Skip to content

quintans/goSQL

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

88 Commits
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

goSQL

a ORM like library in Go (golang) that makes SQL easier to use.

Index

Introduction

(English is not my native language so please bear with me)

goSQL aims to facilitate the convertion between database tables and structs and make easy the use of complex joins. It has no intention of hiding the SQL from the developer and a closer idiom to SQL is also part of the library. Structs can be used as a representation of a table record for CRUD operations but there is no direct dependency between a struct and a table. The fields of a struct are matched with the column alias of the SQL statement to build a result.

This library is not locked to any database vendor. This database abstraction is achieved by what I called Translators. Translators for MySQL, PostgreSQL, FirebirdSQL and Oracle are provided. These Translators can be extended by registering functions to implement functionality not covered by the initial Translators or customize to something specific to a project.

This library is supported by a mapping system that enables you to avoid writing any SQL text, and if you are using an editor with auto-complete it will be easy to write your SQL.

References to your database schema are located in one place, avoiding a major pain when you have to refactor your database.

An example of the syntax is as follows:

var publisher Publisher
store.Query(PUBLISHER).
	All().
	Where(PUBLISHER_C_ID.Matches(2)).
	SelectTo(&publisher)

Short version

store.Retrieve(&publisher, 2)

We are not restricted to the use of structs as demonstrated by the next snippet

var name string
store.Query(PUBLISHER).
	Column(PUBLISHER_C_NAME).
	Where(PUBLISHER_C_ID.Matches(2)).
	SelectInto(&name)

Another example with an update

store.Update(PUBLISHER).Submit(&publisher)

and the shortest version...

store.Modify(&publisher)

To specify the context we do

store.WithContext(context.Background())

Features

  • SQL DSL
  • CRUD actions using structs
  • Simple join declaration
  • Populate struct tree with query results containing joins
  • Embedded Structs
  • Subqueries
  • Automatic setting of primary keys for inserts
  • Optimistic Locking with automatic version increment
  • Database Abstraction
  • Transactions
  • Result Pagination
  • Pre/Post insert/update/delete Struct triggers
  • Support for primitive pointer types like *string, *int64, *float64, *bool, etc
  • Support for types implementing driver.Valuer and sql.Scanner interface, like NullString, type converters, etc
  • Extensible

Tested Databases

  • MariaDB 5.5
  • PostgreSQL 9.2
  • FirebirdSQL 2.5
  • Oracle XE 11g - read this

Dependencies

go 1.1+

Installation

go get github.com/quintans/toolkit

go get github.com/quintans/goSQL

Startup Guide

This guide is based on a MySQL database, so we need to get a database driver. I used the one in https://github.com/go-sql-driver/mysql

So lets get started.

Create the table PUBLISHER in a MySQL database called gosql. Of course the database name can be changed and configured to something else.

CREATE TABLE `PUBLISHER` (
	ID BIGINT NOT NULL AUTO_INCREMENT,
	VERSION INTEGER NOT NULL,
	`NAME` VARCHAR(50),
	`ADDRESS` VARCHAR(255),
	PRIMARY KEY(ID)
)
ENGINE=InnoDB
DEFAULT CHARSET=utf8;

And the code is

package main

import (
	. "github.com/quintans/goSQL/db"
	"github.com/quintans/goSQL/dbx"
	trx "github.com/quintans/goSQL/translators"

	_ "github.com/go-sql-driver/mysql"

	"database/sql"
	"fmt"
)

// the entity
type Publisher struct {
	Id      int64
	Version int64
	Name    string
}

// table description/mapping
var (
	PUBLISHER           = TABLE("PUBLISHER")
	PUBLISHER_C_ID      = PUBLISHER.KEY("ID")          // implicit map to field Id
	PUBLISHER_C_VERSION = PUBLISHER.VERSION("VERSION") // implicit map to field Version
	PUBLISHER_C_NAME    = PUBLISHER.COLUMN("NAME")     // implicit map to field Name
)

// the transaction manager
var TM ITransactionManager

func main() {
	// database configuration
	mydb, err := sql.Open("mysql", "root:root@/gosql?parseTime=true")
	if err != nil {
		panic(err)
	}

	// transaction manager
	tm = NewTransactionManager(mydb, trx.NewMySQL5Translator())

	// get the database context
	store := tm.Store()
	// the target entity
	var publisher Publisher
	// Retrieve
	_, err := store.Retrieve(&publisher, 2)
	if err != nil {
		panic(err)
	}

	fmt.Println(publisher)
}

Source from basic.go.

You can also check out common.go for a lot more examples.

If you need to inject context.Context you can do so with

store := tm.Store().WithContext(ctx)

In the following chapters I will try to explain the several aspects of the library using a set of examples. These examples are supported by tables defined in tables_mysql.sql, a MySQL database sql script.

I will start first by describing the table model and how to map the entities.

Entity Relation Diagram

ER Diagram

Relationships explained:

  • One-to-Many: One PUBLISHER can have many BOOKs and one BOOK has one PUBLISHER.
  • One-to-One: One BOOK has one BOOK_BIN (Hardcover) - binary data is stored in a separated table - and one BOOK_BIN has one BOOK.
  • Many-to-Many: One AUTHOR can have many BOOKs and one BOOK can have many AUTHORs.

Table definition

As seen in the Startup Guide, mapping a table is pretty straight forward.

Declaring a table

var PUBLISHER = TABLE("PUBLISHER")

Declaring a column

var PUBLISHER_C_NAME = PUBLISHER.COLUMN("NAME")     // implicit map to field 'Name''

By default, the result value for this column will be put in the field Name of the target struct. If we wish for a different alias we use the .As(...) at the end resulting in:

var PUBLISHER_C_NAME = PUBLISHER.COLUMN("NAME").As("Other") // map to field 'Other'

The declared alias Other is now the default for all the generated SQL. As all defaults, it can be changed to another value when building a SQL statement.

Besides the regular columns, there are the special columns KEY, VERSION and DELETION.

var PUBLISHER_C_ID       = PUBLISHER.KEY("ID")           // implicit map to field Id
var PUBLISHER_C_VERSION  = PUBLISHER.VERSION("VERSION")  // implicit map to field Version
var PUBLISHER_C_DELETION = PUBLISHER.DELETION("DELETION") // map to field 'Deletion'
  • KEY identifies the column(s) as primary key(s) of a table.
  • VERSION identifies the column used for optimistic locking.
  • DELETION identifies the column used for logic record deletion.

It is not mandatory to map all columns of a table. For the same physical table several logical tables can be created with diferent set of columns. They can even refer to diferent domain values depending on a discriminator column as seen in the Table Discriminator section.

Next we will see how to declare associations. To map associations, we do not think on the multiplicity of the edges, but how to go from A to B. This leaves us only two types of associations: Simple (one-to-one, one-to-many, many-to-one) and Composite (many-to-many) associations. How to use associations is explained in the Query chapter.

Declaring a Simple association

var PUBLISHER_A_BOOKS = PUBLISHER.
			ASSOCIATE(PUBLISHER_C_ID).
			TO(BOOK_C_PUBLISHER_ID).
			As("Books")

In this example, we see the mapping of the relationship between PUBLISHER and BOOK using the column PUBLISHER_C_ID and BOOK_C_PUBLISHER_ID. The .As("Books") part indicates that when transforming a query result to a struct, it should follow the Books field to put the transformation part regarding to the BOOK entity. The Association knows nothing about the multiplicity of its edges. This association only covers going from PUBLISHER to BOOK. If we want to go from BOOK to PUBLISHER we need to declare the reverse association in the table BOOK mapping.

In a association with multiple keys columns we just declare the following:

var PARENT_A_CHILDREN = PARENT.
			ASSOCIATE(PARENT_C_ID1, PARENT_C_ID2).
			TO(CHILD_C_PARENT_ID1, CHILD_C_PARENT_ID2).
			As("Children")

and the reverse association

var CHILD_A_PARENT = CHILD.
			ASSOCIATE(CHILD_C_PARENT_ID1, CHILD_C_PARENT_ID2).
			TO(PARENT_C_ID1, PARENT_C_ID2).
			As("Parent")

Declaring a Composite association.

This kind of associations makes use on an intermediary table, and therefore we need to declare it.

var (
	AUTHOR_BOOK				= TABLE("AUTHOR_BOOK")
	AUTHOR_BOOK_C_AUTHOR_ID	= AUTHOR_BOOK.KEY("AUTHOR_ID") // implicit map to field 'AuthorId'
	AUTHOR_BOOK_C_BOOK_ID	= AUTHOR_BOOK.KEY("BOOK_ID") // implicit map to field 'BookId'
)

And finally the Composite association declaration

var AUTHOR_A_BOOKS = NewM2MAssociation(
	"Books",
	ASSOCIATE(AUTHOR_C_ID).WITH(AUTHOR_BOOK_C_AUTHOR_ID),
	ASSOCIATE(AUTHOR_BOOK_C_BOOK_ID).WITH(BOOK_C_ID),
)

The order of the parameters is very important, because they indicate the direction of the association.

The full definition of the tables and the struct entities used in this document are in entities.go, covering all aspects of table mapping.

Transactions

To wrap operations inside a transaction we do this:

TM.Transaction(func(store IDb) error {
	// put you actions here
});

If an error is returned or a panic occurs, the transaction is rolled back, otherwise is commited.

common.go has several examples of transactions.

Quick CRUD

The following methods are a way to use structs for quick CRUD operations over the database. They have a short syntax but they are not as powerful as the long version, missing certain features like ordering, greater than, etc

For this operations to work the table must be registered with the same alias as the struct name. In this case the table PUBLISHER is registered with the name Publisher.

See the table mappings here

Create

var pub Publisher
pub.Name = ...
(...)
store.Create(&pub)

More detail on inserting with structs can be found here.

Retrieve

store.Retrieve(&publisher, 2)

If there are multiple keys for a table, the supplied keys must be in the same order as they were declared in the table definition.

When using Retrieve, if there is a struct field with the tag sql:"omit" its value will not be retrieved. The strcut Author has this tag in the field Secret.

It is possible to change the associated table name by implemententing the TableNamer interface.

More detail on selecting one instance with structs can be found here.

FindFirst

var book Book
store.FindFirst(&book, Book{PublisherId: ext.Int64(1)})

Retrieves the first record where PublisherId is 1. The same rules of Retrieve() apply.

FindAll

var books []Book
store.FindAll(&books, Book{PublisherId: ext.Int64(2)})

Retrieves all the records where PublisherId is 2. The same rules of Retrieve() apply.

Update

var publisher Publisher
publisher.Name = ...
(...)
store.Modify(&publisher)

There is also another interesting method that does a Insert or an Update, depending on the value of the version field. If the field version is zero or nil an insert is issued, otherwise is an update.

store.Save(&publisher)

More detail on updating with structs can be found here.

Delete

This removes only the book with the matching Id. If the version field is present it is also used in the matching criteria.

var book Book
book.Id = ...
(...)
store.Remove(&book)

The following example removes all records that match the criteria defined by the non zero values of the struct.

store.RemoveAll(Project{StatusCod: ext.String("DEV")})

More detail on deleting with structs can be found here.

Tracking changed fields

When using only struct fields we have no way to select the fields to include in SQL statement. All fields are used (excepts the ones with sql:"omit"), even the ones we did not change. To get around this we need to keep track the changed fields. If there is no changed field, then all fields are used in the SQL statement.

This tracking only applies to updates and inserts.

We track this fields by:

  • adding the anonymous struct db.Marker to the struct entity
  • for every field add a setter
  • inside each setter register the field as changed.

As an example, we show the setter for the field Price of Book.

First Book is defined as:

type Book struct {
	Marker
	(...)
}

And the Price setter is:

func (this *Book) SetPrice(price float64) {
	this.Price = price
	this.Mark("Price") // marks the field for update
}

The normal flow for an update would be:

var book Book
store.Retrieve(&book, 1)
book.SetPrice(book.Price * 0.8) // marked for change
store.Modify(&book)

Since we only changed the Price field, only the PRICE column will be included in the update. (In reality, the VERSION column is also included in the due to the optimistic locking.)

Insert Examples

Simple Insert

insert := Insert(PUBLISHER).
	Columns(PUBLISHER_C_ID, PUBLISHER_C_VERSION, PUBLISHER_C_NAME)
insert.Values(1, 1, "Geek Publications").Execute()
insert.Values(2, 1, "Edições Lusas").Execute()

There is another way of supplying values, for all CRUD operations, and this is by parameter, as seen in the following snippet.

insert.SetParameter("name", "Geek Publications")
insert.Values(1, 1, Param("name")).Execute()

In this example the value for the name parameter is directly supplied in the snippet but it could be an "environment" variable supplied by a custom store for every CRUD operation. One example, could be language (pt, eng, ...) for internationalized text, or channel (web, mobile, ...) for descriptions, etc.

Insert With a Struct

When inserting with a struct, the struct fields are matched with the respective columns.

var pub Publisher
pub.Name = ext.String("Untited Editors")
store.Insert(PUBLISHER).Submit(&pub) // passing as a pointer

The Id and Version fields of the struct are updated if they exist present.

Key fields can be as many as we want. If a key field is of the type (*)int64 and single, it is considered to be a auto generated key.

A shorter version is the quick CRUD operation Create

Insert Returning Generated Key

Any of the above snippets, if the Id field/column is undefined (0 or nil) it returns the generated key by the database.

key, _ := store.Insert(PUBLISHER).
	Columns(PUBLISHER_C_ID, PUBLISHER_C_VERSION, PUBLISHER_C_NAME).
	Values(nil, 1, "New Editions").
	Execute()

Update Examples

Update selected columns with Optimistic lock

store.Update(PUBLISHER).
	Set(PUBLISHER_C_NAME, "Untited Editors"). // column to update
	Set(PUBLISHER_C_VERSION, version + 1).    // increment version
	Where(
	PUBLISHER_C_ID.Matches(1),
	PUBLISHER_C_VERSION.Matches(version),     // old version
).Execute()

Update with struct

When updating with a struct, the struct fields are matched with the respective columns. The presence of a key field is mandatory. If a version column is present its value is also incremented.

var publisher Publisher
publisher.Name = ext.String("Untited Editors")
publisher.Id = ext.Int64(1)      // identifies the record.
publisher.Version = 1   // for optimistic locking
store.Update(PUBLISHER).Submit(&publisher) // passing as a pointer

In this example we are using the publisher.Id as a pointer, but it could be a value. When using values zero is considered unassigned. If that breaks your business logic, use pointers.

The generated SQL will include all matching columns unless the struct has a field tagged with sql:"omit" and its value is a zero value. Generaly sql:"omit" is used for fields with sensitive information, like passwords.

A shorter version is the quick CRUD operation Update

Update with SubQuery

This example shows the use of a subquery to do an update, and also the use of Exists.

sub := store.Query(BOOK).Alias("b").
	Column(AsIs(nil)).
	Where(
	BOOK_C_PUBLISHER_ID.Matches(BOOK_C_ID.For("a")),
	BOOK_C_PRICE.Greater(10),
)

store.Update(PUBLISHER).Alias("a").
	Set(PUBLISHER_C_NAME, Upper(PUBLISHER_C_NAME)).
	Where(Exists(sub)).
	Execute()

Delete Examples

Simple Delete

store.Delete(BOOK).Where(BOOK_C_ID.Matches(2)).Execute()

As we can see the Version column is not taken into account.

Delete with struct

var book Book
book.Id = ext.Int64(2)
book.Version = 1
store.Delete(BOOK).Submit(book)

Althougt we are using Book to execute the delete, any struct with a fields named Id , pointer or not, could be used. Version could also be present. The presence of a key field is mandatory.

A shorter version is the quick CRUD operation Delete

Query Examples

The query operation is by far the richest operation of the ones we have seen.

Query operation that start with Select* retrieve one instance, and those that start with List* returns many instances.

SelectInto

The result of the query is put in the supplied variables. They must be pointers.

var name string
store.Query(PUBLISHER).
	Column(PUBLISHER_C_NAME).
	Where(PUBLISHER_C_ID.Matches(2)).
	SelectInto(&name)

SelectTo

The result of the query is put in the supplied struct pointer.

var publisher Publisher
store.Query(PUBLISHER).
	All().
	Where(PUBLISHER_C_ID.Matches(2)).
	SelectTo(&publisher)

A shorter version is the quick CRUD operation Retrieve

SelectTree

Executes the query and builds a struct tree, reusing previously obtained entities, putting the first element in the supplied struct pointer. When a new entity is needed, the cache is checked to see if there is one instance for this entity, and if found it will use it instead. Since the struct instances are going to be reused it is mandatory that all the structs participating in the result tree implement the toolkit.Hasher interface. Returns true if a result was found, false otherwise.

var publisher Publisher
store.Query(PUBLISHER).
	All().
	Outer(PUBLISHER_A_BOOKS).
	Fetch(). // add all columns off Book in the query
	Where(PUBLISHER_C_ID.Matches(2)).
	SelectTree(&publisher)

SelectFlatTree

Executes the query and builds a flat struct tree putting the first element in the supplied struct pointer.

A flat tree means that one entity will have at most one child, even if related by a one-to-many association. Ex: 1 publisher -> 1 book -> 1 author

Each element of the tree is always a new instance even if representing the same entity. This is most useful to display results in a table. Since the struct instances are not going to be reused it is not mandatory that the structs implement the toolkit.Hasher interface. Returns true if a result was found, false otherwise.

var publisher Publisher
store.Query(PUBLISHER).
	All().
	Outer(PUBLISHER_A_BOOKS).
	Fetch(). // add all columns off Book in the query
	Where(PUBLISHER_C_ID.Matches(2)).
	SelectFlatTree(&publisher)

List

Executes a query, putting the result into a slice.

var books []Book  // []*Book is also valid
store.Query(BOOK).
	All().
	Where(BOOK_C_NAME.Like("%book"))
	List(&books)

As a side note, the WHERE condition can be declared outside the query. Imagine the following example. I want to list published books that have a certain name, that I do not know at runtime. I only want to include the name condition only if the suppliyed name filter in not empty. The code could be something like this:

c := BOOK_C_PUBLISHED.IsNull().Not()
// bookName is some input field unknow at runtime
if bookName != "" {
	// insenstive case like
	c = c.And(T.USER_C_NAME.ILike("%" + bookName + "%"))
}

var names []string  // []*string is also valid
store.Query(BOOK).
	Column(BOOK_C_NAME).
	Where(c).
	List(&names)

ListSimple

Lists simple variables using a closure to assemble the result list, or to do some work. The types for scanning are supplied by the instances parameter.

var names []string
var name string
store.Query(PUBLISHER).
	Column(PUBLISHER_C_NAME).
	ListSimple(func() {
	names = append(names, name)
}, &name)

ListInto

This method is very similar to the previous one, but relies on reflection. A function is used to build the result list. The types for row scanning are supplied by the function parameters. The parameters can be pointers or not. The function can be declared with a return value. This is useful if a returning struct needs to be preprocessed.

names := store.Query(PUBLISHER).
	Column(PUBLISHER_C_NAME).
	ListInto(func(name *string) string {
		return *name;
})

Union has an example of result processing.

We can also use a function accepting a struct, returning or not another struct, to do some processing as shown below.

books := make([]*Book, 0)
store.Query(BOOK).
	All().
	ListInto(func(book *Book) {
	books = append(books, book)
})

The target entity is determined by the receiving type of the function.

ListOf

Another way of executing the above query would be

store.Query(BOOK).
	All().
	ListOf((*Book)(nil))

ListOf returns a collection.Collection interface. There are cases where we might wish to work with a generic list. Later on we will see another use for this new data type.

To traverse the results we use the following code

for e := books.Enumerator(); e.HasNext(); {
	book := e.Next().(*Book)
	// do something
}

ListFlatTree

Executes a query, putting the result in the slice, passed as an argument.

var publishers []*Publisher
store.Query(PUBLISHER).
	All().
	Outer(PUBLISHER_A_BOOKS).
	Fetch(). // add all columns off book in the query
	Where(PUBLISHER_C_ID.Matches(2)).
	ListFlatTree(&publishers)

The slice only contains the head of the tree. A new instance is created for every returned row. This useful if we wish to return a tabular result to use in tables, whithout having to create a struct specificaly for that purpose.

The same can be achieved by the following code.

publishers := make([]*Publisher, 0)
store.Query(PUBLISHER).
	All().
	Outer(PUBLISHER_A_BOOKS).
	Fetch(). // add all columns off book in the query
	Where(PUBLISHER_C_ID.Matches(2)).
	ListFlatTree(func(publisher *Publisher) {
	publishers = append(publishers, publisher)
})

This is useful if we would like to do aditional logic, when building the array. The responsability of building the result is delegated to the receiving function. Union has an example of result processing.

There is another query function named ListFlatTreeOf with the same behaviour as ListFlatTree but returns a collection.

ListTreeOf

Executes a query and transform the results into a tree with the head with the passed struct type. It matches the result column alias with the struct field name, building a struct tree.

When creating the result, previouly fetched entities will be searched and reused if found. A search in a hash collection is faster than search in a slice and that is why a hash that implements the collection.Collectioninterface is returned instead of the classic slice. For a struct to be used in a hash, the struct must implement the toolkit.Hasher interface, as is the case of Book.

publishers, _ := store.Query(PUBLISHER).
	All().
	Outer(PUBLISHER_A_BOOKS).
	Fetch(). // add all columns off book in the query
	Where(PUBLISHER_C_ID.Matches(2)).
	ListTreeOf((*Publisher)(nil))

for e := publishers.Enumerator(); e.HasNext(); {
	publisher := e.Next().(*Publisher)
	// do something here
}

Case Statement

In the following examples we will demonstrate how to declare a Simple CASE statement and a Searched CASE statement.

Simple CASE

Sum all books where the book named "Scrapbook" costs 10 and the others cost 20.

var sale float64
store := TM.Store()
_, err := store.Query(BOOK).
	Column(
	Sum(
		Case(BOOK_C_NAME).
			When("Scrapbook").Then(10).
			Else(AsIs(20)). // showing off AsIs(): value is written as is to the query
			End(),
	),
).SelectInto(&sale)

Searched CASE

Classify the cost of each book.

var dtos []struct {
	Name           string
	Classification string
}

err := store.Query(BOOK).
	Column(BOOK_C_NAME).
	Column(
	If(BOOK_C_PRICE.Greater(20)).Then("expensive").
		If(BOOK_C_PRICE.Range(10, 20)).Then("normal").
		Else("cheap").
		End(),
).As("Classification"). // maps to struct field Classification
	List(&dtos)

Column Subquery

For this example we will use the following struct which will hold the result for each row.

type Dto struct {
	Name      string
	OtherName string
	Value     float64
}

The struct does not represent any table.

The following query gets the name of the publisher and the sum of the prices of books for each publisher (using a subquery to sum), building the result as a slice of Dto.

subquery := store.Query(BOOK).Alias("b").
	Column(Sum(BOOK_C_PRICE)).
	Where(
	BOOK_C_PUBLISHER_ID.Matches(PUBLISHER_C_ID.For("p")),
)

var dtos []*Dto
store.Query(PUBLISHER).Alias("p").
	Column(PUBLISHER_C_NAME).
	Column(subquery).As("Value").
	List(&dtos)

Notice that when I use the subquery variable an alias "Value" is defined. This alias matches with a struct field in Dto. In this query the PUBLISHER_C_NAME column as no associated alias, so the default column alias is used.

Where Subquery

In this example I get a list of records with the name of the Publisher, the name and price of every Book, where the price is lesser or equal than 10. The result is put in a slice of Dto instances. For this a subquery is used in the where clause.

subquery := store.Query(BOOK).
	Distinct().
	Column(BOOK_C_PUBLISHER_ID).
	Where(
	BOOK_C_PRICE.LesserOrMatch(10),
)

var dtos []*Dto
store.Query(PUBLISHER).
	Column(PUBLISHER_C_NAME).
	Inner(PUBLISHER_A_BOOKS).
	Include(BOOK_C_NAME).As("OtherName").
	Include(BOOK_C_PRICE).As("Value").
	Join().
	Where(PUBLISHER_C_ID.In(subquery)).
	List(&dtos)

Joins

The concepts of joins was already introduced in the section SelectTree where we can see the use of an outer join. In the context of goSQL, a Join is seen has a path of associations that goes from the main table to the target table. Along the way we can apply constraints and/or include columns from the participating tables. These paths can overlap without problem because they are seen as isolated from one another. When declaring several paths only when a path deviates from previous path it starts contributing to the SQL generation. Joins can be Outer or Inner and can have constraints applyied to the target table of the last added asscoiation through the use of the function On(). To mark the end of a join definition we use the function Join() or Fetch(). Both process the join but the latter includes in the query all columns from all the tables of the joins. Fetch() is used when a struct tree is desired.

Ex: list all publishers that had a book published before 2013

var publishers []*Publisher
store.Query(PUBLISHER).
	All().
	Distinct().
	Inner(PUBLISHER_A_BOOKS).
	On(BOOK_C_PUBLISHED.Lesser(time.Date(2013, time.January, 1, 0, 0, 0, 0, time.UTC))).
	Join().
	List(&publishers)

The section Where Subquery also shows the use of Include.

The next example executes a (left) outer join and includes ALL columns of the participating tables in the join. The result is a collection of *Publisher structs with its childs in tree.

store.Query(PUBLISHER).
	All().
	Outer(PUBLISHER_A_BOOKS, BOOK_A_AUTHORS).
	Fetch().
	ListTreeOf((*Publisher)(nil))

Fetch() and Join() also reset the path back to the main table, allowing us to declare completly distinct paths.

If we want to retrieve data to populate the tree struct Publisher <- Book -> BookBin, we would have to use two Fetch(), one for each branch, as demonstrated below.

var book Book
store.Query(BOOK).
    All().
    Outer(BOOK_A_PUBLISHER).
    Fetch(). // also resets the path. The next Outer starts AGAIN at BOOK
    Outer(BOOK_A_BOOK_BIN).
    Fetch().
    Where(BOOK_C_ID.Matches(1)).
    SelectTree(&book)

Group By

For this example I will use the struct defined in Column Subquery.

var dtos []*Dto
store.Query(PUBLISHER).
	Column(PUBLISHER_C_NAME).
	Outer(PUBLISHER_A_BOOKS).
	Include(Sum(BOOK_C_PRICE)).As("Value").
	Join().
	GroupByPos(1). // result column position
	List(&dtos)

Having

The criteria used in the Having clause must refer to columns of the Query. This reference is achieved using columns alias. To demonstrate this I will use the following struct which will hold the result for each row.

type PublisherSales struct {
	Name         string
	ThisYear     float64
	PreviousYear float64
}

Struct fields can also be non pointers

var sales []*PublisherSales
store.Query(PUBLISHER).
	Column(PUBLISHER_C_NAME).
	Outer(PUBLISHER_A_BOOKS).
	Include(Sum(BOOK_C_PRICE)).As("ThisYear").
	Join().
	GroupByPos(1).
	Having(Alias("ThisYear").Greater(30)).
	List(&sales)

Order By

List all publishers, ordering ascending by name.

var publishers []*Publisher
store.Query(PUBLISHER).
	All().
	OrderBy(PUBLISHER_C_NAME). // implicit ascending
	List(&publishers)

It is possible to add more orders, and even to order by columns belonging to other tables if joins were present, as seen below.

store.Query(PUBLISHER).
		All().
		Order(PUBLISHER_C_ID).
		Outer(PUBLISHER_A_BOOKS).OrderBy(BOOK_C_ID). // order a column belonging to BOOK
		Outer(BOOK_A_AUTHORS).OrderBy(AUTHOR_C_ID).Desc(). // order a column belonging to AUTHOR
		Fetch(). // this marks the end of the branch and that the results should populate a struct tree
		ListTreeOf((*Publisher)(nil))

The column that the OrderBy refers to, belongs to the table targeted by the last defined association.

If there is no declared association, the column belongs to the driving table, as seen in the first example.

There is also Order if we want to order by a column belonging to the driving table.

Not always we have a well behaved query. We might want a different order without touching the Outer declarations. For that we must use table alias. Below is the same query but using alias.

store.Query(PUBLISHER).
		All().
		Outer(PUBLISHER_A_BOOKS).As("book"). // declares the table alias to use for BOOK
		Outer(BOOK_A_AUTHORS).As("auth"). // declares the table alias to use for AUTHOR
		Fetch().
		Order(PUBLISHER_C_ID). // main table
		OrderAs(BOOK_C_ID.For("book")). // declares the table alias to use
		OrderAs(AUTHOR_C_ID.For("auth")).Desc(). // declares the table alias to use
		ListTreeOf((*Publisher)(nil))

Union

This example list all Publishers and shows side by side the sales of this year and the previous year. The function used in List is responsible for agregating the result.

sales := make([]*PublisherSales, 0)
store.Query(PUBLISHER).
	Column(PUBLISHER_C_ID).
	Column(PUBLISHER_C_NAME).
	Outer(PUBLISHER_A_BOOKS).
	Include(Sum(Coalesce(BOOK_C_PRICE, 0))).As("ThisYear").
	On(
	Range(
		BOOK_C_PUBLISHED,
		time.Date(2013, time.January, 01, 0, 0, 0, 0, time.UTC),
		time.Date(2013, time.December, 31, 23, 59, 59, 1e9-1, time.UTC),
	),
).
	Join().
	Column(AsIs(0)).As("PreviousYear").
	GroupByPos(1).
	UnionAll(
	store.Query(PUBLISHER).Alias("u").
		Column(PUBLISHER_C_ID).
		Column(PUBLISHER_C_NAME).
		Outer(PUBLISHER_A_BOOKS).
		Column(AsIs(0)).As("ThisYear").
		Include(Sum(Coalesce(BOOK_C_PRICE, 0))).As("PreviousYear").
		On(
		Range(
			BOOK_C_PUBLISHED,
			time.Date(2012, time.January, 01, 0, 0, 0, 0, time.UTC),
			time.Date(2012, time.December, 31, 23, 59, 59, 1e9-1, time.UTC),
		),
	).
		Join().
		GroupByPos(1),
).
	List(func(sale *PublisherSales) {
	found := false
	for _, v := range sales {
		if sale.Id == v.Id {
			v.ThisYear += sale.ThisYear
			v.PreviousYear += sale.PreviousYear
			found = true
			break
		}
	}
	if !found {
		sales = append(sales, sale)
	}
})

The alias in the second query is necessary to avoid overlaping replaced parameters between the two queries

Pagination

To paginate the results of a query we use the windowing functions Skip and Limit.

var publishers []*Publisher
store.Query(PUBLISHER).
	All().
	Outer(PUBLISHER_A_BOOKS, BOOK_A_AUTHORS).
	Fetch().
	Order(PUBLISHER_C_NAME).  // implicit ascending
	Skip(2).  // skip the first 2 records
	Limit(3). // limit to 3 records
	ListFlatTree(&publishers)

Converters

Sometimes we may want to store in the database a different representation of the data.

There are two ways of achieveing this.

The native approach

By leveraging the sql.Scanner interface and the driver.Valuer interface we can transform to and from the database. With this approach we can even use the the types that implement the above interfaces in search criterias.

There a couple of minor disadvantages. With this approach we end up polluting the persistence model making it unusuable as a domain model. Another disadvantage is that we have to the raw data that comes from the database. Text can come as a string or as an array of uint8.

type Palette struct {
	EntityBase

	Code  string
	Value *Color
}

type Color struct {
	Red   int
	Green int
	Blue  int
}

func (c *Color) Value() (driver.Value, error) {
	return fmt.Sprintf("%d|%d|%d", c.Red, c.Green, c.Blue), nil
}

func (c *Color) Scan(src interface{}) error {
	s := src.(string) // Postgres is a string but in MySql is []uint8
	rgb := strings.Split(s, "|")
	r, _ := strconv.Atoi(rgb[0])
	g, _ := strconv.Atoi(rgb[1])
	b, _ := strconv.Atoi(rgb[2])
	c.Red = r
	c.Green = g
	c.Blue = b
	return nil
}

Converter approach

Another approach is to use the tag sql:"converter=aName" on the struct field and register on translator the converter, that implements the db.Converter interface.

This approach is less powerful than the previous one because we will not be able to use in search criterias. On the other hand it allows a cleaner model and we don't have to worry about the raw data coming from the database.

type Palette struct {
	EntityBase

	Code  string
	Value *Color `sql:"converter=color"`
}

type Color struct {
	Red   int
	Green int
	Blue  int
}

type ColorConverter struct{}

func (cc ColorConverter) ToDb(in interface{}) (interface{}, error) {
	if in == nil {
		return in, nil
	}
	c := in.(*Color)
	return fmt.Sprintf("%d|%d|%d", c.Red, c.Green, c.Blue), nil
}

func (cc ColorConverter) FromDbInstance() interface{} {
	var s string
	return &s // NB: we need to return a pointer
}

func (cc ColorConverter) FromDb(in interface{}) (interface{}, error) {
	if in == nil {
		return in, nil
	}

	s := in.(*string) // NB: we receive a pointer
	rgb := strings.Split(*s, "|")
	r, _ := strconv.Atoi(rgb[0])
	g, _ := strconv.Atoi(rgb[1])
	b, _ := strconv.Atoi(rgb[2])
	c := &Color{}
	c.Red = r
	c.Green = g
	c.Blue = b
	return &c, nil
}

not forgetting to register the converter

translator.RegisterConverter("color", ColorConverter{})

Embedded Structs

If we want to group columns into a complex type (struct) we just have to tag the field as sql:"embedded". This is handy when working with Value Objects.

Consider the table

CREATE TABLE EMPLOYEE (
	ID SERIAL,
	VERSION INTEGER NOT NULL,
	FIRST_NAME VARCHAR(50),
	LAST_NAME VARCHAR(50),
	PRIMARY KEY(ID)
);

and the structs

type Supervisor struct {
	EntityBase

	FullName FullNameVO `sql:"embedded"`
}

type FullNameVO struct {
	FirstName string
	LastName  string
}

When querying EMPLOYEE table into the Supervisor the data in the columns FIRST_NAME and LAST_NAME will be put in the fields of the FullNameVO struct. The reverse (UPDATE and INSERT) is also true.

Struct Triggers

It is possible to define methods that are called before/after an insert/update/delete/query to the database.

For example, defining the following method will trigger a call for every struct before a insert.

PreInsert(store IDb) error

The same can be done for binding a trigger after an insert using the following signature.

PostInsert(store IDb)

The remaining triggers are:

PreUpdate(store IDb) error
PostUpdate(store IDb)
PreDelete(store IDb) error
PostDelete(store IDb)
PostRetrieve(store IDb)

If an error is returned in a Pre trigger the action is not performed.

To know if a trigger is called inside a transaction use store.InTransaction().

Table Triggers

It is also possible to declare triggers/hooks in the table declaration.

Auditing example:

func init() {
	// pre insert trigger
	BOOK.PreInsertTrigger = func(ins *db.Insert) {
		ins.Set(BOOK_C_VERSION, 1)
		ins.Set(BOOK_C_CREATION, ext.NOW())
		uid, ok := ins.GetDb().GetAttribute(ATTR_USERID)
		if ok {
			ins.Set(BOOK_C_USER_CREATION, uid.(int64))
		}
	}
	// pre update trigger
	BOOK.PreUpdateTrigger = func(upd *db.Update) {
		upd.Set(BOOK_C_MODIFICATION, ext.NOW())
		uid, ok := upd.GetDb().GetAttribute(ATTR_USERID)
		if ok {
			upd.Set(BOOK_C_USER_MODIFICATION, uid.(int64))
		}
	}
}

upd.GetDb() gets a reference to the IDb instance that is unique by transaction.

TODO: explain in more detail

Association Discriminator

An exclusive OR relationship indicates that entity A is related to either entity B or entity C but not both B and C. This is implemented by defining associations with a constraint.

For the example I will use the following database schema:

ER Diagram

As seen in entities.go associations of this type are described as:

PROJECT_A_CONSULTANT = PROJECT.
			ASSOCIATE(PROJECT_C_MANAGER_ID).
			TO(CONSULTANT_C_ID).
			As("Consultant").
			With(PROJECT_C_MANAGER_TYPE, "C")

where the function With declares the constraint applied to this association.

With this in place, its use is the same as regular associations.

store.Query(PROJECT).
	All().
	Inner(PROJECT_A_EMPLOYEE).
	Fetch().
	Order(PROJECT_C_NAME).  // implicit ascending
	ListTreeOf((*Project)(nil))

Table Discriminator

When mapping a table it is possible to declare that the domain of that table only refers to a subset of values of the physical table. This is done by defining a restriction (Discriminator) at the table definition. With this we avoid of having to write a where condition every time we want to refer to a specific domain. Inserts will automatically apply the discriminator.

To demonstrate this I will use a physical table named CATALOG that can hold unrelated information, like gender, eye color, etc. The creation script and table definitions for the next example are at tables_mysql.sql and entities.go respectively.

var statuses []*Status
store.Query(STATUS).
	All().
	List(&statuses)

Custom Functions

The supplied Translators do not have all possible functions of all the databases, but one can register quite easily any missing standard SQL function or even a custom function.

The following steps demonstrates how to add to the Translator, in this case MySQL, and use a function that computes the difference in seconds between two dates.

  1. Define the token name

    const TOKEN_SECONDSDIFF = "SECONDSDIFF"
  2. Register the translation for the token

    translator.RegisterTranslation(
    	TOKEN_SECONDSDIFF,
    	func(dmlType DmlType, token Tokener, tx Translator) string {
    		m := token.GetMembers()
    		return fmt.Sprintf(
    			"TIME_TO_SEC(TIMEDIFF(%s, %s))",
    			tx.Translate(dmlType, m[0]),
    			tx.Translate(dmlType, m[1]),
    		)
    	},
    )
  3. Wrap the token creation in a function for easy use

    func SecondsDiff(left, right interface{}) *Token {
    	return NewToken(TOKEN_SECONDSDIFF, left, right)
    }

Now we are ready to use the new created function, SecondsDiff.

var books []*Book
store.Query(BOOK).
	All().
	Where(
	SecondsDiff(
		time.Date(2013, time.July, 24, 0, 0, 0, 0, time.UTC),
		BOOK_C_PUBLISHED,
	).
		Greater(1000),
).
	List(&books)

Native SQL

It is possible to execute native SQL. The the next example demonstrates the execution of a MySQL query.

// get the database connection
dba := dbx.NewSimpleDBA(TM.Store().GetConnection())
var result []string
_, err := dba.QueryInto("select `name` from `book` where `name` like ?",
	func(name string) { // we calso use pointers: func(name *string)
		result = append(result, name)
	}, "%book")

In the previous example we could have used pointers in the receiving function: func(name *string). The receiving function must have the same number of arguments as the number of columns in the query, in the same order and type. Using native SQL has the drawback of your query not being portable nor easy refactored. For example the prepared statement placeholder for MySQL is '?' while for PostgreSQL is '$1'.

If for some reason you want more control you can use the following.

// get the database connection
dba := dbx.NewSimpleDBA(TM.Store().GetConnection())
result := make([]string, 0)
dba.QueryClosure("select `name` from `book` where `name` like ?",
	func(rows *sql.Rows) error {
		var name string
		if err := rows.Scan(&name); err != nil {
			return err
		}
		result = append(result, name)
		return nil
	}, "%book")

Please see the source code for other methods...

Credits

TODO

  • Add more tests
  • Include more RDBMS
  • Fix code documentation
  • Write idiomatic go