Updated to a newer version of select_trigger.py It seems to work OK with win32 now that Jim changed the way ZServer shutsdown medusa. Also this new select trigger should resolve the sort of permissions problems reported by Alex Rice since it uses a pipe, not a socket file in the /tmp directory.

local timezone name does not appear in the set of recognized time zone
names.

the compiler was in pickly mode.

Got rid of the fatal error check in the module initialization.

Added some needed variable initializations.

Relabeled 'manage_security' from 'Security' to 'Define Permissions'. This makes things clearer and consistent with a related change in App/Management.py

Changed tabs labels so the 'manage_access' is labeled as 'Define Permissions' not 'Security' when the object is being used as a method.

  <dtml-var foo>

Added an entity reference syntax for var:

  &dtml-foo;

As in:

  <a HREF="&dtml-URL1;/manage_workspace">

Added a -D option to make it easier to spell debug mode.

Made enivorment variable settings work with the http server. So now you can do things like ./z2.py -w 8888 SCRIPT_NAME=Zope and actually have the specified SCRIPT_NAME override the normal SCRIPT_NAME. Useful for keeping ZServer behind a proxy.

to SSI syntax, you can now use ordinary HTML syntax, as in:

  <dtml.in objectIds>
     <dtml.var sequence-item>
  </dtml.in>

Both the new syntax and the SSI syntax are allowed in the same document.
In fact, if you were really twisted, you could:

  <dtml.in objectIds>
     <dtml.var sequence-item>
  <!--#/in-->

;)

Also added a missing type initialization for the BTreeItems internal
type.

reference counts of extension classes are increased when instances are
created.  Prior to this change, the class ref counts were only
decreased by subclass destructors.  Base class destructors followed
the Python protocol and didn't DECREF thier classes (types).  (Python
type ref counts have nothing to do with instance life times, since
Python types are static, unlike extension classes, which may be
dynamic.)

This caused the ref counts of instantiatable base classes to grow,
which looked like a memory leak.

Now, the destructor protocol calls for base class destructors to
DECREF the class in addition to deallocating the object.  The subclass
destructor *only* DECREFs the class if it has to deallocate the object
because there was no base class destructor.

This also fixed an logic ordering problem that caused some dynamically
created classes to be destroyed while they were still needed.

It actually seems to work, though there's still a fair amount missing.

raised in external methods.  We're more careful about seeking
standard_error_message - check for existence in the client object, and
if it's not there try aq_parent - and fallback to the previously
resolved error_value, before the canned "Sorry, an error occurred".

form data. The change also eliminates some unneeded
computation.