CONTENTS

NAME

DBIx::Class::Schema::Loader::Base - Base DBIx::Class::Schema::Loader Implementation.

SYNOPSIS

See DBIx::Class::Schema::Loader.

DESCRIPTION

This is the base class for the storage-specific DBIx::Class::Schema::* classes, and implements the common functionality between them.

CONSTRUCTOR OPTIONS

These constructor options are the base options for "loader_options" in DBIx::Class::Schema::Loader. Available constructor options are:

skip_relationships

Skip setting up relationships. The default is to attempt the loading of relationships.

skip_load_external

Skip loading of other classes in @INC. The default is to merge all other classes with the same name found in @INC into the schema file we are creating.

naming

Static schemas (ones dumped to disk) will, by default, use the new-style relationship names and singularized Results, unless you're overwriting an existing dump made by an older version of DBIx::Class::Schema::Loader, in which case the backward compatible RelBuilder will be activated, and the appropriate monikerization used.

Specifying

naming => 'current'

will disable the backward-compatible RelBuilder and use the new-style relationship names along with singularized Results, even when overwriting a dump made with an earlier version.

The option also takes a hashref:

naming => {
    relationships    => 'v8',
    monikers         => 'v8',
    column_accessors => 'v8',
    force_ascii      => 1,
}

or

naming => { ALL => 'v8', force_ascii => 1 }

The keys are:

ALL

Set "relationships", "monikers" and "column_accessors" to the specified value.

relationships

How to name relationship accessors.

monikers

How to name Result classes.

column_accessors

How to name column accessors in Result classes.

force_ascii

For "v8" mode and later, uses String::ToIdentifier::EN instead of String::ToIdentifier::EN::Unicode to force monikers and other identifiers to ASCII.

The values can be:

current

Latest style, whatever that happens to be.

v4

Unsingularlized monikers, has_many only relationships with no _id stripping.

v5

Monikers singularized as whole words, might_have relationships for FKs on UNIQUE constraints, _id stripping for belongs_to relationships.

Some of the _id stripping edge cases in 0.05003 have been reverted for the v5 RelBuilder.

v6

All monikers and relationships are inflected using Lingua::EN::Inflect::Phrase, and there is more aggressive _id stripping from relationship names.

In general, there is very little difference between v5 and v6 schemas.

v7

This mode is identical to v6 mode, except that monikerization of CamelCase table names is also done better (but best in v8.)

CamelCase column names in case-preserving mode will also be handled better for relationship name inflection (but best in v8.) See "preserve_case".

In this mode, CamelCase "column_accessors" are normalized based on case transition instead of just being lowercased, so FooId becomes foo_id.

v8

(EXPERIMENTAL)

The default mode is "v7", to get "v8" mode, you have to specify it in "naming" explicitly until 0.08 comes out.

"monikers" and "column_accessors" are created using String::ToIdentifier::EN::Unicode or String::ToIdentifier::EN if "force_ascii" is set; this is only significant for names with non-\w characters such as ..

CamelCase identifiers with words in all caps, e.g. VLANValidID are supported correctly in this mode.

For relationships, belongs_to accessors are made from column names by stripping postfixes other than _id as well, for example just Id, _?ref, _?cd, _?code and _?num, case insensitively.

preserve

For "monikers", this option does not inflect the table names but makes monikers based on the actual name. For "column_accessors" this option does not normalize CamelCase column names to lowercase column accessors, but makes accessors that are the same names as the columns (with any non-\w chars replaced with underscores.)

singular

For "monikers", singularizes the names using the most current inflector. This is the same as setting the option to "current".

plural

For "monikers", pluralizes the names, using the most current inflector.

Dynamic schemas will always default to the 0.04XXX relationship names and won't singularize Results for backward compatibility, to activate the new RelBuilder and singularization put this in your Schema.pm file:

__PACKAGE__->naming('current');

Or if you prefer to use 0.07XXX features but insure that nothing breaks in the next major version upgrade:

__PACKAGE__->naming('v7');

quiet

If true, will not print the usual Dumping manual schema ... Schema dump completed. messages. Does not affect warnings (except for warnings related to "really_erase_my_files".)

dry_run

If true, don't actually write out the generated files. This can only be used with static schema generation.

generate_pod

By default POD will be generated for columns and relationships, using database metadata for the text if available and supported.

Comment metadata can be stored in two ways.

The first is that you can create two tables named table_comments and column_comments respectively. These tables must exist in the same database and schema as the tables they describe. They both need to have columns named table_name and comment_text. The second one needs to have a column named column_name. Then data stored in these tables will be used as a source of metadata about tables and comments.

(If you wish you can change the name of these tables with the parameters "table_comments_table" and "column_comments_table".)

As a fallback you can use built-in commenting mechanisms. Currently this is only supported for PostgreSQL, Oracle and MySQL. To create comments in PostgreSQL you add statements of the form COMMENT ON TABLE some_table IS '...', the same syntax is used in Oracle. To create comments in MySQL you add COMMENT '...' to the end of the column or table definition. Note that MySQL restricts the length of comments, and also does not handle complex Unicode characters properly.

Set this to 0 to turn off all POD generation.

pod_comment_mode

Controls where table comments appear in the generated POD. Smaller table comments are appended to the NAME section of the documentation, and larger ones are inserted into DESCRIPTION instead. You can force a DESCRIPTION section to be generated with the comment always, only use NAME, or choose the length threshold at which the comment is forced into the description.

name

Use NAME section only.

description

Force DESCRIPTION always.

auto

Use DESCRIPTION if length > "pod_comment_spillover_length", this is the default.

pod_comment_spillover_length

When pod_comment_mode is set to auto, this is the length of the comment at which it will be forced into a separate description section.

The default is 60

table_comments_table

The table to look for comments about tables in. By default table_comments. See "generate_pod" for details.

This must not be a fully qualified name, the table will be looked for in the same database and schema as the table whose comment is being retrieved.

column_comments_table

The table to look for comments about columns in. By default column_comments. See "generate_pod" for details.

This must not be a fully qualified name, the table will be looked for in the same database and schema as the table/column whose comment is being retrieved.

relationship_attrs

Hashref of attributes to pass to each generated relationship, listed by type. Also supports relationship type 'all', containing options to pass to all generated relationships. Attributes set for more specific relationship types override those set in 'all', and any attributes specified by this option override the introspected attributes of the foreign key if any.

For example:

relationship_attrs => {
    has_many   => { cascade_delete => 1, cascade_copy => 1 },
    might_have => { cascade_delete => 1, cascade_copy => 1 },
},

use this to turn DBIx::Class cascades to on on your has_many and might_have relationships, they default to off.

Can also be a coderef, for more precise control, in which case the coderef gets this hash of parameters (as a list):

rel_name        # the name of the relationship
rel_type        # the type of the relationship: 'belongs_to', 'has_many' or 'might_have'
local_source    # the DBIx::Class::ResultSource object for the source the rel is *from*
remote_source   # the DBIx::Class::ResultSource object for the source the rel is *to*
local_table     # the DBIx::Class::Schema::Loader::Table object for the table of the source the rel is from
local_cols      # an arrayref of column names of columns used in the rel in the source it is from
remote_table    # the DBIx::Class::Schema::Loader::Table object for the table of the source the rel is to
remote_cols     # an arrayref of column names of columns used in the rel in the source it is to
attrs           # the attributes that would be set

it should return the new hashref of attributes, or nothing for no changes.

For example:

relationship_attrs => sub {
    my %p = @_;

    say "the relationship name is: $p{rel_name}";
    say "the relationship is a: $p{rel_type}";
    say "the local class is: ",  $p{local_source}->result_class;
    say "the remote class is: ", $p{remote_source}->result_class;
    say "the local table is: ", $p{local_table}->sql_name;
    say "the rel columns in the local table are: ", (join ", ", @{$p{local_cols}});
    say "the remote table is: ", $p{remote_table}->sql_name;
    say "the rel columns in the remote table are: ", (join ", ", @{$p{remote_cols}});

    if ($p{local_table} eq 'dogs' && @{$p{local_cols}} == 1 && $p{local_cols}[0] eq 'name') {
        $p{attrs}{could_be_snoopy} = 1;

        reutrn $p{attrs};
    }
},

These are the default attributes:

has_many => {
    cascade_delete => 0,
    cascade_copy   => 0,
},
might_have => {
    cascade_delete => 0,
    cascade_copy   => 0,
},
belongs_to => {
    on_delete => 'CASCADE',
    on_update => 'CASCADE',
    is_deferrable => 1,
},

For belongs_to relationships, these defaults are overridden by the attributes introspected from the foreign key in the database, if this information is available (and the driver is capable of retrieving it.)

This information overrides the defaults mentioned above, and is then itself overridden by the user's "relationship_attrs" for belongs_to if any are specified.

In general, for most databases, for a plain foreign key with no rules, the values for a belongs_to relationship will be:

on_delete     => 'NO ACTION',
on_update     => 'NO ACTION',
is_deferrable => 0,

In the cases where an attribute is not supported by the DB, a value matching the actual behavior is used, for example Oracle does not support ON UPDATE rules, so on_update is set to NO ACTION. This is done so that the behavior of the schema is preserved when cross deploying to a different RDBMS such as SQLite for testing.

In the cases where the DB does not support DEFERRABLE foreign keys, the value is set to 1 if DBIx::Class has a working $storage->with_deferred_fk_checks. This is done so that the same DBIx::Class code can be used, and cross deployed from and to such databases.

debug

If set to true, each constructive DBIx::Class statement the loader decides to execute will be warn-ed before execution.

db_schema

Set the name of the schema to load (schema in the sense that your database vendor means it).

Can be set to an arrayref of schema names for multiple schemas, or the special value % for all schemas.

For MSSQL, Sybase ASE, and Informix can be set to a hashref of databases as keys and arrays of owners as values, set to the value:

{ '%' => '%' }

for all owners in all databases.

Name clashes resulting from the same table name in different databases/schemas will be resolved automatically by prefixing the moniker with the database and/or schema.

To prefix/suffix all monikers with the database and/or schema, see "moniker_parts".

moniker_parts

The database table names are represented by the DBIx::Class::Schema::Loader::Table class in the loader, the DBIx::Class::Schema::Loader::Table::Sybase class for Sybase ASE and DBIx::Class::Schema::Loader::Table::Informix for Informix.

Monikers are created normally based on just the name property, corresponding to the table name, but can consist of other parts of the fully qualified name of the table.

The "moniker_parts" option is an arrayref of methods on the table class corresponding to parts of the fully qualified table name, defaulting to ['name'], in the order those parts are used to create the moniker name. The parts are joined together using "moniker_part_separator".

The 'name' entry must be present.

Below is a table of supported databases and possible "moniker_parts".

moniker_part_separator

String used to join "moniker_parts" when creating the moniker. Defaults to the empty string. Use :: to get a separate namespace per database and/or schema.

constraint

Only load matching tables.

These can be specified either as a regex (preferably on the qr// form), or as an arrayref of arrayrefs. Regexes are matched against the (unqualified) table name, while arrayrefs are matched according to "moniker_parts".

For example:

db_schema => [qw(some_schema other_schema)],
moniker_parts => [qw(schema name)],
constraint => [
    [ qr/\Asome_schema\z/ => qr/\A(?:foo|bar)\z/ ],
    [ qr/\Aother_schema\z/ => qr/\Abaz\z/ ],
],

In this case only the tables foo and bar in some_schema and baz in other_schema will be dumped.

exclude

Exclude matching tables.

The tables to exclude are specified in the same way as for the "constraint" option.

moniker_map

Overrides the default table name to moniker translation. Either

If the hash entry does not exist, or the function returns a false value, the code falls back to default behavior for that table name.

The default behavior is to split on case transition and non-alphanumeric boundaries, singularize the resulting phrase, then join the titlecased words together. Examples:

Table Name       | Moniker Name
---------------------------------
luser            | Luser
luser_group      | LuserGroup
luser-opts       | LuserOpt
stations_visited | StationVisited
routeChange      | RouteChange

moniker_part_map

Map for overriding the monikerization of individual "moniker_parts". The keys are the moniker part to override, the value is either a hashref or coderef for mapping the corresponding part of the moniker. If a coderef is used, it gets called with the moniker part and the hash key the code ref was found under.

For example:

moniker_part_map => {
    schema => sub { ... },
},

Given the table foo.bar, the code ref would be called with the arguments foo and schema, plus a coderef similar to the one described in "moniker_map".

"moniker_map" takes precedence over this.

col_accessor_map

Same as moniker_map, but for column accessor names. The nested hashref form is traversed according to "moniker_parts", with an extra level at the bottom for the column name. If a coderef is passed, the code is called with the following arguments:

rel_name_map

Similar in idea to moniker_map, but different in the details. It can be a hashref or a code ref.

If it is a hashref, keys can be either the default relationship name, or the moniker. The keys that are the default relationship name should map to the name you want to change the relationship to. Keys that are monikers should map to hashes mapping relationship names to their translation. You can do both at once, and the more specific moniker version will be picked up first. So, for instance, you could have

{
    bar => "baz",
    Foo => {
        bar => "blat",
    },
}

and relationships that would have been named bar will now be named baz except that in the table whose moniker is Foo it will be named blat.

If it is a coderef, it will be passed a hashref of this form:

{
    name           => default relationship name,
    type           => the relationship type eg: C<has_many>,
    local_class    => name of the DBIC class we are building,
    local_moniker  => moniker of the DBIC class we are building,
    local_columns  => columns in this table in the relationship,
    remote_class   => name of the DBIC class we are related to,
    remote_moniker => moniker of the DBIC class we are related to,
    remote_columns => columns in the other table in the relationship,
    # for type => "many_to_many" only:
    link_class     => name of the DBIC class for the link table,
    link_moniker   => moniker of the DBIC class for the link table,
    link_rel_name  => name of the relationship to the link table,
}

In addition it is passed a coderef that can be called with a hashref map.

DBICSL will try to use the value returned as the relationship name.

inflect_plural

Just like "moniker_map" above (can be hash/code-ref, falls back to default if hash key does not exist or coderef returns false), but acts as a map for pluralizing relationship names. The default behavior is to utilize "to_PL" in Lingua::EN::Inflect::Phrase.

inflect_singular

As "inflect_plural" above, but for singularizing relationship names. Default behavior is to utilize "to_S" in Lingua::EN::Inflect::Phrase.

schema_base_class

Base class for your schema classes. Defaults to 'DBIx::Class::Schema'.

schema_components

List of components to load into the Schema class.

result_base_class

Base class for your table classes (aka result classes). Defaults to 'DBIx::Class::Core'.

additional_base_classes

List of additional base classes all of your table classes will use.

left_base_classes

List of additional base classes all of your table classes will use that need to be leftmost.

additional_classes

List of additional classes which all of your table classes will use.

components

List of additional components to be loaded into all of your Result classes. A good example would be InflateColumn::DateTime

result_components_map

A hashref of moniker keys and component values. Unlike "components", which loads the given components into every Result class, this option allows you to load certain components for specified Result classes. For example:

result_components_map => {
    StationVisited => '+YourApp::Schema::Component::StationVisited',
    RouteChange    => [
                          '+YourApp::Schema::Component::RouteChange',
                          'InflateColumn::DateTime',
                      ],
}

You may use this in conjunction with "components".

result_roles

List of Moose roles to be applied to all of your Result classes.

result_roles_map

A hashref of moniker keys and role values. Unlike "result_roles", which applies the given roles to every Result class, this option allows you to apply certain roles for specified Result classes. For example:

result_roles_map => {
    StationVisited => [
                          'YourApp::Role::Building',
                          'YourApp::Role::Destination',
                      ],
    RouteChange    => 'YourApp::Role::TripEvent',
}

You may use this in conjunction with "result_roles".

use_namespaces

This is now the default, to go back to "load_classes" in DBIx::Class::Schema pass a 0.

Generate result class names suitable for "load_namespaces" in DBIx::Class::Schema and call that instead of "load_classes" in DBIx::Class::Schema. When using this option you can also specify any of the options for load_namespaces (i.e. result_namespace, resultset_namespace, default_resultset_class), and they will be added to the call (and the generated result class names adjusted appropriately).

dump_directory

The value of this option is a perl libdir pathname. Within that directory this module will create a baseline manual DBIx::Class::Schema module set, based on what it creates at runtime.

The created schema class will have the same classname as the one on which you are setting this option (and the ResultSource classes will be based on this name as well).

Normally you wouldn't hard-code this setting in your schema class, as it is meant for one-time manual usage.

See "dump_to_dir" in DBIx::Class::Schema::Loader for examples of the recommended way to access this functionality.

dump_overwrite

Deprecated. See "really_erase_my_files" below, which does *not* mean the same thing as the old dump_overwrite setting from previous releases.

really_erase_my_files

Default false. If true, Loader will unconditionally delete any existing files before creating the new ones from scratch when dumping a schema to disk.

The default behavior is instead to only replace the top portion of the file, up to and including the final stanza which contains # DO NOT MODIFY THE FIRST PART OF THIS FILE leaving any customizations you placed after that as they were.

When really_erase_my_files is not set, if the output file already exists, but the aforementioned final stanza is not found, or the checksum contained there does not match the generated contents, Loader will croak and not touch the file.

You should really be using version control on your schema classes (and all of the rest of your code for that matter). Don't blame me if a bug in this code wipes something out when it shouldn't have, you've been warned.

overwrite_modifications

Default false. If false, when updating existing files, Loader will refuse to modify any Loader-generated code that has been modified since its last run (as determined by the checksum Loader put in its comment lines).

If true, Loader will discard any manual modifications that have been made to Loader-generated code.

Again, you should be using version control on your schema classes. Be careful with this option.

omit_version

Omit the package version from the signature comment.

omit_timestamp

Omit the creation timestamp from the signature comment.

custom_column_info

Hook for adding extra attributes to the column_info for a column.

Must be a coderef that returns a hashref with the extra attributes.

Receives the DBIx::Class::Schema::Loader::Table object, column name and column_info.

For example:

custom_column_info => sub {
    my ($table, $column_name, $column_info) = @_;

    if ($column_name eq 'dog' && $column_info->{default_value} eq 'snoopy') {
        return { is_snoopy => 1 };
    }
},

This attribute can also be used to set inflate_datetime on a non-datetime column so it also receives the "datetime_timezone" and/or "datetime_locale".

datetime_timezone

Sets the timezone attribute for DBIx::Class::InflateColumn::DateTime for all columns with the DATE/DATETIME/TIMESTAMP data_types.

datetime_locale

Sets the locale attribute for DBIx::Class::InflateColumn::DateTime for all columns with the DATE/DATETIME/TIMESTAMP data_types.

datetime_undef_if_invalid

Pass a 0 for this option when using MySQL if you DON'T want datetime_undef_if_invalid => 1 in your column info for DATE, DATETIME and TIMESTAMP columns.

The default is recommended to deal with data such as 00/00/00 which sometimes ends up in such columns in MySQL.

config_file

File in Perl format, which should return a HASH reference, from which to read loader options.

preserve_case

Normally database names are lowercased and split by underscore, use this option if you have CamelCase database names.

Drivers for case sensitive databases like Sybase ASE or MSSQL with a case-sensitive collation will turn this option on unconditionally.

NOTE: "naming" = v8 is highly recommended with this option as the semantics of this mode are much improved for CamelCase database names.

"naming" = v7 or greater is required with this option.

qualify_objects

Set to true to prepend the "db_schema" to table names for __PACKAGE__->table calls, and to some other things like Oracle sequences.

This attribute is automatically set to true for multi db_schema configurations, unless explicitly set to false by the user.

use_moose

Creates Schema and Result classes that use Moose, MooseX::NonMoose and MooseX::MarkAsMethods (or namespace::autoclean, see below). The default content after the md5 sum also makes the classes immutable.

It is safe to upgrade your existing Schema to this option.

only_autoclean

By default, we use MooseX::MarkAsMethods to remove imported functions from your generated classes. It uses namespace::autoclean to do this, after telling your object's metaclass that any operator overloads in your class are methods, which will cause namespace::autoclean to spare them from removal.

This prevents the "Hey, where'd my overloads go?!" effect.

If you don't care about operator overloads (or if you know your Moose is at at least version 2.1400, where MooseX::MarkAsMethods is no longer necessary), enabling this option falls back to just using namespace::autoclean itself.

If none of the above made any sense, or you don't have some pressing need to only use namespace::autoclean, leaving this set to the default is just fine.

col_collision_map

This option controls how accessors for column names which collide with perl methods are named. See "COLUMN ACCESSOR COLLISIONS" for more information.

This option takes either a single sprintf format or a hashref of strings which are compiled to regular expressions that map to sprintf formats.

Examples:

col_collision_map => 'column_%s'

col_collision_map => { '(.*)' => 'column_%s' }

col_collision_map => { '(foo).*(bar)' => 'column_%s_%s' }

rel_collision_map

Works just like "col_collision_map", but for relationship names/accessors rather than column names/accessors.

The default is to just append _rel to the relationship name, see "RELATIONSHIP NAME COLLISIONS".

uniq_to_primary

Automatically promotes the largest unique constraints with non-nullable columns on tables to primary keys, assuming there is only one largest unique constraint.

allow_extra_m2m_cols

Generate many_to_many relationship bridges even if the link table has extra columns other than the foreign keys. The primary key must still equal the union of the foreign keys.

filter_generated_code

An optional hook that lets you filter the generated text for various classes through a function that change it in any way that you want. The function will receive the type of file, schema or result, class and code; and returns the new code to use instead. For instance you could add custom comments, or do anything else that you want.

The option can also be set to a string, which is then used as a filter program, e.g. perltidy.

If this exists but fails to return text matching /\bpackage\b/, no file will be generated.

filter_generated_code => sub {
    my ($type, $class, $text) = @_;
    ...
    return $new_code;
}

You can also use this option to set perltidy markers in your generated classes. This will leave the generated code in the default format, but will allow you to tidy your classes at any point in future, without worrying about changing the portions of the file which are checksummed, since perltidy will just ignore all text between the markers.

filter_generated_code => sub {
    return "#<<<\n$_[2]\n#>>>";
}

METHODS

None of these methods are intended for direct invocation by regular users of DBIx::Class::Schema::Loader. Some are proxied via DBIx::Class::Schema::Loader.

new

Constructor for DBIx::Class::Schema::Loader::Base, used internally by DBIx::Class::Schema::Loader.

load

Does the actual schema-construction work.

rescan

Arguments: schema

Rescan the database for changes. Returns a list of the newly added table monikers.

The schema argument should be the schema class or object to be affected. It should probably be derived from the original schema_class used during "load".

get_dump_filename

Arguments: class

Returns the full path to the file for a class that the class has been or will be dumped to. This is a file in a temp dir for a dynamic schema.

tables

Returns a sorted list of loaded tables, using the original database table names.

monikers

Returns a hashref of loaded table to moniker mappings. There will be two entries for each table, the original name and the "normalized" name, in the case that the two are different (such as databases that like uppercase table names, or preserve your original mixed-case definitions, or what-have-you).

classes

Returns a hashref of table to class mappings. In some cases it will contain multiple entries per table for the original and normalized table names, as above in "monikers".

generated_classes

Returns an arrayref of classes that were actually generated (i.e. not skipped because there were no changes).

NON-ENGLISH DATABASES

If you use the loader on a database with table and column names in a language other than English, you will want to turn off the English language specific heuristics.

To do so, use something like this in your loader options:

naming           => { monikers => 'v4' },
inflect_singular => sub { "$_[0]_rel" },
inflect_plural   => sub { "$_[0]_rel" },

COLUMN ACCESSOR COLLISIONS

Occasionally you may have a column name that collides with a perl method, such as can. In such cases, the default action is to set the accessor of the column spec to undef.

You can then name the accessor yourself by placing code such as the following below the md5:

__PACKAGE__->add_column('+can' => { accessor => 'my_can' });

Another option is to use the "col_collision_map" option.

RELATIONSHIP NAME COLLISIONS

In very rare cases, you may get a collision between a generated relationship name and a method in your Result class, for example if you have a foreign key called belongs_to.

This is a problem because relationship names are also relationship accessor methods in DBIx::Class.

The default behavior is to append _rel to the relationship name and print out a warning that refers to this text.

You can also control the renaming with the "rel_collision_map" option.

SEE ALSO

DBIx::Class::Schema::Loader, dbicdump

AUTHORS

See "AUTHORS" in DBIx::Class::Schema::Loader.

LICENSE

This library is free software; you can redistribute it and/or modify it under the same terms as Perl itself.