Using Hooks

"Hooks" can be used to influence the behavior of the Pyramid framework in various ways.

Changing the Not Found View

When Pyramid can't map a URL to view code, it invokes a Not Found View, which is a view callable. The default Not Found View can be overridden through application configuration.

If your application uses imperative configuration, you can replace the Not Found View by using the pyramid.config.Configurator.add_notfound_view() method:

1def notfound(request):
2    return Response('Not Found', status='404 Not Found')
3
4def main(globals, **settings):
5    config = Configurator()
6    config.add_notfound_view(notfound)

The Not Found View callable is a view callable like any other.

If your application instead uses pyramid.view.view_config decorators and a scan, you can replace the Not Found View by using the pyramid.view.notfound_view_config decorator:

1from pyramid.view import notfound_view_config
2
3@notfound_view_config()
4def notfound(request):
5    return Response('Not Found', status='404 Not Found')
6
7def main(globals, **settings):
8    config = Configurator()
9    config.scan()

This does exactly what the imperative example above showed.

Your application can define multiple Not Found Views if necessary. Both pyramid.config.Configurator.add_notfound_view() and pyramid.view.notfound_view_config take most of the same arguments as pyramid.config.Configurator.add_view and pyramid.view.view_config, respectively. This means that Not Found Views can carry predicates limiting their applicability. For example:

 1from pyramid.view import notfound_view_config
 2
 3@notfound_view_config(request_method='GET')
 4def notfound_get(request):
 5    return Response('Not Found during GET', status='404 Not Found')
 6
 7@notfound_view_config(request_method='POST')
 8def notfound_post(request):
 9    return Response('Not Found during POST', status='404 Not Found')
10
11def main(globals, **settings):
12    config = Configurator()
13    config.scan()

The notfound_get view will be called when a view could not be found and the request method was GET. The notfound_post view will be called when a view could not be found and the request method was POST.

Like any other view, the Not Found View must accept at least a request parameter, or both context and request. The request is the current request representing the denied action. The context (if used in the call signature) will be the instance of the HTTPNotFound exception that caused the view to be called.

Both pyramid.config.Configurator.add_notfound_view() and pyramid.view.notfound_view_config can be used to automatically redirect requests to slash-appended routes. See Redirecting to Slash-Appended Routes for examples.

Here's some sample code that implements a minimal Not Found View callable:

1from pyramid.httpexceptions import HTTPNotFound
2
3def notfound(request):
4    return HTTPNotFound()

Note

When a Not Found View callable is invoked, it is passed a request. The exception attribute of the request will be an instance of the HTTPNotFound exception that caused the Not Found View to be called. The value of request.exception.message will be a value explaining why the Not Found exception was raised. This message has different values depending on whether the pyramid.debug_notfound environment setting is true or false.

Note

When a Not Found View callable accepts an argument list as described in Alternate View Callable Argument/Calling Conventions, the context passed as the first argument to the view callable will be the HTTPNotFound exception instance. If available, the resource context will still be available as request.context.

Warning

The Not Found View callables are only invoked when a HTTPNotFound exception is raised. If the exception is returned from a view then it will be treated as a regular response object and it will not trigger the custom view.

Changing the Forbidden View

When Pyramid can't authorize execution of a view based on the authorization policy in use, it invokes a forbidden view. The default forbidden response has a 403 status code and is very plain, but the view which generates it can be overridden as necessary.

The forbidden view callable is a view callable like any other. The view configuration which causes it to be a "forbidden" view consists of using the pyramid.config.Configurator.add_forbidden_view() API or the pyramid.view.forbidden_view_config decorator.

For example, you can add a forbidden view by using the pyramid.config.Configurator.add_forbidden_view() method to register a forbidden view:

1def forbidden(request):
2    return Response('forbidden')
3
4def main(globals, **settings):
5    config = Configurator()
6    config.add_forbidden_view(forbidden)

If instead you prefer to use decorators and a scan, you can use the pyramid.view.forbidden_view_config decorator to mark a view callable as a forbidden view:

1from pyramid.view import forbidden_view_config
2
3@forbidden_view_config()
4def forbidden(request):
5    return Response('forbidden')
6
7def main(globals, **settings):
8    config = Configurator()
9    config.scan()

Like any other view, the forbidden view must accept at least a request parameter, or both context and request. If a forbidden view callable accepts both context and request, the HTTP Exception is passed as context. The context as found by the router when the view was denied (which you normally would expect) is available as request.context. The request is the current request representing the denied action.

Here's some sample code that implements a minimal forbidden view:

1from pyramid.view import view_config
2from pyramid.response import Response
3
4def forbidden_view(request):
5    return Response('forbidden')

Note

When a forbidden view callable is invoked, it is passed a request. The exception attribute of the request will be an instance of the HTTPForbidden exception that caused the forbidden view to be called. The value of request.exception.message will be a value explaining why the forbidden exception was raised, and request.exception.result will be extended information about the forbidden exception. These messages have different values depending on whether the pyramid.debug_authorization environment setting is true or false.

Warning

The forbidden view callables are only invoked when a HTTPForbidden exception is raised. If the exception is returned from a view then it will be treated as a regular response object and it will not trigger the custom view.

Changing the Request Factory

Whenever Pyramid handles a request from a WSGI server, it creates a request object based on the WSGI environment it has been passed. By default, an instance of the pyramid.request.Request class is created to represent the request object.

The class (a.k.a., "factory") that Pyramid uses to create a request object instance can be changed by passing a request_factory argument to the constructor of the configurator. This argument can be either a callable or a dotted Python name representing a callable.

1from pyramid.request import Request
2
3class MyRequest(Request):
4    pass
5
6config = Configurator(request_factory=MyRequest)

If you're doing imperative configuration, and you'd rather do it after you've already constructed a configurator, it can also be registered via the pyramid.config.Configurator.set_request_factory() method:

1from pyramid.config import Configurator
2from pyramid.request import Request
3
4class MyRequest(Request):
5    pass
6
7config = Configurator()
8config.set_request_factory(MyRequest)

Adding Methods or Properties to a Request Object

New in version 1.4.

Since each Pyramid application can only have one request factory, changing the request factory is not that extensible, especially if you want to build composable features (e.g., Pyramid add-ons and plugins).

A lazy property can be registered to the request object via the pyramid.config.Configurator.add_request_method() API. This allows you to specify a callable that will be available on the request object, but will not actually execute the function until accessed.

Warning

This will silently override methods and properties from request factory that have the same name.

 1from pyramid.config import Configurator
 2
 3def total(request, *args):
 4    return sum(args)
 5
 6def prop(request):
 7    print("getting the property")
 8    return "the property"
 9
10config = Configurator()
11config.add_request_method(total)
12config.add_request_method(prop, reify=True)

In the above example, total is added as a method. However, prop is added as a property and its result is cached per-request by setting reify=True. This way, we eliminate the overhead of running the function multiple times.

>>> request.total(1, 2, 3)
6
>>> request.prop
getting the property
'the property'
>>> request.prop
'the property'

To not cache the result of request.prop, set property=True instead of reify=True.

Here is an example of passing a class to Configurator.add_request_method:

 1from pyramid.config import Configurator
 2from pyramid.decorator import reify
 3
 4class ExtraStuff(object):
 5
 6    def __init__(self, request):
 7        self.request = request
 8
 9    def total(self, *args):
10        return sum(args)
11
12    # use @property if you don't want to cache the result
13    @reify
14    def prop(self):
15        print("getting the property")
16        return "the property"
17
18config = Configurator()
19config.add_request_method(ExtraStuff, 'extra', reify=True)

We attach and cache an object named extra to the request object.

>>> request.extra.total(1, 2, 3)
6
>>> request.extra.prop
getting the property
'the property'
>>> request.extra.prop
'the property'

Changing the Response Factory

New in version 1.6.

Whenever Pyramid returns a response from a view, it creates a response object. By default, an instance of the pyramid.response.Response class is created to represent the response object.

The factory that Pyramid uses to create a response object instance can be changed by passing a pyramid.interfaces.IResponseFactory argument to the constructor of the configurator. This argument can be either a callable or a dotted Python name representing a callable.

The factory takes a single positional argument, which is a Request object. The argument may be None.

1from pyramid.response import Response
2
3class MyResponse(Response):
4    pass
5
6config = Configurator(response_factory=lambda r: MyResponse())

If you're doing imperative configuration and you'd rather do it after you've already constructed a configurator, it can also be registered via the pyramid.config.Configurator.set_response_factory() method:

1from pyramid.config import Configurator
2from pyramid.response import Response
3
4class MyResponse(Response):
5    pass
6
7config = Configurator()
8config.set_response_factory(lambda r: MyResponse())

Using the Before Render Event

Subscribers to the pyramid.events.BeforeRender event may introspect and modify the set of renderer globals before they are passed to a renderer. This event object iself has a dictionary-like interface that can be used for this purpose. For example:

1from pyramid.events import subscriber
2from pyramid.events import BeforeRender
3
4@subscriber(BeforeRender)
5def add_global(event):
6    event['mykey'] = 'foo'

An object of this type is sent as an event just before a renderer is invoked.

If a subscriber attempts to add a key that already exists in the renderer globals dictionary, a KeyError is raised. This limitation is enforced because event subscribers do not possess any relative ordering. The set of keys added to the renderer globals dictionary by all pyramid.events.BeforeRender subscribers and renderer globals factories must be unique.

The dictionary returned from the view is accessible through the rendering_val attribute of a BeforeRender event.

Suppose you return {'mykey': 'somevalue', 'mykey2': 'somevalue2'} from your view callable, like so:

1from pyramid.view import view_config
2
3@view_config(renderer='some_renderer')
4def myview(request):
5    return {'mykey': 'somevalue', 'mykey2': 'somevalue2'}

rendering_val can be used to access these values from the BeforeRender object:

1from pyramid.events import subscriber
2from pyramid.events import BeforeRender
3
4@subscriber(BeforeRender)
5def read_return(event):
6    # {'mykey': 'somevalue'} is returned from the view
7    print(event.rendering_val['mykey'])

See the API documentation for the BeforeRender event interface at pyramid.interfaces.IBeforeRender.

Using Response Callbacks

Unlike many other web frameworks, Pyramid does not eagerly create a global response object. Adding a response callback allows an application to register an action to be performed against whatever response object is returned by a view, usually in order to mutate the response.

The pyramid.request.Request.add_response_callback() method is used to register a response callback.

A response callback is a callable which accepts two positional parameters: request and response. For example:

1def cache_callback(request, response):
2    """Set the cache_control max_age for the response"""
3    if request.exception is not None:
4        response.cache_control.max_age = 360
5request.add_response_callback(cache_callback)

No response callback is called if an unhandled exception happens in application code, or if the response object returned by a view callable is invalid. Response callbacks are, however, invoked when a exception view is rendered successfully. In such a case, the request.exception attribute of the request when it enters a response callback will be an exception object instead of its default value of None.

Response callbacks are called in the order they're added (first-to-most-recently-added). All response callbacks are called before the NewResponse event is sent. Errors raised by response callbacks are not handled specially. They will be propagated to the caller of the Pyramid router application.

A response callback has a lifetime of a single request. If you want a response callback to happen as the result of every request, you must re-register the callback into every new request (perhaps within a subscriber of a NewRequest event).

Using Finished Callbacks

A finished callback is a function that will be called unconditionally by the Pyramid router at the very end of request processing. A finished callback can be used to perform an action at the end of a request unconditionally.

The pyramid.request.Request.add_finished_callback() method is used to register a finished callback.

A finished callback is a callable which accepts a single positional parameter: request. For example:

1import logging
2
3log = logging.getLogger(__name__)
4
5def log_callback(request):
6    """Log information at the end of request"""
7    log.debug('Request is finished.')
8request.add_finished_callback(log_callback)

Finished callbacks are called in the order they're added (first-to-most-recently-added). Finished callbacks (unlike a response callback) are always called, even if an exception happens in application code that prevents a response from being generated.

The set of finished callbacks associated with a request are called very late in the processing of that request; they are essentially the very last thing called by the router before a request "ends". They are called after response processing has already occurred in a top-level finally: block within the router request processing code. As a result, mutations performed to the request provided to a finished callback will have no meaningful effect, because response processing will have already occurred, and the request's scope will expire almost immediately after all finished callbacks have been processed.

Errors raised by finished callbacks are not handled specially. They will be propagated to the caller of the Pyramid router application.

A finished callback has a lifetime of a single request. If you want a finished callback to happen as the result of every request, you must re-register the callback into every new request (perhaps within a subscriber of a NewRequest event).

Changing the Traverser

The default traversal algorithm that Pyramid uses is explained in The Traversal Algorithm. Though it is rarely necessary, this default algorithm can be swapped out selectively for a different traversal pattern via configuration.

1from pyramid.config import Configurator
2from myapp.traversal import Traverser
3config = Configurator()
4config.add_traverser(Traverser)

In the example above, myapp.traversal.Traverser is assumed to be a class that implements the following interface:

 1class Traverser(object):
 2    def __init__(self, root):
 3        """ Accept the root object returned from the root factory """
 4
 5    def __call__(self, request):
 6        """ Return a dictionary with (at least) the keys ``root``,
 7        ``context``, ``view_name``, ``subpath``, ``traversed``,
 8        ``virtual_root``, and ``virtual_root_path``.  These values are
 9        typically the result of a resource tree traversal.  ``root``
10        is the physical root object, ``context`` will be a resource
11        object, ``view_name`` will be the view name used (a string),
12        ``subpath`` will be a sequence of strings that
13        followed the view name but were not traversed, ``traversed``
14        will be a sequence of strings that were traversed
15        (including the virtual root path, if any) ``virtual_root``
16        will be a resource object representing the virtual root (or the
17        physical root if traversal was not performed), and
18        ``virtual_root_path`` will be a sequence representing the
19        virtual root path (a sequence of strings) or ``None`` if
20        traversal was not performed.
21
22        Extra keys for special purpose functionality can be added as
23        necessary.
24
25        All values returned in the dictionary will be made available
26        as attributes of the ``request`` object.
27        """

More than one traversal algorithm can be active at the same time. For instance, if your root factory returns more than one type of object conditionally, you could claim that an alternative traverser adapter is "for" only one particular class or interface. When the root factory returned an object that implemented that class or interface, a custom traverser would be used. Otherwise the default traverser would be used. For example:

1from myapp.traversal import Traverser
2from myapp.resources import MyRoot
3from pyramid.config import Configurator
4config = Configurator()
5config.add_traverser(Traverser, MyRoot)

If the above stanza was added to a Pyramid __init__.py file's main function, Pyramid would use the myapp.traversal.Traverser only when the application root factory returned an instance of the myapp.resources.MyRoot object. Otherwise it would use the default Pyramid traverser to do traversal.

Changing How pyramid.request.Request.resource_url() Generates a URL

When you add a traverser as described in Changing the Traverser, it's often convenient to continue to use the pyramid.request.Request.resource_url() API. However, since the way traversal is done will have been modified, the URLs it generates by default may be incorrect when used against resources derived from your custom traverser.

If you've added a traverser, you can change how resource_url() generates a URL for a specific type of resource by adding a call to pyramid.config.Configurator.add_resource_url_adapter().

For example:

1from myapp.traversal import ResourceURLAdapter
2from myapp.resources import MyRoot
3
4config.add_resource_url_adapter(ResourceURLAdapter, MyRoot)

In the above example, the myapp.traversal.ResourceURLAdapter class will be used to provide services to resource_url() any time the resource passed to resource_url is of the class myapp.resources.MyRoot. The resource_iface argument MyRoot represents the type of interface that must be possessed by the resource for this resource url factory to be found. If the resource_iface argument is omitted, this resource URL adapter will be used for all resources.

The API that must be implemented by a class that provides IResourceURL is as follows:

 1class MyResourceURL(object):
 2    """ An adapter which provides the virtual and physical paths of a
 3        resource
 4    """
 5    def __init__(self, resource, request):
 6        """ Accept the resource and request and set self.physical_path and
 7        self.virtual_path """
 8        self.virtual_path =  some_function_of(resource, request)
 9        self.virtual_path_tuple =  some_function_of(resource, request)
10        self.physical_path =  some_other_function_of(resource, request)
11        self.physical_path_tuple =  some_function_of(resource, request)

The default context URL generator is available for perusal as the class pyramid.traversal.ResourceURL in the traversal module.

See pyramid.config.Configurator.add_resource_url_adapter() for more information.

Changing How Pyramid Treats View Responses

New in version 1.1.

It is possible to control how Pyramid treats the result of calling a view callable on a per-type basis by using a hook involving pyramid.config.Configurator.add_response_adapter() or the response_adapter decorator.

Pyramid, in various places, adapts the result of calling a view callable to the IResponse interface to ensure that the object returned by the view callable is a "true" response object. The vast majority of time, the result of this adaptation is the result object itself, as view callables written by "civilians" who read the narrative documentation contained in this manual will always return something that implements the IResponse interface. Most typically, this will be an instance of the pyramid.response.Response class or a subclass. If a civilian returns a non-Response object from a view callable that isn't configured to use a renderer, they will typically expect the router to raise an error. However, you can hook Pyramid in such a way that users can return arbitrary values from a view callable by providing an adapter which converts the arbitrary return value into something that implements IResponse.

For example, if you'd like to allow view callables to return bare string objects (without requiring a renderer to convert a string to a response object), you can register an adapter which converts the string to a Response:

1from pyramid.response import Response
2
3def string_response_adapter(s):
4    response = Response(s)
5    return response
6
7# config is an instance of pyramid.config.Configurator
8
9config.add_response_adapter(string_response_adapter, str)

Likewise, if you want to be able to return a simplified kind of response object from view callables, you can use the IResponse hook to register an adapter to the more complex IResponse interface:

 1from pyramid.response import Response
 2
 3class SimpleResponse(object):
 4    def __init__(self, body):
 5        self.body = body
 6
 7def simple_response_adapter(simple_response):
 8    response = Response(simple_response.body)
 9    return response
10
11# config is an instance of pyramid.config.Configurator
12
13config.add_response_adapter(simple_response_adapter, SimpleResponse)

If you want to implement your own Response object instead of using the pyramid.response.Response object in any capacity at all, you'll have to make sure that the object implements every attribute and method outlined in pyramid.interfaces.IResponse and you'll have to ensure that it uses zope.interface.implementer(IResponse) as a class decorator.

1from pyramid.interfaces import IResponse
2from zope.interface import implementer
3
4@implementer(IResponse)
5class MyResponse(object):
6    # ... an implementation of every method and attribute
7    # documented in IResponse should follow ...

When an alternate response object implementation is returned by a view callable, if that object asserts that it implements IResponse (via zope.interface.implementer(IResponse)) , an adapter needn't be registered for the object; Pyramid will use it directly.

An IResponse adapter for webob.reponse.Response (as opposed to pyramid.response.Response) is registered by Pyramid by default at startup time, as by their nature, instances of this class (and instances of subclasses of the class) will natively provide IResponse. The adapter registered for webob.reponse.Response simply returns the response object.

Instead of using pyramid.config.Configurator.add_response_adapter(), you can use the pyramid.response.response_adapter decorator:

1from pyramid.response import Response
2from pyramid.response import response_adapter
3
4@response_adapter(str)
5def string_response_adapter(s):
6    response = Response(s)
7    return response

The above example, when scanned, has the same effect as:

config.add_response_adapter(string_response_adapter, str)

The response_adapter decorator will have no effect until activated by a scan.

Using a View Mapper

The default calling conventions for view callables are documented in the Views chapter. You can change the way users define view callables by employing a view mapper.

A view mapper is an object that accepts a set of keyword arguments and which returns a callable. The returned callable is called with the view callable object. The returned callable should itself return another callable which can be called with the "internal calling protocol" (context, request).

You can use a view mapper in a number of ways:

  • by setting a __view_mapper__ attribute (which is the view mapper object) on the view callable itself

  • by passing the mapper object to pyramid.config.Configurator.add_view() (or its declarative and decorator equivalents) as the mapper argument

  • by registering a default view mapper

Here's an example of a view mapper that emulates (somewhat) a Pylons "controller". The mapper is initialized with some keyword arguments. Its __call__ method accepts the view object (which will be a class). It uses the attr keyword argument it is passed to determine which attribute should be used as an action method. The wrapper method it returns accepts (context, request) and returns the result of calling the action method with keyword arguments implied by the matchdict after popping the action out of it. This somewhat emulates the Pylons style of calling action methods with routing parameters pulled out of the route matching dict as keyword arguments.

 1# framework
 2
 3class PylonsControllerViewMapper(object):
 4    def __init__(self, **kw):
 5        self.kw = kw
 6
 7    def __call__(self, view):
 8        attr = self.kw['attr']
 9        def wrapper(context, request):
10            matchdict = request.matchdict.copy()
11            matchdict.pop('action', None)
12            inst = view(request)
13            meth = getattr(inst, attr)
14            return meth(**matchdict)
15        return wrapper
16
17class BaseController(object):
18    __view_mapper__ = PylonsControllerViewMapper

A user might make use of these framework components like so:

 1# user application
 2
 3from pyramid.response import Response
 4from pyramid.config import Configurator
 5import pyramid_handlers
 6from wsgiref.simple_server import make_server
 7
 8class MyController(BaseController):
 9    def index(self, id):
10        return Response(id)
11
12if __name__ == '__main__':
13    config = Configurator()
14    config.include(pyramid_handlers)
15    config.add_handler('one', '/{id}', MyController, action='index')
16    config.add_handler('two', '/{action}/{id}', MyController)
17    server.make_server('0.0.0.0', 8080, config.make_wsgi_app())
18    server.serve_forever()

The pyramid.config.Configurator.set_view_mapper() method can be used to set a default view mapper (overriding the superdefault view mapper used by Pyramid itself).

A single view registration can use a view mapper by passing the mapper as the mapper argument to add_view().

Registering Configuration Decorators

Decorators such as view_config don't change the behavior of the functions or classes they're decorating. Instead when a scan is performed, a modified version of the function or class is registered with Pyramid.

You may wish to have your own decorators that offer such behaviour. This is possible by using the Venusian package in the same way that it is used by Pyramid.

By way of example, let's suppose you want to write a decorator that registers the function it wraps with a Zope Component Architecture "utility" within the application registry provided by Pyramid. The application registry and the utility inside the registry is likely only to be available once your application's configuration is at least partially completed. A normal decorator would fail as it would be executed before the configuration had even begun.

However, using Venusian, the decorator could be written as follows:

 1import venusian
 2from mypackage.interfaces import IMyUtility
 3
 4class registerFunction(object):
 5
 6    def __init__(self, path):
 7        self.path = path
 8
 9    def register(self, scanner, name, wrapped):
10        registry = scanner.config.registry
11        registry.getUtility(IMyUtility).register(
12            self.path, wrapped)
13
14    def __call__(self, wrapped):
15        venusian.attach(wrapped, self.register)
16        return wrapped

This decorator could then be used to register functions throughout your code:

1@registerFunction('/some/path')
2def my_function():
3    do_stuff()

However, the utility would only be looked up when a scan was performed, enabling you to set up the utility in advance:

 1from zope.interface import implementer
 2
 3from wsgiref.simple_server import make_server
 4from pyramid.config import Configurator
 5from mypackage.interfaces import IMyUtility
 6
 7@implementer(IMyUtility)
 8class UtilityImplementation:
 9
10    def __init__(self):
11        self.registrations = {}
12
13    def register(self, path, callable_):
14        self.registrations[path] = callable_
15
16if __name__ == '__main__':
17    config = Configurator()
18    config.registry.registerUtility(UtilityImplementation())
19    config.scan()
20    app = config.make_wsgi_app()
21    server = make_server('0.0.0.0', 8080, app)
22    server.serve_forever()

For full details, please read the Venusian documentation.

Registering Tweens

New in version 1.2: Tweens

A tween (a contraction of the word "between") is a bit of code that sits between the Pyramid router's main request handling function and the upstream WSGI component that uses Pyramid as its "app". This is a feature that may be used by Pyramid framework extensions to provide, for example, Pyramid-specific view timing support bookkeeping code that examines exceptions before they are returned to the upstream WSGI application. Tweens behave a bit like WSGI middleware, but they have the benefit of running in a context in which they have access to the Pyramid request, response, and application registry, as well as the Pyramid rendering machinery.

Creating a Tween

To create a tween, you must write a "tween factory". A tween factory must be a globally importable callable which accepts two arguments: handler and registry. handler will be either the main Pyramid request handling function or another tween. registry will be the Pyramid application registry represented by this Configurator. A tween factory must return the tween (a callable object) when it is called.

A tween is called with a single argument, request, which is the request created by Pyramid's router when it receives a WSGI request. A tween should return a response, usually the one generated by the downstream Pyramid application.

You can write the tween factory as a simple closure-returning function:

 1def simple_tween_factory(handler, registry):
 2    # one-time configuration code goes here
 3
 4    def simple_tween(request):
 5        # code to be executed for each request before
 6        # the actual application code goes here
 7
 8        response = handler(request)
 9
10        # code to be executed for each request after
11        # the actual application code goes here
12
13        return response
14
15    return simple_tween

Alternatively, the tween factory can be a class with the __call__ magic method:

 1class simple_tween_factory(object):
 2    def __init__(self, handler, registry):
 3        self.handler = handler
 4        self.registry = registry
 5
 6        # one-time configuration code goes here
 7
 8    def __call__(self, request):
 9        # code to be executed for each request before
10        # the actual application code goes here
11
12        response = self.handler(request)
13
14        # code to be executed for each request after
15        # the actual application code goes here
16
17        return response

You should avoid mutating any state on the tween instance. The tween is invoked once per request and any shared mutable state needs to be carefully handled to avoid any race conditions.

The closure style performs slightly better and enables you to conditionally omit the tween from the request processing pipeline (see the following timing tween example), whereas the class style makes it easier to have shared mutable state and allows subclassing.

Here's a complete example of a tween that logs the time spent processing each request:

 1# in a module named myapp.tweens
 2
 3import time
 4from pyramid.settings import asbool
 5import logging
 6
 7log = logging.getLogger(__name__)
 8
 9def timing_tween_factory(handler, registry):
10    if asbool(registry.settings.get('do_timing')):
11        # if timing support is enabled, return a wrapper
12        def timing_tween(request):
13            start = time.time()
14            try:
15                response = handler(request)
16            finally:
17                end = time.time()
18                log.debug('The request took %s seconds' %
19                          (end - start))
20            return response
21        return timing_tween
22    # if timing support is not enabled, return the original
23    # handler
24    return handler

In the above example, the tween factory defines a timing_tween tween and returns it if asbool(registry.settings.get('do_timing')) is true. It otherwise simply returns the handler which it was given. The registry.settings attribute is a handle to the deployment settings provided by the user (usually in an .ini file). In this case, if the user has defined a do_timing setting and that setting is True, the user has said they want to do timing, so the tween factory returns the timing tween; it otherwise just returns the handler it has been provided, preventing any timing.

The example timing tween simply records the start time, calls the downstream handler, logs the number of seconds consumed by the downstream handler, and returns the response.

Registering an Implicit Tween Factory

Once you've created a tween factory, you can register it into the implicit tween chain using the pyramid.config.Configurator.add_tween() method using its dotted Python name.

Here's an example of registering a tween factory as an "implicit" tween in a Pyramid application:

1from pyramid.config import Configurator
2config = Configurator()
3config.add_tween('myapp.tweens.timing_tween_factory')

Note that you must use a dotted Python name as the first argument to pyramid.config.Configurator.add_tween(); this must point at a tween factory. You cannot pass the tween factory object itself to the method: it must be dotted Python name that points to a globally importable object. In the above example, we assume that a timing_tween_factory tween factory was defined in a module named myapp.tweens, so the tween factory is importable as myapp.tweens.timing_tween_factory.

When you use pyramid.config.Configurator.add_tween(), you're instructing the system to use your tween factory at startup time unless the user has provided an explicit tween list in their configuration. This is what's meant by an "implicit" tween. A user can always elect to supply an explicit tween list, reordering or disincluding implicitly added tweens. See Explicit Tween Ordering for more information about explicit tween ordering.

If more than one call to pyramid.config.Configurator.add_tween() is made within a single application configuration, the tweens will be chained together at application startup time. The first tween factory added via add_tween will be called with the Pyramid exception view tween factory as its handler argument, then the tween factory added directly after that one will be called with the result of the first tween factory as its handler argument, and so on, ad infinitum until all tween factories have been called. The Pyramid router will use the outermost tween produced by this chain (the tween generated by the very last tween factory added) as its request handler function. For example:

1from pyramid.config import Configurator
2
3config = Configurator()
4config.add_tween('myapp.tween_factory1')
5config.add_tween('myapp.tween_factory2')

The above example will generate an implicit tween chain that looks like this.

INGRESS (implicit)
myapp.tween_factory2
myapp.tween_factory1
pyramid.tweens.excview_tween_factory (implicit)
MAIN (implicit)

Suggesting Implicit Tween Ordering

By default, as described above, the ordering of the chain is controlled entirely by the relative ordering of calls to pyramid.config.Configurator.add_tween(). However, the caller of add_tween can provide an optional hint that can influence the implicit tween chain ordering by supplying under or over (or both) arguments to add_tween(). These hints are only used when an explicit tween ordering is not used. See Explicit Tween Ordering for a description of how to set an explicit tween ordering.

Allowable values for under or over (or both) are:

  • None (the default),

  • a dotted Python name to a tween factory: a string representing the predicted dotted name of a tween factory added in a call to add_tween in the same configuration session,

  • one of the constants pyramid.tweens.MAIN, pyramid.tweens.INGRESS, or pyramid.tweens.EXCVIEW, or

  • an iterable of any combination of the above. This allows the user to specify fallbacks if the desired tween is not included, as well as compatibility with multiple other tweens.

Effectively, over means "closer to the request ingress than" and under means "closer to the main Pyramid application than". You can think of an onion with outer layers over the inner layers, the application being under all the layers at the center.

For example, the following call to add_tween() will attempt to place the tween factory represented by myapp.tween_factory directly "above" (in ptweens order) the main Pyramid request handler.

1import pyramid.tweens
2
3config.add_tween('myapp.tween_factory', over=pyramid.tweens.MAIN)

The above example will generate an implicit tween chain that looks like this.

INGRESS (implicit)
pyramid.tweens.excview_tween_factory (implicit)
myapp.tween_factory
MAIN (implicit)

Likewise, calling the following call to add_tween() will attempt to place this tween factory "above" the main handler but "below" a separately added tween factory:

1import pyramid.tweens
2
3config.add_tween('myapp.tween_factory1',
4                 over=pyramid.tweens.MAIN)
5config.add_tween('myapp.tween_factory2',
6                 over=pyramid.tweens.MAIN,
7                 under='myapp.tween_factory1')

The above example will generate an implicit tween chain that looks like this:

INGRESS (implicit)
pyramid.tweens.excview_tween_factory (implicit)
myapp.tween_factory1
myapp.tween_factory2
MAIN (implicit)

Specifying neither over nor under is equivalent to specifying under=INGRESS.

If all options for under (or over) cannot be found in the current configuration, it is an error. If some options are specified purely for compatibility with other tweens, just add a fallback of MAIN or INGRESS. For example, under=('someothertween', 'someothertween2', INGRESS). This constraint will require the tween to be located under the someothertween tween, the someothertween2 tween, and INGRESS. If any of these is not in the current configuration, this constraint will only organize itself based on the tweens that are present.

Explicit Tween Ordering

Implicit tween ordering is obviously only best-effort. Pyramid will attempt to provide an implicit order of tweens as best it can using hints provided by calls to add_tween(). But because it's only best-effort, if very precise tween ordering is required, the only surefire way to get it is to use an explicit tween order. The deploying user can override the implicit tween inclusion and ordering implied by calls to add_tween() entirely by using the pyramid.tweens settings value. When used, this settings value must be a list of Python dotted names which will override the ordering (and inclusion) of tween factories in the implicit tween chain. For example:

1[app:main]
2use = egg:MyApp
3pyramid.reload_templates = true
4pyramid.debug_authorization = false
5pyramid.debug_notfound = false
6pyramid.debug_routematch = false
7pyramid.debug_templates = true
8pyramid.tweens = myapp.my_cool_tween_factory
9                 pyramid.tweens.excview_tween_factory

In the above configuration, calls made during configuration to pyramid.config.Configurator.add_tween() are ignored, and the user is telling the system to use the tween factories he has listed in the pyramid.tweens configuration setting (each is a dotted Python name which points to a tween factory) instead of any tween factories added via pyramid.config.Configurator.add_tween(). The first tween factory in the pyramid.tweens list will be used as the producer of the effective Pyramid request handling function; it will wrap the tween factory declared directly "below" it, ad infinitum. The "main" Pyramid request handler is implicit, and always "at the bottom".

Note

Pyramid's own exception view handling logic is implemented as a tween factory function: pyramid.tweens.excview_tween_factory(). If Pyramid exception view handling is desired, and tween factories are specified via the pyramid.tweens configuration setting, the pyramid.tweens.excview_tween_factory() function must be added to the pyramid.tweens configuration setting list explicitly. If it is not present, Pyramid will not perform exception view handling.

Tween Conflicts and Ordering Cycles

Pyramid will prevent the same tween factory from being added to the tween chain more than once using configuration conflict detection. If you wish to add the same tween factory more than once in a configuration, you should either: (a) use a tween factory that is a separate globally importable instance object from the factory that it conflicts with; (b) use a function or class as a tween factory with the same logic as the other tween factory it conflicts with, but with a different __name__ attribute; or (c) call pyramid.config.Configurator.commit() between calls to pyramid.config.Configurator.add_tween().

If a cycle is detected in implicit tween ordering when over and under are used in any call to add_tween, an exception will be raised at startup time.

Displaying Tween Ordering

The ptweens command-line utility can be used to report the current implicit and explicit tween chains used by an application. See ptweens: Displaying "Tweens".

Adding a Custom View, Route, or Subscriber Predicate

New in version 1.4.

View and Route Predicates

View and route predicates used during configuration allow you to narrow the set of circumstances under which a view or route will match. For example, the request_method view predicate can be used to ensure a view callable is only invoked when the request's method is POST:

@view_config(request_method='POST')
def someview(request):
    ...

Likewise, a similar predicate can be used as a route predicate:

config.add_route('name', '/foo', request_method='POST')

Many other built-in predicates exists (request_param, and others). You can add custom predicates to the list of available predicates by using one of pyramid.config.Configurator.add_view_predicate() or pyramid.config.Configurator.add_route_predicate(). The former adds a view predicate, the latter a route predicate.

When using one of those APIs, you pass a name and a factory to add a predicate during Pyramid's configuration stage. For example:

config.add_view_predicate('content_type', ContentTypePredicate)

The above example adds a new predicate named content_type to the list of available predicates for views. This will allow the following view configuration statement to work:

1@view_config(content_type='File')
2def aview(request): ...

The first argument to pyramid.config.Configurator.add_view_predicate(), the name, is a string representing the name that is expected to be passed to view_config (or its imperative analogue add_view).

The second argument is a view or route predicate factory, or a dotted Python name which refers to a view or route predicate factory. A view or route predicate factory is most often a class with a constructor (__init__), a text method, a phash method, and a __call__ method. For example:

 1class ContentTypePredicate(object):
 2    def __init__(self, val, info):
 3        self.val = val
 4
 5    def text(self):
 6        return 'content_type = %s' % (self.val,)
 7
 8    phash = text
 9
10    def __call__(self, context, request):
11        return request.content_type == self.val

The constructor of an pyramid.interfaces.IPredicateFactory` takes two arguments: val and info. The val argument will be the argument passed to view_config (or add_view). In the example above, it will be the string File. The second argument, info, will be an pyramid.interfaces.IPredicateInfo instance created relative to the action configuring the predicate. This means the info.package value is the package where the action is invoked passing in val and info.maybe_dotted is also relative to this package.

The text method must return a string. It should be useful to describe the behavior of the predicate in error messages.

The phash method must return a string or a sequence of strings. It's most often the same as text, as long as text uniquely describes the predicate's name and the value passed to the constructor. If text is more general, or doesn't describe things that way, phash should return a string with the name and the value serialized. The result of phash is not seen in output anywhere, it just informs the uniqueness constraints for view configuration.

The __call__ method differs depending on whether the predicate is used as a view predicate or a route predicate:

  • When used as a route predicate, the __call__ signature is (info, request). The info object is a dictionary containing two keys: match and route. info['match'] is the matchdict containing the patterns matched in the route pattern. info['route'] is the pyramid.interfaces.IRoute object for the current route.

  • When used as a view predicate, the __call__ signature is (context, request). The context is the result of traversal performed using either the route's root factory or the app's default root factory.

In all cases the __call__ method is expected to return True or False.

It is possible to use the same predicate factory as both a view predicate and as a route predicate, but they'll need to handle the info or context argument specially (many predicates do not need this argument) and you'll need to call add_view_predicate and add_route_predicate separately with the same factory.

Subscriber Predicates

Subscriber predicates work almost exactly like view and route predicates. They narrow the set of circumstances in which a subscriber will be called. There are several minor differences between a subscriber predicate and a view or route predicate:

  • There are no default subscriber predicates. You must register one to use one.

  • The __call__ method of a subscriber predicate accepts a single event object instead of a context and a request.

  • Not every subscriber predicate can be used with every event type. Some subscriber predicates will assume a certain event type.

Here's an example of a subscriber predicate that can be used in conjunction with a subscriber that subscribes to the pyramid.events.NewRequest event type.

 1class RequestPathStartsWith(object):
 2    def __init__(self, val, info):
 3        self.val = val
 4
 5    def text(self):
 6        return 'request_path_startswith = %s' % (self.val,)
 7
 8    phash = text
 9
10    def __call__(self, event):
11        return event.request.path.startswith(self.val)

Once you've created a subscriber predicate, it may be registered via pyramid.config.Configurator.add_subscriber_predicate(). For example:

config.add_subscriber_predicate(
    'request_path_startswith', RequestPathStartsWith)

Once a subscriber predicate is registered, you can use it in a call to pyramid.config.Configurator.add_subscriber() or to pyramid.events.subscriber. Here's an example of using the previously registered request_path_startswith predicate in a call to add_subscriber():

1# define a subscriber in your code
2
3def yosubscriber(event):
4    event.request.yo = 'YO!'
5
6# and at configuration time
7
8config.add_subscriber(yosubscriber, NewRequest,
9                      request_path_startswith='/add_yo')

Here's the same subscriber/predicate/event-type combination used via subscriber.

1from pyramid.events import subscriber
2
3@subscriber(NewRequest, request_path_startswith='/add_yo')
4def yosubscriber(event):
5    event.request.yo = 'YO!'

In either of the above configurations, the yosubscriber callable will only be called if the request path starts with /add_yo. Otherwise the event subscriber will not be called.

Note that the request_path_startswith subscriber you defined can be used with events that have a request attribute, but not ones that do not. So, for example, the predicate can be used with subscribers registered for pyramid.events.NewRequest and pyramid.events.ContextFound events, but it cannot be used with subscribers registered for pyramid.events.ApplicationCreated because the latter type of event has no request attribute. The point being, unlike route and view predicates, not every type of subscriber predicate will necessarily be applicable for use in every subscriber registration. It is not the responsibility of the predicate author to make every predicate make sense for every event type; it is the responsibility of the predicate consumer to use predicates that make sense for a particular event type registration.

View Derivers

New in version 1.7.

Every URL processed by Pyramid is matched against a custom view pipeline. See Request Processing for how this works. The view pipeline itself is built from the user-supplied view callable, which is then composed with view derivers. A view deriver is a composable element of the view pipeline which is used to wrap a view with added functionality. View derivers are very similar to the decorator argument to pyramid.config.Configurator.add_view(), except that they have the option to execute for every view in the application.

It is helpful to think of a view deriver as middleware for views. Unlike tweens or WSGI middleware which are scoped to the application itself, a view deriver is invoked once per view in the application, and can use configuration options from the view to customize its behavior.

Built-in View Derivers

There are several built-in view derivers that Pyramid will automatically apply to any view. Below they are defined in order from furthest to closest to the user-defined view callable:

secured_view

Enforce the permission defined on the view. This element is a no-op if no permission is defined. Note there will always be a permission defined if a default permission was assigned via pyramid.config.Configurator.set_default_permission() unless the view is an exception view.

This element will also output useful debugging information when pyramid.debug_authorization is enabled.

csrf_view

Used to check the CSRF token provided in the request. This element is a no-op if require_csrf view option is not True. Note there will always be a require_csrf option if a default value was assigned via pyramid.config.Configurator.set_default_csrf_options() unless the view is an exception view.

owrapped_view

Invokes the wrapped view defined by the wrapper option.

http_cached_view

Applies cache control headers to the response defined by the http_cache option. This element is a no-op if the pyramid.prevent_http_cache setting is enabled or the http_cache option is None.

decorated_view

Wraps the view with the decorators from the decorator option.

rendered_view

Adapts the result of the view callable into a response object. Below this point the result may be any Python object.

mapped_view

Applies the view mapper defined by the mapper option or the application's default view mapper to the view callable. This is always the closest deriver to the user-defined view and standardizes the view pipeline interface to accept (context, request) from all previous view derivers.

Warning

Any view derivers defined under the rendered_view are not guaranteed to receive a valid response object. Rather they will receive the result from the view mapper which is likely the original response returned from the view. This is possibly a dictionary for a renderer but it may be any Python object that may be adapted into a response.

Custom View Derivers

It is possible to define custom view derivers which will affect all views in an application. There are many uses for this, but most will likely be centered around monitoring and security. In order to register a custom view deriver, you should create a callable that conforms to the pyramid.interfaces.IViewDeriver interface, and then register it with your application using pyramid.config.Configurator.add_view_deriver(). The callable should accept the view to be wrapped and the info object which is an instance of pyramid.interfaces.IViewDeriverInfo. For example, below is a callable that can provide timing information for the view pipeline:

 1import time
 2
 3def timing_view(view, info):
 4    if info.options.get('timed'):
 5        def wrapper_view(context, request):
 6            start = time.time()
 7            response = view(context, request)
 8            end = time.time()
 9            response.headers['X-View-Performance'] = '%.3f' % (end - start,)
10            return response
11        return wrapper_view
12    return view
13
14timing_view.options = ('timed',)
15
16config.add_view_deriver(timing_view)

The setting of timed on the timing_view signifies to Pyramid that timed is a valid view_config keyword argument now. The timing_view custom view deriver as registered above will only be active for any view defined with a timed=True value passed as one of its view_config keywords.

For example, this view configuration will not be a timed view:

1@view_config(route_name='home')
2def home(request):
3    return Response('Home')

But this view will have timing information added to the response headers:

1@view_config(route_name='home', timed=True)
2def home(request):
3    return Response('Home')

View derivers are unique in that they have access to most of the options passed to pyramid.config.Configurator.add_view() in order to decide what to do, and they have a chance to affect every view in the application.

Exception Views and View Derivers

A view deriver has the opportunity to wrap any view, including an exception view. In general this is fine, but certain view derivers may wish to avoid doing certain things when handling exceptions. For example, the csrf_view and secured_view built-in view derivers will not perform security checks on exception views unless explicitly told to do so.

You can check for info.exception_only on the pyramid.interfaces.IViewDeriverInfo object when wrapping the view to determine whether you are wrapping an exception view or a normal view.

Ordering View Derivers

By default, every new view deriver is added between the decorated_view and rendered_view built-in derivers. It is possible to customize this ordering using the over and under options. Each option can use the names of other view derivers in order to specify an ordering. There should rarely be a reason to worry about the ordering of the derivers except when the deriver depends on other operations in the view pipeline.

Both over and under may also be iterables of constraints. For either option, if one or more constraints was defined, at least one must be satisfied, else a pyramid.exceptions.ConfigurationError will be raised. This may be used to define fallback constraints if another deriver is missing.

Two sentinel values exist, pyramid.viewderivers.INGRESS and pyramid.viewderivers.VIEW, which may be used when specifying constraints at the edges of the view pipeline. For example, to add a deriver at the start of the pipeline you may use under=INGRESS.

It is not possible to add a view deriver under the mapped_view as the view mapper is intimately tied to the signature of the user-defined view callable. If you simply need to know what the original view callable was, it can be found as info.original_view on the provided pyramid.interfaces.IViewDeriverInfo object passed to every view deriver.

Warning

The default constraints for any view deriver are over='rendered_view' and under='decorated_view'. When escaping these constraints you must take care to avoid cyclic dependencies between derivers. For example, if you want to add a new view deriver before secured_view then simply specifying over='secured_view' is not enough, because the default is also under decorated view there will be an unsatisfiable cycle. You must specify a valid under constraint as well, such as under=INGRESS to fall between INGRESS and secured_view at the beginning of the view pipeline.