use strict;
use warnings;
+no warnings 'qw';
use Carp;
use Hash::Merge qw//;
},
}, 'SQLA::Tree Behavior' );
-
-# Parser states for _recurse_parse()
-use constant PARSE_TOP_LEVEL => 0;
-use constant PARSE_IN_EXPR => 1;
-use constant PARSE_IN_PARENS => 2;
-use constant PARSE_RHS => 3;
-use constant PARSE_IN_FUNC => 4;
-
my $op_look_ahead = '(?: (?= [\s\)\(\;] ) | \z)';
-my $op_look_behind = '(?: (?<= [\s\)\(] ) | \A )';
+my $op_look_behind = '(?: (?<= [\,\s\)\(] ) | \A )';
+
my $quote_left = qr/[\`\'\"\[]/;
my $quote_right = qr/[\`\'\"\]]/;
+my $placeholder_re = qr/(?: \? | \$\d+ )/x;
+
# These SQL keywords always signal end of the current expression (except inside
# of a parenthesized subexpression).
# Format: A list of strings that will be compiled to extended syntax ie.
)',
'ON',
'WHERE',
- 'VALUES',
+ '(?: DEFAULT \s+ )? VALUES',
'EXISTS',
'GROUP \s+ BY',
'HAVING',
'ORDER \s+ BY',
+ 'SKIP',
+ 'FIRST',
'LIMIT',
'OFFSET',
'FOR',
'UNION',
'INTERSECT',
'EXCEPT',
+ 'BEGIN \s+ WORK',
+ 'COMMIT',
+ 'ROLLBACK \s+ TO \s+ SAVEPOINT',
+ 'ROLLBACK',
+ 'SAVEPOINT',
+ 'RELEASE \s+ SAVEPOINT',
'RETURNING',
'ROW_NUMBER \s* \( \s* \) \s+ OVER',
);
-my $exp_start_re = join ("\n\t|\n", @expression_start_keywords );
-$exp_start_re = qr/ $op_look_behind (?i: $exp_start_re ) $op_look_ahead /xo;
+my $expr_start_re = join ("\n\t|\n", @expression_start_keywords );
+$expr_start_re = qr/ $op_look_behind (?i: $expr_start_re ) $op_look_ahead /x;
# These are binary operator keywords always a single LHS and RHS
# * AND/OR are handled separately as they are N-ary
{ "(?: (?<= [\\w\\s] | $quote_right ) | \\A )" . quotemeta ($_) . "(?: (?= [\\w\\s] | $quote_left ) | \\z )" }
@math_op_keywords
);
-$math_re = qr/$math_re/xo;
+$math_re = qr/$math_re/x;
sub _math_op_re { $math_re }
my $binary_op_re = '(?: NOT \s+)? (?:' . join ('|', qw/IN BETWEEN R?LIKE/) . ')';
-$binary_op_re = "(?: $op_look_behind (?i: $binary_op_re ) $op_look_ahead ) \n\t|\n $math_re";
-$binary_op_re = qr/$binary_op_re/xo;
+$binary_op_re = join "\n\t|\n",
+ "$op_look_behind (?i: $binary_op_re ) $op_look_ahead",
+ $math_re,
+ $op_look_behind . 'IS (?:\s+ NOT)?' . "(?= \\s+ NULL \\b | $op_look_ahead )",
+;
+$binary_op_re = qr/$binary_op_re/x;
sub _binary_op_re { $binary_op_re }
-
-my $tokenizer_re = join("\n\t|\n",
- $exp_start_re,
+my $all_known_re = join("\n\t|\n",
+ $expr_start_re,
$binary_op_re,
- "$op_look_behind (?i: AND|OR|NOT ) $op_look_ahead",
- (map { quotemeta $_ } qw/( ) ? */),
+ "$op_look_behind (?i: AND|OR|NOT|\\* ) $op_look_ahead",
+ (map { quotemeta $_ } qw/, ( )/),
+ $placeholder_re,
);
-#this one *is* capturing
-$tokenizer_re = qr/ \s* ( $tokenizer_re ) \s* /x;
+$all_known_re = qr/$all_known_re/x;
+
+#this one *is* capturing for the split below
+# splits on whitespace if all else fails
+my $tokenizer_re = qr/ \s* ( $all_known_re ) \s* | \s+ /x;
+
+# Parser states for _recurse_parse()
+use constant PARSE_TOP_LEVEL => 0;
+use constant PARSE_IN_EXPR => 1;
+use constant PARSE_IN_PARENS => 2;
+use constant PARSE_IN_FUNC => 3;
+use constant PARSE_RHS => 4;
+
+my $expr_term_re = qr/ ^ (?: $expr_start_re | \) ) $/x;
+my $rhs_term_re = qr/ ^ (?: $expr_term_re | $binary_op_re | (?i: AND | OR | NOT | \, ) ) $/x;
+my $func_start_re = qr/^ (?: \* | $placeholder_re | \( ) $/x;
my %indents = (
select => 0,
join => 1,
'left join' => 1,
on => 2,
+ having => 0,
'group by' => 0,
'order by' => 0,
set => 1,
into => 1,
values => 1,
+ limit => 1,
+ offset => 1,
+ skip => 1,
+ first => 1,
);
my %profiles = (
indent_amount => 2,
newline => "\n",
colormap => {},
- indentmap => { %indents },
+ indentmap => \%indents,
eval { require Term::ANSIColor }
? do {
my $c = \&Term::ANSIColor::color;
+
+ my $red = [$c->('red') , $c->('reset')];
+ my $cyan = [$c->('cyan') , $c->('reset')];
+ my $green = [$c->('green') , $c->('reset')];
+ my $yellow = [$c->('yellow') , $c->('reset')];
+ my $blue = [$c->('blue') , $c->('reset')];
+ my $magenta = [$c->('magenta'), $c->('reset')];
+ my $b_o_w = [$c->('black on_white'), $c->('reset')];
(
- placeholder_surround => [$c->('black on_cyan'), $c->('reset')],
+ placeholder_surround => [$c->('black on_magenta'), $c->('reset')],
colormap => {
- select => [$c->('red'), $c->('reset')],
- 'insert into' => [$c->('red'), $c->('reset')],
- update => [$c->('red'), $c->('reset')],
- 'delete from' => [$c->('red'), $c->('reset')],
-
- set => [$c->('cyan'), $c->('reset')],
- from => [$c->('cyan'), $c->('reset')],
-
- where => [$c->('green'), $c->('reset')],
- values => [$c->('yellow'), $c->('reset')],
-
- join => [$c->('magenta'), $c->('reset')],
- 'left join' => [$c->('magenta'), $c->('reset')],
- on => [$c->('blue'), $c->('reset')],
-
- 'group by' => [$c->('yellow'), $c->('reset')],
- 'order by' => [$c->('yellow'), $c->('reset')],
+ 'begin work' => $b_o_w,
+ commit => $b_o_w,
+ rollback => $b_o_w,
+ savepoint => $b_o_w,
+ 'rollback to savepoint' => $b_o_w,
+ 'release savepoint' => $b_o_w,
+
+ select => $red,
+ 'insert into' => $red,
+ update => $red,
+ 'delete from' => $red,
+
+ set => $cyan,
+ from => $cyan,
+
+ where => $green,
+ values => $yellow,
+
+ join => $magenta,
+ 'left join' => $magenta,
+ on => $blue,
+
+ 'group by' => $yellow,
+ having => $yellow,
+ 'order by' => $yellow,
+
+ skip => $green,
+ first => $green,
+ limit => $green,
+ offset => $green,
}
);
} : (),
indent_amount => 2,
newline => "\n",
colormap => {},
- indentmap => { %indents },
+ indentmap => \%indents,
},
html => {
fill_in_placeholders => 1,
'insert into' => ['<span class="insert-into">' , '</span>'],
update => ['<span class="select">' , '</span>'],
'delete from' => ['<span class="delete-from">' , '</span>'],
- where => ['<span class="where">' , '</span>'],
+
+ set => ['<span class="set">', '</span>'],
from => ['<span class="from">' , '</span>'],
+
+ where => ['<span class="where">' , '</span>'],
+ values => ['<span class="values">', '</span>'],
+
join => ['<span class="join">' , '</span>'],
+ 'left join' => ['<span class="left-join">','</span>'],
on => ['<span class="on">' , '</span>'],
+
'group by' => ['<span class="group-by">', '</span>'],
+ having => ['<span class="having">', '</span>'],
'order by' => ['<span class="order-by">', '</span>'],
- set => ['<span class="set">', '</span>'],
- into => ['<span class="into">', '</span>'],
- values => ['<span class="values">', '</span>'],
+
+ skip => ['<span class="skip">', '</span>'],
+ first => ['<span class="first">', '</span>'],
+ limit => ['<span class="limit">', '</span>'],
+ offset => ['<span class="offset">', '</span>'],
+
+ 'begin work' => ['<span class="begin-work">', '</span>'],
+ commit => ['<span class="commit">', '</span>'],
+ rollback => ['<span class="rollback">', '</span>'],
+ savepoint => ['<span class="savepoint">', '</span>'],
+ 'rollback to savepoint' => ['<span class="rollback-to-savepoint">', '</span>'],
+ 'release savepoint' => ['<span class="release-savepoint">', '</span>'],
},
- indentmap => { %indents },
+ indentmap => \%indents,
},
none => {
colormap => {},
my $args = shift || {};
my $profile = delete $args->{profile} || 'none';
+
+ die "No such profile '$profile'!" unless exists $profiles{$profile};
+
my $data = $merger->merge( $profiles{$profile}, $args );
bless $data, $class
# tokenize string, and remove all optional whitespace
my $tokens = [];
foreach my $token (split $tokenizer_re, $s) {
- push @$tokens, $token if (length $token) && ($token =~ /\S/);
+ push @$tokens, $token if (
+ defined $token
+ and
+ length $token
+ and
+ $token =~ /\S/
+ );
}
-
- my $tree = $self->_recurse_parse($tokens, PARSE_TOP_LEVEL);
- return $tree;
+ $self->_recurse_parse($tokens, PARSE_TOP_LEVEL);
}
sub _recurse_parse {
or
($state == PARSE_IN_PARENS && $lookahead eq ')')
or
- ($state == PARSE_IN_EXPR && $lookahead =~ qr/ ^ (?: $exp_start_re | \) ) $ /x )
+ ($state == PARSE_IN_EXPR && $lookahead =~ $expr_term_re )
or
- ($state == PARSE_RHS && $lookahead =~ qr/ ^ (?: $exp_start_re | $binary_op_re | (?i: AND | OR | NOT ) | \) ) $ /x )
+ ($state == PARSE_RHS && $lookahead =~ $rhs_term_re )
or
- ($state == PARSE_IN_FUNC && $lookahead ne '(')
+ ($state == PARSE_IN_FUNC && $lookahead !~ $func_start_re) # if there are multiple values - the parenthesis will switch the $state
) {
return $left||();
}
$left = $left ? [$left, [PAREN => [$right||()] ]]
: [PAREN => [$right||()] ];
}
- # AND/OR
- elsif ($token =~ /^ (?: OR | AND ) $/xi ) {
- my $op = uc $token;
+ # AND/OR and LIST (,)
+ elsif ($token =~ /^ (?: OR | AND | \, ) $/xi ) {
+ my $op = ($token eq ',') ? 'LIST' : uc $token;
+
my $right = $self->_recurse_parse($tokens, PARSE_IN_EXPR);
# Merge chunks if logic matches
if (ref $right and $op eq $right->[0]) {
- $left = [ (shift @$right ), [$left, map { @$_ } @$right] ];
+ $left = [ (shift @$right ), [$left||(), map { @$_ } @$right] ];
}
else {
- $left = [$op => [$left, $right]];
+ $left = [$op => [ $left||(), $right||() ]];
}
}
# binary operator keywords
$left = [$op => [$left, $right] ];
}
# expression terminator keywords (as they start a new expression)
- elsif ( $token =~ / ^ $exp_start_re $ /x ) {
+ elsif ( $token =~ / ^ $expr_start_re $ /x ) {
my $op = uc $token;
my $right = $self->_recurse_parse($tokens, PARSE_IN_EXPR);
- $left = $left ? [ $left, [$op => [$right] ]]
- : [ $op => [$right] ];
+ $left = $left ? [ $left, [$op => [$right||()] ]]
+ : [ $op => [$right||()] ];
}
# NOT
elsif ( $token =~ /^ NOT $/ix ) {
: [ $op => [$right] ];
}
- # generic function
- elsif (@$tokens && $tokens->[0] eq '(') {
- my $right = $self->_recurse_parse($tokens, PARSE_IN_FUNC);
-
- $left = $left ? [ $left, [ $token => [$right||()] ]]
- : [ $token => [$right||()] ];
+ elsif ( $token =~ $placeholder_re) {
+ $left = $left ? [ $left, [ PLACEHOLDER => [ $token ] ] ]
+ : [ PLACEHOLDER => [ $token ] ];
}
- # literal (eat everything on the right until RHS termination)
+ # we're now in "unknown token" land - start eating tokens until
+ # we see something familiar
else {
- my $right = $self->_recurse_parse ($tokens, PARSE_RHS);
- $left = $left ? [ $left, [LITERAL => [join ' ', $token, $self->unparse($right)||()] ] ]
- : [ LITERAL => [join ' ', $token, $self->unparse($right)||()] ];
+ my $right;
+
+ # check if the current token is an unknown op-start
+ if (@$tokens and $tokens->[0] =~ $func_start_re) {
+ $right = [ $token => [ $self->_recurse_parse($tokens, PARSE_IN_FUNC) || () ] ];
+ }
+ else {
+ $right = [ LITERAL => [ $token ] ];
+ }
+
+ $left = $left ? [ $left, $right ]
+ : $right;
}
}
}
$before = $self->newline . $self->indent($depth + $self->indentmap->{lc $keyword});
}
$before = '' if $depth == 0 and defined $starters{lc $keyword};
- return [$before, ' '];
+ return [$before, ''];
}
sub indent { ($_[0]->indent_string||'') x ( ( $_[0]->indent_amount || 0 ) * $_[1] ) }
my ($self, $bindargs) = @_;
if ($self->fill_in_placeholders) {
- my $val = pop @{$bindargs} || '';
+ my $val = shift @{$bindargs} || '';
+ my $quoted = $val =~ s/^(['"])(.*)\1$/$2/;
my ($left, $right) = @{$self->placeholder_surround};
$val =~ s/\\/\\\\/g;
$val =~ s/'/\\'/g;
- return qq('$left$val$right')
+ $val = qq('$val') if $quoted;
+ return qq($left$val$right)
}
return '?'
}
+# FIXME - terrible name for a user facing API
sub unparse {
- my ($self, $tree, $bindargs, $depth) = @_;
+ my ($self, $tree, $bindargs) = @_;
+ $self->_parenthesis_unroll($tree);
+ $self->_unparse($tree, [@{$bindargs||[]}], 0);
+}
- $depth ||= 0;
+sub _unparse {
+ my ($self, $tree, $bindargs, $depth) = @_;
if (not $tree or not @$tree) {
return '';
}
if (ref $car) {
- return join ('', map $self->unparse($_, $bindargs, $depth), @$tree);
+ return join (' ', map $self->_unparse($_, $bindargs, $depth), @$tree);
}
elsif ($car eq 'LITERAL') {
- if ($cdr->[0] eq '?') {
- return $self->fill_in_placeholder($bindargs)
- }
return $cdr->[0];
}
+ elsif ($car eq 'PLACEHOLDER') {
+ return $self->fill_in_placeholder($bindargs);
+ }
elsif ($car eq 'PAREN') {
- return '(' .
- join(' ',
- map $self->unparse($_, $bindargs, $depth + 2), @{$cdr}) .
- ($self->_is_key($cdr)?( $self->newline||'' ).$self->indent($depth + 1):'') . ') ';
+ return sprintf ('( %s )',
+ join (' ', map { $self->_unparse($_, $bindargs, $depth + 2) } @{$cdr} )
+ .
+ ($self->_is_key($cdr)
+ ? ( $self->newline||'' ) . $self->indent($depth + 1)
+ : ''
+ )
+ );
}
elsif ($car eq 'AND' or $car eq 'OR' or $car =~ / ^ $binary_op_re $ /x ) {
- return join (" $car ", map $self->unparse($_, $bindargs, $depth), @{$cdr});
+ return join (" $car ", map $self->_unparse($_, $bindargs, $depth), @{$cdr});
+ }
+ elsif ($car eq 'LIST' ) {
+ return join (', ', map $self->_unparse($_, $bindargs, $depth), @{$cdr});
}
else {
my ($l, $r) = @{$self->pad_keyword($car, $depth)};
- return sprintf "$l%s %s$r", $self->format_keyword($car), $self->unparse($cdr, $bindargs, $depth);
+
+ return sprintf "$l%s%s%s$r",
+ $self->format_keyword($car),
+ ( ref $cdr eq 'ARRAY' and ref $cdr->[0] eq 'ARRAY' and $cdr->[0][0] and $cdr->[0][0] eq 'PAREN' )
+ ? '' # mysql--
+ : ' '
+ ,
+ $self->_unparse($cdr, $bindargs, $depth),
+ ;
}
}
+# All of these keywords allow their parameters to be specified with or without parenthesis without changing the semantics
+my @unrollable_ops = (
+ 'ON',
+ 'WHERE',
+ 'GROUP \s+ BY',
+ 'HAVING',
+ 'ORDER \s+ BY',
+);
+my $unrollable_ops_re = join ' | ', @unrollable_ops;
+$unrollable_ops_re = qr/$unrollable_ops_re/xi;
+
+sub _parenthesis_unroll {
+ my $self = shift;
+ my $ast = shift;
+
+ #return if $self->parenthesis_significant;
+ return unless (ref $ast and ref $ast->[1]);
+
+ my $changes;
+ do {
+ my @children;
+ $changes = 0;
+
+ for my $child (@{$ast->[1]}) {
+ # the current node in this loop is *always* a PAREN
+ if (not ref $child or not $child->[0] eq 'PAREN') {
+ push @children, $child;
+ next;
+ }
+
+ # unroll nested parenthesis
+ while ( @{$child->[1]} && $child->[1][0][0] eq 'PAREN') {
+ $child = $child->[1][0];
+ $changes++;
+ }
+
+ # if the parenthesis are wrapped around an AND/OR matching the parent AND/OR - open the parenthesis up and merge the list
+ if (
+ ( $ast->[0] eq 'AND' or $ast->[0] eq 'OR')
+ and
+ $child->[1][0][0] eq $ast->[0]
+ ) {
+ push @children, @{$child->[1][0][1]};
+ $changes++;
+ }
+
+ # if the parent operator explcitly allows it nuke the parenthesis
+ elsif ( $ast->[0] =~ $unrollable_ops_re ) {
+ push @children, $child->[1][0];
+ $changes++;
+ }
+
+ # only *ONE* LITERAL or placeholder element
+ elsif (
+ @{$child->[1]} == 1 && (
+ $child->[1][0][0] eq 'LITERAL'
+ or
+ $child->[1][0][0] eq 'PLACEHOLDER'
+ )
+ ) {
+ push @children, $child->[1][0];
+ $changes++;
+ }
+
+ # only one element in the parenthesis which is a binary op
+ # and has exactly two grandchildren
+ # the only time when we can *not* unroll this is when both
+ # the parent and the child are mathops (in which case we'll
+ # break precedence) or when the child is BETWEEN (special
+ # case)
+ elsif (
+ @{$child->[1]} == 1
+ and
+ $child->[1][0][0] =~ SQL::Abstract::Tree::_binary_op_re()
+ and
+ $child->[1][0][0] ne 'BETWEEN'
+ and
+ @{$child->[1][0][1]} == 2
+ and
+ ! (
+ $child->[1][0][0] =~ SQL::Abstract::Tree::_math_op_re()
+ and
+ $ast->[0] =~ SQL::Abstract::Tree::_math_op_re()
+ )
+ ) {
+ push @children, $child->[1][0];
+ $changes++;
+ }
+
+ # a function binds tighter than a mathop - see if our ancestor is a
+ # mathop, and our content is:
+ # a single non-mathop child with a single PAREN grandchild which
+ # would indicate mathop ( nonmathop ( ... ) )
+ # or a single non-mathop with a single LITERAL ( nonmathop foo )
+ # or a single non-mathop with a single PLACEHOLDER ( nonmathop ? )
+ elsif (
+ @{$child->[1]} == 1
+ and
+ @{$child->[1][0][1]} == 1
+ and
+ $ast->[0] =~ SQL::Abstract::Tree::_math_op_re()
+ and
+ $child->[1][0][0] !~ SQL::Abstract::Tree::_math_op_re
+ and
+ (
+ $child->[1][0][1][0][0] eq 'PAREN'
+ or
+ $child->[1][0][1][0][0] eq 'LITERAL'
+ or
+ $child->[1][0][1][0][0] eq 'PLACEHOLDER'
+ )
+ ) {
+ push @children, $child->[1][0];
+ $changes++;
+ }
+
+
+ # otherwise no more mucking for this pass
+ else {
+ push @children, $child;
+ }
+ }
+
+ $ast->[1] = \@children;
+
+ } while ($changes);
+
+}
+
sub format { my $self = shift; $self->unparse($self->parse($_[0]), $_[1]) }
1;
=pod
+=head1 NAME
+
+SQL::Abstract::Tree - Represent SQL as an AST
+
=head1 SYNOPSIS
my $sqla_tree = SQL::Abstract::Tree->new({ profile => 'console' });