPath to avoid hiding bugs in user code.

* Several problems with PyPy were fixed by switching to Cython 0.23.

3.4.4

==================

Bugs fixed
----------

* An ElementTree compatibility test added in lxml 3.4.3 that failed in
Python 3.4+ was removed again.

3.4.3

==================

Bugs fixed
----------

* Expression cache in ElementPath was ignored.  Fix by Changaco.

* LP1426868: Passing a default namespace and a prefixed namespace mapping
as nsmap into ``xmlfile.element()`` raised a ``TypeError``.

* LP1421927: DOCTYPE system URLs were incorrectly quoted when containing
double quotes.  Patch by Olli Pottonen.

* LP1419354: meta-redirect URLs were incorrectly processed by
``iterlinks()`` if preceded by whitespace.

3.4.2

==================

Bugs fixed
----------

* LP1415907: Crash when creating an XMLSchema from a non-root element
of an XML document.

* LP1369362: HTML cleaning failed when hitting processing instructions
with pseudo-attributes.

* ``CDATA()`` wrapped content was rejected for tail text.

* CDATA sections were not serialised as tail text of the top-level element.

3.4.1

==================

Features added
--------------

* New ``htmlfile`` HTML generator to accompany the incremental ``xmlfile``
serialisation API.  Patch by Burak Arslan.

Bugs fixed
----------

* ``lxml.sax.ElementTreeContentHandler`` did not initialise its superclass.

3.4.0

==================

Features added
--------------

* ``xmlfile(buffered=False)`` disables output buffering and flushes the
content after each API operation (starting/ending element blocks or writes).
A new method ``xf.flush()`` can alternatively be used to explicitly flush
the output.

* ``lxml.html.document_fromstring`` has a new option ``ensure_head_body=True``
which will add an empty head and/or body element to the result document if
missing.

* ``lxml.html.iterlinks`` now returns links inside meta refresh tags.

* New ``XMLParser`` option ``collect_ids=False`` to disable ID hash table
creation.  This can substantially speed up parsing of documents with many
different IDs that are not used.

* The parser uses per-document hash tables for XML IDs.  This reduces the
load of the global parser dict and speeds up parsing for documents with
many different IDs.

* ``ElementTree.getelementpath(element)`` returns a structural ElementPath
expression for the given element, which can be used for lookups later.

* ``xmlfile()`` accepts a new argument ``close=True`` to close file(-like)
objects after writing to them.  Before, ``xmlfile()`` only closed the file
if it had opened it internally.

* Allow "bytearray" type for ASCII text input.

Bugs fixed
----------

Other changes
-------------

* LP400588: decoding errors have become hard errors even in recovery mode.
Previously, they could lead to an internal tree representation in a mixed
encoding state, which lead to very late errors or even silently incorrect
behaviour during tree traversal or serialisation.

* Requires Python 2.6, 2.7, 3.2 or later. No longer supports
Python 2.4, 2.5 and 3.1, use lxml 3.3.x for those.

* Requires libxml2 2.7.0 or later and libxslt 1.1.23 or later,
use lxml 3.3.x with older versions.

3.3.6

==================

Bugs fixed
----------

* Prevent tree cycle creation when adding Elements as siblings.

* LP1361948: crash when deallocating Element siblings without parent.

* LP1354652: crash when traversing internally loaded documents in XSLT
extension functions.

3.3.5

==================

Bugs fixed
----------

* HTML cleaning could fail to strip javascript links that mix control
characters into the link scheme.

3.3.4

==================

Features added
--------------

* Source line numbers above 65535 are available on Elements when
using libxml2 2.9 or later.

Bugs fixed
----------

* ``lxml.html.fragment_fromstring()`` failed for bytes input in Py3.

Other changes
-------------

3.3.3

==================

Bugs fixed
----------

* LP1287118: Crash when using Element subtypes with ``__slots__``.

Other changes
-------------

* The internal classes ``_LogEntry`` and ``_Attrib`` can no longer be
subclassed from Python code.

3.3.2

==================

Bugs fixed
----------

* The properties ``resolvers`` and ``version``, as well as the methods
``set_element_class_lookup()`` and ``makeelement()``, were lost from
``iterparse`` objects in 3.3.0.

* LP1222132: instances of ``XMLSchema``, ``Schematron`` and ``RelaxNG``
did not clear their local ``error_log`` before running a validation.

* LP1238500: lxml.doctestcompare mixed up "expected" and "actual" in
attribute values.

* Some file I/O tests were failing in MS-Windows due to non-portable temp
file usage.  Initial patch by Gabi Davar.

* LP910014: duplicate IDs in a document were not reported by DTD validation.

* LP1185332: ``tostring(method="html")`` did not use HTML serialisation
semantics for trailing tail text.  Initial patch by Sylvain Viollon.

* LP1281139: ``.attrib`` value of Comments lost its mutation methods
in 3.3.0.  Even though it is empty and immutable, it should still
provide the same interface as that returned for Elements.

3.3.1

==================

Features added
--------------

Bugs fixed
----------

* LP1014290: HTML documents parsed with ``parser.feed()`` failed to find
elements during tag iteration.

* LP1273709: Building in PyPy failed due to missing support for
``PyUnicode_Compare()`` and ``PyByteArray_*()`` in PyPy's C-API.

* LP1274413: Compilation in MSVC failed due to missing "stdint.h" standard
header file.

* LP1274118: iterparse() failed to parse BOM prefixed files.

Other changes
-------------

3.3.0

==================

Features added
--------------

Bugs fixed
----------

* The heuristic that distinguishes file paths from URLs was tightened
to produce less false negatives.

Other changes
-------------

3.3.0beta5

=======================

Features added
--------------

* The PEP 393 unicode parsing support gained a fallback for wchar strings
which might still be somewhat common on Windows systems.

Bugs fixed
----------

* Several error handling problems were fixed throughout the code base that
could previously lead to exceptions being silently swallowed or not
properly reported.

* The C-API function ``appendChild()`` is now deprecated as it does not
propagate exceptions (its return type is ``void``).  The new function
``appendChildToElement()`` was added as a safe replacement.

* Passing a string into ``fromstringlist()`` raises an exception instead of
parsing the string character by character.

Other changes
-------------

* Document cleanup code was simplified using the new GC features in
Cython 0.20.

3.3.0beta4

=======================

Features added
--------------

Bugs fixed
----------

* The (empty) value returned by the ``attrib`` property of Entity and Comment
objects was mutable.

* Element class lookup wasn't available for the new pull parsers or when using
a custom parser target.

* Setting Element attributes on instantiation with both the ``attrib`` argument
and keyword arguments could modify the mapping passed as ``attrib``.

* LP1266171: DTDs instantiated from internal/external subsets (i.e. through
the docinfo property) lost their attribute declarations.

Other changes
-------------

* Built with Cython 0.20pre (gitrev 012ae82eb) to prepare support for
Python 3.4.

3.3.0beta3

=======================

Features added
--------------

* Unicode string parsing was optimised for Python 3.3 (PEP 393).

Bugs fixed
----------

* HTML parsing of Unicode strings could misdecode the input on some platforms.

* Crash in xmlfile() when closing open elements out of order in an error case.

Other changes
-------------

3.3.0beta2

=======================

Features added
--------------

* ``iterparse()`` supports the ``recover`` option.

Bugs fixed
----------

* Crash in ``iterparse()`` for HTML parsing.

* Crash in target parsing with attributes.

Other changes
-------------

* The safety check in the read-only tree implementation (e.g. used by
``PythonElementClassLookup``) raises a more appropriate ``ReferenceError``
for illegal access after tree disposal instead of an ``AssertionError``.
This should only impact test code that specifically checks the original
behaviour.

3.3.0beta1

=======================

Features added
--------------

* New option ``handle_failures`` in ``make_links_absolute()`` and
``resolve_base_href()`` (lxml.html) that enables ignoring or
discarding links that fail to parse as URLs.

* New parser classes ``XMLPullParser`` and ``HTMLPullParser`` for
incremental parsing, as implemented for ElementTree in Python 3.4.

* ``iterparse()`` enables recovery mode by default for HTML parsing
(``html=True``).

Bugs fixed
----------

* LP1255132: crash when trying to run validation over non-Element (e.g.
comment or PI).

* Error messages in the log and in exception messages that originated
from libxml2 could accidentally be picked up from preceding warnings
instead of the actual error.

* The ``ElementMaker`` in lxml.objectify did not accept a dict as
argument for adding attributes to the element it's building. This
works as in lxml.builder now.

* LP1228881: ``repr(XSLTAccessControl)`` failed in Python 3.

* Raise ``ValueError`` when trying to append an Element to itself or
to one of its own descendants, instead of running into an infinite
loop.

* LP1206077: htmldiff discarded whitespace from the output.

* Compressed plain-text serialisation to file-like objects was broken.

* lxml.html.formfill: Fix textarea form filling.
The textarea used to be cleared before the new content was set,
which removed the name attribute.


Other changes
-------------

* Some basic API classes use freelists internally for faster
instantiation.  This can speed up some ``iterparse()`` scenarios,
for example.

* ``iterparse()`` was rewritten to use the new ``*PullParser``
classes internally instead of being a parser itself.

3.2.5

==================

Features added
--------------

Bugs fixed
----------

* Crash in xmlfile() when closing open elements out of order in an error case.

* Crash in target parsing with attributes.

* LP1255132: crash when trying to run validation over non-Element (e.g.
comment or PI).

Other changes
-------------

3.2.4

==================

Features added
--------------

Bugs fixed
----------

* Memory leak when creating an XPath evaluator in a thread.

* LP1228881: ``repr(XSLTAccessControl)`` failed in Python 3.

* Raise ``ValueError`` when trying to append an Element to itself or
to one of its own descendants.

* LP1206077: htmldiff discarded whitespace from the output.

* Compressed plain-text serialisation to file-like objects was broken.

Other changes
-------------

3.2.3

==================

Bugs fixed
----------

* Fix support for Python 2.4 which was lost in 3.2.2.

3.2.2

==================

Features added
--------------

Bugs fixed
----------

* LP1185701: spurious XMLSyntaxError after finishing iterparse().

* Crash in lxml.objectify during xsi annotation.

Other changes
-------------

* Return values of user provided element class lookup methods are now
validated against the type of the XML node they represent to prevent
API class mismatches.

3.2.1

==================

Features added
--------------

* The methods ``apply_templates()`` and ``process_children()`` of XSLT
extension elements have gained two new boolean options ``elements_only``
and ``remove_blank_text`` that discard either all strings or whitespace-only
strings from the result list.

Bugs fixed
----------

* When moving Elements to another tree, the namespace cleanup mechanism
no longer drops namespace prefixes from attributes for which it finds
a default namespace declaration, to prevent them from appearing as
unnamespaced attributes after serialisation.

* Returning non-type objects from a custom class lookup method could lead
to a crash.

* Instantiating and using subtypes of Comments and ProcessingInstructions
crashed.

Other changes
-------------

3.2.0

==================

Features added
--------------

Bugs fixed
----------

* LP690319: Leading whitespace could change the behaviour of the string
parsing functions in ``lxml.html``.

* LP599318: The string parsing functions in ``lxml.html`` are more robust
in the face of uncommon HTML content like framesets or missing body tags.
Patch by Stefan Seelmann.

* LP712941: I/O errors while trying to access files with paths that contain
non-ASCII characters could raise ``UnicodeDecodeError`` instead of properly
reporting the ``IOError``.

* LP673205: Parsing from in-memory strings disabled network access in the
default parser and made subsequent attempts to parse from a URL fail.

* LP971754: lxml.html.clean appends 'nofollow' to 'rel' attributes instead
of overwriting the current value.

* LP715687: lxml.html.clean no longer discards scripts that are explicitly
allowed by the user provided whitelist.  Patch by Christine Koppelt.

Other changes
-------------

3.1.2

==================

Features added
--------------

Bugs fixed
----------

* LP1136509: Passing attributes through the namespace-unaware API of
the sax bridge (i.e. the ``handler.startElement()`` method) failed
with a ``TypeError``.  Patch by Mike Bayer.

* LP1123074: Fix serialisation error in XSLT output when converting
the result tree to a Unicode string.

* GH105: Replace illegal usage of ``xmlBufLength()`` in libxml2 2.9.0
by properly exported API function ``xmlBufUse()``.

Other changes
-------------

3.1.1

==================

Features added
--------------

Bugs fixed
----------

* LP1160386: Write access to ``lxml.html.FormElement.fields`` raised
an AttributeError in Py3.

* Illegal memory access during cleanup in incremental xmlfile writer.

Other changes
-------------

* The externally useless class ``lxml.etree._BaseParser`` was removed
from the module dict.

3.1.0

==================

Features added
--------------

* GH89: lxml.html.clean allows overriding the set of attributes that it
considers 'safe'.  Patch by Francis Devereux.

Bugs fixed
----------

* LP1104370: ``copy.copy(el.attrib)`` raised an exception.  It now returns
a copy of the attributes as a plain Python dict.

* GH95: When used with namespace prefixes, the  ``el.find*()`` methods
always used the first namespace mapping that was provided for each
path expression instead of using the one that was actually passed
in for the current run.

* LP1092521, GH91: Fix undefined C symbol in Python runtimes compiled
without threading support.  Patch by Ulrich Seidl.

Other changes
-------------

3.1beta1

=====================

Features added
--------------

* New build-time option ``--with-unicode-strings`` for Python 2 that
makes the API always return Unicode strings for names and text
instead of byte strings for plain ASCII content.

* New incremental XML file writing API ``etree.xmlfile()``.

* E factory in lxml.objectify is callable to simplify the creation of
tags with non-identifier names without having to resort to getattr().

Bugs fixed
----------

* When starting from a non-namespaced element in lxml.objectify, searching
for a child without explicitly specifying a namespace incorrectly found
namespaced elements with the requested local name, instead of restricting
the search to non-namespaced children.

* GH85: Deprecation warnings were fixed for Python 3.x.

* GH33: lxml.html.fromstring() failed to accept bytes input in Py3.

* LP1080792: Static build of libxml2 2.9.0 failed due to missing file.

Other changes
-------------

* The externally useless class ``_ObjectifyElementMakerCaller`` was
removed from the module API of lxml.objectify.

* LP1075622: lxml.builder is faster for adding text to elements with
many children.  Patch by Anders Hammarquist.

3.0.2

==================

Features added
--------------

Bugs fixed
----------

* Fix crash during interpreter shutdown by switching to Cython 0.17.3 for building.

Other changes
-------------

3.0.1

==================

Features added
--------------

Bugs fixed
----------

* LP1065924: Element proxies could disappear during garbage collection
in PyPy without proper cleanup.

* GH71: Failure to work with libxml2 2.6.x.

* LP1065139: static MacOS-X build failed in Py3.

Other changes
-------------

3.0

================

Features added
--------------

Bugs fixed
----------

* End-of-file handling was incorrect in iterparse() when reading from
a low-level C file stream and failed in libxml2 2.9.0 due to its
improved consistency checks.

Other changes
-------------

* The build no longer uses Cython by default unless the generated C files
are missing.  To use Cython, pass the option "--with-cython".  To ignore
the fatal build error when Cython is required but not available (e.g. to
run special setup.py commands that do not actually run a build), pass
"--without-cython".

3.0beta1

=====================

Features added
--------------

* Python level access to (optional) libxml2 memory debugging features
to simplify debugging of memory leaks etc.

Bugs fixed
----------

* Fix a memory leak in XPath by switching to Cython 0.17.1.

* Some tests were adapted to work with PyPy.

Other changes
-------------

* The code was adapted to work with the upcoming libxml2 2.9.0 release.

3.0alpha2

======================

Features added
--------------

* The ``.iter()`` method of elements now accepts ``tag`` arguments like
``"{*}name"`` to search for elements with a given local name in any
namespace. With this addition, all combinations of wildcards now work
as expected:
``"{ns}name"``, ``"{}name"``, ``"{*}name"``, ``"{ns}*"``, ``"{}*"``
and ``"{*}*"``.  Note that ``"name"`` is equivalent to ``"{}name"``,
but ``"*"`` is ``"{*}*"``.
The same change applies to the ``.getiterator()``, ``.itersiblings()``,
``.iterancestors()``, ``.iterdescendants()``, ``.iterchildren()``
and ``.itertext()`` methods;the ``strip_attributes()``,
``strip_elements()`` and ``strip_tags()`` functions as well as the
``iterparse()`` class.  Patch by Simon Sapin.

* C14N allows specifying the inclusive prefixes to be promoted
to top-level during exclusive serialisation.

Bugs fixed
----------

* Passing long Unicode strings into the ``feed()`` parser interface
failed to read the entire string.

Other changes
-------------

3.0alpha1

======================

Features added
--------------

* Initial support for building in PyPy (through cpyext).

* DTD objects gained an API that allows read access to their
declarations.

* ``xpathgrep.py`` gained support for parsing line-by-line (e.g.
from grep output) and for surrounding the output with a new root
tag.

* ``E-factory`` in ``lxml.builder`` accepts subtypes of known data
types (such as string subtypes) when building elements around them.

* Tree iteration and ``iterparse()`` with a selective ``tag``
argument supports passing a set of tags.  Tree nodes will be
returned by the iterators if they match any of the tags.

Bugs fixed
----------

* The ``.find*()`` methods in ``lxml.objectify`` no longer use XPath
internally, which makes them faster in many cases (especially when
short circuiting after a single or couple of elements) and fixes
some behavioural differences compared to ``lxml.etree``.  Note that
this means that they no longer support arbitrary XPath expressions
but only the subset that the ``ElementPath`` language supports.
The previous implementation was also redundant with the normal
XPath support, which can be used as a replacement.

* ``el.find('*')`` could accidentally return a comment or processing
instruction that happened to be in the wrong spot.  (Same for the
other ``.find*()`` methods.)

* The error logging is less intrusive and avoids a global setup where
possible.

* Fixed undefined names in html5lib parser.

* ``xpathgrep.py`` did not work in Python 3.

* ``Element.attrib.update()`` did not accept an ``attrib`` of
another Element as parameter.

* For subtypes of ``ElementBase`` that make the ``.text`` or ``.tail``
properties immutable (as in objectify, for example), inserting text
when creating Elements through the E-Factory feature of the class
constructor would fail with an exception, stating that the text
cannot be modified.

Other changes
--------------

* The code base was overhauled to properly use 'const' where the API
of libxml2 and libxslt requests it.  This also has an impact on the
public C-API of lxml itself, as defined in ``etreepublic.pxd``, as
well as the provided declarations in the ``lxml/includes/`` directory.
Code that uses these declarations may have to be adapted.  On the
plus side, this fixes several C compiler warnings, also for user
code, thus making it easier to spot real problems again.

* The functionality of "lxml.cssselect" was moved into a separate PyPI
package called "cssselect".  To continue using it, you must install
that package separately.  The "lxml.cssselect" module is still
available and provides the same interface, provided the "cssselect"
package can be imported at runtime.

* Element attributes passed in as an ``attrib`` dict or as keyword
arguments are now sorted by (namespaced) name before being created
to make their order predictable for serialisation and iteration.
Note that adding or deleting attributes afterwards does not take
that order into account, i.e. setting a new attribute appends it
after the existing ones.

* Several classes that are for internal use only were removed
from the ``lxml.etree`` module dict:
``_InputDocument, _ResolverRegistry, _ResolverContext, _BaseContext,
_ExsltRegExp, _IterparseContext, _TempStore, _ExceptionContext,
__ContentOnlyElement, _AttribIterator, _NamespaceRegistry,
_ClassNamespaceRegistry, _FunctionNamespaceRegistry,
_XPathFunctionNamespaceRegistry, _ParserDictionaryContext,
_FileReaderContext, _ParserContext, _PythonSaxParserTarget,
_TargetParserContext, _ReadOnlyProxy, _ReadOnlyPIProxy,
_ReadOnlyEntityProxy, _ReadOnlyElementProxy, _OpaqueNodeWrapper,
_OpaqueDocumentWrapper, _ModifyContentOnlyProxy,
_ModifyContentOnlyPIProxy, _ModifyContentOnlyEntityProxy,
_AppendOnlyElementProxy, _SaxParserContext, _FilelikeWriter,
_ParserSchemaValidationContext, _XPathContext,
_XSLTResolverContext, _XSLTContext, _XSLTQuotedStringParam``

* Several internal classes can no longer be inherited from:
``_InputDocument, _ResolverRegistry, _ExsltRegExp, _ElementUnicodeResult,
_IterparseContext, _TempStore, _AttribIterator, _ClassNamespaceRegistry,
_XPathFunctionNamespaceRegistry, _ParserDictionaryContext,
_FileReaderContext, _PythonSaxParserTarget, _TargetParserContext,
_ReadOnlyPIProxy, _ReadOnlyEntityProxy, _OpaqueDocumentWrapper,
_ModifyContentOnlyPIProxy, _ModifyContentOnlyEntityProxy,
_AppendOnlyElementProxy, _FilelikeWriter, _ParserSchemaValidationContext,
_XPathContext, _XSLTResolverContext, _XSLTContext, _XSLTQuotedStringParam,
_XSLTResultTree, _XSLTProcessingInstruction``

2.3.6

==================

Features added
--------------

Bugs fixed
----------

* Passing long Unicode strings into the ``feed()`` parser interface
failed to read the entire string.

Other changes
--------------

2.3.5

==================

Features added
--------------

Bugs fixed
----------

* Crash when merging text nodes in ``element.remove()``.

* Crash in sax/target parser when reporting empty doctype.

Other changes
--------------

2.3.4

==================

Features added
--------------

Bugs fixed
----------

* Crash when building an nsmap (Element property) with empty
namespace URIs.

* Crash due to race condition when errors (or user messages) occur
during threaded XSLT processing.

* XSLT stylesheet compilation could ignore compilation errors.

Other changes
--------------

2.3.3

==================

Features added
--------------

* ``lxml.html.tostring()`` gained new serialisation options
``with_tail`` and ``doctype``.

Bugs fixed
----------

* Fixed a crash when using ``iterparse()`` for HTML parsing and
requesting start events.

* Fixed parsing of more selectors in cssselect.  Whitespace before
pseudo-elements and pseudo-classes is significant as it is a
descendant combinator.
"E :pseudo" should parse the same as "E \*:pseudo", not "E:pseudo".
Patch by Simon Sapin.

* lxml.html.diff no longer raises an exception when hitting
'img' tags without 'src' attribute.

Other changes
--------------

2.3.2

==================

Features added
--------------

* ``lxml.objectify.deannotate()`` has a new boolean option
``cleanup_namespaces`` to remove the objectify namespace
declarations (and generally clean up the namespace declarations)
after removing the type annotations.

* ``lxml.objectify`` gained its own ``SubElement()`` function as a
copy of ``etree.SubElement`` to avoid an otherwise redundant import
of ``lxml.etree`` on the user side.

Bugs fixed
----------

* Fixed the "descendant" bug in cssselect a second time (after a first
fix in lxml 2.3.1).  The previous change resulted in a serious
performance regression for the XPath based evaluation of the
translated expression.  Note that this breaks the usage of some of
the generated XPath expressions as XSLT location paths that
previously worked in 2.3.1.

* Fixed parsing of some selectors in cssselect. Whitespace after combinators
">", "+" and "~" is now correctly ignored. Previously it was parsed as
a descendant combinator. For example, "div> .foo" was parsed the same as
"div>* .foo" instead of "div>.foo". Patch by Simon Sapin.

Other changes
--------------

2.3.1

==================

Features added
--------------

* New option ``kill_tags`` in ``lxml.html.clean`` to remove specific
tags and their content (i.e. their whole subtree).

* ``pi.get()`` and ``pi.attrib`` on processing instructions to parse
pseudo-attributes from the text content of processing instructions.

* ``lxml.get_include()`` returns a list of include paths that can be
used to compile external C code against lxml.etree.  This is
specifically required for statically linked lxml builds when code
needs to compile against the exact same header file versions as lxml
itself.

* ``Resolver.resolve_file()`` takes an additional option
``close_file`` that configures if the file(-like) object will be
closed after reading or not.  By default, the file will be closed,
as the user is not expected to keep a reference to it.

Bugs fixed
----------

* HTML cleaning didn't remove 'data:' links.

* The html5lib parser integration now uses the 'official'
implementation in html5lib itself, which