# vim: set sw=2 ft=perl:
package DBIx::Class::Sims;
use 5.010_001;
use strictures 2;
our $VERSION = '0.301000';
{
# Do **NOT** import a clone() function into the DBIx::Class::Schema namespace
# because that will override DBIC's clone() method and break all the things.
package MyCloner;
sub clone {
my ($data) = @_;
if (ref($data) eq 'HASH') {
return {
map { $_ => clone($data->{$_}) }
keys %$data
};
}
elsif (ref($data) eq 'ARRAY') {
return [
map { clone($_) }
@$data
];
}
return $data;
}
}
use DDP;
use Data::Walk qw( walk );
use DateTime;
use DBIx::Class::TopoSort ();
use Hash::Merge qw( merge );
use List::Util qw( first );
use List::MoreUtils qw( natatime );
use Scalar::Util qw( blessed reftype );
{
# The aliases in this block are done at BEGIN time so that the ::Types class
# can use them when it is loaded through `use`.
my @sim_names;
my %sim_types;
my @sim_matchers;
sub set_sim_type {
shift;
my $types = shift // '';
if (ref($types) eq 'HASH') {
while ( my ($name, $meth) = each(%$types) ) {
next unless ref($meth) eq 'CODE';
$sim_types{$name} = $meth;
push @sim_names, $name;
}
}
elsif (ref($types) eq 'ARRAY') {
foreach my $item (@$types) {
next unless ref($item->[2]) eq 'CODE';
push @sim_names, $item->[0];
push @sim_matchers, [ qr/^$item->[1]$/, $item->[2] ];
}
}
return;
}
BEGIN { *set_sim_types = \&set_sim_type; }
sub __find_sim_type {
my ($str) = @_;
unless (exists $sim_types{$str}) {
my $item = first { $str =~ $_->[0] } @sim_matchers;
if ($item) {
$sim_types{$str} = $item->[1];
}
}
return $sim_types{$str};
}
sub sim_type {
shift;
# If no specific type requested, then return the complete list of all
# registered types.
return sort @sim_names if @_ == 0;
return __find_sim_type($_[0]) if @_ == 1;
return map { __find_sim_type($_) } @_;
}
BEGIN { *sim_types = \&sim_type; }
}
use DBIx::Class::Sims::Types;
use DBIx::Class::Sims::Runner;
use DBIx::Class::Sims::Util;
sub add_sims {
my $class = shift;
my ($schema, $source, @remainder) = @_;
my $rsrc = $schema->source($source);
my $it = natatime(2, @remainder);
while (my ($column, $sim_info) = $it->()) {
my $col_info = $schema->source($source)->column_info($column) // next;
$col_info->{sim} = merge(
$col_info->{sim} // {},
$sim_info // {},
);
}
return;
}
*add_sim = \&add_sims;
sub load_sims {
my $self = shift;
my $schema;
if (ref($self) && $self->isa('DBIx::Class::Schema')) {
$schema = $self;
}
else {
$schema = shift(@_);
}
my ($spec_proto, $opts_proto) = @_;
$spec_proto = MyCloner::clone($spec_proto // {});
$opts_proto = MyCloner::clone($opts_proto // {});
my $spec = massage_input($schema, normalize_input($spec_proto));
my $opts = normalize_input($opts_proto);
# 1. Ensure the belongs_to relationships are in $reqs
# 2. Set the rel_info as the leaf in $reqs
my $reqs = normalize_input($opts->{constraints} // {});
# 2: Create the rows in toposorted order
my $hooks = $opts->{hooks} // {};
$hooks->{preprocess} //= sub {};
$hooks->{postprocess} //= sub {};
# Create a lookup of the items passed in so we can return them back.
my $initial_spec = {};
foreach my $name (keys %$spec) {
my $normalized = DBIx::Class::Sims::Util->normalize_aoh($spec->{$name});
unless ($normalized) {
warn "Skipping $name - I don't know what to do!\n";
delete $spec->{$name};
next;
}
$spec->{$name} = $normalized;
foreach my $item (@{$spec->{$name}}) {
$initial_spec->{$name}{$item} = 1;
}
}
my ($rows, $additional) = ({}, {});
if (keys %{$spec}) {
# Yes, this invokes srand() twice, once in implicitly in rand() and once
# again right after. But, that's okay. We don't care what the seed is and
# this allows DBIC to be called multiple times in the same process in the
# same second without problems.
$additional->{seed} = $opts->{seed} //= rand(time & $$);
srand($opts->{seed});
my @toposort = DBIx::Class::TopoSort->toposort(
$schema,
%{$opts->{toposort} // {}},
);
my $runner = DBIx::Class::Sims::Runner->new(
parent => $self,
schema => $schema,
toposort => \@toposort,
initial_spec => $initial_spec,
spec => $spec,
hooks => $hooks,
reqs => $reqs,
);
$rows = eval {
$runner->run();
}; if ($@) {
$additional->{error} = $@;
if ($opts->{die_on_failure} // 1) {
warn "SEED: $opts->{seed}\n";
die $@;
}
}
$additional->{created} = $runner->{created};
$additional->{duplicates} = $runner->{duplicates};
# Force a reload from the database of every row we're returning.
foreach my $item (values %$rows) {
$_->discard_changes for @$item;
}
}
if (wantarray) {
return ($rows, $additional);
}
else {
return $rows;
}
}
use YAML::Any qw( LoadFile Load );
sub normalize_input {
my ($proto) = @_;
if ( ref($proto) ) {
return $proto;
}
# Doing a stat on a filename with a newline throws an error.
my $x = eval {
no warnings;
if ( -e $proto ) {
return LoadFile($proto);
}
};
return $x if $x;
return Load($proto);
}
sub massage_input {
my ($schema, $struct) = @_;
my $dtp = $schema->storage->datetime_parser;
walk({
preprocess => sub {
# Don't descend into the weeds. Only do the things we care about.
return if grep { blessed($_) } @_;
return unless grep { reftype($_) } @_;
return @_;
},
wanted => sub {
return unless (reftype($_)//'') eq 'HASH' && !blessed($_);
foreach my $k ( keys %$_ ) {
my $t = $_;
# Expand the dot-naming convention.
while ( $k =~ /([^.]*)\.(.*)/ ) {
$t->{$1} = { $2 => delete($t->{$k}) };
$t = $t->{$1}; $k = $2;
}
# Handle DateTime values passed to us.
if (defined $t->{$k}) {
if ( $t->{$k} =~ /^(\d\d\d\d)-(\d\d)-(\d\d)T(\d\d):(\d\d):(\d\d)$/ ) {
# format_datetime() requires a DateTime object. This may be a
# string, therefore hoist it if need-be.
unless (blessed($t->{$k})) {
$t->{$k} = DateTime->new(
year => $1, month => $2, day => $3,
hour => $4, minute => $5, second => $6,
);
}
$t->{$k} = $dtp->format_datetime($t->{$k});
}
}
}
},
}, $struct);
return $struct;
}
1;
__END__
=head1 NAME
DBIx::Class::Sims - The addition of simulating data to DBIx::Class
=head1 SYNOPSIS (CLASS VERSION)
DBIx::Class::Sims->add_sims(
$schema, 'source_name',
address => { type => 'us_address' },
zip_code => { type => 'us_zipcode' },
# ...
);
my $rows = DBIx::Class::Sims->load_sims($schema, {
Table1 => [
{}, # Take sims or default values for everything
{ # Override some values, take sim values for others
column1 => 20,
column2 => 'something',
},
],
});
=head1 SYNOPSIS (COMPONENT VERSION)
Within your schema class:
__PACKAGE__->load_components('Sims');
Within your resultsources, specify the sims generation rules for columns that
need specified.
__PACKAGE__->add_columns(
...
address => {
data_type => 'varchar',
is_nullable => 1,
data_length => 10,
sim => { type => 'us_address' },
},
zipcode => {
data_type => 'varchar',
is_nullable => 1,
data_length => 10,
sim => { type => 'us_zipcode' },
},
column1 => {
data_type => 'int',
is_nullable => 0,
sim => {
min => 10,
max => 20,
},
},
column2 => {
data_type => 'varchar',
is_nullable => 1,
data_length => 10,
default_value => 'foobar',
},
...
);
Later:
$schema->deploy({
add_drop_table => 1,
});
my $rows = $schema->load_sims({
Table1 => [
{}, # Take sims or default values for everything
{ # Override some values, take sim values for others
column1 => 20,
column2 => 'something',
},
],
});
=head1 PURPOSE
Generating test data for non-simplistic databases is extremely hard, especially
as the schema grows and changes. Designing scenarios B<should> be doable by only
specifying the minimal elements actually used in the test with the test being
resilient to any changes in the schema that don't affect the elements specified.
This includes changes like adding a new parent table, new required child tables,
and new non-NULL columns to the table being tested.
With Sims, you specify only what you care about. Any required parent rows are
automatically generated. If a row requires a certain number of child rows (all
artists must have one or more albums), that can be set as well. If a column must
have specific data in it (a US zipcode or a range of numbers), you can specify
that in the table definition.
And, in all cases, you can override anything.
=head1 DESCRIPTION
This is a L<DBIx::Class> component that adds a few methods to your
L<DBIx::Class::Schema> object. These methods make it much easier to create data
for testing purposes (though, obviously, it's not limited to just test data).
Alternately, it can be used as a class method vs. a component, if that fits your
needs better.
=head1 METHODS
=head2 load_sims
C<< $rv, $addl? = $schema->load_sims( $spec, ?$opts ) >>
C<< $rv, $addl? = DBIx::Class::Sims->load_sims( $schema, $spec, ?$opts ) >>
This method will load the rows requested in C<$spec>, plus any additional rows
necessary to make those rows work. This includes any parent rows (as defined by
C<belongs_to>) and per any constraints defined in C<$opts->{constraints}>. If
need-be, you can pass in hooks (as described below) to manipulate the data.
load_sims does all of its work within a call to L<DBIx::Class::Schema/txn_do>.
If anything goes wrong, load_sims will rethrow the error after the transaction
is rolled back.
This, of course, assumes that the tables you are working with support
transactions. (I'm looking at you, MyISAM!) If they do not, that is on you.
=head3 Return value
This returns one or two values, depending on if you call load_sims in a scalar
or array context.
The first value is a hash of arrays of hashes. This will match the C<$spec>,
except that where the C<$spec> has a requested set of things to make, the return
will have the DBIx::Class::Row objects that were created.
Note that you do not get back the objects for anything other than the objects
specified at the top level.
This second value is a hashref with additional items that may be useful. It may
contain:
=over 4
=item * error
This will contain any error that happened while trying to create the rows.
This is most useful when C<< die_on_failure >> is set to 0.
=item * seed
This is the random seed that was used in this run. If you set the seed in the
opts parameter in the load_sims call, it will be that value. Otherwise, it will
be set to a usefully random value for you. It will be different every time even
if you call load_sims multiple times within the same process in the same second.
=item * created
This is a hashref containing a count of each source that was created. This is
different from the first return value in that this lists everything created, not
just what was requested. It also only has counts, not the actual rows.
=item * duplicates
This is a hashref containing a list for each source of all the duplicates that
were found when creating rows for that source. For each duplicate found, there
will be an entry that specifies the criteria used to find that duplicate and the
row in the database that was found.
The list will be ordered by when the duplicate was found, but that ordering will
B<NOT> be stable across different runs unless the same C<< seed >> is used.
=back
=head2 set_sim_type
C<< $class_or_obj->set_sim_type({ $name => $handler, ... }); >>
C<< $class_or_obj->set_sim_type([ [ $name, $regex, $handler ], ... ]); >>
This method will set the handler for the C<$name> sim type. The C<$handler> must
be a reference to a subroutine. You may pass in as many name/handler pairs as you
like.
You may alternately pass in an arrayref of triplets. This allows you to use a
regex to match the provided type. C<$name> will be returned when the user
introspects the list of loaded sim types. C<$regex> will be used when finding the
type to handle this column. C<$handler> must be a reference to a subroutine.
You cannot set pairs and triplets in the same invocation.
This method may be called as a class or object method.
This method returns nothing.
C<set_sim_types()> is an alias to this method.
=head2 sim_types
C<< $class_or_obj->sim_types(); >>
This method will return a sorted list of all registered sim types.
This method may be called as a class or object method.
=head1 SPECIFICATION
The specification can be passed along as a filename that contains YAML or JSON,
a string that contains YAML or JSON, or as a hash of arrays of hashes. The
structure should look like:
{
ResultSourceName => [
{
column => $value,
column => $value,
relationship => $parent_object,
relationship => {
column => $value,
},
'relationship.column' => $value,
'rel1.rel2.rel3.column' => $value,
},
],
}
If a column is a belongs_to relationship name, then the row associated with that
relationship specifier will be used. This is how you would specify a specific
parent-child relationship. (Otherwise, a random choice will be made as to which
parent to use, creating one as necessary if possible.) The dots will be followed
as far as necessary.
If a column's value is a hashref, then that will be treated as a sim entry.
Example:
{
Artist => [
{
name => { type => 'us_name' },
},
],
}
That will use the provided sim type 'us_name'. This will override any sim entry
specified on the column. See L</SIM ENTRY> for more information.
Note: Before 0.300800, this behavior was triggered by a reference to a hashref.
That will still work, but is deprecated, throws a warning, and will be removed
in a future release.
Columns that have not been specified will be populated in one of two ways. The
first is if the database has a default value for it. Otherwise, you can specify
the C<sim> key in the column_info for that column. This is a new key that is not
used by any other component. See L</SIM ENTRY> for more information.
(Please see L<DBIx::Class::ResultSource/add_columns> for details on column_info)
B<NOTE>: The keys of the outermost hash are resultsource names. The keys within
the row-specific hashes are either columns or relationships. Not resultsources.
=head2 Reuse wherever possible
The Sims's normal behavior is to attempt to reuse whenever possible. The theory
is that if you didn't say you cared about something, you do B<NOT> care about
that thing.
=head3 Unique constraints
If a source has unique constraints defined, the Sims will use them to determine
if a new row with these values I<can> be created or not. If a row already
exists with these values for the unique constraints, then that row will be used
instead of creating a new one.
This is B<REGARDLESS> of the values for the non-unique-constraint rows.
=head3 Forcing creation of a parent
If you do not specify values for a parent (i.e., belongs_to), then the first row
for that parent will be be used. If you don't care what values the parent has,
but you care that a different parent is used, then you can set the __META__ key
as follows:
$schema->load_sims({
Album => {
artist => { __META__ => { create => 1 } },
name => 'Some name',
}
})
This will force the creation of a parent instead of reusing the parent.
B<NOTE>: If the simmed values within the parent's class would result in values
that are the same across a unique constraint with an existing row, then that
row will be used. This just bypasses the "attempt to use the first parent".
=head2 Alternatives
=head3 Hard-coded number of things
If you only want N of a thing, not really caring just what the column values end
up being, you can take a shortcut:
{
ResultSourceName => 3,
}
That will create 3 of that thing, taking all the defaults and sim'ed options as
exist.
This will also work if you want 3 of a child via a has_many relationship. For
example, you can do:
{
Artist => {
name => 'Someone Famous',
albums => 240,
},
}
That will create 240 different albums for that artist, all with the defaults.
=head3 Just one thing
If you are creating one of a thing and setting some of the values, you can skip
the arrayref and pass the hashref directly.
{
ResultSourceName => {
column => $value,
column => $value,
relationship => {
column => $value,
},
'relationship.column' => $value,
'rel1.rel2.rel3.column' => $value,
},
}
And that will work exactly as expected.
=head3 References
Let's say you have a table that's a child of two other tables. You can specify
that relationship as follows:
{
Parent1 => 1,
Parent2 => {
Child => {
parent1 => \"Parent1[0]",
},
},
}
That's a reference to a string with the tablename as a pseudo-array, then the
index into that array. This only works for rows that you are going to return
back from the C<< load_sims() >> call.
This also only works for belongs_to relationships. Since all parents are created
before all children, the Sims cannot back-reference into children.
=head2 Notes
=over 4
=item * Multiply-specified children
Sometimes, you will have a table with more than one parent (q.v. t/t5.t for an
example of this). If you specify a row for each parent and, in each parent,
specify a child with the same characteristics, only one child will be created.
The assumption is that you meant the same row.
This does B<not> apply to creating multiple rows with the same characteristics
as children of the same parent. The assumption is that you meant to do that.
=back
=head1 OPTS
There are several possible options.
=head2 constraints
The constraints can be passed along as a filename that contains YAML or JSON, a
string that contains YAML or JSON, or as a hash of arrays of hashes. The
structure should look like:
{
Person => {
addresses => 2,
},
}
All the C<belongs_to> relationships are automatically added to the constraints.
You can add additional constraints, as needed. The most common use for this will
be to add required child rows. For example, C<< Person->has_many('addresses') >>
would normally mean that if you create a Person, no Address rows would be
created. But, we could specify a constraint that says "Every person must have
at least 2 addresses." Now, whenever a Person is created, two Addresses will be
added along as well, if they weren't already created through some other
specification.
=head2 die_on_failure
If set to 0, this will prevent a die when creating a row. Instead, you will be
responsible for checking C<< $additional->{error} >> yourself.
This defaults to 1.
=head2 seed
If set, this will be the srand() seed used for this invocation.
=head2 toposort
This is passed directly to the call to C<< DBIx::Class::TopoSort->toposort >>.
=head2 hooks
Most people will never need to use this. But, some schema definitions may have
reasons that prevent a clean simulating with this module. For example, there may
be application-managed sequences. To that end, you may specify the following
hooks:
=over 4
=item * preprocess
This receives C<$name, $source, $spec> and expects nothing in return. C<$spec>
is the hashref that will be passed to C<<$schema->resultset($name)->create()>>.
This hook is expected to modify C<$spec> as needed.
=item * postprocess
This receives C<$name, $source, $row> and expects nothing in return. This hook
is expected to modify the newly-created row object as needed.
=back
=head1 SIM ENTRY
To control how a column's values are simulated, add a "sim" entry in the
column_info for that column. The sim entry is a hash that can have the followingkeys:
=over 4
=item * value / values
This behaves just like default_value would behave, but doesn't require setting a
default value on the column.
sim => {
value => 'The value to always use',
},
This can be either a string, number, or an arrayref of strings or numbers. If it
is an arrayref, then a random choice from that array will be selected.
=item * type
This labels the column as having a certain type. A type is registered using
L</set_sim_type>. The type acts as a name for a function that's used to generate
the value. See L</Types> for more information.
=item * min / max
If the column is numeric, then the min and max bound the random value generated.
If the column is a string, then the min and max are the length of the random
value generated.
=item * func
This is a function that is provided the column info. Its return value is used to
populate the column.
=item * null_chance
If the column is nullable I<and> this is set I<and> it is a number between 0 and
1, then if C<rand()> is less than that number, the column will be set to null.
Otherwise, the standard behaviors will apply.
If the column is B<not> nullable, this setting is ignored.
=back
(Please see L<DBIx::Class::ResultSource/add_columns> for details on column_info)
=head2 Types
The handler for a sim type will receive the column info (as defined in
L<DBIx::Class::ResultSource/add_columns>). From that, the handler returns the
value that will be used for this column.
Please see L<DBIx::Class::Sims::Types> for the list of included sim types.
=head1 SEQUENCE OF EVENTS
When an item is created, the following actions are taken (in this order):
=over 4
=item 1 The columns are fixed up.
This is where generated values are generated. After this is done, all the values
that will be inserted into the database are now available.
q.v. L</SIM ENTRY> for more information.
=item 1 The preprocess hook fires.
You can modify the hashref as necessary. This includes potentially changing what
parent and/or child rows to associate with this row.
=item 1 All foreign keys are resolved.
If it's a parent relationship, the parent row will be found or created. All
parent rows will go through the same sequence of events as described here.
If it's a child relationship, creation of the child rows will be deferred until
later.
=item 1 The row is found or created.
It might be found by unique constraint or created.
=item 1 All child relationships are handled
Because they're a child relationship, they are deferred until the time that
model is handled in the toposorted graph. They are not created now because they
might associate with a different parent that has not been created yet.
=item 1 The postprocess hook fires.
Note that any child rows are not guaranteed to exist yet.
=back
=head1 TODO
=head2 Multi-column types
In some applications, columns like "state" and "zipcode" are correlated. Values
for one must be legal for the value in the other. The Sims currently has no way
of generating correlated columns like this.
This is most useful for saying "These 6 columns should be a coherent address".
=head2 Allow a column to reference other columns
Sometimes, a column should alter its behavior based on other columns. A fullname
column may have the firstname and lastname columns concatenated, with other
things thrown in. Or, a zipcode column should only generate a zipcode that're
legal for the state.
=head1 BUGS/SUGGESTIONS
This module is hosted on Github at
L<https://github.com/robkinyon/dbix-class-sims>. Pull requests are strongly
encouraged.
=head1 DBIx::Class::Fixtures
L<DBIx::Class::Fixtures> is another way to load data into a database. Unlike
this module, L<DBIx::Class::Fixtures> approaches the problem by loading the same
data every time. This is complementary because some tables (such as lookup
tables of countries) want to be seeded with the same data every time. The ideal
solution would be to have a set of tables loaded with fixtures and another set
of tables loaded with sims.
=head1 SEE ALSO
L<DBIx::Class>, L<DBIx::Class::Fixtures>
=head1 AUTHOR
Rob Kinyon <rob.kinyon@gmail.com>
=head1 LICENSE
Copyright (c) 2013 Rob Kinyon. All Rights Reserved.
This is free software, you may use it and distribute it under the same terms
as Perl itself.
=cut