==================
Developing Loaders
==================

If you want to use jinja in a bigger application and you want to change
the way jinja looks up templates the best way is to write a loader.

This allows you for example to load different template sets depending on
the setting of the user etc.


BaseLoader
==========

In ``jinja.loader`` is a class called ``BaseLoader`` you can use to implement
your own loader. A common loader looks like this::

    class Loader(object):

        def load(self, name, parent=None):
            """This method isn't allowed to cache the data"""
            raise NotImplementedError()

        def load_and_compile(self, name, lib=None, parent=None):
            """Get's called when the template requires an nodelist
            to render on."""
            template = self.load(name, parent)
            lexer = Lexer(template)
            parser = Parser(lexer.tokenize(), self, lib)
            return parser.parse()

        def load_and_compile_uncached(self, name, lib=None, parent=None):
            """Get's called for the extends tag to get a fresh
            nodelist to manipulate."""
            return self.load_and_compile(name, lib, parent)

``load`` has to return a unicode object with the template source in.
``load_and_compile`` and ``load_and_compile_uncached`` has to return
parsed nodelists. ``name`` is the name of the template the user
wants the loader to import. ``parent`` is the name of the template
the loader gets triggered.

Normally this is ``None``. But when you load a template using
``{% extends %}`` or ``{% include %}`` ``parent`` is the name of the
template containing the tag.

``load_and_compile``/``load_and_compile_uncached`` take another
argument called ``lib`` which is the reference of the template lib
the parser should use. If ``lib`` is ``None`` the parser will use
the default lib ``jinja.lib.stdlib``.


load_and_compile vs load_and_compile_uncached
=============================================

Since jinja 0.8 there are two different loader functions. A cached one
and an uncached one. The uncached one gets called when the template
requires a fresh copy of a nodelist which it can use to modify and
pickle afterwards. In case of the stdlib these are the ``extends`` and
``include`` tags.

The normal ``cached`` version doesn't have to cache. For example the
``FileSystemLoader`` doesn't provide caching. Because of this
the ``load_and_compile_uncached`` method automatically calls the
``load_and_compile`` method with the given arguments when you inherit
from the ``BaseLoader`` and don't overwrite ``BaseLoader``.