= Querying in Sequel

This guide is based on http://guides.rubyonrails.org/active_record_querying.html

== Purpose of this Guide

Sequel is a flexible and powerful database library
that supports a wide variety of different querying methods.  This guide
aims to be a introduction to Sequel's querying support.

While you can use raw SQL with Sequel, a large part of the
advantage you get from using Sequel is Sequel's ability to abstract
SQL from you and give you a pure-ruby interface. Sequel also ships with
a {core_extensions extension}[rdoc-ref:doc/core_extensions.rdoc],
which adds methods to core ruby types to work with Sequel.

== Retrieving Objects

Sequel provides a few separate methods for retrieving objects from the
database.  The underlying method is Sequel::Dataset#each, which yields each
row as the Sequel::Database provides it.  However, while Dataset#each can and
often is used directly, in many cases there is a more convenient retrieval
method you can use.

=== Sequel::Dataset

If you are new to Sequel and aren't familiar with Sequel, you should probably
read the {"Dataset Basics" guide}[rdoc-ref:doc/dataset_basics.rdoc],
then come back here.

=== Retrieving a Single Object

Sequel offers quite a few ways to to retrieve a single object.

==== Using a Primary Key [Sequel::Model]

The <tt>Sequel::Model.[]</tt> is the easiest method to use to find a model instance
by its primary key value:

  # Find artist with primary key (id) 1
  artist = Artist[1]
  # SELECT * FROM artists WHERE (id = 1)
  # => #<Artist @values={:name=>"YJM", :id=>1}>

If there is no record with the given primary key, nil will be returned. If you want
to raise an exception if no record is found, you can use <tt>Sequel::Model.with_pk!</tt>:

  artist = Artist.with_pk!(1)

==== Using +first+

If you want the first record in the dataset,
<tt>Sequel::Dataset#first</tt> is probably the most obvious method to use:

  artist = Artist.first
  # SELECT * FROM artists LIMIT 1
  # => #<Artist @values={:name=>"YJM", :id=>1}>

Any options you pass to +first+ will be used as a filter:

  artist = Artist.first(name: 'YJM')
  # SELECT * FROM artists WHERE (name = 'YJM') LIMIT 1
  # => #<Artist @values={:name=>"YJM", :id=>1}>
  
  artist = Artist.first(Sequel.like(:name, 'Y%'))
  # SELECT * FROM artists WHERE (name LIKE 'Y%' ESCAPE '\') LIMIT 1
  # => #<Artist @values={:name=>"YJM", :id=>1}>
  
If there is no matching row, +first+ will return nil.  If you want to
raise an exception instead, use <tt>first!</tt>.

<tt>Sequel::Dataset#[]</tt> is basically an alias for +first+, except it
requires an argument:

  DB[:artists][{name: 'YJM'}]
  # SELECT * FROM artists WHERE (name = 'YJM') LIMIT 1
  # => {:name=>"YJM", :id=>1}

Note that while Model.[] allows you to pass a primary key directly,
Dataset#[] does not (unless it is a model dataset).

==== Using +last+

If you want the last record in the dataset,
<tt>Sequel::Dataset#last</tt> is an obvious method to use.  +last+ requires the
dataset be ordered, unless the dataset is a model dataset in which case +last+
will do a reverse order by the primary key field:

  artist = Artist.last
  # SELECT * FROM artists ORDER BY id DESC LIMIT 1
  # => #<Artist @values={:name=>"YJM", :id=>1}>

Note:

1. +last+ is equivalent to running a +reverse.first+ query, in other words it reverses the order of the dataset and then calls +first+.  This is why +last+ raises a Sequel::Error when there is no order on a plain dataset - because it will provide the same record as +first+, and most users will find that confusing.
2. +last+ is not necessarily going to give you the last record in the dataset unless you give the dataset an unambiguous order.
3. +last+ will ignore +limit+ if chained together in a query because it sets a limit of 1 if no arguments are given.

==== Retrieving a Single Column Value

Sometimes, instead of wanting an entire row, you only want the value of
a specific column.  For this <tt>Sequel::Dataset#get</tt> is the method
you want:

  artist_name = Artist.get(:name)
  # SELECT name FROM artists LIMIT 1
  # => "YJM"

==== Retrieving a Multiple Column Values

If you want the value for multiple columns, you can pass an array to
<tt>Sequel::Dataset#get</tt>:

  artist_id, artist_name = Artist.get([:id, :name])
  # SELECT id, name FROM artists LIMIT 1
  # => [1, "YJM"]

=== Retrieving Multiple Objects

==== As an Array of Hashes or Model Objects

In many cases, you want an array of all of the rows associated with the
dataset, in which case <tt>Sequel::Dataset#all</tt> is the method you
want to use:

  artists = Artist.all
  # SELECT * FROM artists
  # => [#<Artist @values={:name=>"YJM", :id=>1}>,
  #     #<Artist @values={:name=>"AS", :id=>2}>]

==== Using an Enumerable Interface

<tt>Sequel::Dataset</tt> uses an Enumerable Interface, so it provides a
method named each that yields hashes or model objects as they are retrieved
from the database:

  Artist.each{|x| p x.name}
  # SELECT * FROM artists
  "YJM"
  "AS"

This means that all of the methods in the Enumerable module are available,
such as +map+:

  artist_names = Artist.map{|x| x.name}
  # SELECT * FROM artists
  # => ["YJM", "AS"]

==== As an Array of Column Values

Sequel also has an extended +map+ method that takes an argument.  If you
provide an argument to +map+, it will return an array of values for the
given column.  So the previous example can be handled more easily with:

  artist_names = Artist.map(:name)
  # SELECT * FROM artists
  # => ["YJM", "AS"]

One difference between these two ways of returning an array of values is
that providing +map+ with an argument is really doing:

  artist_names = Artist.map{|x| x[:name]} # not x.name

Note that regardless of whether you provide +map+ with an argument, it
does not modify the columns selected.  If you only want to select a
single column and return an array of the columns values, you can use
+select_map+:

  artist_names = Artist.select_map(:name)
  # SELECT name FROM artists
  # => ["YJM", "AS"]

It's also common to want to order such a map, so Sequel provides a
+select_order_map+ method as well:

  artist_names = Artist.select_order_map(:name)
  # SELECT name FROM artists ORDER BY name
  # => ["AS", "YJM"]

In all of these cases, you can provide an array of column symbols and
an array of arrays of values will be returned:

  artist_names = Artist.select_map([:id, :name])
  # SELECT id, name FROM artists
  # => [[1, "YJM"], [2, "AS"]]

==== As a Hash

Sequel makes it easy to take an SQL query and return it as a ruby hash,
using the +as_hash+ method:

  artist_names = Artist.as_hash(:id, :name)
  # SELECT * FROM artists
  # => {1=>"YJM", 2=>"AS"}

As you can see, the +as_hash+ method uses the first symbol as the key
and the second symbol as the value.  So if you swap the two arguments the hash
will have its keys and values transposed:

  artist_names = Artist.as_hash(:name, :id)
  # SELECT * FROM artists
  # => {"YJM"=>1, "AS"=>2}

Now what if you have multiple values for the same key?  By default, +as_hash+
will just have the last matching value.  If you care about all matching values,
use +to_hash_groups+, which makes the values of the array an array of matching
values, in the order they were received:

  artist_names = Artist.to_hash_groups(:name, :id)
  # SELECT * FROM artists
  # => {"YJM"=>[1, 10, ...], "AS"=>[2, 20, ...]}

If you only provide one argument to +as_hash+, it uses the entire hash
or model object as the value:

  artist_names = DB[:artists].as_hash(:name)
  # SELECT * FROM artists
  # => {"YJM"=>{:id=>1, :name=>"YJM"}, "AS"=>{:id=>2, :name=>"AS"}}

and +to_hash_groups+ works similarly:

  artist_names = DB[:artists].to_hash_groups(:name)
  # SELECT * FROM artists
  # => {"YJM"=>[{:id=>1, :name=>"YJM"}, {:id=>10, :name=>"YJM"}], ...}

Model datasets have a +as_hash+ method that can be called without any
arguments, in which case it will use the primary key as the key and
the model object as the value.  This can be used to easily create an
identity map:

  artist_names = Artist.as_hash
  # SELECT * FROM artists
  # => {1=>#<Artist @values={:id=>1, :name=>"YGM"}>,
  #     2=>#<Artist @values={:id=>2, :name=>"AS"}>}

There is no equivalent handling to +to_hash_groups+, since there would
only be one matching record, as the primary key must be unique.

Note that +as_hash+ never modifies the columns selected.  However, just
like Sequel has a +select_map+ method to modify the columns selected and
return an array, Sequel also has a +select_hash+ method to modify the
columns selected and return a hash:

  artist_names = Artist.select_hash(:name, :id)
  # SELECT name, id FROM artists
  # => {"YJM"=>1, "AS"=>2}

Likewise, +select_hash_groups+ also exists:

  artist_names = Artist.select_hash_groups(:name, :id)
  # SELECT name, id FROM artists
  # => {"YJM"=>[1, 10, ...], "AS"=>[2, 20, ...]}

== Modifying datasets 

Note that the retrieval methods discussed above just return
the row(s) included in the existing dataset.  In most cases,
you aren't interested in every row in a table, but in a subset
of the rows, based on some criteria.  In Sequel, filtering
the dataset is generally done separately than retrieving
the records.

There are really two types of dataset methods that you will
be using:

1. Methods that return row(s), discussed above
2. Methods that return modified datasets, discussed below

Sequel datasets are frozen and use a method chaining, functional style API
that returns modified datasets.  Let's start with a simple example.

This is a basic dataset that includes all records in the
table +artists+:

  ds1 = DB[:artists]
  # SELECT * FROM artists

Let's say we are only interested in the artists whose names
start with "A":

  ds2 = ds1.where(Sequel.like(:name, 'A%'))
  # SELECT * FROM artists WHERE (name LIKE 'A%' ESCAPE '\')

Here we see that +where+ returns a dataset that adds a +WHERE+
clause to the query.  It's important to note that +where+ does
not modify the receiver:

  ds1
  # SELECT * FROM artists
  ds2
  # SELECT * FROM artists WHERE (name LIKE 'A%' ESCAPE '\')
 
In Sequel, dataset methods do not modify the dataset itself, so you can freely use the dataset in multiple
places without worrying that its usage in one place will affect its usage
in another place.  This is what is meant by a functional style API.

Let's say we only want to select the id and name columns, and that
we want to order by name:

  ds3 = ds2.order(:name).select(:id, :name)
  # SELECT id, name FROM artists WHERE (name LIKE 'A%' ESCAPE '\') ORDER BY name
  
Note how you don't need to assign the returned value of order to a variable,
and then call select on that.  Because order just returns a dataset, you can
call select directly on the returned dataset.  This is what is meant by a
method chaining API.

Also note how you can call methods that modify different clauses in any order.
In this case, the WHERE clause was added first, then the ORDER clause, then the
SELECT clause was modified.  This makes for a flexible API, where you can modify
any part of the query at any time.

== Filters

Filtering is probably the most common dataset modifying action done in Sequel.
Both the +where+ and +filter+ methods filter the dataset by modifying the
dataset's WHERE clause.  Both accept a wide variety of input formats, discussed
below.

=== Hashes

The most common format for providing filters is via a hash.  In general, Sequel
treats conditions specified with a hash as equality, inclusion, or identity.  What type
of condition is used depends on the values in the hash.

Unless Sequel has special support for the value's class, it uses a simple
equality statement:

  Artist.where(id: 1)
  # SELECT * FROM artists WHERE (id = 1)

  Artist.where(name: 'YJM')
  # SELECT * FROM artists WHERE (name = 'YJM')

For arrays, Sequel uses the IN operator with a value list:

  Artist.where(id: [1, 2])
  # SELECT * FROM artists WHERE (id IN (1, 2))

For datasets, Sequel uses the IN operator with a subselect:

  Artist.where(id: Album.select(:artist_id))
  # SELECT * FROM artists WHERE (id IN (
  #   SELECT artist_id FROM albums))

For boolean values such as nil, true, and false, Sequel uses the IS operator:

  Artist.where(id: nil)
  # SELECT * FROM artists WHERE (id IS NULL)

For ranges, Sequel uses a pair of inequality statements:

  Artist.where(id: 1..5)
  # SELECT * FROM artists WHERE ((id >= 1) AND (id <= 5))

  Artist.where(id: 1...5)
  # SELECT * FROM artists WHERE ((id >= 1) AND (id < 5))

Finally, for regexps, Sequel uses an SQL regular expression.  Note that this
is probably only supported on PostgreSQL and MySQL.

  Artist.where(name: /JM$/)
  # SELECT * FROM artists WHERE (name ~ 'JM$')

If there are multiple arguments in the hash, the filters are ANDed together:

  Artist.where(id: 1, name: /JM$/)
  # SELECT * FROM artists WHERE ((id = 1) AND (name ~ 'JM$'))

This works the same as if you used two separate +where+ calls:

  Artist.where(id: 1).where(name: /JM$/)
  # SELECT * FROM artists WHERE ((id = 1) AND (name ~ 'JM$'))

=== Array of Two Element Arrays

If you use an array of two element arrays, it is treated as a hash.  The only
advantage to using an array of two element arrays is that it allows you to
duplicate keys, so you can do:

  Artist.where([[:name, /JM$/], [:name, /^YJ/]])
  # SELECT * FROM artists WHERE ((name ~ 'JM$')) AND ((name ~ '^YJ'))

=== Virtual Row Blocks

If a block is passed to a filter, it is treated as a virtual row block:

  Artist.where{id > 5}
  # SELECT * FROM artists WHERE (id > 5)

You can learn more about virtual row blocks in the {"Virtual Rows" guide}[rdoc-ref:doc/virtual_rows.rdoc].

You can provide both regular arguments and a block, in which case the results
will be ANDed together:

  Artist.where(name: 'A'...'M'){id > 5}
  # SELECT * FROM artists WHERE ((name >= 'A') AND (name < 'M') AND (id > 5))

Using virtual row blocks, what you can do with single entry hash or an array with
a single two element array can also be done using the =~ method:

  Artist.where{id =~ 5}
  # SELECT * FROM artists WHERE (id = 5)

=== Symbols

If you have a boolean column in the database, and you want only true
values, you can just provide the column symbol to filter:

  Artist.where(:retired)
  # SELECT * FROM artists WHERE retired

=== SQL::Expression

Sequel has a DSL that allows easily creating SQL expressions.  These SQL
expressions are instances of subclasses of Sequel::SQL::Expression. You've
already seen an example earlier:

  Artist.where(Sequel.like(:name, 'Y%'))
  # SELECT * FROM artists WHERE name LIKE 'Y%' ESCAPE '\'

In this case Sequel.like returns a Sequel::SQL::BooleanExpression object,
which is used directly in the filter.

You can use the DSL to create arbitrarily complex expressions.  SQL::Expression
objects can be created via singleton methods on the Sequel module.  The most common
method is Sequel.[], which takes any object and wraps it in a SQL::Expression
object.  In most cases, the SQL::Expression returned supports the & operator for
+AND+, the | operator for +OR+, and the ~ operator for inversion:

  Artist.where(Sequel.like(:name, 'Y%') & (Sequel[{b: 1}] | Sequel.~(c: 3)))
  # SELECT * FROM artists WHERE ((name LIKE 'Y%' ESCAPE '\') AND ((b = 1) OR (c != 3)))

You can combine these expression operators with the virtual row support:

  Artist.where{(a > 1) & ~((b(c) < 1) | d)}
  # SELECT * FROM artists WHERE ((a > 1) AND (b(c) >= 1) AND NOT d)

Note the use of parentheses when using the & and | operators, as they have lower
precedence than other operators.  The following will not work:

  Artist.where{a > 1 & ~(b(c) < 1 | d)}
  # Raises a TypeError

=== Strings with Placeholders

Assuming you want to get your hands dirty and use SQL fragments in filters, Sequel allows you
to do so if you explicitly mark the strings as literal strings using +Sequel.lit+. You can
use placeholders in the string and pass arguments for the placeholders:

  Artist.where(Sequel.lit("name LIKE ?", 'Y%'))
  # SELECT * FROM artists WHERE (name LIKE 'Y%')

This is the most common type of placeholder, where each question mark is substituted
with the next argument:

  Artist.where(Sequel.lit("name LIKE ? AND id = ?", 'Y%', 5))
  # SELECT * FROM artists WHERE (name LIKE 'Y%' AND id = 5)

You can also use named placeholders with a hash, where the named placeholders use
colons before the placeholder names:

  Artist.where(Sequel.lit("name LIKE :name AND id = :id", name: 'Y%', id: 5))
  # SELECT * FROM artists WHERE (name LIKE 'Y%' AND id = 5)

You don't have to provide any placeholders if you don't want to:

  Artist.where(Sequel.lit("id = 2"))
  # SELECT * FROM artists WHERE id = 2

However, if you are using any untrusted input, you should definitely be using placeholders.
In general, unless you are hardcoding values in the strings, you should use placeholders.
You should never pass a string that has been built using interpolation, unless you are
sure of what you are doing.

  Artist.where(Sequel.lit("id = #{params[:id]}")) # Don't do this!
  Artist.where(Sequel.lit("id = ?", params[:id])) # Do this instead
  Artist.where(id: params[:id].to_i)              # Even better

=== Inverting

You may be wondering how to specify a not equals condition in Sequel, or the NOT IN
operator.  Sequel has generic support for inverting conditions, so to write a not
equals condition, you write an equals condition, and invert it:

  Artist.where(id: 5).invert
  # SELECT * FROM artists WHERE (id != 5)

Note that +invert+ inverts the entire filter:

  Artist.where(id: 5).where{name > 'A'}.invert
  # SELECT * FROM artists WHERE ((id != 5) OR (name <= 'A'))

In general, +invert+ is used rarely, since +exclude+ allows you to invert only specific
filters:

  Artist.exclude(id: 5)
  # SELECT * FROM artists WHERE (id != 5)
  
  Artist.where(id: 5).exclude{name > 'A'}
  # SELECT * FROM artists WHERE ((id = 5) AND (name <= 'A')

So to do a NOT IN with an array:

  Artist.exclude(id: [1, 2])
  # SELECT * FROM artists WHERE (id NOT IN (1, 2))

Or to use the NOT LIKE operator:

  Artist.exclude(Sequel.like(:name, '%J%'))
  # SELECT * FROM artists WHERE (name NOT LIKE '%J%' ESCAPE '\')

You can use Sequel.~ to negate expressions:

  Artist.where(Sequel.~(id: 5))
  # SELECT * FROM artists WHERE id != 5

On Sequel expression objects, you can use ~ to negate them:
 
  Artist.where(~Sequel.like(:name, '%J%'))
  # SELECT * FROM artists WHERE (name NOT LIKE '%J%' ESCAPE '\')

You can use !~ on Sequel expressions to create negated expressions:

  Artist.where{id !~ 5}
  # SELECT * FROM artists WHERE (id != 5)

=== Removing

To remove all existing filters, use +unfiltered+:

  Artist.where(id: 1).unfiltered
  # SELECT * FROM artists

== Ordering

Sequel offers quite a few methods to manipulate the SQL ORDER BY clause.  The
most basic of these is +order+:

  Artist.order(:id)
  # SELECT * FROM artists ORDER BY id

You can specify multiple arguments to order by more than one column:

  Album.order(:artist_id, :id)
  # SELECT * FROM album ORDER BY artist_id, id

Note that unlike +where+, +order+ replaces an existing order, it does not
append to an existing order:

  Artist.order(:id).order(:name)
  # SELECT * FROM artists ORDER BY name
  
If you want to add a column to the end of the existing order:

  Artist.order(:id).order_append(:name)
  # SELECT * FROM artists ORDER BY id, name

If you want to add a column to the beginning of the existing order:

  Artist.order(:id).order_prepend(:name)
  # SELECT * FROM artists ORDER BY name, id

=== Reversing

Just like you can invert an existing filter, you can reverse an existing
order, using +reverse+ without an order:

  Artist.order(:id).reverse
  # SELECT FROM artists ORDER BY id DESC

Alternatively, you can provide reverse with the order:

  Artist.reverse(:id)
  # SELECT FROM artists ORDER BY id DESC

To specify a single entry be reversed, <tt>Sequel.desc</tt> can be used:

  Artist.order(Sequel.desc(:id))
  # SELECT FROM artists ORDER BY id DESC

This allows you to easily use both ascending and descending orders:

  Artist.order(:name, Sequel.desc(:id))
  # SELECT FROM artists ORDER BY name, id DESC

=== Removing

Just like you can remove filters with +unfiltered+, you can remove
orders with +unordered+:

  Artist.order(:name).unordered
  # SELECT * FROM artists

== Selected Columns

Sequel offers a few methods to manipulate the columns selected.  As
you may be able to guess, the main method used is +select+:

  Artist.select(:id, :name)
  # SELECT id, name FROM artists

You just specify all of the columns that you are selecting as
arguments to the method.

If you are dealing with model objects, you'll want to include the
primary key if you want to update or destroy the object.  You'll
also want to include any keys (primary or foreign) related to
associations you plan to use.

If a column is not selected, and you attempt to access it, you will
get nil:

  artist = Artist.select(:name).first
  # SELECT name FROM artists LIMIT 1

  artist[:id]
  # => nil

Like +order+, +select+ replaces the existing selected columns:

  Artist.select(:id).select(:name)
  # SELECT name FROM artists

To add to the existing selected columns, use +select_append+:

  Artist.select(:id).select_append(:name)
  # SELECT id, name FROM artists

To remove specifically selected columns, and default back to all
columns, use +select_all+:

  Artist.select(:id).select_all
  # SELECT * FROM artists

To select all columns from a given table, provide an argument to
+select_all+:

  Artist.select_all(:artists)
  # SELECT artists.* FROM artists

=== Distinct

To treat duplicate rows as a single row when retrieving the records,
use +distinct+:

  Artist.distinct.select(:name)
  # SELECT DISTINCT name FROM artists

Note that DISTINCT is a separate SQL clause, it's not a function
that you pass to select.

== Limit and Offset

You can limit the dataset to a given number of rows using +limit+:

  Artist.limit(5)
  # SELECT * FROM artists LIMIT 5

You can provide a second argument to +limit+ to specify an offset:

  Artist.limit(5, 10)
  # SELECT * FROM artists LIMIT 5 OFFSET 10

You can also call the +offset+ method separately:

  Artist.limit(5).offset(10)
  # SELECT * FROM artists LIMIT 5 OFFSET 10

Either of these would return the 11th through 15th records in the original
dataset.

To remove a limit and offset from a dataset, use +unlimited+:

  Artist.limit(5, 10).unlimited
  # SELECT * FROM artists

== Grouping

The SQL GROUP BY clause is used to combine multiple rows based on
the values of a given group of columns.

To modify the GROUP BY clause of the SQL statement, you use +group+:

  Album.group(:artist_id)
  # SELECT * FROM albums GROUP BY artist_id

You can remove an existing grouping using +ungrouped+:

  Album.group(:artist_id).ungrouped
  # SELECT * FROM albums

If you want to add a column to the end of the existing grouping columns:

  Album.group(:artist_id).group_append(:name)
  # SELECT * FROM albums GROUP BY artist_id, name
 
A common use of grouping is to count based on the number of grouped rows,
and Sequel provides a +group_and_count+ method to make this easier:

  Album.group_and_count(:artist_id)
  # SELECT artist_id, count(*) AS count FROM albums GROUP BY artist_id

This will return the number of albums for each artist_id.

If you want to select and group on the same columns, you can use +select_group+:

  Album.select_group(:artist_id)
  # SELECT artist_id FROM albums GROUP BY artist_id

Usually you would add a +select_append+ call after that, to add some sort of
aggregation:

  Album.select_group(:artist_id).select_append{sum(num_tracks).as(tracks)}
  # SELECT artist_id, sum(num_tracks) AS tracks FROM albums GROUP BY artist_id

== Having

The SQL HAVING clause is similar to the WHERE clause, except that
filters the results after the grouping has been applied, instead of
before.  One possible use is if you only wanted to return artists
who had at least 10 albums:

  Album.group_and_count(:artist_id).having{count.function.* >= 10}
  # SELECT artist_id, count(*) AS count FROM albums
  # GROUP BY artist_id HAVING (count(*) >= 10)

Both the WHERE clause and the HAVING clause are removed by +unfiltered+:

  Album.group_and_count(:artist_id).having{count.function.* >= 10}.
   where(:name.like('A%')).unfiltered
  # SELECT artist_id, count(*) AS count FROM albums GROUP BY artist_id

== Joins

Sequel has support for many different SQL join types.
The underlying method used is +join_table+:

  Album.join_table(:inner, :artists, id: :artist_id)
  # SELECT * FROM albums
  # INNER JOIN artists ON (artists.id = albums.artist_id)

In most cases, you won't call +join_table+ directly, as Sequel provides
shortcuts for all common (and most uncommon) join types. For example
+join+ does an inner join:

  Album.join(:artists, id: :artist_id)
  # SELECT * FROM albums
  # INNER JOIN artists ON (artists.id = albums.artist_id)

And +left_join+ does a LEFT JOIN:

  Album.left_join(:artists, id: :artist_id)
  # SELECT * FROM albums
  # LEFT JOIN artists ON (artists.id = albums.artist_id)

=== Table/Dataset to Join

For all of these specialized join methods, the first argument is
generally the name of the table to which you are joining.  However, you
can also provide a dataset, in which case a subselect is used:

  Album.join(Artist.where{name < 'A'}, id: :artist_id)
  # SELECT * FROM albums
  # INNER JOIN (SELECT * FROM artists WHERE (name < 'A')) AS t1
  #  ON (t1.id = albums.artist_id)

=== Join Conditions

The second argument to the specialized join methods is the conditions 
to use when joining, which is similar to a filter expression, with
a few minor exceptions.

==== Implicit Qualification

A hash used as the join conditions operates similarly to a filter,
except that unqualified symbol keys are automatically qualified
with the table from the first argument, and unqualified symbol values
are automatically qualified with the last table joined (or the first
table in the dataset if there hasn't been a previous join):

  Album.join(:artists, id: :artist_id)
  # SELECT * FROM albums
  # INNER JOIN artists ON (artists.id = albums.artist_id)

Note how the +id+ symbol is automatically qualified with +artists+,
while the +artist_id+ symbol is automatically qualified with +albums+.

Because Sequel uses the last joined table for implicit qualifications
of values, you can do things like:

  Album.join(:artists, id: :artist_id).
   join(:members, artist_id: :id)
  # SELECT * FROM albums
  # INNER JOIN artists ON (artists.id = albums.artist_id)
  # INNER JOIN members ON (members.artist_id = artists.id)

Note that when joining to the +members+ table, +artist_id+ is qualified
with +members+ and +id+ is qualified with +artists+.

While a good default, implicit qualification is not always correct:

  Album.join(:artists, id: :artist_id).
   join(:tracks, album_id: :id)
  # SELECT * FROM albums
  # INNER JOIN artists ON (artists.id = albums.artist_id)
  # INNER JOIN tracks ON (tracks.album_id = artists.id)

Note here how +id+ is qualified with +artists+ instead of +albums+. This
is wrong as the foreign key <tt>tracks.album_id</tt> refers to <tt>albums.id</tt>, not
<tt>artists.id</tt>.  To fix this, you need to explicitly qualify when joining:

  Album.join(:artists, id: :artist_id).
   join(:tracks, album_id: Sequel[:albums][:id])
  # SELECT * FROM albums
  # INNER JOIN artists ON (artists.id = albums.artist_id)
  # INNER JOIN tracks ON (tracks.album_id = albums.id)

Just like in filters, an array of two element arrays is treated the same
as a hash, but allows for duplicate keys:

  Album.join(:artists, [[:id, :artist_id], [:id, 1..5]])
  # SELECT * FROM albums INNER JOIN artists
  #  ON ((artists.id = albums.artist_id)
  #   AND (artists.id >= 1) AND (artists.id <= 5))

And just like in the hash case, unqualified symbol elements in the
array are implicitly qualified.

By default, Sequel only qualifies unqualified symbols in the conditions.  However,
You can provide an options hash with a <tt>qualify: :deep</tt> option to do a deep
qualification, which can qualify subexpressions.  For example, let's say you are doing
a JOIN using case insensitive string comparison:

  Album.join(:artists, {Sequel.function(:lower, :name) =>
                        Sequel.function(:lower, :artist_name)},
             qualify: :deep)
  # SELECT * FROM albums INNER JOIN artists
  #  ON (lower(artists.name) = lower(albums.artist_name))

Note how the arguments to lower were qualified correctly in both cases.

==== USING Joins

The most common type of join conditions is a JOIN ON, as displayed
above.  However, the SQL standard allows for join conditions to be
specified with JOIN USING, assuming the column name is the same in
both tables.

For example, if instead of having a primary
column named +id+ in all of your tables, you use +artist_id+ in your
+artists+ table and +album_id+ in your +albums+ table, you could do:

  Album.join(:artists, [:artist_id])
  # SELECT * FROM albums INNER JOIN artists USING (artist_id)

See here how you specify the USING columns as an array of symbols.

==== NATURAL Joins

NATURAL joins take it one step further than USING joins, by assuming
that all columns with the same names in both tables should be
used for joining:

  Album.natural_join(:artists)
  # SELECT * FROM albums NATURAL JOIN artists

In this case, you don't even need to specify any conditions.

==== Join Blocks

You can provide a block to any of the join methods that accept
conditions.  This block should accept 3 arguments: the table alias
for the table currently being joined, the table alias for the last
table joined (or first table), and an array of previous
<tt>Sequel::SQL::JoinClause</tt>s.

This allows you to qualify columns similar to how the implicit
qualification works, without worrying about the specific aliases
being used.  For example, let's say you wanted to join the albums
and artists tables, but only want albums where the artist's name
comes before the album's name.

  Album.join(:artists, id: :artist_id) do |j, lj, js|
    Sequel[j][:name] < Sequel[lj][:name]
  end
  # SELECT * FROM albums INNER JOIN artists
  # ON ((artists.id = albums.artist_id)
  # AND (artists.name < albums.name))

Because greater than can't be expressed with a hash in Sequel, you
need to use a block and qualify the tables manually.

== From

In general, the FROM table is the first clause populated when creating
a dataset.  For a standard Sequel::Model, the dataset already has the
FROM clause populated, and the most common way to create datasets is
with the <tt>Database#[]</tt> method, which populates the FROM clause.

However, you can modify the tables you are selecting FROM using +from+:

  Album.from(:albums, :old_albums)
  # SELECT * FROM albums, old_albums

Be careful with this, as multiple tables in the FROM clause use a cross
join by default, so the number of rows will be number of albums times the
number of old albums.

Using multiple FROM tables and setting conditions in the WHERE clause is
an old-school way of joining tables:

  DB.from(:albums, :artists).where{{artists[:id]=>albums[:artist_id]}}
  # SELECT * FROM albums, artists WHERE (artists.id = albums.artist_id)

=== Using the current dataset in a subselect

In some cases, you may want to wrap the current dataset in a subselect.
Here's an example using +from_self+:

  Album.order(:artist_id).limit(100).from_self.group(:artist_id)
  # SELECT * FROM (SELECT * FROM albums ORDER BY artist_id LIMIT 100)
  #  AS t1 GROUP BY artist_id

This is different than without +from_self+:

  Album.order(:artist_id).limit(100).group(:artist_id)
  # SELECT * FROM albums GROUP BY artist_id ORDER BY name LIMIT 100

Without +from_self+, you are doing the grouping, and limiting the number
of grouped records returned to 100.  So assuming you have albums by more
than 100 artists, you'll end up with 100 results.

With +from_self+, you are limiting the number of records before grouping.
So if the artist with the lowest id had 100 albums, you'd get 1 result,
not 100.

== Locking for Update

Sequel allows you to easily add a FOR UPDATE clause to your queries so
that the records returned can't be modified by another query until the
current transaction commits.  You just use the +for_update+ dataset
method when returning the rows:

  DB.transaction do
    album = Album.for_update.first(id: 1)
    # SELECT * FROM albums WHERE (id = 1) FOR UPDATE
    album.num_tracks += 1
    album.save
  end

This will ensure that no other connection modifies the row between when you select
it and when the transaction ends.

=== Optimistic Locking

One of the model plugins that ships with Sequel is an optimistic locking plugin, which provides
a database independent way to detect and raise an error if two different connections
modify the same row.  It's useful for things like web forms where you cannot keep a
transaction open while the user is looking at the form, because of the web's
stateless nature.

== Custom SQL

Sequel makes it easy to use custom SQL for the query by providing it to the <tt>Database#[]</tt>
method as a string:

  DB["SELECT * FROM artists"]
  # SELECT * FROM artists

You can also use the +with_sql+ dataset method to return a dataset that uses that
exact SQL:

  DB[:albums].with_sql("SELECT * FROM artists")
  # SELECT * FROM artists

With either of these methods, you can use placeholders:

  DB["SELECT * FROM artists WHERE id = ?", 5]
  # SELECT * FROM artists WHERE id = 5

  DB[:albums].with_sql("SELECT * FROM artists WHERE id = :id", id: 5)
  # SELECT * FROM artists WHERE id = 5

Note that if you specify the dataset using custom SQL, you can still call the dataset
modification methods, but in many cases they will appear to have no affect:

  DB["SELECT * FROM artists"].select(:name).order(:id)
  # SELECT * FROM artists

You can use the implicit_subquery extension to automatically wrap queries that use
custom SQL in subqueries if a method is called that would modify the SQL:

  DB.extension :implicit_subquery
  DB["SELECT * FROM artists"].select(:name).order(:id)
  # SELECT name FROM (SELECT * FROM artists) AS t1 ORDER BY id"

If you must drop down to using custom SQL, it's recommended that you only do so for
specific parts of a query.  For example, if the reason you are using custom SQL is
to use a custom operator in the database in the SELECT clause:

  DB["SELECT name, (foo !@# ?) AS baz FROM artists", 'bar']

it's better to use Sequel's DSL, and use a literal string for the custom operator:

  DB[:artists].select(:name, Sequel.lit("(foo !@# ?)", 'bar').as(:baz))

That way Sequel's method chaining still works, and it increases Sequel's ability to
introspect the code.

== Checking for Records

If you just want to know whether the current dataset would return any rows, use <tt>empty?</tt>:

  Album.empty?
  # SELECT 1 FROM albums LIMIT 1
  # => false

  Album.where(id: 0).empty?
  # SELECT 1 FROM albums WHERE (id = 0) LIMIT 1
  # => true

  Album.where(Sequel.like(:name, 'R%')).empty?
  # SELECT 1 FROM albums WHERE (name LIKE 'R%' ESCAPE '\') LIMIT 1
  # => false

== Aggregate Calculations

The SQL standard defines a few helpful methods to get aggreate information about
datasets, such as +count+, +sum+, +avg+, +min+, and +max+.  There are dataset methods
for each of these aggregate functions.

+count+ just returns the number of records in the dataset.

  Album.count
  # SELECT count(*) AS count FROM albums LIMIT 1
  # => 2
 
If you pass an expression to count, it will return the number of records where
that expression in not NULL:

  Album.count(:artist_id)
  # SELECT count(artist_id) AS count FROM albums LIMIT 1
  # => 1

The other methods take a column argument and call the aggregate function with
the argument:

  Album.sum(:id)
  # SELECT sum(id) AS sum FROM albums LIMIT 1
  # => 3

  Album.avg(:id)
  # SELECT avg(id) AS avg FROM albums LIMIT 1
  # => 1.5

  Album.min(:id)
  # SELECT min(id) AS min FROM albums LIMIT 1
  # => 1

  Album.max(:id)
  # SELECT max(id) AS max FROM albums LIMIT 1
  # => 2