Policy Injection is similar to aspect-oriented programming in that it helps to break away cross-cutting concerns (authorization, logging) from the actual business logic. And yes, I basically copied that from the documentation.

Here is what policy injection means for me. I can build business logic, which performs only what I want it to. So I can have a function called AddUser and all that is done is the actual operation of adding that user. I can then attach policies to that function, which specify how to authorize and log it. This has a couple effects. One, it reduces the amount of code which in turn cleans up the code. Second, it allows me to customize cross-cutting concerns for different projects. If I have an administrative application, I need to authorize differently than a user application. Policy injection makes this possible without dirtying up the code.

After the initial few hours of getting used to using it, I am a huge proponent. I definitely recommend checking it out.

At first, I started working on my own implementation again, which provided a simple, in code, mapping technique. I quickly realized that mapping business entity relationships wasn’t easy though. After a couple hours of staring at my monitor hoping I could come up with an elegant solution, I decided to look at some 3rd party mappers. I had heard about NHibernate before and decided to give it a shot.

I spent about two days prototyping part of my database in NHibernate and my first impression is wow. Everything I could think of adding in an o/r mapper exists in NHibernate. Lazy loading, relationships, inheritance, component mapping, and so much more is possible. Over the past few weeks, my entire data access layer has been converted to NHibernate with overwhelming success. Like any library, it has its own quarks, but for the most part it speeds up development time while adding lots of power to my applications. I would recommend it to anyone.

NHibernate uses XML documents or .NET attributes to map an entity class to a database. I never thought I would use XML again after painful VB6 experiences, but I decided to give it another shot. To my surprise, I loved it. It allows for really clean separation of the database and code, which basically means cleaner code and a happier Brian. The only really annoying part of NHibernate is session management, but I’ll save that for part 2.

After a few days of digging though, I found a few great articles (Choosing Data Containers for .NET, On the Way to Mastering ASP.NET: Introducing Custom Entity Classes) on why not to use DataSets, as well as why you should. DataSets were designed for rapid application development, and seems to be Microsoft’s answer to the Rails ActiveRecord pattern. It is a quick and dirty way to build the data access layer for web applications. To be more specific, by quick I mean it allows version 1.0 of a product to be released quickly, because a data abstraction already exists. By dirty, I mean it isn’t an abstraction created for your application, it is a data abstraction created for any application.

Pro’s of ADO.net

• Fast development time
• In place architecture.
  Developing custom data access solutions can be very powerful, and very risky. A bad design can lead to bad performance and is inherently error prone
• Feature rich
  • Disconnected data
  • Concurrency
  • Many more…
• Easily serializable

Con’s of DataSets

• Large memory footprint
• Not object oriented.
  DataSet’s represent relation data, just like a database. But I don’t want an in-memory representation of the Employee table, I want Employee objects. Scott Hanselman sums it up pretty nicely:

  A DataSet is an object, right? But it’s not a Domain Object, it’s not an “Apple” or “Orange” – it’s an object of type “DataSet.” A DataSet is a bowl (one that knows about the backing Data Store). A DataSet is an object that knows how to HOLD Rows and Columns. It’s an object that knows a LOT about the Database. But I don’t want to return bowls. I want to return Domain Objects, like “Apples.”

• Not abstracted.
  Directly connected with database data and structure. So, any change to the database schema, could trickle down to the presentation layer.

Conclusion

ADO.NET is a huge improvement from old ADO, but don’t buy into the entire framework. Microsoft is trying to push DataSets but they might not always be the best solution. For enterprise applications, I don’t think DataSet’s are the answer. On the other hand, for smaller web application, it might be a perfect fit.

I initially had low expectations, expecting 1.0 to just clean up 0.7 code, with minimal feature additions. After I downloaded it, I found out how wrong I was. The 1.0 release fixed every negative comment I had about 0.7. Although it’s not perfect, it’s a lot better than it used to be.

The Good..

• Variables for defining which row and rowset classes to use in Zend_Db_Table. This enables custom row and rowset classes which I believe is a must-have.
• Support for compound primary keys
• Ability to cache the table meta data and prevent unnecessary database access

The Bad…

• Application level support for foreign keys, namely cascading deletes. This feature is awesome since a lot of the tables I work with are myisam and don’t support cascading deletes. Too bad it doesn’t truly cascade. If you delete a single record, it will cascade that delete to all related tables, but no cascading after that. Basically, a worthless feature that causes more confusion than it helps. Just use MySQL’s Innodb storage engine and clean up the code
• Complexity. Lots of classes and lots going on.

For many applications, the date created and date modified information for data is very useful, but annoying to set. Since Active Record abstracts all access to a database, these fields can be set easily and transparently. Many other implementations of the Active Record object already support this, so its not a new feature, but definitely a needed feature. The main issue is what timezone the server is in since NOW() doesn’t return GMT. It’s obvious that GMT is the preferred storage format of this data, so maybe passing a GMT date string from GMT is best.

Another feature that I just thought of is soft deletes. Basically a soft delete is where the data isn’t actually deleted, but a flag is set to show that it is deleted. This is very useful if you need to keep old data around for restores, or if you just want to have a record of everything. One of our new hires MikeT brought it up to me since his project has a soft delete requirement.

Hopefully I can get these added into my implementation fairly quickly since they are so simple.  Even though features like this are just the tip of the iceberg, they are crucial for creating a viable Active Record design.

The webmail application I work on is very different than many other Web 2.0 applications, which is why most of the existing frameworks don’t work for us.

• There are no page refreshes and therefore:
  • Most drawing is done by Javascript
  • All data transfered is JSON encoded
• Email, Calendar, Tasks, Contacts, Settings are all combined into 1 application which means:
  • Large memory footprint
  • Javascript performance very important
  • On the fly loading of HTML, Javascript, and CSS is a must

For the past few months, I have been thinking about what I really need in a web framework. Just like Rails, an MVC framework would need to be the base. In Part 1, I will just talk about the MVC foundation for this framework.

Model

An ActiveRecord type object would provide the MySQL data abstraction. I have mentioned that I have been working on my own implementation for a while, so I won’t go into too much detail on this. Basically, for a complicated web application, most existing model abstractions don’t work very well. With complicated JOIN’s, WHERE’s, and subquerys, something different must be used.

View

A templating mechanism must exist to have a complete web framework. Current view abstractions won’t work in webmail. Since webmail never refreshes, going to the server every time we need to populate a template is troublesome. The templating mechanism in the better web framework would allow a template to be used in PHP or in Javascript.

If a template is used in Javascript, once it is downloaded once, it wouldn’t have to be downloaded again. The hard part about a Javascript template is the inability to have loops and the need for slow regular expression parsing.

If a template request in Javascript has to make a server request every time a template needs to be populated, speed becomes an issue. This is fine for large templates, but what about small widget templates? Making a server request for a small element increases 1. latency and 2. server load which in turn increases latency even more.

Controller

Controllers are the backbone of a web framework. I like how Rail works with a controller class where each data request is mapped to a function call. This seems effective to me and needs little modification.

Conclusion

I know what you must be thinking: Your MVC design isn’t that much different than Rails. Well, that’s true, but remember this is part 1. The MVC design is just the foundation of a better web framework. Rails is popular because it is useful, so it is definitely a good starting point for developing a new framework.

Shout out to Mike for his work on Javascript templates.

At Webmail.us, I have decided to use my side-project hours to work on my own version of the active record pattern. It has been a slow going process, since I rarely find free hours to work on it, but I have finally created a workable version. But, before we get to the inner workings of my implementation, I need to give credit where credit is due.

Rails Active Record:

Since the first active record pattern I encountered was the Rails implementation, I decided to start there with my first PHP implementation. It wasn’t long before I realized the PHP vs. Ruby differences were going to be a serious problem.

I wanted the find functions to be static, because an object shouldn’t have to populate itself after it has been instantiated. Unfortunately, in PHP, static methods aren’t capable of polymorphism (which makes sense since $this isn’t applicable). Therefore, overloading them becomes impossible.

CakePHP solved this by making all the find functions non-static, but I didn’t like this solution either, because it doesn’t make sense for the previously stated reasons. Then I found …..

Zend DB:

The Zend framework, while still very new, has some powerful features, namely the DB section. When I first saw this implementation of the active record pattern, I knew this was the starting point for my own version.

Zend DB has three main classes, Zend_Db_Table, Zend_Db_Table_Row, and Zend_Db_Table_Rowset. This design is much more flexible, and makes more sense to me than Rails’ and CakePHP’s implementations. Each database object is abstracted to its own classes, aka tables, rows, and rowsets (or arrays of rows).

At first, I almost decided to use this implementation and forgo the long process of creating my own. But, it fell short of my requirements for a few different reasons. The biggest problem was that the Row and RowSet class names are hard-coded in the Table class. This means overloading them is impossible:

1
2
3
4
5

return new Zend_Db_Table_Row(array(
    'db'    => $this->_db,
    'table' => $this,
    'data'  => $this->_fetch('Row', $where, $order, 1),
));

It also uses prepared statements, which I have never used and don’t really have a desire to. They are definitely slower than straight MySQL calls and I don’t have the mysqli extension installed so that is out (Although I should probably work on getting that extension installed).