X-Git-Url: http://git.shadowcat.co.uk/gitweb/gitweb.cgi?p=scpubgit%2FQ-Branch.git;a=blobdiff_plain;f=lib%2FSQL%2FAbstract%2FTree.pm;h=3791fe9d2e94b69ccf025fb1cf4dd9d18b407327;hp=f6076ba3881988c6f4ae957097bd23966c76f4dd;hb=cf5b7ab163f8ac123ebc9bb1156e79646cd5bd2f;hpb=1c33db5d0ccacdfc4d6028ec596b0f34a045471c diff --git a/lib/SQL/Abstract/Tree.pm b/lib/SQL/Abstract/Tree.pm index f6076ba..3791fe9 100644 --- a/lib/SQL/Abstract/Tree.pm +++ b/lib/SQL/Abstract/Tree.pm @@ -1,38 +1,10 @@ package SQL::Abstract::Tree; -use strict; -use warnings; +use Moo; 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 fill_in_placeholders - placeholder_surround -); -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' ); +use Carp; +use Sub::Quote 'quote_sub'; my $op_look_ahead = '(?: (?= [\s\)\(\;] ) | \z)'; my $op_look_behind = '(?: (?<= [\,\s\)\(] ) | \A )'; @@ -50,10 +22,10 @@ my $placeholder_re = qr/(?: \? | \$\d+ )/x; my @expression_start_keywords = ( 'SELECT', 'UPDATE', + 'SET', 'INSERT \s+ INTO', 'DELETE \s+ FROM', 'FROM', - 'SET', '(?: (?: (?: (?: LEFT | RIGHT | FULL ) \s+ )? @@ -64,11 +36,11 @@ my @expression_start_keywords = ( 'ON', 'WHERE', '(?: DEFAULT \s+ )? VALUES', - 'EXISTS', 'GROUP \s+ BY', 'HAVING', 'ORDER \s+ BY', 'SKIP', + 'FETCH', 'FIRST', 'LIMIT', 'OFFSET', @@ -83,7 +55,6 @@ my @expression_start_keywords = ( 'SAVEPOINT', 'RELEASE \s+ SAVEPOINT', 'RETURNING', - 'ROW_NUMBER \s* \( \s* \) \s+ OVER', ); my $expr_start_re = join ("\n\t|\n", @expression_start_keywords ); @@ -92,46 +63,55 @@ $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 +# * BETWEEN without parentheses around the ANDed arguments (which +# makes it a non-binary op) is detected and accommodated in # _recurse_parse() +# * 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_re = join ("\n\t|\n", map +# testing as one is tighter than the other, plus alphanum cmp ops have different +# look ahead/behind (e.g. "x"="y" ) +my @alphanum_cmp_op_keywords = (qw/< > != <> = <= >= /); +my $alphanum_cmp_op_re = join ("\n\t|\n", map { "(?: (?<= [\\w\\s] | $quote_right ) | \\A )" . quotemeta ($_) . "(?: (?= [\\w\\s] | $quote_left ) | \\z )" } - @math_op_keywords + @alphanum_cmp_op_keywords ); -$math_re = qr/$math_re/x; - -sub _math_op_re { $math_re } +$alphanum_cmp_op_re = qr/$alphanum_cmp_op_re/x; - -my $binary_op_re = '(?: NOT \s+)? (?:' . join ('|', qw/IN BETWEEN R?LIKE/) . ')'; +my $binary_op_re = '(?: NOT \s+)? (?:' . join ('|', qw/IN BETWEEN [RI]?LIKE REGEXP/) . ')'; $binary_op_re = join "\n\t|\n", - "$op_look_behind (?i: $binary_op_re ) $op_look_ahead", - $math_re, + "$op_look_behind (?i: $binary_op_re | AS ) $op_look_ahead", + $alphanum_cmp_op_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 $rno_re = qr/ROW_NUMBER \s* \( \s* \) \s+ OVER/ix; + +my $unary_op_re = 'NOT \s+ EXISTS | NOT | ' . $rno_re; +$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 $all_known_re = join("\n\t|\n", +my $tokenizer_re = join("\n\t|\n", $expr_start_re, $binary_op_re, - "$op_look_behind (?i: AND|OR|NOT ) $op_look_ahead", - (map { quotemeta $_ } qw/, ( ) */), + $unary_op_re, + $asc_desc_re, + $and_or_re, + $op_look_behind . ' \* ' . $op_look_ahead, + (map { quotemeta $_ } qw/, ( )/), $placeholder_re, ); -$all_known_re = qr/$all_known_re/x; - -#this one *is* capturing for the split below +# this one *is* capturing for the split below # splits on whitespace if all else fails -my $tokenizer_re = qr/ \s* ( $all_known_re ) \s* | \s+ /x; +# has to happen before the composing qr's are anchored (below) +$tokenizer_re = qr/ \s* ( $tokenizer_re ) \s* | \s+ /x; # Parser states for _recurse_parse() use constant PARSE_TOP_LEVEL => 0; @@ -139,10 +119,32 @@ 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 | (?i: AND | OR | NOT | \, ) ) $/x; -my $func_start_re = qr/^ (?: \* | $placeholder_re | \( ) $/x; +my $expr_term_re = qr/$expr_start_re | \)/x; +my $rhs_term_re = qr/ $expr_term_re | $binary_op_re | $unary_op_re | $asc_desc_re | $and_or_re | \, /x; +my $all_std_keywords_re = qr/ $rhs_term_re | \( | $placeholder_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, + $alphanum_cmp_op_re, + $binary_op_re, + $unary_op_re, + $asc_desc_re, + $and_or_re, + $expr_term_re, + $rhs_term_re, + $all_std_keywords_re, +) { + $_ = qr/ \A $_ \z /x; +} + +# what can be bunched together under one MISC in an AST +my $compressable_node_re = qr/^ \- (?: MISC | LITERAL | PLACEHOLDER ) $/x; my %indents = ( select => 0, @@ -166,18 +168,33 @@ my %indents = ( first => 1, ); -my %profiles = ( - console => { - fill_in_placeholders => 1, - placeholder_surround => ['?/', ''], - indent_string => ' ', - indent_amount => 2, - newline => "\n", - colormap => {}, - indentmap => \%indents, - - eval { require Term::ANSIColor } - ? do { + +has [qw( + newline indent_string indent_amount fill_in_placeholders placeholder_surround +)] => (is => 'ro'); + +has [qw( indentmap colormap )] => ( is => 'ro', default => quote_sub('{}') ); + +# class global is in fact desired +my $merger; + +sub BUILDARGS { + my $class = shift; + my $args = ref $_[0] eq 'HASH' ? $_[0] : {@_}; + + if (my $p = delete $args->{profile}) { + my %extra_args; + if ($p eq 'console') { + %extra_args = ( + fill_in_placeholders => 1, + placeholder_surround => ['?/', ''], + indent_string => ' ', + indent_amount => 2, + newline => "\n", + colormap => {}, + indentmap => \%indents, + + ! ( eval { require Term::ANSIColor } ) ? () : do { my $c = \&Term::ANSIColor::color; my $red = [$c->('red') , $c->('reset')]; @@ -188,7 +205,7 @@ my %profiles = ( my $magenta = [$c->('magenta'), $c->('reset')]; my $b_o_w = [$c->('black on_white'), $c->('reset')]; ( - placeholder_surround => [q(') . $c->('black on_magenta'), $c->('reset') . q(')], + placeholder_surround => [$c->('black on_magenta'), $c->('reset')], colormap => { 'begin work' => $b_o_w, commit => $b_o_w, @@ -222,79 +239,86 @@ my %profiles = ( offset => $green, } ); - } : (), - }, - console_monochrome => { - fill_in_placeholders => 1, - placeholder_surround => ['?/', ''], - indent_string => ' ', - indent_amount => 2, - newline => "\n", - colormap => {}, - indentmap => \%indents, - }, - html => { - fill_in_placeholders => 1, - placeholder_surround => ['', ''], - indent_string => ' ', - indent_amount => 2, - newline => "
\n", - colormap => { - 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, - }, - none => { - colormap => {}, - indentmap => {}, - }, -); - -sub new { - my $class = shift; - my $args = shift || {}; - - my $profile = delete $args->{profile} || 'none'; + }, + ); + } + elsif ($p eq 'console_monochrome') { + %extra_args = ( + fill_in_placeholders => 1, + placeholder_surround => ['?/', ''], + indent_string => ' ', + indent_amount => 2, + newline => "\n", + indentmap => \%indents, + ); + } + elsif ($p eq 'html') { + %extra_args = ( + fill_in_placeholders => 1, + placeholder_surround => ['', ''], + indent_string => ' ', + indent_amount => 2, + newline => "
\n", + colormap => { map { + (my $class = $_) =~ s/\s+/-/g; + ( $_ => [ qq||, '' ] ) + } ( + keys %indents, + qw(commit rollback savepoint), + 'begin work', 'rollback to savepoint', 'release savepoint', + ) }, + indentmap => \%indents, + ); + } + elsif ($p eq 'none') { + # nada + } + else { + croak "No such profile '$p'"; + } - die "No such profile '$profile'!" unless exists $profiles{$profile}; + # see if we got any duplicates and merge if needed + if (scalar grep { exists $args->{$_} } keys %extra_args) { + # heavy-duty merge + $args = ($merger ||= do { + require Hash::Merge; + my $m = Hash::Merge->new; + + $m->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' ); + + $m; + })->merge(\%extra_args, $args ); - my $data = $merger->merge( $profiles{$profile}, $args ); + } + else { + $args = { %extra_args, %$args }; + } + } - bless $data, $class + $args; } sub parse { my ($self, $s) = @_; + return [] unless defined $s; + # tokenize string, and remove all optional whitespace my $tokens = []; foreach my $token (split $tokenizer_re, $s) { @@ -306,102 +330,173 @@ sub parse { $token =~ /\S/ ); } - $self->_recurse_parse($tokens, PARSE_TOP_LEVEL); + + 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 && $lookahead =~ $expr_term_re ) + ($state == PARSE_IN_EXPR && $tokens->[0] =~ $expr_term_re ) or - ($state == PARSE_RHS && $lookahead =~ $rhs_term_re ) + ($state == PARSE_RHS && $tokens->[0] =~ $rhs_term_re ) or - ($state == PARSE_IN_FUNC && $lookahead !~ $func_start_re) # if there are multiple values - the parenthesis will switch the $state + ($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); + + push @left, [ '-PAREN' => \@right ]; + } - $left = $left ? [$left, [PAREN => [$right||()] ]] - : [PAREN => [$right||()] ]; + # AND/OR + elsif ($token =~ $and_or_re) { + my $op = uc $token; + + my @right = $self->_recurse_parse($tokens, PARSE_IN_EXPR); + + # Merge chunks if "logic" matches + @left = [ $op => [ @left, (@right and $op eq $right[0][0]) + ? @{ $right[0][1] } + : @right + ] ]; } - # AND/OR and LIST (,) - elsif ($token =~ /^ (?: OR | AND | \, ) $/xi ) { - my $op = ($token eq ',') ? 'LIST' : uc $token; - my $right = $self->_recurse_parse($tokens, PARSE_IN_EXPR); + # LIST (,) + elsif ($token eq ',') { + + my @right = $self->_recurse_parse($tokens, PARSE_LIST_ELT); + + # deal with malformed lists ( foo, bar, , baz ) + @right = [] unless @right; - # Merge chunks if logic matches - if (ref $right and $op eq $right->[0]) { - $left = [ (shift @$right ), [$left||(), map { @$_ } @$right] ]; + @right = [ -MISC => [ @right ] ] if @right > 1; + + if (!@left) { + @left = [ -LIST => [ [], @right ] ]; + } + elsif ($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 ( $token =~ /^ $binary_op_re $ /x ) { + 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, $right] ]; + push @left, [$op => [ (@left ? pop @left : ''), @right ]]; } - # expression terminator keywords (as they start a new expression) - elsif ( $token =~ / ^ $expr_start_re $ /x ) { + + # 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||()] ]; + + # normalize RNO explicitly + $op = 'ROW_NUMBER() OVER' if $op =~ /^$rno_re$/; + + my @right = $self->_recurse_parse($tokens, PARSE_RHS); + + push @left, [ $op => \@right ]; } - # NOT - 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, [ $op => \@right ]; + } + # a '?' + elsif ($token =~ $placeholder_re) { + push @left, [ -PLACEHOLDER => [ $token ] ]; } - elsif ( $token =~ $placeholder_re) { - $left = $left ? [ $left, [ PLACEHOLDER => [ $token ] ] ] - : [ PLACEHOLDER => [ $token ] ]; + + # check if the current token is an unknown op-start + elsif (@$tokens and ($tokens->[0] eq '(' or $tokens->[0] =~ $placeholder_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 + # we see something familiar, OR in the case of RHS (binop) stop + # after the first token + # Also stop processing when we could end up with an unknown func else { - 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) || () ] ]; + my @lits = [ -LITERAL => [$token] ]; + + unshift @lits, pop @left if @left == 1; + + unless ( $state == PARSE_RHS ) { + while ( + @$tokens + and + $tokens->[0] !~ $all_std_keywords_re + and + ! (@$tokens > 1 and $tokens->[1] eq '(') + ) { + push @lits, [ -LITERAL => [ shift @$tokens ] ]; + } } - else { - $right = [ LITERAL => [ $token ] ]; + + @lits = [ -MISC => [ @lits ] ] if @lits > 1; + + push @left, @lits; + } + + # compress -LITERAL -MISC and -PLACEHOLDER pieces into a single + # -MISC container + if (@left > 1) { + my $i = 0; + while ($#left > $i) { + if ($left[$i][0] =~ $compressable_node_re and $left[$i+1][0] =~ $compressable_node_re) { + splice @left, $i, 2, [ -MISC => [ + map { $_->[0] eq '-MISC' ? @{$_->[1]} : $_ } (@left[$i, $i+1]) + ]]; + } + else { + $i++; + } } + } + + return @left if $state == PARSE_RHS; - $left = $left ? [ $left, $right ] - : $right; + # deal with post-fix operators + if (@$tokens) { + # asc/desc + if ($tokens->[0] =~ $asc_desc_re) { + @left = [ ('-' . uc (shift @$tokens)) => [ @left ] ]; + } } } } @@ -448,9 +543,11 @@ sub fill_in_placeholder { 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 '?' @@ -469,52 +566,261 @@ sub _unparse { return ''; } - my ($car, $cdr) = @{$tree}[0,1]; + # FIXME - needs a config switch to disable + $self->_parenthesis_unroll($tree); + + my ($op, $args) = @{$tree}[0,1]; - if (! defined $car or (! ref $car and ! defined $cdr) ) { + 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 $car) { + if (ref $op) { return join (' ', map $self->_unparse($_, $bindargs, $depth), @$tree); } - elsif ($car eq 'LITERAL') { - return $cdr->[0]; + elsif ($op eq '-LITERAL') { # literal has different sig + return $args->[0]; } - elsif ($car eq 'PLACEHOLDER') { + elsif ($op eq '-PLACEHOLDER') { return $self->fill_in_placeholder($bindargs); } - elsif ($car eq 'PAREN') { - return sprintf ('(%s)', - join (' ', map { $self->_unparse($_, $bindargs, $depth + 2) } @{$cdr} ) + elsif ($op eq '-PAREN') { + return sprintf ('( %s )', + join (' ', map { $self->_unparse($_, $bindargs, $depth + 2) } @{$args} ) . - ($self->_is_key($cdr) + ($self->_is_key($args) ? ( $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}); + elsif ($op eq 'AND' or $op eq 'OR' or $op =~ $binary_op_re ) { + return join (" $op ", map $self->_unparse($_, $bindargs, $depth), @{$args}); + } + 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 ($car eq 'LIST' ) { - return join (', ', map $self->_unparse($_, $bindargs, $depth), @{$cdr}); + elsif ($op =~ qr/^-(ASC|DESC)$/ ) { + my $dir = $1; + return join (' ', (map $self->_unparse($_, $bindargs, $depth), @{$args}), $dir); } else { - my ($l, $r) = @{$self->pad_keyword($car, $depth)}; - return sprintf "$l%s %s$r", $self->format_keyword($car), $self->_unparse($cdr, $bindargs, $depth); + my ($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; + } + + my $parent_op = $ast->[0]; + + # unroll nested parenthesis + while ( $parent_op ne 'IN' and @{$child->[1]} == 1 and $child->[1][0][0] eq '-PAREN') { + $child = $child->[1][0]; + $changes++; + } + + # set to CHILD in the case of PARENT ( CHILD ) + # but NOT in the case of PARENT( CHILD1, CHILD2 ) + my $single_child_op = (@{$child->[1]} == 1) ? $child->[1][0][0] : ''; + + my $child_op_argc = $single_child_op ? scalar @{$child->[1][0][1]} : undef; + + my $single_grandchild_op + = ( $child_op_argc||0 == 1 and ref $child->[1][0][1][0] eq 'ARRAY' ) + ? $child->[1][0][1][0][0] + : '' + ; + + # if the parent operator explicitly allows it AND the child isn't a subselect + # nuke the parenthesis + if ($parent_op =~ $unrollable_ops_re and $single_child_op ne 'SELECT') { + push @children, @{$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 ( + $single_child_op eq $parent_op + and + ( $parent_op eq 'AND' or $parent_op eq 'OR') + ) { + push @children, @{$child->[1][0][1]}; + $changes++; + } + + # only *ONE* LITERAL or placeholder element + # as an AND/OR/NOT argument + elsif ( + ( $single_child_op eq '-LITERAL' or $single_child_op eq '-PLACEHOLDER' ) + and + ( $parent_op eq 'AND' or $parent_op eq 'OR' or $parent_op eq 'NOT' ) + ) { + push @children, @{$child->[1]}; + $changes++; + } + + # an AND/OR expression with only one binop in the parenthesis + # with exactly two grandchildren + # the only time when we can *not* unroll this is when both + # the parent and the child are mathops (in which case we'll + # break precedence) or when the child is BETWEEN (special + # case) + elsif ( + ($parent_op eq 'AND' or $parent_op eq 'OR') + and + $single_child_op =~ $binary_op_re + and + $single_child_op ne 'BETWEEN' + and + $child_op_argc == 2 + and + ! ( + $single_child_op =~ $alphanum_cmp_op_re + and + $parent_op =~ $alphanum_cmp_op_re + ) + ) { + push @children, @{$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 ( + $single_child_op + and + $parent_op =~ $alphanum_cmp_op_re + and + $single_child_op !~ $alphanum_cmp_op_re + and + $child_op_argc == 1 + and + ( + $single_grandchild_op eq '-PAREN' + or + $single_grandchild_op eq '-LITERAL' + or + $single_grandchild_op eq '-PLACEHOLDER' + ) + ) { + push @children, @{$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 + # and except for the case of RNO, where the double are explicit syntax + elsif ( + $parent_op ne 'ROW_NUMBER() OVER' + and + $single_child_op + and + $single_child_op ne 'NOT' + and + $child_op_argc == 1 + and + $single_grandchild_op eq '-PAREN' + ) { + push @children, @{$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($_[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' }); @@ -580,7 +886,7 @@ structure of the returned tree. It may be stable at some point, but not yet. =head2 unparse - $sqlat->parse($tree_structure, \@bindargs) + $sqlat->unparse($tree_structure, \@bindargs) Transform "tree" into SQL, applying various transforms on the way.