Skip to content

Query Result Mapping (QRM)

go-jet edited this page Nov 1, 2022 · 13 revisions

Contents

How scan works?

Query and QueryContext statement methods perform scan and grouping of each row result to arbitrary destination structure.

  • Query(db qrm.DB, destination interface{}) error - executes statements over database connection db(or transaction) and stores row results in destination.
  • QueryContext(db qrm.DB, context context.Context, destination interface{}) error - executes statement with a context over database connection db(or transaction) and stores row result in destination.

Destination can be either a pointer to struct or a pointer to slice of structs.

The easiest way to understand how scan works is by an example. Lets say we want to retrieve list of cities, with list of customers for each city, and address for each customer. For simplicity we will narrow the choice to 'London' and 'York'.

stmt := 
    SELECT(
        City.CityID, 
        City.City,
        Address.AddressID, 
        Address.Address,
        Customer.CustomerID, 
        Customer.LastName,
    ).FROM(
        City.
            INNER_JOIN(Address, Address.CityID.EQ(City.CityID)).
            INNER_JOIN(Customer, Customer.AddressID.EQ(Address.AddressID)).
    ).
    WHERE(City.City.EQ(String("London")).OR(City.City.EQ(String("York")))).
    ORDER_BY(City.CityID, Address.AddressID, Customer.CustomerID)

Debug sql of above statement:

SELECT city.city_id AS "city.city_id",
     city.city AS "city.city",
     address.address_id AS "address.address_id",
     address.address AS "address.address",
     customer.customer_id AS "customer.customer_id",
     customer.last_name AS "customer.last_name"
FROM dvds.city
     INNER JOIN dvds.address ON (address.city_id = city.city_id)
     INNER JOIN dvds.customer ON (customer.address_id = address.address_id)
WHERE (city.city = 'London') OR (city.city = 'York')
ORDER BY city.city_id, address.address_id, customer.customer_id;

Note that every column is aliased by default. Format is "table_name.column_name".

Above statement will produce following result set:

row city.city_id city.city address.address_id address.address customer.customer_id customer.last_name
1 312 "London" 256 "1497 Yuzhou Drive" 252 "Hoffman"
2 312 "London" 517 "548 Uruapan Street" 512 "Vines"
3 589 "York" 502 "1515 Korla Way" 497 "Sledge"

Lets execute statement and scan result set to destination dest:

var dest []struct {
   model.City

   Customers []struct{
       model.Customer

       Address model.Address
   }
}

err := stmt.Query(db, &dest)

Note that camel case of result set column names(aliases) is the same as model type name.field name. For instance city.city_id -> City.CityID. This is being used to find appropriate column for each destination model field. It is not an error if there is not a column for each destination model field. Table and column names does not have to be in snake case.

Query uses reflection to introspect destination type structure, and result set column names(aliases), to find appropriate destination field for result set column. Every new destination struct object is cached by his and all the parents primary key. So grouping to work correctly at least table primary keys has to appear in query result set. If there is no primary key in a result set row number is used as grouping condition(which is always unique).
For instance, after row 1 is processed, two objects are stored to cache:

Key:                                        Object:
(City(312))                              -> (*struct { model.City; Customers []struct { model.Customer; Address model.Address } })
(City(312)),(Customer(252),Address(256)) -> (*struct { model.Customer; Address model.Address })

After row 2 processing only one new object is stored to cache, because city with city_id 312 is already in cache.

Key:                                        Object:
(City(312))                              -> pulled from cache
(City(312)),(Customer(512),Address(517)) -> (*struct { model.Customer; Address model.Address })

Lets print dest as a json, to visualize Query result:

[
	{
		"CityID": 312,
		"City": "London",
		"CountryID": 0,
		"LastUpdate": "0001-01-01T00:00:00Z",
		"Customers": [
			{
				"CustomerID": 252,
				"StoreID": 0,
				"FirstName": "",
				"LastName": "Hoffman",
				"Email": null,
				"AddressID": 0,
				"Activebool": false,
				"CreateDate": "0001-01-01T00:00:00Z",
				"LastUpdate": null,
				"Active": null,
				"Address": {
					"AddressID": 256,
					"Address": "1497 Yuzhou Drive",
					"Address2": null,
					"District": "",
					"CityID": 0,
					"PostalCode": null,
					"Phone": "",
					"LastUpdate": "0001-01-01T00:00:00Z"
				}
			},
			{
				"CustomerID": 512,
				"StoreID": 0,
				"FirstName": "",
				"LastName": "Vines",
				"Email": null,
				"AddressID": 0,
				"Activebool": false,
				"CreateDate": "0001-01-01T00:00:00Z",
				"LastUpdate": null,
				"Active": null,
				"Address": {
					"AddressID": 517,
					"Address": "548 Uruapan Street",
					"Address2": null,
					"District": "",
					"CityID": 0,
					"PostalCode": null,
					"Phone": "",
					"LastUpdate": "0001-01-01T00:00:00Z"
				}
			}
		]
	},
	{
		"CityID": 589,
		"City": "York",
		"CountryID": 0,
		"LastUpdate": "0001-01-01T00:00:00Z",
		"Customers": [
			{
				"CustomerID": 497,
				"StoreID": 0,
				"FirstName": "",
				"LastName": "Sledge",
				"Email": null,
				"AddressID": 0,
				"Activebool": false,
				"CreateDate": "0001-01-01T00:00:00Z",
				"LastUpdate": null,
				"Active": null,
				"Address": {
					"AddressID": 502,
					"Address": "1515 Korla Way",
					"Address2": null,
					"District": "",
					"CityID": 0,
					"PostalCode": null,
					"Phone": "",
					"LastUpdate": "0001-01-01T00:00:00Z"
				}
			}
		]
	}
]

All the fields missing source column in result set are initialized with empty value. City of London has two customers, which is the product of object reuse in ROW 2 processing.

Custom model types

Destinations are not limited to just generated model types, any destination will work, as long as projection name corresponds to model type name.field name. Only letters are compared, and cases(lowercase, uppercase, CamelCase,...) are ignored.
Go struct field has to be public for scan to work.
Field type can be of any base Go lang type, plus any type that implements sql.Scanner interface (UUID, decimal.Decimal{}, ...).

Lets rewrite above example to use custom model types instead generated ones:

// Address struct has the same name and fields as auto-generated model struct
type Address struct {
    ID  	 int32 `sql:"primary_key"`
    AddressLine  string
}

type MyCustomer struct {
    ID         int32 `sql:"primary_key"`
    LastName   *string

    Address Address
}

type MyCity struct {
    ID     int32 `sql:"primary_key"`
    Name   string

    Customers []MyCustomer
}

dest2 := []MyCity{}

stmt2 := 
    SELECT(
        City.CityID.AS("my_city.id"),                 // snake case
        City.City.AS("myCity.Name"),                  // camel case
        Address.AddressID,                            // No need for aliasing. 
        Address.Address,                              // Default aliasing still works.  
        Customer.CustomerID.AS("My_Customer.id"),      //mixed case
        Customer.LastName.AS("my customer.last name"), //with spaces
    ).FROM(
        City.
            INNER_JOIN(Address, Address.CityID.EQ(City.CityID)).
            INNER_JOIN(Customer, Customer.AddressID.EQ(Address.AddressID)),
    ).WHERE(
        City.City.EQ(String("London")).OR(City.City.EQ(String("York"))),
    ).ORDER_BY(
        City.CityID, Address.AddressID, Customer.CustomerID,
    )

err := stmt2.Query(db, &dest2)

Destination type names and field names are now changed. Every type has 'My' prefix, every primary key column is named ID, LastName is now string pointer, etc.
Now, since we use custom types with changed field identifiers, each column must have an alias for the query mapping to work.
For instance: City.CityID.AS("my_city.id") -> MyCity.ID, City.City.AS("myCity.Name") -> MyCity.Name , etc.

Table names, column names and aliases doesn't have to be in a snake case. CamelCase, PascalCase or some other mixed space is also supported, but it is strongly recommended to use snake case for database identifiers.

Json of new destination is also changed:

[
	{
		"ID": 312,
		"Name": "London",
		"Customers": [
			{
				"ID": 252,
				"LastName": "Hoffman",
				"Address": {
					"ID": 256,
					"AddressLine": "1497 Yuzhou Drive"
				}
			},
			{
				"ID": 512,
				"LastName": "Vines",
				"Address": {
					"ID": 517,
					"AddressLine": "548 Uruapan Street"
				}
			}
		]
	},
	{
		"ID": 589,
		"Name": "York",
		"Customers": [
			{
				"ID": 497,
				"LastName": "Sledge",
				"Address": {
					"ID": 502,
					"AddressLine": "1515 Korla Way"
				}
			}
		]
	}
]

Anonymous custom types

There is no need to create new named type every time. The destination type can be declared inline without naming of any new type.

var dest []struct {
    CityID int32 `sql:"primary_key"`
    CityName   string

    Customers []struct {
        CustomerID int32 `sql:"primary_key"`
        LastName   string

        Address struct {
            AddressID   int32 `sql:"primary_key"`
            AddressLine string
        }
    }
}

stmt := 
    SELECT(
        City.CityID.AS("city_id"),
        City.City.AS("city_name"),
        Customer.CustomerID.AS("customer_id"),
        Customer.LastName.AS("last_name"),
        Address.AddressID.AS("address_id"),
        Address.Address.AS("address_line"),
    ).FROM(
        City.
            INNER_JOIN(Address, Address.CityID.EQ(City.CityID)).
            INNER_JOIN(Customer, Customer.AddressID.EQ(Address.AddressID)).
    )
    WHERE(City.City.EQ(String("London")).OR(City.City.EQ(String("York")))).
    ORDER_BY(City.CityID, Address.AddressID, Customer.CustomerID)

err := stmt.Query(db, &dest)

Aliasing is now simplified. Alias contains only (column/field) name. On the other hand, we can not have 3 fields named ID, because aliases must be unique.

Tagging model type fields

Desired mapping can be set the other way around as well, by tagging destination fields and types.

var dest []struct {
    CityID   int32 `sql:"primary_key" alias:"city.city_id"`
    CityName string `alias:"city.city"`

    Customers []struct {
        // because the whole struct is refering to 'customer.*' (see below tag),
        // we can just use 'alias:"customer_id"`' instead of 'alias:"customer.customer_id"`'
        CustomerID int32 `sql:"primary_key" alias:"customer_id"` 
        LastName   *string `alias:"last_name"`                   

        Address struct {
            AddressID   int32 `sql:"primary_key" alias:"AddressId"` // camel case for alias will work as well
            AddressLine string `alias:"address.address"`            // full alias will work as well
        } `alias:"address.*"`                                       // struct is now refering to all address.* columns

    } `alias:"customer.*"`                                          // struct is now refering to all  customer.* columns
}

stmt := 
    SELECT(
        City.CityID,
        City.City,
        Customer.CustomerID,
        Customer.LastName,
        Address.AddressID,
        Address.Address,
    ).FROM(
        City.
            INNER_JOIN(Address, Address.CityID.EQ(City.CityID)).
            INNER_JOIN(Customer, Customer.AddressID.EQ(Address.AddressID)).
    ).
    WHERE(City.City.EQ(String("London")).OR(City.City.EQ(String("York")))).
    ORDER_BY(City.CityID, Address.AddressID, Customer.CustomerID)

err := stmt.Query(db, &dest)

This kind of mapping is more complicated than in previous examples, and it should avoided and used only when there is no alternative. Usually this is the case in two scenarios:

1) Self join
var dest []struct{
    model.Employee

    Manager *model.Employee `alias:"Manager.*"` //or just `alias:"Manager"
}

manager := Employee.AS("Manager")

stmt := 
    SELECT(
        Employee.EmployeeId,
        Employee.FirstName,
        manager.EmployeeId,
        manager.FirstName,
    ).FROM(
        Employee.
          LEFT_JOIN(manager, Employee.ReportsTo.EQ(manager.EmployeeId)).
    )

This example could also be written without tag alias, by just introducing of a new type type Manager model.Employee.

2) Slices of go base types (int32, float64, string, ...)
var dest struct {
    model.Film
    
    InventoryIDs []int32 `alias:"inventory.inventory_id"`
}

Combining autogenerated and custom model types

It is allowed to combine autogenerated and custom model types. For instance:

type MyCustomer struct {
    ID         int32 `sql:"primary_key"`
    LastName   string

    Address    model.Address                  //model.Address is autogenerated model type
}

type MyCity struct {
    ID     int32 `sql:"primary_key"`
    Name   string

    Customers []MyCustomer
}

Specifying primary keys

Model types generated from database views does not contain any field with primary_key tag. Because there is no primary_key field those types can not be used as a grouping destination. For instance:

var dest []struct {
    model.ActorInfo       // <- view model file, without `primary_key` fields
    Films []model.Film    
}

Querying into above destination would not give correct result, because Films slice does not know to which ActorInfo it relates.
To overcome this issue, we have to manually specify primary keys for view model types.

var dest []struct {                               // ID is a field name in model.ActorInfo
    model.ActorInfo     `sql:"primary_key=ID"`    // coma separated list of field names
    Films []model.Film    
}

Above tag can be used to set new primary key fields on a model type with already defined primary key fields.