Upgrading code from older versions of Quixote
=============================================

This document lists backward-incompatible changes in Quixote, and
explains how to update application code to work with the newer
version.

Changes from 0.6.1 to 1.0
-------------------------

Sessions
********

A leading underscore was removed from the ``Session`` attributes
``__remote_address``, ``__creation_time``, and ``__access_time``.  If
you have pickled ``Session`` objects you will need to upgrade them
somehow.  Our preferred method is to write a script that unpickles each
object, renames the attributes and then re-pickles it.



Changes from 0.6 to 0.6.1
-------------------------

``_q_exception_handler`` now called if exception while traversing
*****************************************************************

``_q_exception_handler`` hooks will now be called if an exception is
raised during the traversal process.  Quixote 0.6 had a bug that caused
``_q_exception_handler`` hooks to only be called if an exception was
raised after the traversal completed.



Changes from 0.5 to 0.6
-----------------------

``_q_getname`` renamed to ``_q_lookup``
***************************************

The ``_q_getname`` special function was renamed to ``_q_lookup``,
because that name gives a clearer impression of the function's
purpose.  In 0.6, ``_q_getname`` still works but will trigger a
warning.


Form Framework Changes
**********************

The ``quixote.form.form`` module was changed from a .ptl file to a .py
file.  You should delete or move the existing ``quixote/`` directory
in ``site-packages`` before running ``setup.py``, or at least delete
the old ``form.ptl`` and ``form.ptlc`` files.

The widget and form classes in the ``quixote.form`` package now return
``htmltext`` instances.  Applications that use forms and widgets will
likely have to be changed to use the ``[html]`` template type to avoid
over-escaping of HTML special characters.

Also, the constructor arguments to ``SelectWidget`` and its subclasses have
changed.  This only affects applications that use the form framework
located in the ``quixote.form`` package.

In Quixote 0.5, the ``SelectWidget`` constructor had this signature::

     def __init__ (self, name, value=None,
                   allowed_values=None,
                   descriptions=None,
                   size=None,
                   sort=0):

``allowed_values`` was the list of objects that the user could choose,
and ``descriptions`` was a list of strings that would actually be
shown to the user in the generated HTML.

In Quixote 0.6, the signature has changed slightly::

     def __init__ (self, name, value=None,
                   allowed_values=None,
                   descriptions=None,
                   options=None,
                   size=None,
                   sort=0):

The ``quote`` argument is gone, and the ``options`` argument has been 
added.  If an ``options`` argument is provided, ``allowed_values`` 
and ``descriptions`` must not be supplied.

The ``options`` argument, if present, must be a list of tuples with
1,2, or 3 elements, of the form ``(value:any, description:any,
key:string)``.

    * ``value`` is the object that will be returned if the user chooses 
      this item, and must always be supplied.
  
    * ``description`` is a string or htmltext instance which will be
      shown to the user in the generated HTML.  It will be passed
      through the htmlescape() functions, so for an ordinary string
      special characters such as '&' will be converted to '&'.
      htmltext instances will be left as they are.

    * If supplied, ``key`` will be used in the value attribute 
      of the option element (``<option value="...">``).  
      If not supplied, keys will be generated; ``value`` is checked for a 
      ``_p_oid`` attribute and if present, that string is used; 
      otherwise the description is used.

In the common case, most applications won't have to change anything,
though the ordering of selection items may change due to the
difference in how keys are generated.


File Upload Changes
*******************

Quixote 0.6 introduces new support for HTTP upload requests.  Any HTTP
request with a Content-Type of "multipart/form-data" -- which is
generally only used for uploads -- is now represented by
HTTPUploadRequest, a subclass of HTTPRequest, and the uploaded files
themselves are represented by Upload objects.  

Whenever an HTTP request has a Content-Type of "multipart/form-data",
an instance of HTTPUploadRequest is created instead of HTTPRequest.
Some of the fields in the request are presumably uploaded files and
might be quite large, so HTTPUploadRequest will read all of the fields
supplied in the request body and write them out to temporary files;
the temporary files are written in the directory specified by the
UPLOAD_DIR configuration variable.

Once the temporary files have been written, the HTTPUploadRequest
object is passed to a function or PTL template, just like an ordinary
request.  The difference between HTTPRequest and HTTPUploadRequest 
is that all of the form variables are represented as Upload objects.
Upload objects have three attributes:

``orig_filename``
  the filename supplied by the browser.
``base_filename``
  a stripped-down version of orig_filename with unsafe characters removed.
  This could be used when writing uploaded data to a permanent location.
``tmp_filename``
  the path of the temporary file containing the uploaded data for this field.

Consult upload.txt for more information about handling file uploads.
 

Refactored `Publisher` Class
****************************

Various methods in the `Publisher` class were rearranged.  If your
application subclasses Publisher, you may need to change your code
accordingly.

  * ``parse_request()`` no longer creates the HTTPRequest object; 
    instead a new method, ``create_request()``,  handles this, 
    and can be overridden as required.

    As a result, the method signature has changed from 
    ``parse_request(stdin, env)`` to ``parse_request(request)``.

  * The ``Publisher.publish()`` method now catches exceptions raised 
    by ``parse_request()``.


Changes from 0.4 to 0.5
-----------------------

Session Management Changes
**************************

The Quixote session management interface underwent lots of change and
cleanup with Quixote 0.5.  It was previously undocumented (apart from
docstrings in the code), so we thought that this was a good opportunity
to clean up the interface.  Nevertheless, those brave souls who got
session management working just by reading the code are in for a bit of
suffering; this brief note should help clarify things.  The definitive
documentation for session management is session-mgmt.txt -- you should
start there.


Attribute renamings and pickled objects
+++++++++++++++++++++++++++++++++++++++

Most attributes of the standard Session class were made private in order
to reduce collisions with subclasses.  The downside is that pickled
Session objects will break.  You might want to (temporarily) modify
session.py and add this method to Session::

    def __setstate__ (self, dict):
        # Update for attribute renamings made in rev. 1.51.2.3
        # (between Quixote 0.4.7 and 0.5).
        self.__dict__.update(dict)
        if hasattr(self, 'remote_address'):
            self.__remote_address = self.remote_address
            del self.remote_address
        if hasattr(self, 'creation_time'):
            self.__creation_time = self.creation_time
            del self.creation_time
        if hasattr(self, 'access_time'):
            self.__access_time = self.access_time
            del self.access_time
        if hasattr(self, 'form_tokens'):
            self._form_tokens = self.form_tokens
            del self.form_tokens

However, if your sessions were pickled via ZODB, this may not work.  (It
didn't work for us.)  In that case, you'll have to add something like
this to your class that inherits from both ZODB's Persistent and
Quixote's Session::

    def __setstate__ (self, dict):
        # Blechhh!  This doesn't work if I put it in Quixote's
        # session.py, so I have to second-guess how Python
        # treats "__" attribute names.
        self.__dict__.update(dict)
        if hasattr(self, 'remote_address'):
            self._Session__remote_address = self.remote_address
            del self.remote_address
        if hasattr(self, 'creation_time'):
            self._Session__creation_time = self.creation_time
            del self.creation_time
        if hasattr(self, 'access_time'):
            self._Session__access_time = self.access_time
            del self.access_time
        if hasattr(self, 'form_tokens'):
            self._form_tokens = self.form_tokens
            del self.form_tokens

It's not pretty, but it worked for us.


Cookie domains and paths
++++++++++++++++++++++++

The session cookie config variables -- ``COOKIE_NAME``,
``COOKIE_DOMAIN``, and ``COOKIE_PATH`` -- have been renamed to
``SESSION_COOKIE_*`` for clarity.

If you previously set the config variable ``COOKIE_DOMAIN`` to the name
of your server, this is most likely no longer necessary -- it's now fine
to leave ``SESSION_COOKIE_DOMAIN`` unset (ie. ``None``), which
ultimately means browsers will only include the session cookie in
requests to the same server that sent it to them in the first place.

If you previously set ``COOKIE_PATH``, then you should probably preserve
your setting as ``SESSION_COOKIE_PATH``.  The default of ``None`` means
that browsers will only send session cookies with requests for URIs
under the URI that originally resulted in the session cookie being sent.
See session-mgmt.txt and RFCs 2109 and 2965.

If you previously set ``COOKIE_NAME``, change it to
``SESSION_COOKIE_NAME``.


$Id: upgrading.txt 23961 2004-04-12 15:51:38Z nascheme $