X-Git-Url: http://git.shadowcat.co.uk/gitweb/gitweb.cgi?a=blobdiff_plain;f=lib%2FSQL%2FAbstract%2FTree.pm;h=a4b01fd9f3891683c27aa74bc1cae07ee2dbc24b;hb=87abf9bc3395870cef72a3f00b7d4cfe18abd980;hp=0110fc5f0e9bfc3b382d8f82b90b1cf5a8d1dc52;hpb=b4e0e260fc40159f0e12b6e98b68ecdcb30b3ab1;p=scpubgit%2FQ-Branch.git
diff --git a/lib/SQL/Abstract/Tree.pm b/lib/SQL/Abstract/Tree.pm
index 0110fc5..a4b01fd 100644
--- a/lib/SQL/Abstract/Tree.pm
+++ b/lib/SQL/Abstract/Tree.pm
@@ -2,120 +2,305 @@ package SQL::Abstract::Tree;
use strict;
use warnings;
+no warnings 'qw';
use Carp;
+use Hash::Merge qw//;
use base 'Class::Accessor::Grouped';
-__PACKAGE__->mk_group_accessors( simple => $_ ) for qw(
- newline indent_string indent_amount colormap indentmap
-);
+__PACKAGE__->mk_group_accessors( simple => qw(
+ newline indent_string indent_amount colormap indentmap fill_in_placeholders
+ placeholder_surround
+));
-# 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;
+my $merger = Hash::Merge->new;
+
+$merger->specify_behavior({
+ SCALAR => {
+ SCALAR => sub { $_[1] },
+ ARRAY => sub { [ $_[0], @{$_[1]} ] },
+ HASH => sub { $_[1] },
+ },
+ ARRAY => {
+ SCALAR => sub { $_[1] },
+ ARRAY => sub { $_[1] },
+ HASH => sub { $_[1] },
+ },
+ HASH => {
+ SCALAR => sub { $_[1] },
+ ARRAY => sub { [ values %{$_[0]}, @{$_[1]} ] },
+ HASH => sub { Hash::Merge::_merge_hashes( $_[0], $_[1] ) },
+ },
+}, 'SQLA::Tree Behavior' );
+
+my $op_look_ahead = '(?: (?= [\s\)\(\;] ) | \z)';
+my $op_look_behind = '(?: (?<= [\,\s\)\(] ) | \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.
+# Format: A list of strings that will be compiled to extended syntax ie.
# /.../x) regexes, without capturing parentheses. They will be automatically
-# anchored to word boundaries to match the whole token).
-my @expression_terminator_sql_keywords = (
+# anchored to op boundaries (excluding quotes) to match the whole token.
+my @expression_start_keywords = (
'SELECT',
+ 'UPDATE',
+ 'SET',
+ 'INSERT \s+ INTO',
+ 'DELETE \s+ FROM',
'FROM',
'(?:
(?:
- (?: \b (?: LEFT | RIGHT | FULL ) \s+ )?
- (?: \b (?: CROSS | INNER | OUTER ) \s+ )?
+ (?: (?: LEFT | RIGHT | FULL ) \s+ )?
+ (?: (?: CROSS | INNER | OUTER ) \s+ )?
)?
JOIN
)',
'ON',
'WHERE',
- 'EXISTS',
+ '(?: DEFAULT \s+ )? VALUES',
'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 $expr_start_re = join ("\n\t|\n", @expression_start_keywords );
+$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
# * so is NOT as being unary
# * BETWEEN without paranthesis around the ANDed arguments (which
# makes it a non-binary op) is detected and accomodated in
# _recurse_parse()
-my $stuff_around_mathops = qr/[\w\s\`\'\"\)]/;
-my @binary_op_keywords = (
- ( map
- {
- ' ^ ' . quotemeta ($_) . "(?= \$ | $stuff_around_mathops ) ",
- " (?<= $stuff_around_mathops)" . quotemeta ($_) . "(?= \$ | $stuff_around_mathops ) ",
- }
- (qw/< > != <> = <= >=/)
- ),
- ( map
- { '\b (?: NOT \s+)?' . $_ . '\b' }
- (qw/IN BETWEEN LIKE/)
- ),
+# * AS is not really an operator but is handled here as it's also LHS/RHS
+
+# this will be included in the $binary_op_re, the distinction is interesting during
+# testing as one is tighter than the other, plus mathops have different look
+# ahead/behind (e.g. "x"="y" )
+my @math_op_keywords = (qw/ - + < > != <> = <= >= /);
+my $math_op_re = join ("\n\t|\n", map
+ { "(?: (?<= [\\w\\s] | $quote_right ) | \\A )" . quotemeta ($_) . "(?: (?= [\\w\\s] | $quote_left ) | \\z )" }
+ @math_op_keywords
);
-
-my $tokenizer_re_str = join("\n\t|\n",
- ( map { '\b' . $_ . '\b' } @expression_terminator_sql_keywords, 'AND', 'OR', 'NOT'),
- @binary_op_keywords,
+$math_op_re = qr/$math_op_re/x;
+
+my $binary_op_re = '(?: NOT \s+)? (?:' . join ('|', qw/IN BETWEEN R?LIKE/) . ')';
+$binary_op_re = join "\n\t|\n",
+ "$op_look_behind (?i: $binary_op_re | AS ) $op_look_ahead",
+ $math_op_re,
+ $op_look_behind . 'IS (?:\s+ NOT)?' . "(?= \\s+ NULL \\b | $op_look_ahead )",
+;
+$binary_op_re = qr/$binary_op_re/x;
+
+my $unary_op_re = '(?: NOT \s+ EXISTS | NOT )';
+$unary_op_re = join "\n\t|\n",
+ "$op_look_behind (?i: $unary_op_re ) $op_look_ahead",
+;
+$unary_op_re = qr/$unary_op_re/x;
+
+my $asc_desc_re = qr/$op_look_behind (?i: ASC | DESC ) $op_look_ahead /x;
+my $and_or_re = qr/$op_look_behind (?i: AND | OR ) $op_look_ahead /x;
+
+my $tokenizer_re = join("\n\t|\n",
+ $expr_start_re,
+ $binary_op_re,
+ $unary_op_re,
+ $asc_desc_re,
+ $and_or_re,
+ "$op_look_behind \\* $op_look_ahead",
+ (map { quotemeta $_ } qw/, ( )/),
+ $placeholder_re,
);
-my $tokenizer_re = qr/ \s* ( $tokenizer_re_str | \( | \) | \? ) \s* /xi;
+# this one *is* capturing for the split below
+# splits on whitespace if all else fails
+# has to happen before the composiign qr's are anchored (below)
+$tokenizer_re = qr/ \s* ( $tokenizer_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;
+use constant PARSE_LIST_ELT => 5;
+
+my $expr_term_re = qr/$expr_start_re | \)/x;
+my $rhs_term_re = qr/ $expr_term_re | $binary_op_re | $unary_op_re | $asc_desc_re | $and_or_re | \, /x;
+my $common_single_args_re = qr/ \* | $placeholder_re /x;
+my $all_std_keywords_re = qr/ $rhs_term_re | \( | $common_single_args_re /x;
+
+# anchor everything - even though keywords are separated by the tokenizer, leakage may occur
+for (
+ $quote_left,
+ $quote_right,
+ $placeholder_re,
+ $expr_start_re,
+ $math_op_re,
+ $binary_op_re,
+ $unary_op_re,
+ $asc_desc_re,
+ $and_or_re,
+ $expr_term_re,
+ $rhs_term_re,
+ $common_single_args_re,
+ $all_std_keywords_re,
+) {
+ $_ = qr/ \A $_ \z /x;
+}
+
-sub _binary_op_keywords { @binary_op_keywords }
my %indents = (
- select => 0,
- where => 1,
- from => 1,
- join => 1,
- on => 2,
- 'group by' => 1,
- 'order by' => 1,
+ select => 0,
+ update => 0,
+ 'insert into' => 0,
+ 'delete from' => 0,
+ from => 1,
+ where => 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 = (
console => {
+ fill_in_placeholders => 1,
+ placeholder_surround => ['?/', ''],
indent_string => ' ',
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_magenta'), $c->('reset')],
+ colormap => {
+ '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,
+ }
+ );
+ } : (),
},
console_monochrome => {
+ fill_in_placeholders => 1,
+ placeholder_surround => ['?/', ''],
indent_string => ' ',
indent_amount => 2,
newline => "\n",
colormap => {},
- indentmap => { %indents },
+ indentmap => \%indents,
},
html => {
+ fill_in_placeholders => 1,
+ placeholder_surround => ['', ''],
indent_string => ' ',
indent_amount => 2,
newline => "
\n",
colormap => {
- select => ['' , ''],
- where => ['' , ''],
- from => ['' , ''],
- join => ['' , ''],
- on => ['' , ''],
- 'group by' => ['', ''],
+ select => ['' , ''],
+ 'insert into' => ['' , ''],
+ update => ['' , ''],
+ 'delete from' => ['' , ''],
+
+ set => ['', ''],
+ from => ['' , ''],
+
+ where => ['' , ''],
+ values => ['', ''],
+
+ join => ['' , ''],
+ 'left join' => ['',''],
+ on => ['' , ''],
+
+ 'group by' => ['', ''],
+ having => ['', ''],
+ 'order by' => ['', ''],
+
+ skip => ['', ''],
+ first => ['', ''],
+ limit => ['', ''],
+ offset => ['', ''],
+
+ 'begin work' => ['', ''],
+ commit => ['', ''],
+ rollback => ['', ''],
+ savepoint => ['', ''],
+ 'rollback to savepoint' => ['', ''],
+ 'release savepoint' => ['', ''],
},
- indentmap => { %indents },
+ indentmap => \%indents,
},
none => {
colormap => {},
@@ -123,24 +308,15 @@ my %profiles = (
},
);
-eval {
- require Term::ANSIColor;
- $profiles{console}->{colormap} = {
- select => [Term::ANSIColor::color('red'), Term::ANSIColor::color('reset')],
- where => [Term::ANSIColor::color('green'), Term::ANSIColor::color('reset')],
- from => [Term::ANSIColor::color('cyan'), Term::ANSIColor::color('reset')],
- join => [Term::ANSIColor::color('magenta'), Term::ANSIColor::color('reset')],
- on => [Term::ANSIColor::color('blue'), Term::ANSIColor::color('reset')],
- 'group by' => [Term::ANSIColor::color('cyan'), Term::ANSIColor::color('reset')],
- 'order by' => [Term::ANSIColor::color('yellow'), Term::ANSIColor::color('reset')],
- };
-};
-
sub new {
- my ($class, $args) = @_;
+ my $class = shift;
+ my $args = shift || {};
my $profile = delete $args->{profile} || 'none';
- my $data = {%{$profiles{$profile}}, %{$args||{}}};
+
+ die "No such profile '$profile'!" unless exists $profiles{$profile};
+
+ my $data = $merger->merge( $profiles{$profile}, $args );
bless $data, $class
}
@@ -151,89 +327,160 @@ sub parse {
# 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;
+ return [ $self->_recurse_parse($tokens, PARSE_TOP_LEVEL) ];
}
sub _recurse_parse {
my ($self, $tokens, $state) = @_;
- my $left;
+ my @left;
while (1) { # left-associative parsing
- my $lookahead = $tokens->[0];
- if ( not defined($lookahead)
+ if ( ! @$tokens
or
- ($state == PARSE_IN_PARENS && $lookahead eq ')')
+ ($state == PARSE_IN_PARENS && $tokens->[0] eq ')')
or
- ($state == PARSE_IN_EXPR && grep { $lookahead =~ /^ $_ $/xi } ('\)', @expression_terminator_sql_keywords ) )
+ ($state == PARSE_IN_EXPR && $tokens->[0] =~ $expr_term_re )
or
- ($state == PARSE_RHS && grep { $lookahead =~ /^ $_ $/xi } ('\)', @expression_terminator_sql_keywords, @binary_op_keywords, 'AND', 'OR', 'NOT' ) )
+ ($state == PARSE_RHS && $tokens->[0] =~ $rhs_term_re )
+ or
+ ($state == PARSE_LIST_ELT && ( $tokens->[0] eq ',' or $tokens->[0] =~ $expr_term_re ) )
) {
- return $left;
+ return @left;
}
my $token = shift @$tokens;
# nested expression in ()
if ($token eq '(' ) {
- my $right = $self->_recurse_parse($tokens, PARSE_IN_PARENS);
- $token = shift @$tokens or croak "missing closing ')' around block " . $self->unparse($right);
- $token eq ')' or croak "unexpected token '$token' terminating block " . $self->unparse($right);
+ my @right = $self->_recurse_parse($tokens, PARSE_IN_PARENS);
+ $token = shift @$tokens or croak "missing closing ')' around block " . $self->unparse(\@right);
+ $token eq ')' or croak "unexpected token '$token' terminating block " . $self->unparse(\@right);
- $left = $left ? [@$left, [PAREN => [$right] ]]
- : [PAREN => [$right] ];
+ push @left, [ '-PAREN' => \@right ];
}
+
# AND/OR
- elsif ($token =~ /^ (?: OR | AND ) $/xi ) {
+ elsif ($token =~ $and_or_re) {
my $op = 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] ];
+ my @right = $self->_recurse_parse($tokens, PARSE_IN_EXPR);
+
+ # Merge chunks if "logic" matches
+ @left = [ $op => [ @left, (@right and $op eq $right[0][0])
+ ? @{ $right[0][1] }
+ : @right
+ ] ];
+ }
+
+ # LIST (,)
+ elsif ($token eq ',') {
+
+ my @right = $self->_recurse_parse($tokens, PARSE_LIST_ELT);
+
+ # deal with malformed lists ( foo, bar, , baz )
+ @right = [] unless @right;
+
+ @right = [ -MISC => [ @right ] ] if @right > 1;
+
+ if (!@left) {
+ @left = [ -LIST => [ [], @right ] ];
+ }
+ elsif ($left[0][0] eq '-LIST') {
+ push @{$left[0][1]}, (@{$right[0]} and $right[0][0] eq '-LIST')
+ ? @{$right[0][1]}
+ : @right
+ ;
}
else {
- $left = [$op => [$left, $right]];
+ @left = [ -LIST => [ @left, @right ] ];
}
}
+
# binary operator keywords
- elsif (grep { $token =~ /^ $_ $/xi } @binary_op_keywords ) {
+ elsif ($token =~ $binary_op_re) {
my $op = uc $token;
- my $right = $self->_recurse_parse($tokens, PARSE_RHS);
+
+ my @right = $self->_recurse_parse($tokens, PARSE_RHS);
# A between with a simple LITERAL for a 1st RHS argument needs a
# rerun of the search to (hopefully) find the proper AND construct
- if ($op eq 'BETWEEN' and $right->[0] eq 'LITERAL') {
- unshift @$tokens, $right->[1][0];
- $right = $self->_recurse_parse($tokens, PARSE_IN_EXPR);
+ if ($op eq 'BETWEEN' and $right[0] eq '-LITERAL') {
+ unshift @$tokens, $right[1][0];
+ @right = $self->_recurse_parse($tokens, PARSE_IN_EXPR);
}
- $left = [$op => [$left, $right] ];
+ @left = [$op => [ @left, @right ]];
}
- # expression terminator keywords (as they start a new expression)
- elsif (grep { $token =~ /^ $_ $/xi } @expression_terminator_sql_keywords ) {
+
+ # unary op keywords
+ elsif ( $token =~ $unary_op_re ) {
my $op = uc $token;
- my $right = $self->_recurse_parse($tokens, PARSE_IN_EXPR);
- $left = $left ? [ $left, [$op => [$right] ]]
- : [ $op => [$right] ];
+ my @right = $self->_recurse_parse ($tokens, PARSE_RHS);
+
+ push @left, [ $op => \@right ];
}
- # NOT (last as to allow all other NOT X pieces first)
- elsif ( $token =~ /^ not $/ix ) {
+
+ # expression terminator keywords
+ elsif ( $token =~ $expr_start_re ) {
my $op = uc $token;
- my $right = $self->_recurse_parse ($tokens, PARSE_RHS);
- $left = $left ? [ @$left, [$op => [$right] ]]
- : [ $op => [$right] ];
+ my @right = $self->_recurse_parse($tokens, PARSE_IN_EXPR);
+ push @left, [ $op => \@right ];
}
- # literal (eat everything on the right until RHS termination)
+
+ # a '?'
+ elsif ( $token =~ $placeholder_re) {
+ push @left, [ -PLACEHOLDER => [ $token ] ];
+ }
+
+ # check if the current token is an unknown op-start
+ elsif (@$tokens and ($tokens->[0] eq '(' or $tokens->[0] =~ $common_single_args_re ) ) {
+ push @left, [ $token => [ $self->_recurse_parse($tokens, PARSE_RHS) ] ];
+ }
+
+ # 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 @lits = [ -LITERAL => [$token] ];
+
+ while (@$tokens and $tokens->[0] !~ $all_std_keywords_re) {
+ push @lits, [ -LITERAL => [ shift @$tokens ] ];
+ }
+
+ if (@left == 1) {
+ unshift @lits, pop @left;
+ }
+
+ @lits = [ -MISC => [ @lits ] ] if @lits > 1;
+
+ push @left, @lits;
+ }
+
+ # deal with post-fix operators (only when sql is sane - i.e. we have one element to apply to)
+ if (@left == 1 and @$tokens) {
+
+ # asc/desc
+ if ($tokens->[0] =~ $asc_desc_re) {
+ my $op = shift @$tokens;
+
+ # if -MISC - this is a literal collection, do not promote asc/desc to an op
+ if ($left[0][0] eq '-MISC') {
+ push @{$left[0][1]}, [ -LITERAL => [ $op ] ];
+ }
+ else {
+ @left = [ ('-' . uc ($op)) => [ @left ] ];
+ }
+ }
}
}
}
@@ -248,65 +495,308 @@ sub format_keyword {
return $keyword
}
-sub whitespace {
+my %starters = (
+ select => 1,
+ update => 1,
+ 'insert into' => 1,
+ 'delete from' => 1,
+);
+
+sub pad_keyword {
my ($self, $keyword, $depth) = @_;
my $before = '';
if (defined $self->indentmap->{lc $keyword}) {
$before = $self->newline . $self->indent($depth + $self->indentmap->{lc $keyword});
}
- $before = '' if $depth == 0 and lc $keyword eq 'select';
- return [$before, ' '];
+ $before = '' if $depth == 0 and defined $starters{lc $keyword};
+ return [$before, ''];
}
sub indent { ($_[0]->indent_string||'') x ( ( $_[0]->indent_amount || 0 ) * $_[1] ) }
-sub _is_select {
- my $tree = shift;
+sub _is_key {
+ my ($self, $tree) = @_;
$tree = $tree->[0] while ref $tree;
- defined $tree && lc $tree eq 'select';
+ defined $tree && defined $self->indentmap->{lc $tree};
}
+sub fill_in_placeholder {
+ my ($self, $bindargs) = @_;
+
+ if ($self->fill_in_placeholders) {
+ my $val = shift @{$bindargs} || '';
+ my $quoted = $val =~ s/^(['"])(.*)\1$/$2/;
+ my ($left, $right) = @{$self->placeholder_surround};
+ $val =~ s/\\/\\\\/g;
+ $val =~ s/'/\\'/g;
+ $val = qq('$val') if $quoted;
+ return qq($left$val$right)
+ }
+ return '?'
+}
+
+# FIXME - terrible name for a user facing API
sub unparse {
- my ($self, $tree, $depth) = @_;
+ my ($self, $tree, $bindargs) = @_;
+ $self->_unparse($tree, [@{$bindargs||[]}], 0);
+}
- $depth ||= 0;
+sub _unparse {
+ my ($self, $tree, $bindargs, $depth) = @_;
- if (not $tree ) {
+ if (not $tree or not @$tree) {
return '';
}
- my $car = $tree->[0];
- my $cdr = $tree->[1];
+ # FIXME - needs a config switch to disable
+ $self->_parenthesis_unroll($tree);
- if (ref $car) {
- return join ('', map $self->unparse($_, $depth), @$tree);
+ my ($op, $args) = @{$tree}[0,1];
+
+ if (! defined $op or (! ref $op and ! defined $args) ) {
+ require Data::Dumper;
+ Carp::confess( sprintf ( "Internal error - malformed branch at depth $depth:\n%s",
+ Data::Dumper::Dumper($tree)
+ ) );
+ }
+
+ if (ref $op) {
+ return join (' ', map $self->_unparse($_, $bindargs, $depth), @$tree);
+ }
+ elsif ($op eq '-LITERAL') { # literal has different sig
+ return $args->[0];
+ }
+ elsif ($op eq '-PLACEHOLDER') {
+ return $self->fill_in_placeholder($bindargs);
}
- elsif ($car eq 'LITERAL') {
- return $cdr->[0];
+ elsif ($op eq '-PAREN') {
+ return sprintf ('( %s )',
+ join (' ', map { $self->_unparse($_, $bindargs, $depth + 2) } @{$args} )
+ .
+ ($self->_is_key($args)
+ ? ( $self->newline||'' ) . $self->indent($depth + 1)
+ : ''
+ )
+ );
}
- elsif ($car eq 'PAREN') {
- return '(' .
- join(' ',
- map $self->unparse($_, $depth + 2), @{$cdr}) .
- (_is_select($cdr)?( $self->newline||'' ).$self->indent($depth + 1):'') . ') ';
+ elsif ($op eq 'AND' or $op eq 'OR' or $op =~ $binary_op_re ) {
+ return join (" $op ", map $self->_unparse($_, $bindargs, $depth), @{$args});
}
- elsif ($car eq 'OR' or $car eq 'AND' or (grep { $car =~ /^ $_ $/xi } @binary_op_keywords ) ) {
- return join (" $car ", map $self->unparse($_, $depth), @{$cdr});
+ elsif ($op eq '-LIST' ) {
+ return join (', ', map $self->_unparse($_, $bindargs, $depth), @{$args});
+ }
+ elsif ($op eq '-MISC' ) {
+ return join (' ', map $self->_unparse($_, $bindargs, $depth), @{$args});
+ }
+ elsif ($op =~ qr/^-(ASC|DESC)$/ ) {
+ my $dir = $1;
+ return join (' ', (map $self->_unparse($_, $bindargs, $depth), @{$args}), $dir);
}
else {
- my ($l, $r) = @{$self->whitespace($car, $depth)};
- return sprintf "$l%s %s$r", $self->format_keyword($car), $self->unparse($cdr, $depth);
+ my ($l, $r) = @{$self->pad_keyword($op, $depth)};
+
+ my $rhs = $self->_unparse($args, $bindargs, $depth);
+
+ return sprintf "$l%s$r", join(
+ ( ref $args eq 'ARRAY' and @{$args} == 1 and $args->[0][0] eq '-PAREN' )
+ ? '' # mysql--
+ : ' '
+ ,
+ $self->format_keyword($op),
+ (length $rhs ? $rhs : () ),
+ );
+ }
+}
+
+# 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',
+ 'I?LIKE',
+);
+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 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 (! ref $child or ! @$child or $child->[0] ne '-PAREN') {
+ push @children, $child;
+ next;
+ }
+
+ # unroll nested parenthesis
+ while ( @{$child->[1]} == 1 and $child->[1][0][0] eq '-PAREN') {
+ $child = $child->[1][0];
+ $changes++;
+ }
+
+ # if the parent operator explcitly allows it nuke the parenthesis
+ if ( $ast->[0] =~ $unrollable_ops_re ) {
+ push @children, @{$child->[1]};
+ $changes++;
+ }
+
+ # if the parenthesis are wrapped around an AND/OR matching the parent AND/OR - open the parenthesis up and merge the list
+ elsif (
+ @{$child->[1]} == 1
+ and
+ ( $ast->[0] eq 'AND' or $ast->[0] eq 'OR')
+ and
+ $child->[1][0][0] eq $ast->[0]
+ ) {
+ push @children, @{$child->[1][0][1]};
+ $changes++;
+ }
+
+ # only *ONE* LITERAL or placeholder element
+ # as an AND/OR/NOT argument
+ elsif (
+ @{$child->[1]} == 1 && (
+ $child->[1][0][0] eq '-LITERAL'
+ or
+ $child->[1][0][0] eq '-PLACEHOLDER'
+ ) && (
+ $ast->[0] eq 'AND' or $ast->[0] eq 'OR' or $ast->[0] eq 'NOT'
+ )
+ ) {
+ push @children, @{$child->[1]};
+ $changes++;
+ }
+
+ # an AND/OR expression with only one binop in the parenthesis
+ # with exactly two grandchildren
+ # the only time when we can *not* unroll this is when both
+ # the parent and the child are mathops (in which case we'll
+ # break precedence) or when the child is BETWEEN (special
+ # case)
+ elsif (
+ @{$child->[1]} == 1
+ and
+ ($ast->[0] eq 'AND' or $ast->[0] eq 'OR')
+ and
+ $child->[1][0][0] =~ $binary_op_re
+ and
+ $child->[1][0][0] ne 'BETWEEN'
+ and
+ @{$child->[1][0][1]} == 2
+ and
+ ! (
+ $child->[1][0][0] =~ $math_op_re
+ and
+ $ast->[0] =~ $math_op_re
+ )
+ ) {
+ push @children, @{$child->[1]};
+ $changes++;
+ }
+
+ # a function binds tighter than a mathop - see if our ancestor is a
+ # mathop, and our content is:
+ # a single non-mathop child with a single PAREN grandchild which
+ # would indicate mathop ( nonmathop ( ... ) )
+ # or a single non-mathop with a single LITERAL ( nonmathop foo )
+ # or a single non-mathop with a single PLACEHOLDER ( nonmathop ? )
+ elsif (
+ @{$child->[1]} == 1
+ and
+ @{$child->[1][0][1]} == 1
+ and
+ $ast->[0] =~ $math_op_re
+ and
+ $child->[1][0][0] !~ $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]};
+ $changes++;
+ }
+
+ # a construct of ... ( somefunc ( ... ) ) ... can safely lose the outer parens
+ # except for the case of ( NOT ( ... ) ) which has already been handled earlier
+ elsif (
+ @{$child->[1]} == 1
+ and
+ @{$child->[1][0][1]} == 1
+ and
+ $child->[1][0][0] ne 'NOT'
+ and
+ ref $child->[1][0][1][0] eq 'ARRAY'
+ and
+ $child->[1][0][1][0][0] eq '-PAREN'
+ ) {
+ push @children, @{$child->[1]};
+ $changes++;
+ }
+
+
+ # otherwise no more mucking for this pass
+ else {
+ push @children, $child;
+ }
+ }
+
+ $ast->[1] = \@children;
+
+ } while ($changes);
+}
+
+sub _strip_asc_from_order_by {
+ my ($self, $ast) = @_;
+
+ return $ast if (
+ ref $ast ne 'ARRAY'
+ or
+ $ast->[0] ne 'ORDER BY'
+ );
+
+
+ my $to_replace;
+
+ if (@{$ast->[1]} == 1 and $ast->[1][0][0] eq '-ASC') {
+ $to_replace = [ $ast->[1][0] ];
+ }
+ elsif (@{$ast->[1]} == 1 and $ast->[1][0][0] eq '-LIST') {
+ $to_replace = [ grep { $_->[0] eq '-ASC' } @{$ast->[1][0][1]} ];
}
+
+ @$_ = @{$_->[1][0]} for @$to_replace;
+
+ $ast;
}
-sub format { my $self = shift; $self->unparse($self->parse(@_)) }
+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' });
@@ -317,3 +807,116 @@ sub format { my $self = shift; $self->unparse($self->parse(@_)) }
# FROM foo
# WHERE foo.a > 2
+=head1 METHODS
+
+=head2 new
+
+ my $sqla_tree = SQL::Abstract::Tree->new({ profile => 'console' });
+
+ $args = {
+ profile => 'console', # predefined profile to use (default: 'none')
+ fill_in_placeholders => 1, # true for placeholder population
+ placeholder_surround => # The strings that will be wrapped around
+ [GREEN, RESET], # populated placeholders if the above is set
+ indent_string => ' ', # the string used when indenting
+ indent_amount => 2, # how many of above string to use for a single
+ # indent level
+ newline => "\n", # string for newline
+ colormap => {
+ select => [RED, RESET], # a pair of strings defining what to surround
+ # the keyword with for colorization
+ # ...
+ },
+ indentmap => {
+ select => 0, # A zero means that the keyword will start on
+ # a new line
+ from => 1, # Any other positive integer means that after
+ on => 2, # said newline it will get that many indents
+ # ...
+ },
+ }
+
+Returns a new SQL::Abstract::Tree object. All arguments are optional.
+
+=head3 profiles
+
+There are four predefined profiles, C, C, C,
+and C. Typically a user will probably just use C or
+C, but if something about a profile bothers you, merely
+use the profile and override the parts that you don't like.
+
+=head2 format
+
+ $sqlat->format('SELECT * FROM bar WHERE x = ?', [1])
+
+Takes C<$sql> and C<\@bindargs>.
+
+Returns a formatting string based on the string passed in
+
+=head2 parse
+
+ $sqlat->parse('SELECT * FROM bar WHERE x = ?')
+
+Returns a "tree" representing passed in SQL. Please do not depend on the
+structure of the returned tree. It may be stable at some point, but not yet.
+
+=head2 unparse
+
+ $sqlat->unparse($tree_structure, \@bindargs)
+
+Transform "tree" into SQL, applying various transforms on the way.
+
+=head2 format_keyword
+
+ $sqlat->format_keyword('SELECT')
+
+Currently this just takes a keyword and puts the C stuff around it.
+Later on it may do more and allow for coderef based transforms.
+
+=head2 pad_keyword
+
+ my ($before, $after) = @{$sqlat->pad_keyword('SELECT')};
+
+Returns whitespace to be inserted around a keyword.
+
+=head2 fill_in_placeholder
+
+ my $value = $sqlat->fill_in_placeholder(\@bindargs)
+
+Removes last arg from passed arrayref and returns it, surrounded with
+the values in placeholder_surround, and then surrounded with single quotes.
+
+=head2 indent
+
+Returns as many indent strings as indent amounts times the first argument.
+
+=head1 ACCESSORS
+
+=head2 colormap
+
+See L
+
+=head2 fill_in_placeholders
+
+See L
+
+=head2 indent_amount
+
+See L
+
+=head2 indent_string
+
+See L
+
+=head2 indentmap
+
+See L
+
+=head2 newline
+
+See L
+
+=head2 placeholder_surround
+
+See L
+