X-Git-Url: http://git.shadowcat.co.uk/gitweb/gitweb.cgi?a=blobdiff_plain;f=lib%2FSQL%2FAbstract%2FTree.pm;h=9edc1d722dd41ef7cebaf9ef386b686e312c965f;hb=a4d17ff1e4ca5f981aacbeab10b0efa93ac047d9;hp=8f92ce90963402076ec1540e7f9921fe7a11faed;hpb=ad591616b7132657eccc0e6369dd786c7b6704e6;p=dbsrgits%2FSQL-Abstract.git diff --git a/lib/SQL/Abstract/Tree.pm b/lib/SQL/Abstract/Tree.pm index 8f92ce9..9edc1d7 100644 --- a/lib/SQL/Abstract/Tree.pm +++ b/lib/SQL/Abstract/Tree.pm @@ -68,6 +68,7 @@ my @expression_start_keywords = ( 'HAVING', 'ORDER \s+ BY', 'SKIP', + 'FETCH', 'FIRST', 'LIMIT', 'OFFSET', @@ -82,7 +83,6 @@ 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 ); @@ -91,30 +91,32 @@ $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_op_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_op_re = qr/$math_op_re/x; +$alphanum_cmp_op_re = qr/$alphanum_cmp_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, + $alphanum_cmp_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 )'; +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", ; @@ -129,14 +131,14 @@ my $tokenizer_re = join("\n\t|\n", $unary_op_re, $asc_desc_re, $and_or_re, - "$op_look_behind \\* $op_look_ahead", + $op_look_behind . ' \* ' . $op_look_ahead, (map { quotemeta $_ } qw/, ( )/), $placeholder_re, ); # 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) +# 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() @@ -149,8 +151,7 @@ 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; +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 ( @@ -158,20 +159,20 @@ for ( $quote_right, $placeholder_re, $expr_start_re, - $math_op_re, + $alphanum_cmp_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; } - +# what can be bunched together under one MISC in an AST +my $compressable_node_re = qr/^ \- (?: MISC | LITERAL | PLACEHOLDER ) $/x; my %indents = ( select => 0, @@ -419,12 +420,16 @@ sub _recurse_parse { @right = $self->_recurse_parse($tokens, PARSE_IN_EXPR); } - @left = [$op => [ @left, @right ]]; + push @left, [$op => [ (@left ? pop @left : ''), @right ]]; } # unary op keywords elsif ( $token =~ $unary_op_re ) { my $op = uc $token; + + # normalize RNO explicitly + $op = 'ROW_NUMBER() OVER' if $op =~ /^$rno_re$/; + my @right = $self->_recurse_parse ($tokens, PARSE_RHS); push @left, [ $op => \@right ]; @@ -444,44 +449,61 @@ sub _recurse_parse { } # check if the current token is an unknown op-start - elsif (@$tokens and ($tokens->[0] eq '(' or $tokens->[0] =~ $common_single_args_re ) ) { + elsif (@$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 @lits = [ -LITERAL => [$token] ]; - while (@$tokens and $tokens->[0] !~ $all_std_keywords_re) { - push @lits, [ -LITERAL => [ shift @$tokens ] ]; - } + unshift @lits, pop @left if @left == 1; - if (@left == 1) { - unshift @lits, pop @left; - } + 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 ] ]; + } + } @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 ] ]; + # 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, $i+1]) + ]]; } else { - @left = [ ('-' . uc ($op)) => [ @left ] ]; + $i++; } } } + + return @left if $state == PARSE_RHS; + + # deal with post-fix operators + if (@$tokens) { + # asc/desc + if ($tokens->[0] =~ $asc_desc_re) { + @left = [ ('-' . uc (shift @$tokens)) => [ @left ] ]; + } + } } } @@ -590,16 +612,23 @@ sub _unparse { 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->pad_keyword($op, $depth)}; - return sprintf "$l%s%s%s$r", - $self->format_keyword($op), + + 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-- : ' ' , - $self->_unparse($args, $bindargs, $depth), - ; + $self->format_keyword($op), + (length $rhs ? $rhs : () ), + ); } } @@ -634,25 +663,38 @@ sub _parenthesis_unroll { next; } + my $parent_op = $ast->[0]; + # unroll nested parenthesis - while ( @{$child->[1]} == 1 and $child->[1][0][0] eq '-PAREN') { + while ( $parent_op ne 'IN' and @{$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 ) { + # 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]}; $changes++; } # if the parenthesis are wrapped around an AND/OR matching the parent AND/OR - open the parenthesis up and merge the list elsif ( - @{$child->[1]} == 1 - and - ( $ast->[0] eq 'AND' or $ast->[0] eq 'OR') - and - $child->[1][0][0] eq $ast->[0] + $single_child_op eq $parent_op + and + ( $parent_op eq 'AND' or $parent_op eq 'OR') ) { push @children, @{$child->[1][0][1]}; $changes++; @@ -661,13 +703,9 @@ sub _parenthesis_unroll { # only *ONE* LITERAL or placeholder element # as an AND/OR/NOT argument elsif ( - @{$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' - ) + ( $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]}; $changes++; @@ -680,20 +718,18 @@ sub _parenthesis_unroll { # break precedence) or when the child is BETWEEN (special # case) elsif ( - @{$child->[1]} == 1 - and - ($ast->[0] eq 'AND' or $ast->[0] eq 'OR') + ($parent_op eq 'AND' or $parent_op eq 'OR') and - $child->[1][0][0] =~ $binary_op_re + $single_child_op =~ $binary_op_re and - $child->[1][0][0] ne 'BETWEEN' + $single_child_op ne 'BETWEEN' and - @{$child->[1][0][1]} == 2 + $child_op_argc == 2 and ! ( - $child->[1][0][0] =~ $math_op_re + $single_child_op =~ $alphanum_cmp_op_re and - $ast->[0] =~ $math_op_re + $parent_op =~ $alphanum_cmp_op_re ) ) { push @children, @{$child->[1]}; @@ -707,26 +743,44 @@ sub _parenthesis_unroll { # 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 + $single_child_op and - @{$child->[1][0][1]} == 1 + $parent_op =~ $alphanum_cmp_op_re and - $ast->[0] =~ $math_op_re + $single_child_op !~ $alphanum_cmp_op_re and - $child->[1][0][0] !~ $math_op_re + $child_op_argc == 1 and ( - $child->[1][0][1][0][0] eq '-PAREN' + $single_grandchild_op eq '-PAREN' or - $child->[1][0][1][0][0] eq '-LITERAL' + $single_grandchild_op eq '-LITERAL' or - $child->[1][0][1][0][0] eq '-PLACEHOLDER' + $single_grandchild_op eq '-PLACEHOLDER' ) ) { push @children, @{$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]}; + $changes++; + } + # otherwise no more mucking for this pass else { @@ -739,6 +793,30 @@ sub _parenthesis_unroll { } while ($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; @@ -814,7 +892,7 @@ 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.