4 stable releases
2.0.0 | May 6, 2021 |
---|---|
1.1.0 | Apr 19, 2021 |
1.0.1 | Apr 24, 2020 |
1.0.0 | Mar 27, 2020 |
#1173 in Data structures
436 downloads per month
Used in transit_model
17KB
189 lines
relational_types
Modeling the relations between objects. See documentation to know more.
lib.rs
:
Modeling the relations between objects.
By default, feature relational_types_procmacro
is enabled, exposing macros to
help build relations. See documentation of the crate relational_types_procmacro
for more information.
This module defines types for modeling the relations between
objects, and use them thanks to the GetCorresponding
custom
derive.
Let's clarify that with an example. Suppose that Bike
s have a
Brand
. Bike
s also have an Owner
, and these Owner
s have a
Job
. Bike
s also have a Kind
.
Brand - Bike - Owner - Job
|
Kind
Let's defines these relations and use them a bit:
#[derive(Default, GetCorresponding)]
pub struct World {
brands_to_bikes: OneToMany<Brand, Bike>,
owners_to_bikes: OneToMany<Owner, Bike>,
jobs_to_owners: OneToMany<Job, Owner>,
kinds_to_bikes: OneToMany<Kind, Bike>,
}
let world = World::default();
let mbk: Idx<Brand> = get_mbk_brand();
let owners_with_mbk: IdxSet<Owner> = world.get_corresponding_from_idx(mbk);
let jobs_with_mbk: IdxSet<Job> = world.get_corresponding(&owners_with_mbk);
println!(
"{} owners with {} different jobs own a bike of the brand MBK.",
owners_with_mbk.len(),
jobs_with_mbk.len()
);
First, we want to model the relations between the object. One bike
has a brand, and a brand has several bikes (hopefully). Thus, we
use a OneToMany<Bike, Brand>
to model this relation.
We repeat this process to model every relation. We obtain without
too much effort the World
struct.
The GetCorresponding
derive looks at each field of the World
struct, keeping the fields containing _to_
with a type with 2
generics, and interpret that as a relation. For example,
bikes_to_brands: OneToMany<Bike, Brand>
is a relation between
Bike
and Brand
. Using all the relations, it generates a graph,
compute the shortest path between all the types, and generate an
impl GetCorresponding
for each feasible path.
These impl GetCorresponding
are used by
World::get_corresponding_from_idx
and World::get_corresponding
that are helpers to explore the World
.
Thus, when we call world.get_corresponding_from_idx(mbk)
for
Owner
, we will use the generated code that, basically, gets all
the Bike
s corresponding to the Brand
MBK, and then gets all
the Owner
s corresponding to these Bike
s.
Imagine that, in our application, we use a lot the Owner->Kind
and Brand->Kind
search. To do these searches, we pass by
Bike
, and there is a lot of Bike
s in our model. Thus, as an
optimization, we want to precompute these relations.
Brand - Bike - Owner - Job
\ | /
`-- Kind --'
The shortcuts Brand - Kind
and Kind - Owner
allow our
optimization, but we now have a problem for the Owner->Brand
search: we can do Owner->Kind->Brand
and Owner->Bike->Brand
with a cost of 2. The first solution is clearly wrong, introduced
by our shortcuts. To fix this problem, we can put a weight of 1.9
on Brand - Kind
and Kind - Owner
. The path
Owner->Kind->Brand
now cost 3.8 and is discarded.
Let's implement that:
#[derive(GetCorresponding)]
pub struct World {
brands_to_bikes: OneToMany<Brand, Bike>,
owners_to_bikes: OneToMany<Owner, Bike>,
jobs_to_owners: OneToMany<Job, Owner>,
kinds_to_bikes: OneToMany<Kind, Bike>,
// shortcuts
#[get_corresponding(weight = "1.9")]
brands_to_kinds: ManyToMany<Brand, Kind>,
#[get_corresponding(weight = "1.9")]
kinds_to_owners: ManyToMany<Kind, Owner>,
}
impl World {
fn new() -> World {
let brands_to_bikes = create_brands_to_bikes();
let owners_to_bikes = create_owners_to_bikes();
let jobs_to_owners = create_jobs_to_owners();
let kinds_to_bikes = create_kinds_to_bikes();
World {
brands_to_kinds: ManyToMany::from_relations_sink(
&brands_to_bikes,
&kinds_to_bikes,
),
kinds_to_owners: ManyToMany::from_relations_sink(
&kinds_to_bikes,
&owners_to_bikes,
),
brands_to_bikes,
owners_to_bikes,
jobs_to_owners,
kinds_to_bikes,
}
}
}
Dependencies
~0.9–1.6MB
~33K SLoC