Context-Oriented Programming: Beyond Layers

©2007Martinv.Löwis
Context-Oriented Programming:
Beyond Layers
Martin v. Löwis
Marcus Denker
Oscar Nierstrasz

Context-oriented Programming: Beyond Layers
Agenda
• Context-dependent Behavior
• Method Layers (PyContext example)
• Implicit Layer Activation
• Case Studies
• Context Variables
• Implementation Notes
2

Context Dependencies
• Programs need to be aware of the context in which they
operate
– what is the state of the environment
– what user is accessing the system
– what mode is the program to be executed in
• Example: current user
– different roles may cause completely different code to be executed
(e.g. administrator may be offered different facilities)
• can be modeled through method layers
– different users acting in the same role access different data
• modeling through method layers is not adequate
• Example: dependency of program output on output device
– In OO system, rendering algorithm spreads over methods of
different classes
3

Method Layers
• addition of a few concepts to object-oriented programming
• layer: group of classes and methods to be used together in
dynamic scope of execution
• layered class: collection of partial definitions of a class, for
different layers
– layered methods: definitions of methods for specific layers
– layered slots: definition of instance attributes for specific layers
• (explicit) layer activation: specification of code block that
runs in the context of a layer
– inside the block, each sent message selects the method defined for
that layer
– nested activation: need to consider multiple layers in sequence
5

Example: User-Agent Header
• Web browsers sent User-Agent header to indicate client
software (e.g. MSIE, Firefox, Safari, etc.)
– "Mozilla/4.0 (compatible; MSIE 7.0; Windows NT 5.1)"
• Web servers sometimes have different behavior depending
on User-Agent header
• Problem: automated web client might need to claim to
operate as a specific user agent
6
Client
HTTP Library
HTTP ServerHTTP
Request
User-Agent

Example: User-Agent Header (2)
• Assumption: client consists of multiple modules, each using
different software layers to access underlying HTTP libraries
– explicitly specifying User-Agent to the library is not possible
• Assumption: client is multi-threaded; different threads may
need to operate in different contexts
– setting User-Agent as a global variable is not possible
• HTTP libraries in Python:
– httplib: direct access to protocol
– urllib: unifying library for http, ftp, ...
7

PyContext: Using Method Layers
• with statement: automatic enter/leave semantics
from useragent import HTTPUserAgent
with HTTPUserAgent("WebCOP"):
print "Using useragent layer"
get1()
get2()
• Importing useragent module automatically defines the layer
and the layered methods
• Disabling layers
from layers import Disabled
with Disabled(Layer):
code
8

Defining Layers
• Inherit from class Layer
– Class can have arbitrary methods, instance variables, etc
class HTTPUserAgent(layers.Layer):
def __init__(self, agent):
self.agent = agent
9

Defining Layered Methods
• Inherit a class (with arbitrary name) from both the layer and
the class to augment
• Define methods with the same name as the original methods
– Each method has automatic second parameter "context" (after self,
before explicit method parameters)
• Decorate each method with either before, after, or instead
• Context: Object indicating the layer activation
– .layer: reference to the layer object
– .result: result of the original method (for after-methods)
– .proceed: callable object denoting the continuation to the original
method (or the next layer)
10

class HTTPConnection(HTTPUserAgent, httplib.HTTPConnection):
# Always add a User-Agent header
@before
def endheaders(self, context):
with layers.Disabled(HTTPUserAgent):
self.putheader("User-Agent", context.layer.agent)
# suppress other User-Agent headers added
@instead
def putheader(self, context, header, value):
if header.lower() == 'user-agent':
return
return context.proceed(header, value)
11

Implicit Activation

Implicit Activation
• Problem: explicit activation still needs to identify point in
code where context might change or where context will be
relevant
• Objective: allow addition of layers which get activated
"automatically"
– specifically, when a condition on the environment changes
• Design issues:
– how can the system tell whether a condition becomes true?
• each layer implements an active method
– when should the active method be evaluated?
• each time a layered method is executed whose meaning depends on
whether the layer is active or not
13

Case Studies

Objective
• We tried to evaluate what aspects of context are common in
application programs today
• Issue: how can we find code that depends on context?
– Starting point: assume caller and callee are designed to run within
the same context
– Starting point: look for traditional examples of context
• Selected case studies: large Python applications/libraries
– Django: web application framework
– Roundup: bug tracker
– SCons: automated build tool
15

Results
• Web applications (Django, Roundup) need to support
concept of "current" request, including authenticated user,
session data, target URL, etc.
• SCons keeps track of context in "environment": information
about the current build goal
• These things were often referred to as "context", or showed
up as pass-through parameters in methods
– Searching for "context" revealed further context-dependent code
fragments
– Searching for pass-through parameters not easily possible with pure
text searching; subject for further study
• Context information often not used to select different pieces
of code, but merely as lookup keys in associative arrays
16

Dynamic Variables

Motivation
• case study results lead to identification of additional concept
for context-oriented programming: Dynamic Variables
• in order to avoid pass-through parameters, a variable holding
context should be set in a caller, and then read in a nested
callee
– similar to dynamic variables in functional languages
– requires careful usage, to avoid old problems with dynamic variables
(unintentional access due to naming collisions)
• require explicit read and write operations
18

Dynamic Variables in PyContext
• Example: current HTTP session
1. Declare dynamic variable
_session = Variable()
2. Obtain current variable (e.g. through helper function)
def current_session():
return _session.get()
3. Setup variable from dynamically-read context
def process_request(request):
session = lookup_session(request)
with _session.set(session):
dispatch_request(request)
19

Implementation Notes
• Method layers:
– Dynamically replace methods with wrappers
• Dynamic variables:
1. perform stack walk: O(stack-depth)
2. use thread-local storage: O(1)
20

Summary
• current applications (in particular webapps) show high
degree of context-awareness
• context-dependency is not made explicit in the code
• layers are a first step to making context explicit
• rehabilitation of dynamic variables necessary to support
common cases of context
21

Context-Oriented Programming: Beyond Layers

More Related Content

What's hot (20)

Viewers also liked (12)

Similar to Context-Oriented Programming: Beyond Layers (20)

More from ESUG (20)

Recently uploaded (20)

Context-Oriented Programming: Beyond Layers