=pod

=head1 NAME

SQL::Statement::Structure - parse and examine structure of SQL queries

=head1 SYNOPSIS

    use SQL::Statement;
    my $sql    = "SELECT a FROM b JOIN c WHERE c=? AND e=7 ORDER BY f DESC LIMIT 5,2";
    my $parser = SQL::Parser->new();
    $parser->{RaiseError}=1;
    $parser->{PrintError}=0;
    $parser->parse("LOAD 'MyLib::MySyntax' ");
    my $stmt = SQL::Statement->new($sql,$parser);
    printf "Command             %s\n",$stmt->command;
    printf "Num of Placeholders %s\n",scalar $stmt->params;
    printf "Columns             %s\n",join( ',', map {$_->name} $stmt->column_defs() );
    printf "Tables              %s\n",join( ',', map {$_->name} $stmt->tables() );
    printf "Where operator      %s\n",join( ',', $stmt->where->op() );
    printf "Limit               %s\n",$stmt->limit();
    printf "Offset              %s\n",$stmt->offset();

    # these will work not before $stmt->execute()
    printf "Order Columns       %s\n",join(',', map {$_->column} $stmt->order() );

=head1 DESCRIPTION

The L<SQL::Statement> module can be used by itself, without DBI and without
a subclass to parse SQL statements and to allow you to examine the structure
of the statement (table names, column names, where clause predicates, etc.).
It will also execute statements using in-memory tables.  That means that
you can create and populate some tables, then query them and fetch the
results of the queries as well as examine the differences between statement
metadata during different phases of prepare, execute, fetch. See the
remainder of this document for a description of how to create and modify
a parser object and how to use it to parse and examine SQL statements.
See L<SQL::Statement> for other uses of the module.

=head1 B<Creating a parser object>

The parser object only needs to be created once per script. It can
then be reused to parse any number of SQL statements. The basic
creation of a parser is this:

    my $parser = SQL::Parser->new();

You can set the error-reporting for the parser the same way you do in DBI:

    $parser->{RaiseError}=1;   # turn on die-on-error behaviour
    $parser->{PrinteError}=1;  # turn on warnings-on-error behaviour

As with DBI, RaiseError defaults to 0 (off) and PrintError defaults to 1 (on).

For many purposes, the built-in SQL syntax should be sufficient. However, if
you need to, you can change the behaviour of the parser by extending the
supported SQL syntax either by loading a file containing definitions; or by
issuing SQL commands that modify the way the parser treats types, keywords,
functions, and operators.

    $parser->parse("LOAD MyLib::MySyntax");
    $parser->parse("CREATE TYPE myDataType");

See L<SQL::Statement::Syntax> for details of the supported SQL syntax and
for methods of extending the syntax.

=head1 B<Parsing SQL statements>

While you only need to define a new SQL::Parser object once per script, you
need to define a new SQL::Statment object once for each statement you want
to parse.

    my $stmt = SQL::Statement->new($sql, $parser);

The call to new() takes two arguments - the SQL string you want to parse,
and the SQL::Parser object you previously created.  The call to new is the
equivalent of a DBI call to prepare() - it parses the SQL into a structure
but does not attempt to execute the SQL unless you explicitly call execute().

=head1 Examining the structure of SQL statements

The following methods can be used to obtain information about a query:

=head2 B<command>

Returns the SQL command. See L<SQL::Statement::Syntax> for supported
command. Example:

    my $command = $stmt->command();

=head2 B<column definitions>

    my $numColumns = $stmt->column_defs();  # Scalar context
    my @columnList = $stmt->column_defs();  # Array context
    my($col1, $col2) = ($stmt->column_defs(0), $stmt->column_defs(1));

This method is used to retrieve column lists. The meaning depends on
the query command:

    SELECT $col1, $col2, ... $colN FROM $table WHERE ...
    UPDATE $table SET $col1 = $val1, $col2 = $val2, ...
        $colN = $valN WHERE ...
    INSERT INTO $table ($col1, $col2, ..., $colN) VALUES (...)

When used without arguments, the method returns a list of the columns
C<$col1>, C<$col2>, ..., C<$colN>, you may alternatively use a column number
as argument. Note that the column list may be empty as in

    INSERT INTO $table VALUES (...)

and in I<CREATE> or I<DROP> statements.

But what does "returning a column" mean? It is returning an
C<SQL::Statement::Util::Column> instance, a class that implements the methods
C<table> and C<name>, both returning the respective scalar. For example,
consider the following statements:

    INSERT INTO foo (bar) VALUES (1)
    SELECT bar FROM foo WHERE ...
    SELECT foo.bar FROM foo WHERE ...

In all these cases exactly one column instance would be returned with

    $col->name() eq 'bar'
    $col->table() eq 'foo'

=head2 B<tables>

    my $tableNum = $stmt->tables();  # Scalar context
    my @tables = $stmt->tables();    # Array context
    my($table1, $table2) = ($stmt->tables(0), $stmt->tables(1));

Similar to C<columns>, this method returns instances of
C<SQL::Statement::Table>. For I<UPDATE>, I<DELETE>, I<INSERT>,
I<CREATE> and I<DROP>, a single table will always be returned.
I<SELECT> statements can return more than one table, in case
of joins. Table objects offer a single method, C<name> which
returns the table name.

=head2 B<params>

    my $paramNum = $stmt->params();  # Scalar context
    my @params = $stmt->params();    # Array context
    my($p1, $p2) = ($stmt->params(0), $stmt->params(1));

The C<params> method returns information about the input parameters
used in a statement. For example, consider the following:

    INSERT INTO foo VALUES (?, ?)

This would return two instances of C<SQL::Statement::Param>. Param objects
implement a single method, C<$param->num()>, which retrieves the parameter
number. (0 and 1, in the above example). As of now, not very useful ... :-)

=head2 B<row_values>

    my $rowValueNum = $stmt->row_values(); # Scalar context
    my @rowValues = $stmt->row_values(0);  # Array context
    my($rval1, $rval2) = ($stmt->row_values(0,0),
                          $stmt->row_values(0,1));

This method is used for statements like

    UPDATE $table SET $col1 = $val1, $col2 = $val2, ...
        $colN = $valN WHERE ...
    INSERT INTO $table (...) VALUES ($val1, $val2, ..., $valN),
                                    ($val1, $val2, ..., $valN)

to read the values C<$val1>, C<$val2>, ... C<$valN>. It returns (lists of)
scalar values or C<SQL::Statement::Param> instances.

=head2 B<order>

    my $orderNum = $stmt->order();   # Scalar context
    my @order = $stmt->order();      # Array context
    my($o1, $o2) = ($stmt->order(0), $stmt->order(1));

In I<SELECT> statements you can use this for looking at the ORDER clause.
Example:

    SELECT * FROM FOO ORDER BY id DESC, name

In this case, C<order> could return 2 instances of C<SQL::Statement::Order>.
You can use the methods C<$o-E<gt>table()>, C<$o-E<gt>column()>,
C<$o-E<gt>direction()> and C<$o-E<gt>desc()> to examine the order object.

=head2 B<limit>

    my $limit = $stmt->limit();

In a SELECT statement you can use a C<LIMIT> clause to implement
cursoring:

    SELECT * FROM FOO LIMIT 5
    SELECT * FROM FOO LIMIT 5, 5
    SELECT * FROM FOO LIMIT 10, 5

These three statements would retrieve the rows C<0..4>, C<5..9>, C<10..14>
of the table FOO, respectively. If no C<LIMIT> clause is used, then the
method C<$stmt-E<gt>limit> returns undef. Otherwise it returns the limit
number (the maximum number of rows) from the statement (C<5> or C<10> for
the statements above).

=head2 B<offset>

    my $offset = $stmt->offset();

If no C<LIMIT> clause is used, then the method C<$stmt-E<gt>limit> returns
I<undef>. Otherwise it returns the offset number (the index of the first row
to be included in the limit clause).

=head2 B<where_hash>

    my $where_hash = $stmt->where_hash();

To manually evaluate the I<WHERE> clause, fetch the topmost where clause node
with the C<where_hash> method. Then evaluate the left-hand and right-hand side
of the operation, perhaps recursively. Once that is done, apply the operator
and finally negate the result, if required.

The where clause nodes have (up to) 4 attributes:

=over 12
 
=item op
 
contains the operator, one of C<AND>, C<OR>, C<=>, C<E<lt>E<gt>>, C<E<gt>=>,
C<E<gt>>, C<E<lt>=>, C<E<lt>>, C<LIKE>, C<CLIKE>, C<IS>, C<IN>, C<BETWEEN> or
a user defined operator, if any.
 
=item arg1

contains the left-hand side of the operator. This can be a scalar value, a
hash containing column or function definition, a parameter definition (hash has
attribute C<type> defined) or another operation (hash has attribute C<op>
defined).

=item arg2

contains the right-hand side of the operator. This can be a scalar value, a
hash containing column or function definition, a parameter definition (hash has
attribute C<type> defined) or another operation (hash has attribute C<op>
defined).

=item neg

contains a TRUE value, if the operation result must be negated after evaluation.

=back

To illustrate the above, consider the following WHERE clause:

    WHERE NOT (id > 2 AND name = 'joe') OR name IS NULL

We can represent this clause by the following tree:

              (id > 2)   (name = 'joe')
                     \   /
          NOT         AND
                         \      (name IS NULL)
                          \    /
                            OR

Thus the WHERE clause would return an SQL::Statement::Op instance with
the op() field set to 'OR'. The arg2() field would return another
SQL::Statement::Op instance with arg1() being the SQL::Statement::Column
instance representing id, the arg2() field containing the value undef
(NULL) and the op() field being 'IS'.

The arg1() field of the topmost Op instance would return an Op instance
with op() eq 'AND' and neg() returning TRUE. The arg1() and arg2()
fields would be Op's representing "id > 2" and "name = 'joe'".

Of course there's a ready-for-use method for WHERE clause evaluation:

The WHERE clause evaluation depends on an object being used for
fetching parameter and column values. Usually this can be an
SQL::Statement::RAM::Table object or SQL::Eval object, but in fact it
can be any object that supplies the methods

    $val = $eval->param($paramNum);
    $val = $eval->column($table, $column);

Once you have such an object, you can call eval_where;

    $match = $stmt->eval_where($eval);

=head2 B<where>

    my $where = $stmt->where();

This method is used to examine the syntax tree of the C<WHERE> clause. It
returns I<undef> (if no C<WHERE> clause was used) or an instance of
L<SQL::Statement::Term>.

The where clause is evaluated automatically on the current selected row of
the table currently worked on when it's C<value()> method is invoked.

C<SQL::Statement> creates the object tree for where clause evaluation
directly after successfully parsing a statement from the given
C<where_clause>, if any.

=head1 Executing and fetching data from SQL statements

=head2 execute

When called from a DBD or other subclass of SQL::Statement, the execute()
method will be executed against whatever datasource (persistent storage) is
supplied by the DBD or the subclass (e.g. CSV files for L<DBD::CSV>, or
BerkeleyDB for L<DBD::DBM>). If you are using L<SQL::Statement> directly
rather than as a subclass, you can call the execute() method and the
statements will be executed() using temporary in-memory tables. When used
directly, like that, you need to create a cache hashref and pass it as the
first argument to execute:

  my $cache  = {};
  my $parser = SQL::Parser->new();
  my $stmt   = SQL::Statement->new('CREATE TABLE x (id INT)',$parser);
  $stmt->execute( $cache );

If you are using a statement with placeholders, those can be passed to
execute after the C<$cache>:

  $stmt      = SQL::Statement->new('INSERT INTO y VALUES(?,?)',$parser);
  $stmt->execute( $cache, 7, 'foo' );

=head2 fetch

Only a single C<fetch()> method is provided - it returns a single row of
data as an arrayref. Use a loop to fetch all rows:

 while (my $row = $stmt->fetch()) {
     # ...
 }

=head2 an example of executing and fetching

 #!/usr/bin/perl -w
 use strict;
 use SQL::Statement;

 my $cache={};
 my $parser = SQL::Parser->new();
 for my $sql(split /\n/,
 "  CREATE TABLE a (b INT)
    INSERT INTO a VALUES(1)
    INSERT INTO a VALUES(2)
    SELECT MAX(b) FROM a  "
 )
 {
    $stmt = SQL::Statement->new($sql,$parser);
    $stmt->execute($cache);
    next unless $stmt->command eq 'SELECT';
    while (my $row=$stmt->fetch)
    {
        print "@$row\n";
    }
 }
 __END__

=head1 AUTHOR & COPYRIGHT

Copyright (c) 2005, Jeff Zucker <jzuckerATcpan.org>, all rights reserved.
Copyright (c) 2009, Jens Rehsack <rehsackATcpan.org>, all rights reserved.

This document may be freely modified and distributed under the same terms
as Perl itself.

=cut