Abstract

Arel is a Relational Algebra for Ruby. It 1) simplifies the generation complex of SQL queries and it 2) adapts to various RDBMS systems. It is intended to be a framework framework; that is, you can build your own ORM with it, focusing on innovative object and collection modeling as opposed to database compatibility and query generation.

Status

For the moment, Arel uses ActiveRecord's connection adapters to connect to the various engines, connection pooling, perform quoting, and do type conversion. On the horizon is the use of DataObjects instead.

The long term goal, following both LINQ and DataMapper, is to have Arel adapt to engines beyond RDBMS, including XML, IMAP, YAML, etc.

A Gentle Introduction

Generating a query with ARel is simple. For example, in order to produce

SELECT * FROM users

you construct a table relation and convert it to sql:

users = Table(:users)
users.to_sql

In fact, you will probably never call #to_sql. Rather, you'll work with data from the table directly. You can iterate through all rows in the users table like this:

users.each { |user| ... }

In other words, Arel relations implement Ruby's Enumerable interface. Let's have a look at a concrete example:

users.first # => { users[:id] => 1, users[:name] => 'bob' }

As you can see, Arel converts the rows from the database into a hash, the values of which are sublimated to the appropriate Ruby primitive (integers, strings, and so forth).

More Sophisticated Queries Relations

Here is a whirlwind tour through the most common relational operators. These will probably cover 80% of all interaction with the database.

First is the 'restriction' operator, where:

users.where(users[:name].eq('amy'))
# => SELECT * FROM users WHERE users.name = 'amy'

What would, in SQL, be part of the SELECT clause is called in Arel a projection:

users.project(users[:id]) # => SELECT users.id FROM users

Joins resemble SQL strongly:

users.join(photos).on(users[:id].eq(photos[:user_id]))
# => SELECT * FROM users INNER JOIN photos ON users.id = photos.user_id

What are called LIMIT and OFFSET in SQL are called take and skip in Arel:

users.take(5) # => SELECT * FROM users LIMIT 5
users.skip(4) # => SELECT * FROM users OFFSET 4

GROUP BY is called group:

users.group(users[:name]) # => SELECT * FROM users GROUP BY name

The best property of the Relational Algebra is its "composability", or closure under all operations. For example, to select AND project, just "chain" the method invocations:

users                                 \
  .where(users[:name].eq('amy'))      \
  .project(users[:id])                \
# => SELECT users.id FROM users WHERE users.name = 'amy'

All operators are chainable in this way, and they are chainable any number of times, in any order.

users.where(users[:name].eq('bob')).where(users[:age].lt(25))

Of course, many of the operators take multiple arguments, so the last example can be written more tersely:

users.where(users[:name].eq('bob'), users[:age].lt(25))

The OR operator is not yet supported. It will work like this:

users.where(users[:name].eq('bob').or(users[:age].lt(25)))

The AND operator will behave similarly.

Finally, most operations take a block form. For example:

Table(:users)                    \
  .where   { |u| u[:id].eq(1) } \
  .project { |u| u[:id] }

This provides a (sometimes) convenient alternative syntax.

The Crazy Features

The examples above are fairly simple and other libraries match or come close to matching the expressiveness of Arel (e.g., Sequel in Ruby).

Complex Joins

Where Arel really shines in its ability to handle complex joins and aggregations. As a first example, let's consider an "adjacency list", a tree represented in a table. Suppose we have a table comments, representing a threaded discussion:

comments = Table(:comments)

And this table has the following attributes:

comments.attributes # => [comments[:id], comments[:body], comments[:parent_id]]

The parent_id column is a foreign key from the comments table to itself. Now, joining a table to itself requires aliasing in SQL. In fact, you may alias in Arel as well:

replies = comments.alias
comments_with_replies = \
  comments.join(replies).on(replies[:parent_id].eq(comments[:id]))
# => SELECT * FROM comments INNER JOIN comments AS comments_2 WHERE comments_2.parent_id = comments.id

The call to #alias is actually optional: Arel will always produce a unique name for every table joined in the relation, and it will always do so deterministically to exploit query caching. Explicit aliasing is more common, however. When you want to extract specific slices of data, aliased tables are a necessity. For example to get just certain columns from the row, treat a row like a hash:

comments_with_replies.first[replies[:body]]

This will return the first comment's reply's body.

If you don't need to extract the data later (for example, you're simply doing a join to find comments that have replies, you don't care what the content of the replies are), the block form may be preferable:

comments.join(comments) { |comments, replies| replies[:parent_id].eq(comments[:id]) }
# => SELECT * FROM comments INNER JOIN comments AS comments_2 WHERE comments_2.parent_id = comments.id

Note that you do NOT want to do something like:

comments.join(comments, comments[:parent_id].eq(comments[:id]))
# => SELECT * FROM comments INNER JOIN comments AS comments_2 WHERE comments.parent_id = comments.id

This does NOT have the same meaning as the previous query, since the comments[:parent_id] reference is effectively ambiguous.

Complex Aggregations

My personal favorite feature of Arel, certainly the most difficult to implement, and possibly only of marginal value, is closure under joining even in the presence of aggregations. This is a feature where the Relational Algebra is fundamentally easier to use than SQL. Think of this as a preview of the kind of radical functionality that is to come, stuff no other "ORM" is doing.

The easiest way to introduce this is in SQL. Your task is to get all users and the count of their associated photos. Let's start from the inside out:

SELECT count(*)
FROM photos
GROUP BY user_id

Now, we'd like to join this with the user table. Naively, you might try to do this:

SELECT users.*, count(photos.id)
FROM users
LEFT OUTER JOIN photos
  ON users.id = photos.user_id
GROUP BY photos.user_id

Of course, this has a slightly different meaning than our intended query. This is actually a fairly advanced topic in SQL so let's see why this doesn't work step by step. Suppose we have these records in our users table:

mysql> select * from users;
+------+--------+
| id   | name   |
+------+--------+
|    1 | hai    |
|    2 | bai    |
|    3 | dumpty |
+------+--------+

And these in the photos table:

mysql> select * from photos;
+------+---------+-----------+
| id   | user_id | camera_id |
+------+---------+-----------+
|    1 |       1 |         1 |
|    2 |       1 |         1 |
|    3 |       1 |         1 |
+------+---------+-----------+

If we perform the above, incorrect query, we get the following:

mysql> select users.*, count(photos.id) from users left outer join photos on users.id = photos.user_id limit 3 group by user_id;
+------+------+------------------+
| id   | name | count(photos.id) |
+------+------+------------------+
|    2 | bai  |                0 |
|    1 | hai  |                3 |
+------+------+------------------+

As you can see, we're completely missing data for user with id 3. dumpty has no photos, neither does bai. But strangely bai appeared and dumpty didn't! The reason is that the GROUP BY clause is aggregating on both tables, not just the photos table. All users without photos have a photos.id of null (thanks to the left outer join). These are rolled up together and an arbitrary user wins. In this case, bai not dumpty.

SELECT users.*, photos_aggregation.cnt
FROM users
LEFT OUTER JOIN (SELECT user_id, count(*) as cnt FROM photos GROUP BY user_id) AS photos_aggregation
  ON photos_aggregation.user_id = users.id