| Rose::DB::Object::ConventionManager(3pm) | User Contributed Perl Documentation | Rose::DB::Object::ConventionManager(3pm) |
Rose::DB::Object::ConventionManager - Provide missing metadata by convention.
package My::Product; use base 'Rose::DB::Object'; __PACKAGE__->meta->setup(columns => [ ... ]); # No table is set above, but look at this: the # convention manager provided one for us. print __PACKAGE__->meta->table; # "products" ## ## See the EXAMPLE section below for a more complete demonstration. ##
Each Rose::DB::Object-derived object has a convention manager that it uses to fill in missing metadata. The convention manager encapsulates a set of rules (conventions) for generating various pieces of metadata in the absence of explicitly specified values: table names, column names, etc.
Each Rose::DB::Object-derived class's convention manager object is stored in the convention_manager attribute of its Rose::DB::Object::Metadata (meta) object. Rose::DB::Object::ConventionManager is the default convention manager class.
The object method documentation below describes both the purpose of each convention manager method and the particular rules that Rose::DB::Object::ConventionManager follows to fulfill that purpose. Subclasses must honor the purpose of each method, but are free to use any rules they choose.
Note well: When reading the descriptions of the rules used by each convention manager method below, remember that only values that are missing will be set by the convention manager. Explicitly providing a value for a piece of metadata obviates the need for the convention manager to generate one.
If insufficient information is available, or if the convention manager simply declines to fulfill a request, undef may be returned from any metadata-generating method.
In the documentation, the adjectives "local" and "foreign" are used to distinguish between the things that belong to the convention manager's class and the class on "the other side" of the inter-table relationship, respectively.
Although the object method documentation below includes all the information required to understand the default conventions, it's also quite spread out. What follows is a summary of the default conventions. Some details have necessarily been omitted or simplified for the sake of brevity, but this summary should give you a good starting point for further exploration.
Here's a brief summary of the default conventions as implemented in Rose::DB::Object::ConventionManager.
(This convention can be overridden via the tables_are_singular method.)
The foreign key's class name is generated by calling related_table_to_class, passing NAME and the convention manager's class as arguments. An attempt is made is load the class. If this fails, the foreign key's class name is not set.
The foreign key's key_columns are only set if both the "local" and "foreign" tables have single-column primary keys. The foreign class's primary key column name is used as the foreign column in the key_columns map. If there is a local column with the same name as the foreign key name, and if that column is aliased (making way for the foreign key method to use that name), then that is used as the local column. If not, then the local column name is generated by joining the foreign key name and the foreign class's primary key column name with an underscore. If no column by that name exists, then the search is abandoned. Example:
Given these pieces:
Name Description Value
--------- -------------------------------- -------
NAME Foreign key name vendor
FCLASS Foreign class My::Vendor
FPK Foreign primary key column name id
Consider column maps in this order:
Value Formula
--------------------- ----------------------
{ vendor => 'id' } { NAME => FPK }
{ vendor_id => 'id' } { <NAME>_<FPK> => FPK }
If there is more than one pair of columns in KEY_COLUMNS, then the name is generated by calling plural_to_singular, passing the table name of the foreign class. The CURRENT_NAME is used if the call to plural_to_singular does not return a true value.
If there is just one pair of columns in KEY_COLUMNS, and if the name of the local column ends with an underscore and the name of the referenced column, then that part of the column name is removed and the remaining string is used as the foreign key name. For example, given the following tables:
CREATE TABLE categories
(
id SERIAL PRIMARY KEY,
...
);
CREATE TABLE products
(
category_id INT REFERENCES categories (id),
...
);
The foreign key name would be "category", which is the name of the referring column ("category_id") with an underscore and the name of the referenced column ("_id") removed from the end of it.
If the foreign key has only one column, but it does not meet the criteria described above, then the name is generated by calling plural_to_singular, passing the table name of the foreign class. The CURRENT_NAME is used if the call to plural_to_singular does not return a true value.
If the name selected using the above techniques is in the USED_NAMES hash, or is the same as that of an existing or potential method in the target class, then the suffixes "_obj" and "_object" are tried in that order. If neither of those suffixes resolves the situation, then ascending numeric suffixes starting with "1" are tried until a unique name is found.
If no table is specified then the table name is derived from the current class name by calling class_to_table_plural.
If tables_are_singular is true, then TABLE is passed to the singular_to_plural method and the result is returned. Otherwise, TABLE is returned as-is.
The default implementation passes the name of the table pointed to by FK through the singular_to_plural method in order to build the name.
If the selected name is the name of an existing or potential method in the target class, then the suffixes "_objs" and "_objects" are tried in that order. If neither of those suffixes resolves the situation, then ascending numeric suffixes starting with "1" are tried until a unique name is found.
If tables_are_singular is true, then TABLE is passed to the singular_to_plural method and the result is used as the name. Otherwise, TABLE is used as-is.
If the selected name is the name of an existing or potential method in the target class, then the suffixes "_objs" and "_objects" are tried in that order. If neither of those suffixes resolves the situation, then ascending numeric suffixes starting with "1" are tried until a unique name is found.
If the selected name is the name of an existing or potential method in the target class, then the suffixes "obj_" and "_object" are tried in that order. If neither of those suffixes resolves the situation, then ascending numeric suffixes starting with "1" are tried until a unique name is found.
If a column named "id" exists, it is selected as the sole primary key column name. If not, the column name generated by joining the return value of class_to_table_singular with "_id" is considered. If no column with that name exists, then the first column (sorted alphabetically) whose type is "serial" is selected. If all of the above fails, then the first column is selected as the primary key column (assuming one exists).
Examples:
My::A->meta->columns(qw(a a_id id));
print My::A->meta->primary_key_columns; # "id"
My::B->meta->columns(qw(b b_id foo));
print My::B->meta->primary_key_columns; # "a_id"
My::D->meta->columns
(
cnt => { type => 'int' },
dub => { type => 'serial' },
foo => { type => 'serial'},
a_id => { type => 'int' }
)
print My::D->meta->primary_key_columns; # "dub"
My::C->meta->columns(qw(foo bar baz));
print My::C->meta->primary_key_columns; # "foo"
If the relationship's type is "one to one" or "many to one", then the relationship's class name is generated by calling related_table_to_class, passing NAME and the convention manager's class as arguments. An attempt is made is load the class. If this fails, the relationship's class name is not set.
The column map for "one to one" and "many to one" relationships is generated using the same rules used to generate key_columns in the auto_foreign_key method.
If the relationship's type is "one to many" then the relationship's class name is generated by calling plural_to_singular on NAME, then passing that value along with the convention manager's class to the related_table_to_class method. An attempt is made is load the class. If this fails, the relationship's class name is not set.
The column map for a "one to many" relationship is only set if both the "local" and "foreign" tables have single-column primary keys. The following ordered list of combinations is considered.
Given:
Local class: My::Product
Foreign class: My::Price
Relationship: prices
Generate these pieces:
Name Description Value
--------- --------------------------------- -------
LTABLE_S Local class_to_table_singular() product
LPK Local primary key column name id
FPK Foreign primary key column name id
Consider column maps in this order:
Value Formula
---------------------- --------------------------
{ id => 'product' } { LPK => LTABLE_S }
{ id => 'product_id' } { LPK => <LTABLE_S>_<PK> }
The first value whose foreign column actually exists in the foreign table is chosen.
If the relationship's type is "many to many" then the relationship's map_class is chosen from a list of possibilities. This list is generated by constructing singular and plural versions of the local and foreign class names (sans prefixes) and then joining them in various ways, all re-prefixed by the class prefix of the convention manager's class. Example:
Given:
Local class: My::Product
Foreign class: My::Color
Relationship: colors
Generate these pieces:
Name Description Value
--------- --------------------------------- -------
PREFIX Local class prefix My::
LCLASS_S Unprefixed local class, singular Product
LCLASS_P Unprefixed local class, plural Products
FCLASS_S Unprefixed foreign class, singular Color
FCLASS_P Unprefixed foreign class, plural Colors
Consider map class names in this order:
Value Formula
--------------- ---------------------
My::ProductsColorsMap <PREFIX><LCLASS_P><FCLASS_P>Map
My::ProductColorMap <PREFIX><LCLASS_S><FCLASS_S>Map
My::ColorsProductsMap <PREFIX><FCLASS_P><LCLASS_P>Map
My::ColorProductMap <PREFIX><FCLASS_S><LCLASS_S>Map
My::ProductsColors <PREFIX><LCLASS_P><FCLASS_P>
My::ProductColors <PREFIX><LCLASS_S><FCLASS_P>
My::ColorsProducts <PREFIX><FCLASS_P><LCLASS_P>
My::ColorProducts <PREFIX><FCLASS_S><LCLASS_P>
My::ColorMap <PREFIX><FCLASS_S>Map
My::ColorsMap <PREFIX><FCLASS_P>Map
My::ProductMap <PREFIX><LCLASS_S>Map
My::ProductsMap <PREFIX><LCLASS_P>Map
The first class found that inherits from Rose::DB::Object and is loaded successfully will be chosen as the relationship's map_class.
Class names are singular and table names are plural. To build the table name, the class prefix is removed from the class name, transitions from lowercase letters or digits to uppercase letters have underscores inserted, and the whole thing is converted to lowercase.
Examples:
Class Table
----------- --------
Product products
My::Product products
My::BigBox big_boxes
My5HatPig my5_hat_pig
Examples:
Class Prefix
----------- --------------
Product <empty string>
My::Product My::
A::B::C::D A::B::C::
To do this, the output of the class_to_table_singular method is passed to a call to the singular_to_plural method. (The CLASS argument, if any, is passed to the call to class_to_table_singular.)
Examples:
Class Table
----------- --------
Product products
My::Product products
My::Box boxes
Examples:
Class Table
----------- --------
Product product
My::Product product
My::Box box
The default implementations returns true if CLASS is derived from Rose::DB::Object and its table name looks like a map table name according to the looks_like_map_table method and the looks_like_map_class method returns either true or undef.
Override this method to control which classes are considered map classes. Note that it may be called several times on the same class at various stages of that class's construction.
The default implementation returns true if CLASS is derived from Rose::DB::Object and has exactly two foreign keys. It returns false (but defined) if CLASS is derived from Rose::DB::Object and has been initialized (or if the foreign keys have been auto-initialized) and the CLASS has no deferred foreign keys. It returns undef otherwise.
The default implementation returns true if TABLE is in one of these forms:
Regex Examples
----------------------- -----------------------------
(\w+_){2,}map pig_toe_map, pig_skin_toe_map
(\w+_)*\w+_(\w+_)*\w+s pig_toes, pig_skin_toe_jams
(\w+_)*\w+s_(\w+_)*\w+s pigs_toes, pig_skins_toe_jams
It returns false otherwise.
Otherwise, the following rules are applied, case-insensitively.
* If STRING ends in "ies", then the "ies" is replaced with "y".
* If STRING ends in "ses" then the "ses" is replaced with "s".
* If STRING matches "/[aeiouy]ss$/i", it is returned unmodified.
For all other cases, the letter "s" is removed from the end of STRING and the result is returned.
To do this, table_to_class is called with TABLE and the class_prefix of LOCAL_CLASS passed as arguments.
Examples:
Table Local Class Related Class
----------- ------------ ----------------
prices My::Product My::Price
big_hats A::B::FooBar A::B::BigHat
a1_steaks Meat A1Steak
* If STRING ends in "x", "ss", or "es", then "es" is appended.
* If STRING ends in "y" then the "y" is replaced with "ies".
* If STRING ends in "s" then it is returned as-is.
* Otherwise, "s" is appended.
If tables_are_singular is true, then TABLE is returned as-is. Otherwise, TABLE is passed to the plural_to_singular method and the result is returned. Otherwise, TABLE is returned as-is.
If tables_are_singular is true, then TABLE is passed to the singular_to_plural method and the result is returned. Otherwise, TABLE is returned as-is.
To do this, any letter that follows an underscore ("_") in the table name is replaced with an uppercase version of itself, and the underscore is removed.
Examples:
Table Prefix Class
----------- ------ -----------
products My:: My::Product
products <none> Product
big_hats My:: My::BigHat
my5_hat_pig <none> My5HatPig
These methods are not part of the public interface, but are supported for use by subclasses. Put another way, given an unknown object that "isa" Rose::DB::Object::Metadata::ConventionManager, there should be no expectation that the following methods exist. But subclasses, which know the exact class from which they inherit, are free to use these methods in order to implement the public API described above.
Much of the richness of a convention manager relies upon the quality of the singular_to_plural and plural_to_singular methods. The default implementations are primitive at best. For example, singular_to_plural will not correctly form the plural of the word "alumnus".
One easy way to improve this is by setting a custom singular_to_plural_function. Here's an example using the handy Lingua::EN::Inflect module:
package My::Product;
...
use Lingua::EN::Inflect;
$cm = __PACKAGE__->meta->convention_manager;
$cm->singular_to_plural_function(\&Lingua::EN::Inflect::PL);
print $cm->singular_to_plural('person'); # "people"
But that's a bit of a pain to do in every single class. An easier way to do it for all of your classes is to make a new Rose::DB::Object::Metadata subclass that overrides the init_convention_manager method, then make a Rose::DB::Object-derived base class that uses your new metadata class. Example:
package My::DB::Metadata;
use Rose::DB::Object::Metadata;
our @ISA = qw(Rose::DB::Object::Metadata);
use Lingua::EN::Inflect;
sub init_convention_manager
{
my $self = shift;
# Let the base class make ths convention manager object
my $cm = $self->SUPER::init_convention_manager(@_);
# Set the new singular-to-plural function
$cm->singular_to_plural_function(\&Lingua::EN::Inflect::PL);
# Return the modified convention manager
return $cm;
}
...
package My::DB::Object;
use My::DB::Metadata;
use Rose::DB::Object;
our @ISA = qw(Rose::DB::Object);
sub meta_class { 'My::DB::Metadata' }
...
package My::Person;
use My::DB::Object;
our @ISA = qw(My::DB::Object);
# The big pay-off: smart plurals!
print __PACKAGE__->meta->table; # "people"
You might wonder why I don't use Lingua::EN::Inflect in Rose::DB::Object::ConventionManager to save you this effort. The answer is that the Lingua::EN::Inflect module adds almost a megabyte of memory overhead on my system. I'd rather not incur that overhead just for the sake of being more clever about naming conventions. Furthermore, as primitive as the default plural-forming is, at least it's deterministic. Guessing what Lingua::EN::Inflect will return is not always easy, and the results can change depending on which version Lingua::EN::Inflect you have installed.
Here's a complete example of nearly all of the major features of Rose::DB::Object::ConventionManager. Let's start with the database schema. (This example uses PostgreSQL, but any supported database with native foreign key support will work.)
CREATE TABLE vendors
(
id SERIAL NOT NULL PRIMARY KEY,
name VARCHAR(255)
);
CREATE TABLE colors
(
code CHAR(3) NOT NULL PRIMARY KEY,
name VARCHAR(255)
);
CREATE TABLE products
(
id SERIAL NOT NULL PRIMARY KEY,
name VARCHAR(255),
vendor_id INT NOT NULL REFERENCES vendors (id)
);
CREATE TABLE prices
(
price_id SERIAL NOT NULL PRIMARY KEY,
product_id INT NOT NULL REFERENCES products (id),
region CHAR(2) NOT NULL DEFAULT 'US',
price DECIMAL(10,2) NOT NULL
);
CREATE TABLE product_colors
(
id SERIAL NOT NULL PRIMARY KEY,
product_id INT NOT NULL REFERENCES products (id),
color_code CHAR(3) NOT NULL REFERENCES colors (code)
);
Now the classes:
# Rose::DB subclass to handle the db connection
package My::DB;
use base 'Rose::DB';
My::DB->register_db
(
type => 'default',
domain => 'default',
driver => 'Pg',
database => 'test',
username => 'postgres',
);
...
# Common Rose::DB::Object-derived base class for the other objects
package My::Object;
use My::DB;
use base 'Rose::DB::Object';
sub init_db { My::DB->new }
...
package My::Price;
use base 'My::Object';
__PACKAGE__->meta->setup
(
columns =>
[
price_id => { type => 'serial', not_null => 1 },
product_id => { type => 'int' },
region => { type => 'char', length => 2, default => 'US' },
price => { type => 'decimal', precision => 10, scale => 2 },
],
foreign_keys => [ 'product' ],
);
...
package My::Vendor;
use base 'My::Object';
__PACKAGE__->meta->setup
(
columns =>
[
id => { type => 'serial', not_null => 1 },
name => { type => 'varchar', length => 255 },
],
);
...
package My::Color;
use base 'My::Object';
__PACKAGE__->meta->setup
(
columns =>
[
code => { type => 'char', length => 3, not_null => 1 },
name => { type => 'varchar', length => 255 },
],
);
...
package My::Product;
use base 'My::Object';
__PACKAGE__->meta->setup
(
columns =>
[
id => { type => 'serial', not_null => 1 },
name => { type => 'varchar', length => 255 },
vendor_id => { type => 'int' },
],
foreign_keys => [ 'vendor' ],
relationships =>
[
prices => { type => 'one to many' },
colors => { type => 'many to many' },
],
);
...
package My::ProductColors;
use base 'My::Object';
__PACKAGE__->meta->setup
(
columns => [ qw(id product_id color_code) ],
foreign_keys => [ 'product', 'color' ],
);
Let's add some data:
INSERT INTO vendors (id, name) VALUES (1, 'V1');
INSERT INTO vendors (id, name) VALUES (2, 'V2');
INSERT INTO products (id, name, vendor_id) VALUES (1, 'A', 1);
INSERT INTO products (id, name, vendor_id) VALUES (2, 'B', 2);
INSERT INTO products (id, name, vendor_id) VALUES (3, 'C', 1);
INSERT INTO prices (product_id, region, price) VALUES (1, 'US', 1.23);
INSERT INTO prices (product_id, region, price) VALUES (1, 'DE', 4.56);
INSERT INTO prices (product_id, region, price) VALUES (2, 'US', 5.55);
INSERT INTO prices (product_id, region, price) VALUES (3, 'US', 5.78);
INSERT INTO prices (product_id, region, price) VALUES (3, 'US', 9.99);
INSERT INTO colors (code, name) VALUES ('CC1', 'red');
INSERT INTO colors (code, name) VALUES ('CC2', 'green');
INSERT INTO colors (code, name) VALUES ('CC3', 'blue');
INSERT INTO colors (code, name) VALUES ('CC4', 'pink');
INSERT INTO product_colors (product_id, color_code) VALUES (1, 'CC1');
INSERT INTO product_colors (product_id, color_code) VALUES (1, 'CC2');
INSERT INTO product_colors (product_id, color_code) VALUES (2, 'CC4');
INSERT INTO product_colors (product_id, color_code) VALUES (3, 'CC2');
INSERT INTO product_colors (product_id, color_code) VALUES (3, 'CC3');
(Be aware that not all databases are smart enough to track explicitly setting serial column values as shown in the INSERT statements above. Subsequent auto-generated serial values may conflict with the explicitly set serial column values already in the table. Values are set explicitly here to make the examples easier to follow. In "real" code, you should let the serial columns populate automatically.)
Finally, the classes in action:
$p = My::Product->new(id => 1)->load;
print $p->vendor->name, "\n"; # "V1"
# "US: 1.23, DE: 4.56"
print join(', ', map { $_->region .': '. $_->price } $p->prices), "\n";
# "red, green"
print join(', ', map { $_->name } $p->colors), "\n";
Using Rose::DB::Object's auto-initialization feature, the Perl code can be reduced to an absurd degree. Given the same database schema and data shown in the example above, consider the following classes:
package My::Auto::Color; use base 'My::Object'; __PACKAGE__->meta->auto_initialize; ... package My::Auto::Price; use base 'My::Object'; __PACKAGE__->meta->auto_initialize; ... package My::Auto::ProductColors; use base 'My::Object'; __PACKAGE__->meta->auto_initialize; ... package My::Auto::Vendor; use base 'My::Object'; __PACKAGE__->meta->auto_initialize; ... package My::Auto::Product; use base 'My::Object'; __PACKAGE__->meta->auto_initialize;
Not a single table, column, foreign key, or relationship is specified, yet everything still works:
$p = My::Auto::Product->new(id => 1)->load;
print $p->vendor->name, "\n"; # "V1"
# "US: 1.23, DE: 4.56"
print join(', ', map { $_->region .': '. $_->price } $p->prices), "\n";
# "red, green"
print join(', ', map { $_->name } $p->colors), "\n";
More precisely, everything still works provided that you load all the of the related modules. For example, if you load "My::Auto::Product" but don't load "My::Auto::Price" (either from within the "My::Auto::Product" class or in your program itself), then the "My::Auto::Product" will not have a "prices()" method (since your program will have no knowledge of the "My::Auto::Price" class). Use the loader if you want to set up a bunch of related classes automatically without worrying about this kind of thing.
Anyway, I don't recommend this kind of extreme approach, but it is an effective demonstration of the power of the convention manager.
John C. Siracusa (siracusa@gmail.com)
Copyright (c) 2010 by John C. Siracusa. All rights reserved. This program is free software; you can redistribute it and/or modify it under the same terms as Perl itself.
| 2022-10-14 | perl v5.34.0 |