Recently, I've been generating Velocity components that should be evaluated at request-time but have at least some of the values they must work with calculated asynchronously when the component is generated. Here's an example that involves localizable content:
<div class="fubars"> $text.get("fubars.per.second", [ $fubarRate ]) </div>So let's say the ResourceBundle has a key in it for fubars.per.second like so
fubars.per.second=Number of Fubars Per Second: {0}If all of the calculation is done at request time, MessageTool would do its thing and this would Just Work. However, if $fubarRate is part of a heavier weight calculation that is done offline, we have to set it. So this is where I use Velocity to generate Velocity code:
#set($fr = '#set($fubarRate = ') #set($fr = "${fr} $measurement.fubarRate)") $frNotice the use of single quotes and double quotes to get the right combination of literal and interpolated evaluation. If my measurement object has a fubarRate property set to 42 then the last line simply outputs
#set($fubarRate = 42)and later, after the generated component gets its request time evaluation, the display is rendered as
<div class="fubars"> Number of Fubars Per Second: 42 </div>
Sure, I could generate my components with the web tier's ResourceBundle to get messages evaluated async as well. This would be 100% baking instead of 90% but it would be bad in other ways:
This separation of baking versus frying ain't new. I advocated it a long time ago in a talk at the O'Reilly Open Source Conference. I was hot on mod_perl and HTML::Mason back then (and, given a Perl environment, I still like them ...but I'd prefer a Java web application environment for i18n hands down), however the same basic ideas hold water using Velocity. At the time, application server misuse was in vogue and hundreds of thousands or even millions of dollars were being poured into "Enterprise Content Management" systems that coupled the CMS functions with those of publishing and request handling. Count that as millions of dollars squandered. There are still people struggling with the legacy of slow and stupid systems that can't be replaced because they spent too much money on it already (yea, what'd Forrest Gump say about stupid?). A few years later, when Aaron Swartz wrote about baking content he was insistent that he didn't care about performance, which is cool. The other benefits of baking that he mentions are perfectly valid. In fact, the publishing system at Salon.com distributes baked goods (HTML::Mason components generated with HTML::Mason components) to the web servers akin to Aaron's call to have something you can just do filesystem operations on. However, that's just the beginning. My maxim is that things that can scale independently should. The users of Bricolage, MovableType and other CMS and blog platforms that separate the management of editorial data, the publish cycle and content serving are enjoying that benefit right now.
java struts velocity i18n l10n CMS Bricolage MovableType mod_perl
( May 16 2005, 11:31:22 AM PDT ) Permalink