$case_sensitive $sql_differ/;
our $case_sensitive = 0;
+our $parenthesis_significant = 0;
our $sql_differ; # keeps track of differing portion between SQLs
our $tb = __PACKAGE__->builder;
# 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 $tokenizer_re_str = join("\n\t|\n",
- ( map { '\b' . $_ . '\b' } @expression_terminator_sql_keywords, 'AND', 'OR' ),
+ ( map { '\b' . $_ . '\b' } @expression_terminator_sql_keywords, 'AND', 'OR', 'NOT'),
( map { q! (?<= [\w\s\`\'\)] ) ! . $_ . q! (?= [\w\s\`\'\(] ) ! } @binary_op_keywords ),
);
# both are an op-list combo
else {
- for my $ast ($left, $right) {
-
- next unless (ref $ast->[1]);
-
- # unroll parenthesis in an elaborate loop
- my $changes;
- do {
-
- my @children;
- $changes = 0;
-
- for my $child (@{$ast->[1]}) {
- if (not ref $child or not $child->[0] eq 'PAREN') {
- push @children, $child;
- next;
- }
-
- # unroll nested parenthesis
- while ($child->[1][0][0] eq 'PAREN') {
- $child = $child->[1][0];
- $changes++;
- }
-
- # if the parenthesis are wrapped around an AND/OR matching the parent AND/OR - open the parenthesis up and merge the list
- if (
- ( $ast->[0] eq 'AND' or $ast->[0] eq 'OR')
- and
- $child->[1][0][0] eq $ast->[0]
- ) {
- push @children, @{$child->[1][0][1]};
- $changes++;
- }
-
- # if the parent operator explcitly allows it nuke the parenthesis
- elsif ( grep { $ast->[0] =~ /^ $_ $/xi } @unrollable_ops ) {
- push @children, $child->[1][0];
- $changes++;
- }
-
- # only one element in the parenthesis which is a binary op with two EXPR sub-children
- elsif (
- @{$child->[1]} == 1
- and
- grep { $child->[1][0][0] =~ /^ $_ $/xi } (@binary_op_keywords)
- and
- $child->[1][0][1][0][0] eq 'EXPR'
- and
- $child->[1][0][1][1][0] eq 'EXPR'
- ) {
- push @children, $child->[1][0];
- $changes++;
- }
-
- # otherwise no more mucking for this pass
- else {
- push @children, $child;
- }
- }
-
- $ast->[1] = \@children;
- } while ($changes);
- }
+ # unroll parenthesis if possible/allowed
+ _parenthesis_unroll ($_) for ($left, $right);
# if operators are different
if ($left->[0] ne $right->[0]) {
}
}
-
sub parse {
my $s = shift;
or
($state == PARSE_IN_EXPR && grep { $lookahead =~ /^ $_ $/xi } ('\)', @expression_terminator_sql_keywords ) )
or
- ($state == PARSE_RHS && grep { $lookahead =~ /^ $_ $/xi } ('\)', @expression_terminator_sql_keywords, @binary_op_keywords, 'AND', 'OR' ) )
+ ($state == PARSE_RHS && grep { $lookahead =~ /^ $_ $/xi } ('\)', @expression_terminator_sql_keywords, @binary_op_keywords, 'AND', 'OR', 'NOT' ) )
) {
return $left;
}
# nested expression in ()
if ($token eq '(') {
my $right = _recurse_parse($tokens, PARSE_IN_PARENS);
- $token = shift @$tokens or croak "missing ')'";
- $token eq ')' or croak "unexpected token : $token";
+ $token = shift @$tokens or croak "missing closing ')' around block " . unparse ($right);
+ $token eq ')' or croak "unexpected token '$token' terminating block " . unparse ($right);
$left = $left ? [@$left, [PAREN => [$right] ]]
: [PAREN => [$right] ];
}
$left = $left ? [@$left, [$op => [$right] ]]
: [[ $op => [$right] ]];
}
+ # NOT (last as to allow all other NOT X pieces first)
+ elsif ( $token =~ /^ not $/ix ) {
+ my $op = uc $token;
+ my $right = _recurse_parse ($tokens, PARSE_RHS);
+ $left = $left ? [ @$left, [$op => [$right] ]]
+ : [[ $op => [$right] ]];
+
+ }
# leaf expression
else {
$left = $left ? [@$left, [EXPR => [$token] ] ]
}
}
+sub _parenthesis_unroll {
+ my $ast = shift;
+
+ return if $parenthesis_significant;
+ return unless (ref $ast and ref $ast->[1]);
+
+ my $changes;
+ do {
+ my @children;
+ $changes = 0;
+
+ for my $child (@{$ast->[1]}) {
+ if (not ref $child or not $child->[0] eq 'PAREN') {
+ push @children, $child;
+ next;
+ }
+
+ # unroll nested parenthesis
+ while ($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 ( grep { $ast->[0] =~ /^ $_ $/xi } @unrollable_ops ) {
+ push @children, $child->[1][0];
+ $changes++;
+ }
+
+ # only one EXPR element in the parenthesis
+ elsif (
+ @{$child->[1]} == 1 && $child->[1][0][0] eq 'EXPR'
+ ) {
+ push @children, $child->[1][0];
+ $changes++;
+ }
+
+ # only one element in the parenthesis which is a binary op with two EXPR sub-children
+ elsif (
+ @{$child->[1]} == 1
+ and
+ grep { $child->[1][0][0] =~ /^ $_ $/xi } (@binary_op_keywords)
+ and
+ $child->[1][0][1][0][0] eq 'EXPR'
+ and
+ $child->[1][0][1][1][0] eq 'EXPR'
+ ) {
+ push @children, $child->[1][0];
+ $changes++;
+ }
+
+ # otherwise no more mucking for this pass
+ else {
+ push @children, $child;
+ }
+ }
+
+ $ast->[1] = \@children;
+
+ } while ($changes);
+}
sub unparse {
my $tree = shift;
is_same_sql_bind is_same_sql is_same_bind
eq_sql_bind eq_sql eq_bind
/];
-
+
my ($sql, @bind) = SQL::Abstract->new->select(%args);
is_same_sql_bind($given_sql, \@given_bind,
Therefore the tests will pass as long as the semantics
is preserved, even if the surface syntax has changed.
-B<Disclaimer> : this is only a half-cooked semantic equivalence;
-parsing is simple-minded, and comparison of SQL abstract syntax trees
-ignores commutativity or associativity of AND/OR operators, Morgan
-laws, etc.
+B<Disclaimer> : the semantic equivalence handling is pretty limited.
+A lot of effort goes into distinguishing significant from
+non-significant parenthesis, including AND/OR operator associativity.
+Currently this module does not support commutativity and more
+intelligent transformations like Morgan laws, etc.
+
+For a good overview of what this test framework is capable of refer
+to C<t/10test.t>
=head1 FUNCTIONS
If true, SQL comparisons will be case-sensitive. Default is false;
+=head2 $parenthesis_significant
+
+If true, SQL comparison will preserve and report difference in nested
+parenthesis. Useful for testing the C<-nest> modifier. Defaults to false;
+
=head2 $sql_differ
When L</eq_sql> returns false, the global variable
Norbert Buchmuller <norbi@nix.hu>
+Peter Rabbitson <ribasushi@cpan.org>
+
=head1 COPYRIGHT AND LICENSE
Copyright 2008 by Laurent Dami.