CHAPTER

TWENTYFIVE

MUCH ADO ABOUT TRAVERSAL

(Or, why you should care about it)

latex-note.png

This chapter was adapted, with permission, from a blog post by Rob Miller, originally published at http://blog.nonsequitarian.org/2010/much-ado-about-traversal/ .

Traversal is an alternative to URL dispatch which allows Pyramid applications to map URLs to code.

latex-note.png

Ex-Zope users who are already familiar with traversal and view lookup conceptually may want to skip directly to the Traversal chapter, which discusses technical details. This chapter is mostly aimed at people who have previous Pylons experience or experience in another framework which does not provide traversal, and need an introduction to the “why” of traversal.

Some folks who have been using Pylons and its Routes-based URL matching for a long time are being exposed for the first time, via Pyramid, to new ideas such as “traversal” and “view lookup” as a way to

273

25. MUCH ADO ABOUT TRAVERSAL

route incoming HTTP requests to callable code. Some of the same folks believe that traversal is hard to understand. Others question its usefulness; URL matching has worked for them so far, why should they even consider dealing with another approach, one which doesn’t fit their brain and which doesn’t provide any immediately obvious value?

You can be assured that if you don’t want to understand traversal, you don’t have to. You can happily build Pyramid applications with only URL dispatch. However, there are some straightforward, real-world use cases that are much more easily served by a traversal-based approach than by a pattern-matching mechanism. Even if you haven’t yet hit one of these use cases yourself, understanding these new ideas is worth the effort for any web developer so you know when you might want to use them. Traversal is actually a straightforward metaphor easily comprehended by anyone who’s ever used a run-of-the-mill file system with folders and files.

25.1 URL Dispatch

Let’s step back and consider the problem we’re trying to solve. An HTTP request for a particular path has been routed to our web application. The requested path will possibly invoke a specific view callable function defined somewhere in our app. We’re trying to determine which callable function, if any, should be invoked for a given requested URL.

Many systems, including Pyramid, offer a simple solution. They offer the concept of “URL matching”. URL matching approaches this problem by parsing the URL path and comparing the results to a set of registered “patterns”, defined by a set of regular expressions, or some other URL path templating syntax. Each pattern is mapped to a callable function somewhere; if the request path matches a specific pattern, the associated function is called. If the request path matches more than one pattern, some conflict resolution scheme is used, usually a simple order precedence so that the first match will take priority over any subsequent matches. If a request path doesn’t match any of the defined patterns, a “404 Not Found” response is returned.

In Pyramid, we offer an implementation of URL matching which we call URL dispatch. Using Pyramid syntax, we might have a match pattern such as /{userid}/photos/{photoid}, mapped to a photo_view() function defined somewhere in our code. Then a request for a path such as /joeschmoe/photos/photo1 would be a match, and the photo_view() function would be invoked to handle the request. Similarly, /{userid}/blog/{year}/{month}/{postid} might map to a blog_post_view() function, so /joeschmoe/blog/2010/12/urlmatching would trigger the function, which presumably would know how to find and render the urlmatching blog post.

274

25.2. HISTORICAL REFRESHER

25.2 Historical Refresher

Now that we’ve refreshed our understanding of URL dispatch, we’ll dig in to the idea of traversal. Before we do, though, let’s take a trip down memory lane. If you’ve been doing web work for a while, you may remember a time when we didn’t have fancy web frameworks like Pylons and Pyramid. Instead, we had general purpose HTTP servers that primarily served files off of a file system. The “root” of a given site mapped to a particular folder somewhere on the file system. Each segment of the request URL path represented a subdirectory. The final path segment would be either a directory or a file, and once the server found the right file it would package it up in an HTTP response and send it back to the client. So serving up a request for /joeschmoe/photos/photo1 literally meant that there was a joeschmoe folder somewhere, which contained a photos folder, which in turn contained a photo1 file. If at any point along the way we find that there is not a folder or file matching the requested path, we return a 404 response.

As the web grew more dynamic, however, a little bit of extra complexity was added. Technologies such as CGI and HTTP server modules were developed. Files were still looked up on the file system, but if the file ended with (for example) .cgi or .php, or if it lived in a special folder, instead of simply sending the file to the client the server would read the file, execute it using an interpreter of some sort, and then send the output from this process to the client as the final result. The server configuration specified which files would trigger some dynamic code, with the default case being to just serve the static file.

25.3 Traversal (aka Resource Location)

Believe it or not, if you understand how serving files from a file system works, you understand traversal. And if you understand that a server might do something different based on what type of file a given request specifies, then you understand view lookup.

The major difference between file system lookup and traversal is that a file system lookup steps through nested directories and files in a file system tree, while traversal steps through nested dictionary-type objects in a resource tree. Let’s take a detailed look at one of our example paths, so we can see what I mean:

The path /joeschmoe/photos/photo1, has four segments: /, joeschmoe, photos and photo1. With file system lookup we might have a root folder (/) containing a nested folder (joeschmoe), which contains another nested folder (photos), which finally contains a JPG file (photo1). With traversal, we instead have a dictionary-like root object. Asking for the joeschmoe key gives us another dictionary-like object. Asking this in turn for the photos key gives us yet another mapping object, which finally (hopefully) contains the resource that we’re looking for within its values, referenced by the photo1 key.

In pure Python terms, then, the traversal or “resource location” portion of satisfying the /joeschmoe/photos/photo1 request will look something like this pseudocode:

275

25. MUCH ADO ABOUT TRAVERSAL

get_root()[’joeschmoe’][’photos’][’photo1’]

get_root() is some function that returns a root traversal resource. If all of the specified keys exist, then the returned object will be the resource that is being requested, analogous to the JPG file that was retrieved in the file system example. If a KeyError is generated anywhere along the way, Pyramid will return 404. (This isn’t precisely true, as you’ll see when we learn about view lookup below, but the basic idea holds.)

25.4 What Is a “Resource”?

“Files on a file system I understand”, you might say. “But what are these nested dictionary things? Where do these objects, these ‘resources’, live? What are they?”

Since Pyramid is not a highly opinionated framework, it makes no restriction on how a resource is implemented; a developer can implement them as he wishes. One common pattern used is to persist all of the resources, including the root, in a database as a graph. The root object is a dictionary-like object. Dictionary-like objects in Python supply a __getitem__ method which is called when key lookup is done. Under the hood, when adict is a dictionary-like object, Python translates adict[’a’] to adict.__getitem__(’a’). Try doing this in a Python interpreter prompt if you don’t believe us:

The dictionary-like root object stores the ids of all of its subresources as keys, and provides a __getitem__ implementation that fetches them. So get_root() fetches the unique root object, while get_root()[’joeschmoe’] returns a different object, also stored in the database, which in turn has its own subresources and __getitem__ implementation, etc. These resources might be persisted in a relational database, one of the many “NoSQL” solutions that are becoming popular these days, or anywhere else, it doesn’t matter. As long as the returned objects provide the dictionary-like API (i.e. as long as they have an appropriately implemented __getitem__ method) then traversal will work.

In fact, you don’t need a “database” at all. You could use plain dictionaries, with your site’s URL structure hard-coded directly in the Python source. Or you could trivially implement a set of objects with

276