Skip to content


Subversion checkout URL

You can clone with
Download ZIP
Browse files


  • Loading branch information...
commit 0da2610b84cd815c70565e531d1ff580353772b5 1 parent 5238163
@tef authored
Showing with 144 additions and 103 deletions.
  1. +134 −102 README.rst
  2. +4 −0 doc/todo
  3. +5 −0 glyph/resource/
  4. +1 −1  tests/
236 README.rst
@@ -1,120 +1,152 @@
-it is like json-rpc with callbacks.
-glyph is a client and server library for getting an object from a server,
-and calling methods on it using the callbacks provided by the server.
-the server offers objects over http, where the url describes the resource
-and its state, and GETing an object returns any contents & callbacks for methods
-glyph uses urls internally to track the state between requests, but
-the server is free to change the mapping without breaking the client.
-this allows glyph to provide duck-typing: clients do not care *what*
-kind of object is returned, as long as it has the right named methods.
-as a result, glyph allows you to grow your api like you grow a website.
+glyph-rpc makes minature websites for your objects, so you can grow your api
+like you grow a website:
+- new methods and objects can be added without breaking clients
- glyph services can redirect/link to other services on other hosts
-- new methods and resources can be added without breaking clients
- glyph can take advantage of http tools like caches and load-balancers
-glyph tries to exploit http rather than simply tunnel requests over it.
-* simple echo server code, running it
-* connecting with the client, calling a method
-* inspecting the contents of the reply
-* urls are object id's - used to create/find resources at server
-* callbacks are links and forms,
-* how it works: router, mapper, resource
- - router
- attach resource classes to url prefixes
- on request, finds resource and its mapper
- uses the mapper to handle the request
- - mappers
- maps http <-> resource class
- url query string is the resource state
- make a resource to handle this request,
- and dispatch to a method
- adds callbacks to
- - resources
- have an index() method that returns contents
- are created per request - represent a handle
- to some state at the server.
-* configuration server?
-the serialization format is an extension of bencoding (from bittorrent).
-it is not language specific, json-alike with a few more convieniences.
-historical reasons mandated the support of bytestrings & unicode data,
-and existing formats (xml/json) required clumsy workarounds. it works
-but i'm not proud to reinvent another wheel.
-json like vocabulary
- - unicode -> u<len bytes>:<utf-8 string>
- - dict -> d<key><value><key><value>....e - sorted ascending by key
- - list -> l<item><item><item><item>....e
- - float -> f<len>:<float in hex - c99 hexadecimal literal format>
- - int -> i<number as text>e
- - true -> T
- - false -> F
- - none -> N
-additonal datatypes
- - byte str -> s<len bytes>:<string>
- - datetime -> D%Y-%m-%dT%H:%M:%S.%f
-xml like vocabulary
- - node -> N<name item><attr item><children item>
- an object with a name, attributes and children
- attributes is nominally a dict. children nominally list
- - extension -> X<item><item><item>
- like a node, but contains hyperlinks.
-todo: timezones, periods?
-todo: standard behaviour on duplicate keys
-expect some tweaks
+glyph-rpc tries to exploit http rather than simply tunnel requests over it.
-glyph evolved from trying to connect processes together, after some bad experiences
-with message queues exploding. http was known and loved throughout the company,
-and yet another ad-hoc rpc system was born.
-in the beginning, there was JSON and POST, and it mostly worked with the notable exception of UnicodeDecodeError.
-it didn't last very long. 8-bit data was a hard requirement, and so bencoding was used instead, with
-a small change to handle utf-8 data as well as bytes.
+The server:
+ import glyph
+ r = glyph.Router() # a wsgi application
+ @r.add()
+ def hello()
+ return "Hello World"
+ s = glyph.Server(r)
+ s.start()
+The client:
+ import glyph
+ server = glyph.get('http://server/')
+ print server.hello()
+There is no stub for the client, just the library.
+Adding a new function is simple:
+ @r.add()
+ def goodbye(name):
+ return "Goodbye " + name
+Or change the functions a little:
+ @r.add()
+ def hello(name="World"):
+ return "Hello "+name
+The client still works, without changes:
+ print server.hello()
+with very little changes to call new methods:
+ print server.hello('dave')
+ print server.goodbye('dave')
+functions can return lists, dicts, sets, byte strings, unicode,
+dates, booleans, ints & floats:
+ @r.add()
+ def woo():
+ return [1,True, None, False, "a",u"b"]
+functions can even return other functions that are mapped,
+through redirection:
+ @r.add()
+ @glyph.redirect()
+ def greeting(lang="en"):
+ if lang == "en":
+ return hello
+the client doesn't care:
+ greet = client.greeting()
+ print greet()
+glyph can map objects too:
+ @r.add()
+ @glyph.redirect()
+ def find_user(name):
+ user_id = database.find_user(name)
+ return User(user_id)
+ @r.add()
+ class User(glyph.Resource):
+ def __init__(self, id):
+ = id
+ def message(self, subject, body):
+ database.send_message(, subject, body)
+ def bio(self):
+ return database.get_bio(
+and the client can get a User and find details:
+ bob = server.find_user('bob')
+ bob.messsage('lol', 'feels good man')
+like before, new methods can be added without breaking old clients.
+unlike before, we can change object internals:
+ @r.add()
+ @glyph.redirect()
+ def find_user(name):
+ user_id, shard = database.find_user(name)
+ return User(user_id, shard)
+ @r.add()
+ class User(glyph.Resource):
+ def __init__(self, id, shard):
+ = id
+ self.shard = shard
+ ...
+Even though the internals have changed, the names haven't, so the client
+works as ever:
+ bob = server.find_user('bob')
+ bob.messsage('lol', 'feels good man')
+underneath all this - glyph maps all of this to http:
+ # by default, a server returns an object with a bunch
+ # of methods that redirect to the mapped obejcts
+ server = glyph.get('http://server/')
-we had a simple server stub that bound methods to urls and client code would call POST(url, {args}).
-and we passed a bunch of urls around to sub-processes in order for them to report things.
-although we had not hard coded urls into the system, the api was still rigid. adding a new method
-required passing in yet another url, or crafting urls per-request with client side logic.
+ # in this case, it will have an attribute 'find_user'
+ # find user is a special sort of object - a form
+ # it has a url, method and arguments attached.
-instead of passing around urls and writing stubs to use them each time, we figured we could pass around links and forms,
-which would *know* how to call the api, and it would fetch these *from* the api server itself.
-the urls contain enough information to make the call on the server end and are opaque to the client.
-we've needed to change the api numerous times since then. adding new methods doesn't break old clients.
-adding new state to the server doesn't break clients. using links and forms to interact with services is pleasant to
-use for the client, and flexible for the server.
+ # when we call server.find_user(...), it submits that form
+ # find_user redirects to a url for User(bob_id, cluster_id)
+ bob = server.find_user('bob')
-of all the terrible code i've written, this worked out pretty well so far.
+ # each object is mapped to a url, which contains the internal state
+ # of the object - i.e /User/?id=bob_id&cluster=cluster_id
+ # similarly, methods are mapped to a url too
+ # bob.message is a form pointing to /User/message?id=bo_id&cluster=cluster_id
+ bob.messsage('lol', 'feels good man')
-notable omissions:
- html/json/xml output
- content type overriding
- authentication handling
+although glyph maps urls to objects on the server side, these urls are
+opaque to the client - the server is free to change them to point to
+other objects, or to add new internal state without breaking the client.
+Client code doesn't need to know how to construct requests, or store all
+of the state needed to make requests - the server tells it, rather than
+the programmer.
4 doc/todo
@@ -134,6 +134,7 @@ todo code:
options support ?
def OPTIONS in mapper, options handling in Handler
+ although roy thinks it is is pretty useless.
http compliance:
set cache-control, expires, etags headers on get
@@ -204,6 +205,9 @@ todo code:
or x-www-urlencoding/form encoding?
to think?
+ uri templates
- shouldn't post to create? form should use get?
5 glyph/resource/
@@ -3,6 +3,7 @@
from werkzeug.wrappers import Response
from werkzeug.exceptions import HTTPException, NotFound, BadRequest, NotImplemented, MethodNotAllowed
from werkzeug.utils import redirect as Redirect
+from werkzeug.datastructures import ResponseCacheControl
from import CONTENT_TYPE, dump, parse, get, form, link, node, embed, ismethod, methodargs
@@ -15,15 +16,19 @@ class Handler(object):
+ CACHE=ResponseCacheControl()
def __init__(self, handler=None):
if handler:
+ self.cache=handler.cache
+ self.cache=self.CACHE
def parse(resource, data):
2  tests/
@@ -3,7 +3,7 @@
-class Test(unittest2.TestCase):
+class EncodingTest(unittest2.TestCase):
def testCase(self):
cases = [

0 comments on commit 0da2610

Please sign in to comment.
Something went wrong with that request. Please try again.