Changelog
---------

Version 1.0dev
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~

- Added limited support for the XInclude ``include`` directive. The
  implemented subset corresponds to the Genshi implementation, except
  Match-templates, which are not made available to the calling
  template. [malthe]

- Use a global template registry for templates on the
  file-system. This makes it inexpensive to have multiple template
  class instances pointing to the same file. [malthe]

- Reimplemented the disk cache to correctly restore all template
  data. This implementation keeps a cache in a pickled format in a
  file next to the original template. [malthe]

- Refactored compilation classes to better separate concerns. [malthe]

- Genshi macros (py:def) are now available globally. [malthe]

- A syntax error is now raised when an interpolation expression is not
  exhausted, e.g. only a part of the string is a valid
  Python-expression. [malthe]

- System variables are now defined in a configuration class. [malthe]

- Improve performance of codegen by not repeatedly calling
  an expensive "flatten" function. [chrism]

- Remove ``safe_render`` implementation detail.  It hid information
  in tracebacks. [chrism]

- Fixed a few more UnicodeDecodeErrors (test contributed by Wiggy).
  In particular, never upcast to unicode during transformation, and
  utf-8 encode Unicode attribute keys and values in Assign expressions
  (e.g. py:attrs). [chrism]

- Implemented TAL global defines. [malthe]

- Added support for variables with global scope. [malthe]

- Curly braces may now be omitted in an expression interpolation if
  the expression is just a variable name; this complies with the
  Genshi syntax. [malthe]

- Fixed off-by-one bug in interpolation routine. [malthe]

- The repeat-clause should not output tail with every iteration. [malthe]

- UTF-8 encode Unicode attribute literals. [chrism]

- Substantially reduced compiler overhead for lxml CDATA
  workaround. [malthe]

- Split out element compiler classes for Genshi and Zope language
  dialects. [malthe]

- Make lxml a setuptools "extra".  To install with lxml support
  (currently required by Genshi), specify "z3c.pt [lxml]" in
  any references you need to make to the package in buildout or
  in setup.py install_requires.  [chrism]

- Add test-nolxml and py-nolxml parts to buildout so the package's
  tests can be run without lxml.  [chrism]  

- CDATA sections are now correctly handled when using the
  ElementTree-parser. [malthe]

- Fixed bug in path-expressions where string instances would be
  (attempted) called. [malthe]

- No longer require default namespace. [malthe]

- Changed source code debug mode files to be named <filename>.py instead of
  <filename>.source.

- Generalized ElementTree-import to allow both Python 2.5's
  ``xml.etree`` module and the standalone ``ElementTree``
  package. [malthe]

- CDATA sections are now correctly preserved when using expression
  interpolation. [malthe]

- Expression results are now validated for XML correctness when the
  compiler is running in debug-mode. [malthe]

- Preserve CDATA sections. [malthe]

- The Genshi interpolation operator ${} should not have its result
  escaped when used in the text or tail regions. [malthe]

- Preliminary support for using ``xml.etree`` as fallback for
  ``lxml.etree``. [malthe]

- Fixed edge case bug where inserting both a numeric entity and a
  literal set of unicode bytes into the same document would cause a
  UnicodeDecodeError. See also
  http://groups.google.com/group/z3c_pt/browse_thread/thread/aea963d25a1778d0?hl=en
  [chrism]

- Static attributes are now properly overriden by py:attr-attributes.
  [malthe]
  
- String-expressions may now contain semi-colons using a double
  semi-colon literal (;;). [malthe]

- Get rid of package-relative magic in constructor of BaseTemplateFile
  in favor of just requiring an absolute path or a path relative
  to getcwd(). Rationale: it didn't work when called from __main__
  when the template was relative to getcwd(), which is the 99% case
  for people first trying it out. [chrism]

- Added support for METAL.
  [malthe]

- Add a TemplateLoader class to have a convenient method to instantiate
  templates. This is similar to the template loaders from other template
  toolkits and makes integration with Pylons a lot simpler.
  [wichert]

- Switch from hardcoding all options in config.py to using parameters
  for the template. This also allows us to use the more logical
  auto_reload flag instead of reusing PROD_MODE, which is also used
  for other purposes.
  [wichert]

- Treat comments, processing instructions, and named entities in the
  source template as "literals", which will be rendered into the
  output unchanged. [chrism]

Version 0.9 - August 7, 2008
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~

- Added support for Genshi-templates.
  [malthe]

- Cleanup and refactoring of translation module.
  [malthe]
  
- If the template source contains a DOCTYPE declaration, output it
  during rendering. [chrism]

- Fixed an error where numeric entities specified in text or tail
  portions of elements would cause a UnicodeDecodeError to be raised
  on systems configured with an 'ascii' default encoding. [chrism]

- Refactored file system based cache a bit and added a simple benchmark for
  the cache. The initial load speed for a template goes down significantly
  with the cache. Compared to zope.pagetemplate we are only 3x slower,
  compared to 50x slower when cooking each template on process startup.

- Got rid entirely of the _escape function and inlined the actual code
  instead. We go up again to 12x for path and 19x for Python expressions :)
  [hannosch]

- Avoid string concatenation and use multiple write statements instead. These
  are faster now, since we use a list append internally.
  [hannosch]

- Inline the _escape function, because function calls are expensive in Python.
  Added missing escaping for Unicode values.
  [fschulze, hannosch]

- When templates are instantiated outside of a class-definition, a
  relative file path will be made absolute using the module path.
  [malthe]

- Simplified the _escape function handling by pulling in the str call into the
  function. Corrected the bigtable hotshot test to only benchmark rendering.

- Replaced the cgi.escape function by an optimized local version, we go up
  to 11x for path and 16x for Python expressions :) In the bigtable benchmark
  the enhancement is more noticable - we are the same speed as spitfire -O1
  templates now and just half the speed of -O3 :))

- Added a new benchmark test called bigtable that produces results which are
  directly comparable to those produced by the bigtable.py benchmark in the
  spitfire project.

- Introduce a new config option called `Z3C_PT_DISABLE_I18N`. If this
  environment variable is set to `true`, the template engine will not call
  into the zope.i18n machinery anymore, but fall back to simple interpolation
  in all cases. In a normal Zope environment that has the whole i18n
  infrastructure set up, this will render the templates about 15x faster than
  normal TAL, instead of only 10x faster at this point.

- Removed the `second rendering` tests from the benchmark suite. Since we
  enable the file cache for the benchmarks, there's no difference between the
  first and second rendering anymore after the cache file has been written.

- Require zope.i18n 3.5 and add support for using its new negotiate function.
  If you use the `zope_i18n_allowed_languages` environment variable the target
  language for a template is only negotiated once per template, instead of
  once for each translate function call. This more than doubles the speed
  and the benchmark is back at 9.2 times faster.

- Extended the i18n handling to respect the passed in translation context to
  the template. Usually this is the request, which is passed on under the
  internal name of `_context` into the render functions. After extending the
  i18n tests to include a negotiator and message catalog the improvement is
  only at 4.5 anymore, as most of the time is spent inside the i18n machinery.

- Added persistent file cache functionality. If the environment variable is
  set, each file system based template will add a directory to the cache
  (currently a SHA-1 of the file's absolute path is used as the folder name)
  and in the folder one file per params for the template (cache filename is
  the hash of the params). Once a template file is initialized, an instance
  local registry is added, which then looks up all cached files and
  pre-populates the registry with the render functions.

- Fixed interpolation edge case bugs.
  [malthe]

- Added new `Z3C_PT_FILECACHE` environment variable pointing to a directory.
  If set, this will be used to cache the compiled files.

- Added a second variation of the repeat clause, using a simple for loop. It
  doesn't support the repeatdict, though and is therefor not used yet. Also
  began work to add introspection facilities to clauses about the variables
  being used in them. The simpler loop causes the benchmarks to go up to a
  10.5 (old 9.5) for path expressions and 14.5 (12.5) for python expressions.
  So the next step is to introduce an optimization phase, that can decide
  which variant of the loops to use.

- Made the debug mode independent from the Python debug mode. You can now
  specify an environment variable called `Z3C_PT_DEBUG` to enable it.

- Added some code in a filecache module that can later be used to write out
  and reload the compiled Python code to and from the file system. We should
  be able to avoid reparsing on Python process restart.

- Simplified the generated _escape code. cgi.escape's second argument is a
  simple boolean and not a list of characters to quote.

- Use a simple list based BufferIO class instead of a cStringIO for the out
  stream. Avoiding the need to encode Unicode data is a bigger win. We do
  not support arbitrarily mixing of Unicode and non-ascii inside the engine.

- Merged two adjacent writes into one inside the Tag clause.

- Applied a bunch of micro-optimizations. ''.join({}) is slightly faster
  than ''.join({}.keys()) and does the same. Avoid a try/except for error
  handling in non-debug mode. Test against 'is None' instead of a boolean
  check for the result of the template registry lookup. Made PROD_MODE
  available defined as 'not DEBUG_MODE' in config.py, so we avoid the 'not'
  in every cook-check.

- Added more benchmark tests for the file variants.

- Optimized 'is None' handling in Tag clause similar to the Write clause.

- Made the _out.write method directly available as _write in all scopes, so
  we avoid the method lookup call each time.

- Optimized 'is None' handling in Write clause.

- Slightly refactored benchmark tests and added tests for the file variants.

- In debug mode the actual source code for file templates is written out to
  a <filename>.source file, to make it easier to inspect it.

- Make debug mode setting explicit in a config.py. Currently it is bound to
  Python's __debug__, which is False when run with -O and otherwise True.

- Use a simplified UnicodeWrite clause for the result of _translate calls,
  as the result value is guaranteed to be Unicode.

- Added benchmark tests for i18n handling.

- Added more tests for i18n attributes handling.

- Don't generate empty mappings for expressions with a trailing semicolon.

- Fixed undefined name 'static' error in i18n attributes handling and added
  quoting to i18n attributes.

- Added condition to the valid attributes on tags in the tal namespace.

- Made sure the traceback from the *first* template exception
  is carried over to __traceback_info__

- Added template source annotations on exceptions raised while
  rendering a template.

Version 0.8 - March 19, 2008
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~

- Added support for 'nocall' and 'not' (for path-expressions).

- Added support for path- and string-expressions.

- Abstracted expression translation engine. Expression implementations
  are now pluggable. Expression name pragmas are supported throughout.

- Formalized expression types

- Added support for 'structure'-keyword for replace and content.

- Result of 'replace' and 'content' is now escaped by default.

- Benchmark is now built as a custom testrunner

Version 0.7 - March 10, 2008
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~

- Added support for comments; expressions are allowed
  inside comments, i.e.

     <!-- ${'Hello World!'} -->

  Comments are always included.
  
Version 0.6 - February 24, 2008
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~

- Added support for text templates; these allow expression
  interpolation in non-XML documents like CSS stylesheets and
  javascript files.

Version 0.5 - February 23, 2008
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~

- Expression interpolation implemented.

Version 0.4 - February 22, 2008
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~

- Engine now uses cStringIO yielding a 2.5x performance
  improvement. Unicode is now handled correctly.

Version 0.3 - December 23, 2007
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~

- Code optimization; bug fixing spree  

- Added ``ViewPageTemplateFile`` class
  
- Added support for i18n
  
- Engine rewrite; improved code generation abstractions 
  
Version 0.2 - December 5, 2007
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~

- Major optimizations to the generated code  

Version 0.1 - December 3, 2007
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~

- First public release

