Bug 290402: Functions to support reading-in a Schema object from the database

Patch by Max Kanat-Alexander <mkanat@bugzilla.org> r=Tomas.Kopal, a=justdave

1 files changed, 168 insertions, 0 deletions

diff --git a/Bugzilla/DB/Schema/Mysql.pm b/Bugzilla/DB/Schema/Mysql.pm
index cc30246cc..1ea1d285a 100644
--- a/Bugzilla/DB/Schema/Mysql.pm
+++ b/Bugzilla/DB/Schema/Mysql.pm
@@ -33,6 +33,57 @@ use Bugzilla::Error;
 
 use base qw(Bugzilla::DB::Schema);
 
+# This is for column_info_to_column, to know when a tinyint is a 
+# boolean and when it's really a tinyint. This only has to be accurate
+# up to and through 2.19.3, because that's the only time we need
+# column_info_to_column.
+#
+# This is basically a hash of tables/columns, with one entry for each column
+# that should be interpreted as a BOOLEAN instead of as an INT1 when
+# reading in the Schema from the disk. The values are discarded; I just
+# used "1" for simplicity.
+use constant BOOLEAN_MAP => {
+    bugs           => {everconfirmed => 1, reporter_accessible => 1,
+                       cclist_accessible => 1, qacontact_accessible => 1,
+                       assignee_accessible => 1},
+    longdescs      => {isprivate => 1, already_wrapped => 1},
+    attachments    => {ispatch => 1, isobsolete => 1, isprivate => 1},
+    flags          => {is_active => 1},
+    flagtypes      => {is_active => 1, is_requestable => 1, 
+                       is_requesteeble => 1, is_multiplicable => 1},
+    fielddefs      => {mailhead => 1, obsolete => 1},
+    bug_status     => {isactive => 1},
+    resolution     => {isactive => 1},
+    bug_severity   => {isactive => 1},
+    priority       => {isactive => 1},
+    rep_platform   => {isactive => 1},
+    op_sys         => {isactive => 1},
+    profiles       => {mybugslink => 1, newemailtech => 1},
+    namedqueries   => {linkinfooter => 1, watchfordiffs => 1},
+    groups         => {isbuggroup => 1, isactive => 1},
+    group_control_map => {entry => 1, membercontrol => 1, othercontrol => 1,
+                          canedit => 1},
+    group_group_map => {isbless => 1},
+    user_group_map => {isbless => 1, isderived => 1},
+    products       => {disallownew => 1},
+    series         => {public => 1},
+    whine_queries  => {onemailperbug => 1},
+    quips          => {approved => 1},
+    setting        => {is_enabled => 1}
+};
+
+# Maps the db_specific hash backwards, for use in column_info_to_column.
+use constant REVERSE_MAPPING => {
+    # Boolean and the SERIAL fields are handled in column_info_to_column,
+    # and so don't have an entry here.
+    TINYINT   => 'INT1',
+    SMALLINT  => 'INT2',
+    MEDIUMINT => 'INT3',
+    INTEGER   => 'INT4',
+    # All the other types have the same name in their abstract version
+    # as in their db-specific version, so no reverse mapping is needed.
+};
+
 #------------------------------------------------------------------------------
 sub _initialize {
 
@@ -110,6 +161,123 @@ sub get_drop_index_ddl {
     return ("DROP INDEX \`$name\` ON $table");
 }
 
+# Converts a DBI column_info output to an abstract column definition.
+# Expects to only be called by Bugzila::DB::Mysql::_bz_build_schema_from_disk,
+# although there's a chance that it will also work properly if called
+# elsewhere.
+sub column_info_to_column {
+    my ($self, $column_info) = @_;
+
+    # Unfortunately, we have to break Schema's normal "no database"
+    # barrier a few times in this function.
+    my $dbh = Bugzilla->dbh;
+
+    my $table = $column_info->{TABLE_NAME};
+    my $col_name = $column_info->{COLUMN_NAME};
+
+    my $column = {};
+
+    ($column->{NOTNULL} = 1) if $column_info->{NULLABLE} == 0;
+
+    if ($column_info->{mysql_is_pri_key}) {
+        # In MySQL, if a table has no PK, but it has a UNIQUE index,
+        # that index will show up as the PK. So we have to eliminate
+        # that possibility.
+        # Unfortunately, the only way to definitely solve this is
+        # to break Schema's standard of not touching the live database
+        # and check if the index called PRIMARY is on that field.
+        my $pri_index = $dbh->bz_index_info_real($table, 'PRIMARY');
+        if ( $pri_index && grep($_ eq $col_name, @{$pri_index->{FIELDS}}) ) {
+            $column->{PRIMARYKEY} = 1;
+        }
+    }
+
+    # MySQL frequently defines a default for a field even when we
+    # didn't explicitly set one. So we have to have some special
+    # hacks to determine whether or not we should actually put
+    # a default in the abstract schema for this field.
+    if (defined $column_info->{COLUMN_DEF}) {
+        # The defaults that MySQL inputs automatically are usually
+        # something that would be considered "false" by perl, either
+        # a 0 or an empty string. (Except for ddatetime and decimal
+        # fields, which have their own special auto-defaults.)
+        #
+        # Here's how we handle this: If it exists in the schema
+        # without a default, then we don't use the default. If it
+        # doesn't exist in the schema, then we're either going to
+        # be dropping it soon, or it's a custom end-user column, in which
+        # case having a bogus default won't harm anything.
+        my $schema_column = $self->get_column($table, $col_name);
+        unless ( (!$column_info->{COLUMN_DEF} 
+                  || $column_info->{COLUMN_DEF} eq '0000-00-00 00:00:00'
+                  || $column_info->{COLUMN_DEF} eq '0.00')
+                && $schema_column 
+                && !exists $schema_column->{DEFAULT}) {
+            
+            my $default = $column_info->{COLUMN_DEF};
+            # Schema uses '0' for the defaults for decimal fields. 
+            $default = 0 if $default =~ /^0\.0+$/;
+            # If we're not a number, we're a string and need to be
+            # quoted.
+            $default = $dbh->quote($default) if !($default =~ /^(-)?(\d+)(.\d+)?$/);
+            $column->{DEFAULT} = $default;
+        }
+    }
+
+    my $type = $column_info->{TYPE_NAME};
+
+    # Certain types of columns need the size/precision appended.
+    if ($type =~ /CHAR$/ || $type eq 'DECIMAL') {
+        # This is nicely lowercase and has the size/precision appended.
+        $type = $column_info->{mysql_type_name};
+    }
+
+    # If we're a tinyint, we could be either a BOOLEAN or an INT1.
+    # Only the BOOLEAN_MAP knows the difference.
+    elsif ($type eq 'TINYINT' && exists BOOLEAN_MAP->{$table}
+           && exists BOOLEAN_MAP->{$table}->{$col_name}) {
+        $type = 'BOOLEAN';
+        if (exists $column->{DEFAULT}) {
+            $column->{DEFAULT} = $column->{DEFAULT} ? 'TRUE' : 'FALSE';
+        }
+    }
+
+    # We also need to check if we're an auto_increment field.
+    elsif ($type =~ /INT/) {
+        # Unfortunately, the only way to do this in DBI is to query the
+        # database, so we have to break the rule here that Schema normally
+        # doesn't touch the live DB.
+        my $ref_sth = $dbh->prepare(
+            "SELECT $col_name FROM $table LIMIT 1");
+        $ref_sth->execute;
+        if ($ref_sth->{mysql_is_auto_increment}->[0]) {
+            if ($type eq 'MEDIUMINT') {
+                $type = 'MEDIUMSERIAL';
+            }
+            elsif ($type eq 'SMALLINT') {
+                $type = 'SMALLSERIAL';
+            } 
+            else {
+                $type = 'INTSERIAL';
+            }
+        }
+        $ref_sth->finish;
+
+    }
+
+    # For all other db-specific types, check if they exist in 
+    # REVERSE_MAPPING and use the type found there.
+    if (exists REVERSE_MAPPING->{$type}) {
+        $type = REVERSE_MAPPING->{$type};
+    }
+
+    $column->{TYPE} = $type;
+
+    #print "$table.$col_name: " . Data::Dumper->Dump([$column]) . "\n";
+
+    return $column;
+}
+
 sub get_rename_column_ddl {
     my ($self, $table, $old_name, $new_name) = @_;
     my $def = $self->get_type_ddl($self->get_column($table, $old_name));