package SQL::Translator::Diff; ## SQLT schema diffing code use strict; use warnings; use Data::Dumper; use Carp::Clan qw/^SQL::Translator/; use SQL::Translator::Schema::Constants; use Sub::Quote qw(quote_sub); use Moo; has ignore_index_names => (is => 'rw',); has ignore_constraint_names => (is => 'rw',); has ignore_view_sql => (is => 'rw',); has ignore_proc_sql => (is => 'rw',); has output_db => (is => 'rw',); has source_schema => (is => 'rw',); has target_schema => (is => 'rw',); has case_insensitive => (is => 'rw',); has no_batch_alters => (is => 'rw',); has ignore_missing_methods => (is => 'rw',); has sqlt_args => ( is => 'rw', lazy => 1, default => quote_sub '{}', ); has tables_to_drop => ( is => 'rw', lazy => 1, default => quote_sub '[]', ); has tables_to_create => ( is => 'rw', lazy => 1, default => quote_sub '[]', ); has table_diff_hash => ( is => 'rw', lazy => 1, default => quote_sub '{}', ); my @diff_arrays = qw/ tables_to_drop tables_to_create /; my @diff_hash_keys = qw/ constraints_to_create constraints_to_drop indexes_to_create indexes_to_drop fields_to_create fields_to_alter fields_to_rename fields_to_drop table_options table_renamed_from /; sub schema_diff { # use Data::Dumper; ## we are getting instructions on how to turn the source into the target ## source == original, target == new (hmm, if I need to comment this, should I rename the vars again ??) ## _schema isa SQL::Translator::Schema ## _db is the name of the producer/db it came out of/into ## results are formatted to the source preferences my ($source_schema, $source_db, $target_schema, $output_db, $options) = @_; $options ||= {}; my $obj = SQL::Translator::Diff->new({ %$options, source_schema => $source_schema, target_schema => $target_schema, output_db => $output_db }); $obj->compute_differences->produce_diff_sql; } my $warned_dep; sub BUILD { my ($self, $args) = @_; for my $deprecated (qw/producer_options producer_args/) { if ($args->{$deprecated}) { carp "$deprecated is deprecated -- it does not go straight to the producer, it goes to the internal sqlt object. Please use sqlt_args, which reflects how it's used" unless $warned_dep++; $self->sqlt_args({ %{ $args->{$deprecated} }, %{ $self->sqlt_args } }); } } if (!$self->output_db) { $self->output_db($args->{source_db}); } } sub compute_differences { my ($self) = @_; my $target_schema = $self->target_schema; my $source_schema = $self->source_schema; my $producer_class = "SQL::Translator::Producer::@{[$self->output_db]}"; eval "require $producer_class"; die $@ if $@; if (my $preprocess = $producer_class->can('preprocess_schema')) { $preprocess->($source_schema); $preprocess->($target_schema); } my %src_tables_checked = (); my @tar_tables = sort { $a->name cmp $b->name } $target_schema->get_tables; ## do original/source tables exist in target? for my $tar_table (@tar_tables) { my $tar_table_name = $tar_table->name; my $src_table; $self->table_diff_hash->{$tar_table_name} = { map { $_ => [] } @diff_hash_keys }; if (my $old_name = $tar_table->extra('renamed_from')) { $src_table = $source_schema->get_table($old_name, $self->case_insensitive); if ($src_table) { $self->table_diff_hash->{$tar_table_name}{table_renamed_from} = [ [ $src_table, $tar_table ] ]; } else { delete $tar_table->extra->{renamed_from}; carp qq#Renamed table can't find old table "$old_name" for renamed table\n#; } } else { $src_table = $source_schema->get_table($tar_table_name, $self->case_insensitive); } unless ($src_table) { ## table is new ## add table(s) later. push @{ $self->tables_to_create }, $tar_table; next; } my $src_table_name = $src_table->name; $src_table_name = lc $src_table_name if $self->case_insensitive; $src_tables_checked{$src_table_name} = 1; $self->diff_table_options($src_table, $tar_table); ## Compare fields, their types, defaults, sizes etc etc $self->diff_table_fields($src_table, $tar_table); $self->diff_table_indexes($src_table, $tar_table); $self->diff_table_constraints($src_table, $tar_table); } # end of target_schema->get_tables loop for my $src_table ($source_schema->get_tables) { my $src_table_name = $src_table->name; $src_table_name = lc $src_table_name if $self->case_insensitive; push @{ $self->tables_to_drop }, $src_table unless $src_tables_checked{$src_table_name}; } return $self; } sub produce_diff_sql { my ($self) = @_; my $target_schema = $self->target_schema; my $source_schema = $self->source_schema; my $tar_name = $target_schema->name; my $src_name = $source_schema->name; my $producer_class = "SQL::Translator::Producer::@{[$self->output_db]}"; eval "require $producer_class"; die $@ if $@; # Map of name we store under => producer method name my %func_map = ( constraints_to_create => 'alter_create_constraint', constraints_to_drop => 'alter_drop_constraint', indexes_to_create => 'alter_create_index', indexes_to_drop => 'alter_drop_index', fields_to_create => 'add_field', fields_to_alter => 'alter_field', fields_to_rename => 'rename_field', fields_to_drop => 'drop_field', table_options => 'alter_table', table_renamed_from => 'rename_table', ); my @diffs; if (!$self->no_batch_alters && (my $batch_alter = $producer_class->can('batch_alter_table'))) { # Good - Producer supports batch altering of tables. foreach my $table (sort keys %{ $self->table_diff_hash }) { my $tar_table = $target_schema->get_table($table) || $source_schema->get_table($table); push @diffs, $batch_alter->( $tar_table, { map { $func_map{$_} => $self->table_diff_hash->{$table}{$_} } keys %func_map }, $self->sqlt_args ); } } else { # If we have any table renames we need to do those first; my %flattened_diffs; foreach my $table (sort keys %{ $self->table_diff_hash }) { my $table_diff = $self->table_diff_hash->{$table}; for (@diff_hash_keys) { push(@{ $flattened_diffs{ $func_map{$_} } ||= [] }, @{ $table_diff->{$_} }); } } push @diffs, map({ if (@{ $flattened_diffs{$_} || [] }) { my $meth = $producer_class->can($_); $meth ? map { map { $_ ? "$_" : () } $meth->((ref $_ eq 'ARRAY' ? @$_ : $_), $self->sqlt_args); } @{ $flattened_diffs{$_} } : $self->ignore_missing_methods ? "-- $producer_class cant $_" : die "$producer_class cant $_"; } else { () } } qw/rename_table alter_drop_constraint alter_drop_index drop_field add_field alter_field rename_field alter_create_index alter_create_constraint alter_table/), ; } if (my @tables = @{ $self->tables_to_create }) { my $translator = SQL::Translator->new( producer_type => $self->output_db, add_drop_table => 0, no_comments => 1, # TODO: sort out options %{ $self->sqlt_args } ); $translator->producer_args->{no_transaction} = 1; my $schema = $translator->schema; $schema->add_table($_) for @tables; unshift @diffs, # Remove begin/commit here, since we wrap everything in one. grep { $_ !~ /^(?:COMMIT|START(?: TRANSACTION)?|BEGIN(?: TRANSACTION)?)/ } $producer_class->can('produce')->($translator); } if (my @tables_to_drop = @{ $self->{tables_to_drop} || [] }) { my $meth = $producer_class->can('drop_table'); push @diffs, $meth ? (map { $meth->($_, $self->sqlt_args) } @tables_to_drop) : $self->ignore_missing_methods ? "-- $producer_class cant drop_table" : die "$producer_class cant drop_table"; } if (@diffs) { unshift @diffs, "BEGIN"; push @diffs, "\nCOMMIT"; } else { @diffs = ("-- No differences found"); } if (@diffs) { if ($self->output_db !~ /^(?:MySQL|SQLite|PostgreSQL)$/) { unshift(@diffs, "-- Output database @{[$self->output_db]} is untested/unsupported!!!"); } my @return = map { $_ ? ($_ =~ /;\s*\z/xms ? $_ : "$_;\n\n") : "\n" } ("-- Convert schema '$src_name' to '$tar_name':", @diffs); return wantarray ? @return : join('', @return); } return undef; } sub diff_table_indexes { my ($self, $src_table, $tar_table) = @_; my (%checked_indices); INDEX_CREATE: for my $i_tar ($tar_table->get_indices) { for my $i_src ($src_table->get_indices) { if ($i_tar->equals($i_src, $self->case_insensitive, $self->ignore_index_names)) { $checked_indices{$i_src} = 1; next INDEX_CREATE; } } push @{ $self->table_diff_hash->{$tar_table}{indexes_to_create} }, $i_tar; } INDEX_DROP: for my $i_src ($src_table->get_indices) { next if !$self->ignore_index_names && $checked_indices{$i_src}; for my $i_tar ($tar_table->get_indices) { next INDEX_DROP if $i_src->equals($i_tar, $self->case_insensitive, $self->ignore_index_names); } push @{ $self->table_diff_hash->{$tar_table}{indexes_to_drop} }, $i_src; } } sub diff_table_constraints { my ($self, $src_table, $tar_table) = @_; my (%checked_constraints); CONSTRAINT_CREATE: for my $c_tar ($tar_table->get_constraints) { for my $c_src ($src_table->get_constraints) { # This is a bit of a hack - needed for renaming tables to work local $c_src->{table} = $tar_table; if ($c_tar->equals($c_src, $self->case_insensitive, $self->ignore_constraint_names)) { $checked_constraints{$c_src} = 1; next CONSTRAINT_CREATE; } } push @{ $self->table_diff_hash->{$tar_table}{constraints_to_create} }, $c_tar; } CONSTRAINT_DROP: for my $c_src ($src_table->get_constraints) { # This is a bit of a hack - needed for renaming tables to work local $c_src->{table} = $tar_table; next if !$self->ignore_constraint_names && $checked_constraints{$c_src}; for my $c_tar ($tar_table->get_constraints) { next CONSTRAINT_DROP if $c_src->equals($c_tar, $self->case_insensitive, $self->ignore_constraint_names); } push @{ $self->table_diff_hash->{$tar_table}{constraints_to_drop} }, $c_src; } } sub diff_table_fields { my ($self, $src_table, $tar_table) = @_; # List of ones we've renamed from so we don't drop them my %renamed_source_fields; for my $tar_table_field ($tar_table->get_fields) { my $f_tar_name = $tar_table_field->name; if (my $old_name = $tar_table_field->extra->{renamed_from}) { my $src_table_field = $src_table->get_field($old_name, $self->case_insensitive); unless ($src_table_field) { carp qq#Renamed column can't find old column "@{[$src_table->name]}.$old_name" for renamed column\n#; delete $tar_table_field->extra->{renamed_from}; } else { push @{ $self->table_diff_hash->{$tar_table}{fields_to_rename} }, [ $src_table_field, $tar_table_field ]; $renamed_source_fields{$old_name} = 1; next; } } my $src_table_field = $src_table->get_field($f_tar_name, $self->case_insensitive); unless ($src_table_field) { push @{ $self->table_diff_hash->{$tar_table}{fields_to_create} }, $tar_table_field; next; } # field exists, something changed. This is a bit complex. Parsers can # normalize types, but only some of them do, so compare the normalized and # parsed types for each field to each other if ( !$tar_table_field->equals($src_table_field, $self->case_insensitive) && !$tar_table_field->equals($src_table_field->parsed_field, $self->case_insensitive) && !$tar_table_field->parsed_field->equals($src_table_field, $self->case_insensitive) && !$tar_table_field->parsed_field->equals($src_table_field->parsed_field, $self->case_insensitive)) { # Some producers might need src field to diff against push @{ $self->table_diff_hash->{$tar_table}{fields_to_alter} }, [ $src_table_field, $tar_table_field ]; next; } } # Now check to see if any fields from src_table need to be dropped for my $src_table_field ($src_table->get_fields) { my $f_src_name = $src_table_field->name; next if $renamed_source_fields{$f_src_name}; my $tar_table_field = $tar_table->get_field($f_src_name, $self->case_insensitive); unless ($tar_table_field) { push @{ $self->table_diff_hash->{$tar_table}{fields_to_drop} }, $src_table_field; next; } } } sub diff_table_options { my ($self, $src_table, $tar_table) = @_; my $cmp = sub { my ($a_name, undef, $b_name, undef) = (%$a, %$b); $a_name cmp $b_name; }; # Need to sort the options so we don't get spurious diffs. my (@src_opts, @tar_opts); @src_opts = sort $cmp $src_table->options; @tar_opts = sort $cmp $tar_table->options; # If there's a difference, just re-set all the options push @{ $self->table_diff_hash->{$tar_table}{table_options} }, $tar_table unless $src_table->_compare_objects(\@src_opts, \@tar_opts); } # support producer_options as an alias for sqlt_args for legacy code. sub producer_options { my $self = shift; return $self->sqlt_args(@_); } # support producer_args as an alias for sqlt_args for legacy code. sub producer_args { my $self = shift; return $self->sqlt_args(@_); } 1; __END__ =head1 NAME SQL::Translator::Diff - determine differences between two schemas =head1 DESCRIPTION Takes two input SQL::Translator::Schemas (or SQL files) and produces ALTER statements to make them the same =head1 SNYOPSIS Simplest usage: use SQL::Translator::Diff; my $sql = SQL::Translator::Diff::schema_diff($source_schema, 'MySQL', $target_schema, 'MySQL', $options_hash) OO usage: use SQL::Translator::Diff; my $diff = SQL::Translator::Diff->new({ output_db => 'MySQL', source_schema => $source_schema, target_schema => $target_schema, %$options_hash, })->compute_differences->produce_diff_sql; =head1 OPTIONS =over =item B Match indexes based on types and fields, ignoring name. =item B Match constrains based on types, fields and tables, ignoring name. =item B Which producer to use to produce the output. =item B Ignore case of table, field, index and constraint names when comparing =item B Produce each alter as a distinct C statement even if the producer supports the ability to do all alters for a table as one statement. =item B If the diff would need a method that is missing from the producer, just emit a comment showing the method is missing, rather than dieing with an error =item B Hash of extra arguments passed to L and the below L. =back =head1 PRODUCER FUNCTIONS The following producer functions should be implemented for completeness. If any of them are needed for a given diff, but not found, an error will be thrown. =over =item * C =item * C =item * C =item * C =item * C =item * C =item * C =item * C =item * C =item * C =item * C (optional) =item * C (optional) If the producer supports C, it will be called with the table to alter and a hash, the keys of which will be the method names listed above; values will be arrays of fields or constraints to operate on. In the case of the field functions that take two arguments this will appear as an array reference. I.e. the hash might look something like the following: { alter_create_constraint => [ $constraint1, $constraint2 ], add_field => [ $field ], alter_field => [ [$old_field, $new_field] ] } =item * C (optional) C is called by the Diff code to allow the producer to normalize any data it needs to first. For example, the MySQL producer uses this method to ensure that FK constraint names are unique. Basicaly any changes that need to be made to produce the SQL file for the schema should be done here, so that a diff between a parsed SQL file and (say) a parsed DBIx::Class::Schema object will be sane. (As an aside, DBIx::Class, for instance, uses the presence of a C function on the producer to know that it can diff between the previous SQL file and its own internal representation. Without this method on th producer it will diff the two SQL files which is slower, but known to work better on old-style producers.) =back =head1 AUTHOR Original Author(s) unknown. Refactor/re-write and more comprehensive tests by Ash Berlin C<< ash@cpan.org >>. Redevelopment sponsored by Takkle Inc. =cut