Re: [Jython-dev] PyServlet destroy, ns

[Robert W. B. <rb...@di...>]

> [Robert]
> >On an unrelated note, module global variables are common in the Jython
> >world.  Is there an advantage to PyServlet if each servlet has it's own
> >namespace to accomodate this?
>
> [Finn]
> > Well, is there an advantage? I considered it when I added PyServlet to
> > the repository, but for smallish apps, the application namespace is an
> > advantage IMHO. For larger apps, the application probably have to deal
> > with more controlled namespaces anyway.
>
> [Robert]
> >Very true about the namespaces.  I'm not sure about
> >advantages, but currently module globals work- until another PyServlet
> >tramples them.  This seemed in need of a fix.
>
> Each deployed PyServlet class have each own namespace (stored in
> PythonInterpreter.locals). It is the .py files stored under the same
> <context> that share the same globals namespace.

Sorry for the confusion.  As noted in my reply to David, my use of
PyServlet was wrong.  Before PyServlet, I used a java class called
PyMapping, and referred to the .py files as the PyServlets.  This lingo
has created the confusion.  Above, PyServlet was meant as .py file.

> >The choices looked like
> >disallowing globals, or adding individual namespaces,
>
> In addition to the shared globals, each .py servlet already have its own
> namespace available from within the doXXX() and service() methods as
> self.__dict__.

Yes, that's assumed.  The question was whether python users expect safe
module globals in Jython applications.  My guess was that they would, but
it sounds like I'm mistaken.

> >This depends on whether adding the unique namespaces was thought to be a
> >good thing to begin with.  If not, it doesn't matter.  If so,
> >implementing them looked like one of three options:
> >
> >1. an interpreter instance per PyServlet- overkill?
>
> Done already.

Again, meant as .py file - sorry.

> >2. extract and insert servlet and namespace into and from the interpreter
> >for each request- messy?
>
> The implementation wouldn't be messy. The interp.execfile() call in
> loadServlet should instead be done as a
>
>    PyObject ns = new PyStringMap()
>    __builtin__.execfile(path, ns, ns)
>
> The request handling will then automaticly use this namespace when the
> py service method is called from PyServlet.service().
>
> >3. execute the file, class, and servlet methods with Py.exec methods-
> >visually appealing, but not common practice, maybe for a
> >reason I thought :)
>
> Each Py.exec (and execfile) cause a classload. That should be avoided
> unless absolutely necessary.

Very useful to know- thank you.


-Robert