Proteus

A simple tool for generating an application's data access layer.

Purpose

Proteus makes your SQL queries type-safe and prevents SQL injection attacks. It processes structs with struct tags on function fields to generate Go functions at runtime. These functions map input parameters to SQL query parameters and optionally map the output parameters to the output of your SQL queries.

In addition to being type-safe, Proteus also prevents SQL injection by generating prepared statements from your SQL queries. Even dynamic in clauses are converted into injection-proof prepared statements.

Proteus is not an ORM; it does not generate SQL. It just automates away the boring parts of interacting with databases in Go.

Quick Start

Define a struct that contains function fields and tags to indicate the query and the parameter names:

type ProductDaoCtx struct {
	FindByID                      func(ctx context.Context, q proteus.ContextQuerier, id int) (Product, error)                                     `proq:"select * from Product where id = :id:" prop:"id"`
	Update                        func(ctx context.Context, e proteus.ContextExecutor, p Product) (int64, error)                                   `proq:"update Product set name = :p.Name:, cost = :p.Cost: where id = :p.Id:" prop:"p"`
	FindByNameAndCost             func(ctx context.Context, q proteus.ContextQuerier, name string, cost float64) ([]Product, error)                `proq:"select * from Product where name=:name: and cost=:cost:" prop:"name,cost"`
	FindByIDMap                   func(ctx context.Context, q proteus.ContextQuerier, id int) (map[string]interface{}, error)                      `proq:"select * from Product where id = :id:" prop:"id"`
	UpdateMap                     func(ctx context.Context, e proteus.ContextExecutor, p map[string]interface{}) (int64, error)                    `proq:"update Product set name = :p.Name:, cost = :p.Cost: where id = :p.Id:" prop:"p"`
	FindByNameAndCostMap          func(ctx context.Context, q proteus.ContextQuerier, name string, cost float64) ([]map[string]interface{}, error) `proq:"select * from Product where name=:name: and cost=:cost:" prop:"name,cost"`
	Insert                        func(ctx context.Context, e proteus.ContextExecutor, id int, name string, cost *float64) (int64, error)          `proq:"insert into product(id, name, cost) values(:id:, :name:, :cost:)" prop:"id,name,cost"`
	FindByIDSlice                 func(ctx context.Context, q proteus.ContextQuerier, ids []int) ([]Product, error)                                `proq:"select * from Product where id in (:ids:)" prop:"ids"`
	FindByIDSliceAndName          func(ctx context.Context, q proteus.ContextQuerier, ids []int, name string) ([]Product, error)                   `proq:"select * from Product where name = :name: and id in (:ids:)" prop:"ids,name"`
	FindByIDSliceNameAndCost      func(ctx context.Context, q proteus.ContextQuerier, ids []int, name string, cost *float64) ([]Product, error)    `proq:"select * from Product where name = :name: and id in (:ids:) and (cost is null or cost = :cost:)" prop:"ids,name,cost"`
	FindByIDSliceCostAndNameSlice func(ctx context.Context, q proteus.ContextQuerier, ids []int, names []string, cost *float64) ([]Product, error) `proq:"select * from Product where id in (:ids:) and (cost is null or cost = :cost:) and name in (:names:)" prop:"ids,names,cost"`
	FindByNameAndCostUnlabeled    func(ctx context.Context, q proteus.ContextQuerier, name string, cost float64) ([]Product, error)                `proq:"select * from Product where name=:$1: and cost=:$2:"`
}

The first input parameter is of type context.Context, and the second input parameter is of type proteus.ContextExecutor or proteus.ContextQuerier:

// ContextQuerier defines the interface of a type that runs a SQL query with a context
type ContextQuerier interface {
	// QueryContext executes a query that returns rows, typically a SELECT.
	// The args are for any placeholder parameters in the query.
	QueryContext(ctx context.Context, query string, args ...interface{}) (*sql.Rows, error)
}

// ContextExecutor defines the interface of a type that runs a SQL exec with a context
type ContextExecutor interface {
	// ExecContext executes a query without returning any rows.
	// The args are for any placeholder parameters in the query.
	ExecContext(ctx context.Context, query string, args ...interface{}) (sql.Result, error)
}

The remaining input parameters can be primitives, structs, maps of string to interface{}, or slices.

For queries, return types can be:

empty
a single value being returned (a primitive, struct, or map of string to interface{})
a single value that's a slice of primitive, struct, or a map of string to interface{}
a primitive, struct, or map of string to interface{} and an error
a slice of primitive, struct, or a map of string to interface{} and an error

For insert/updates, return types can be:

empty
an int64 that indicates the number of rows affected
a sql.Result
an int64 that indicates the number of rows affected and an error
a sql.Result and an error

The proq struct tag stores the query. You place variable substitutions between : s. Proteus allows you to refer to fields in maps and structs, as well as elements in arrays or slices using . as a path separator. If you have a struct like this:

type Person struct {
    Name string
    Address Address
    Pets []Pet
} 

type Pet struct {
    Name string
    Species string
}

type Address struct {
    Street string
    City string
    State string
}

You can write a query like this:

insert into person(name, city, pet1_name, pet2_name) values (:p.Name:, :p.Address.City:, :p.Pets.0.Name:, :p.Pets.1.Name:)

Note that the index for an array or slice must be an int literal and the key for a map must be a string.

2. If you want to map response fields to a struct, define a struct with struct tags to indicate the mapping:

type Product struct {
	Id   int     `prof:"id"`
	Name string  `prof:"name"`
	Cost float64 `prof:"cost"`
}

3. Pass an instance of the Dao struct to the proteus.ShouldBuild function:

var productDao = ProductDao{}

func init() {
	err := proteus.ShouldBuild(context.Background(), &productDao, proteus.Sqlite)
	if err != nil {
		panic(err)
	}
}

4. Open a connection to a SQL database:

	db := setupDb()
	defer db.Close()

5. Make calls to the function fields in your Proteus-populated struct:

    ctx := context.Background()
	fmt.Println(productDao.FindById(ctx, db, 10))
	p := Product{10, "Thingie", 56.23}
	fmt.Println(productDao.Update(ctx, db, p))
	fmt.Println(productDao.FindById(ctx, db, 10))
	fmt.Println(productDao.FindByNameAndCost(ctx, db, "fred", 54.10))
	fmt.Println(productDao.FindByNameAndCost(ctx, db, "Thingie", 56.23))

	//using a map of [string]interface{} works too!
	fmt.Println(productDao.FindByIdMap(ctx, db, 10))
	fmt.Println(productDao.FindByNameAndCostMap(ctx, db, "Thingie", 56.23))

	fmt.Println(productDao.FindById(ctx, db, 11))
	m := map[string]interface{}{
		"Id":   11,
		"Name": "bobbo",
		"Cost": 12.94,
	}
	fmt.Println(productDao.UpdateMap(ctx, db, m))
	fmt.Println(productDao.FindById(ctx, db, 11))

	fmt.Println(productDao.FindByIDSlice(ctx, db, []int{1, 3, 5}))
	fmt.Println(productDao.FindByIDSliceAndName(ctx, db, []int{1, 3, 5}, "person1"))
	fmt.Println(productDao.FindByIDSliceNameAndCost(ctx, db, []int{1, 3, 5}, "person3", nil))
	fmt.Println(productDao.FindByIDSliceCostAndNameSlice(ctx, db, []int{1, 3, 5}, []string{"person3", "person5"}, nil))

Proteus without the context

If you are using an older database driver that does not work with the context.Context, Proteus provides support for them as well:

type ProductDao struct {
	FindByID                      func(q proteus.Querier, id int) (Product, error)                                     `proq:"select * from Product where id = :id:" prop:"id"`
	Update                        func(e proteus.Executor, p Product) (int64, error)                                   `proq:"update Product set name = :p.Name:, cost = :p.Cost: where id = :p.Id:" prop:"p"`
	FindByNameAndCost             func(q proteus.Querier, name string, cost float64) ([]Product, error)                `proq:"select * from Product where name=:name: and cost=:cost:" prop:"name,cost"`
	FindByIDMap                   func(q proteus.Querier, id int) (map[string]interface{}, error)                      `proq:"select * from Product where id = :id:" prop:"id"`
	UpdateMap                     func(e proteus.Executor, p map[string]interface{}) (int64, error)                    `proq:"update Product set name = :p.Name:, cost = :p.Cost: where id = :p.Id:" prop:"p"`
	FindByNameAndCostMap          func(q proteus.Querier, name string, cost float64) ([]map[string]interface{}, error) `proq:"select * from Product where name=:name: and cost=:cost:" prop:"name,cost"`
	Insert                        func(e proteus.Executor, id int, name string, cost *float64) (int64, error)          `proq:"insert into product(id, name, cost) values(:id:, :name:, :cost:)" prop:"id,name,cost"`
	FindByIDSlice                 func(q proteus.Querier, ids []int) ([]Product, error)                                `proq:"select * from Product where id in (:ids:)" prop:"ids"`
	FindByIDSliceAndName          func(q proteus.Querier, ids []int, name string) ([]Product, error)                   `proq:"select * from Product where name = :name: and id in (:ids:)" prop:"ids,name"`
	FindByIDSliceNameAndCost      func(q proteus.Querier, ids []int, name string, cost *float64) ([]Product, error)    `proq:"select * from Product where name = :name: and id in (:ids:) and (cost is null or cost = :cost:)" prop:"ids,name,cost"`
	FindByIDSliceCostAndNameSlice func(q proteus.Querier, ids []int, names []string, cost *float64) ([]Product, error) `proq:"select * from Product where id in (:ids:) and (cost is null or cost = :cost:) and name in (:names:)" prop:"ids,names,cost"`
	FindByNameAndCostUnlabeled    func(q proteus.Querier, name string, cost float64) ([]Product, error)                `proq:"select * from Product where name=:$1: and cost=:$2:"`
}

In this case, the first input parameter is either of type proteus.Executor or proteus.Querier:

// Executor runs queries that modify the data store.
type Executor interface {
	// Exec executes a query without returning any rows.
	// The args are for any placeholder parameters in the query.
	Exec(query string, args ...interface{}) (sql.Result, error)
}

// Querier runs queries that return Rows from the data store
type Querier interface {
	// Query executes a query that returns rows, typically a SELECT.
	// The args are for any placeholder parameters in the query.
	Query(query string, args ...interface{}) (*sql.Rows, error)
}

Use proteus.Build to generate your DAO functions:

var productDao = ProductDao{}

func init() {
	err := proteus.Build(&productDao, proteus.Sqlite)
	if err != nil {
		panic(err)
	}
}

Struct Tags

Proteus generates implementations of DAO functions by examining struct tags and parameters types on function fields in a struct. The following are the recognized struct tags:

proq - The query. Returns single entity or list of entities
prop - The parameter names. Should be in the order of the function parameters (skipping over the first Executor or Querier parameter)

The prop struct tag is optional. If it is not supplied, the query must contain positional parameters ($1, $2, etc.) instead of named parameters. For example:

type ProductDaoS struct {
	FindById             func(ctx context.Context, q proteus.ContextQuerier, id int) (Product, error)                      `proq:"select * from Product where id = :$1:"`
	Update               func(ctx context.Context, e proteus.ContextExecutor, p Product) (int64, error)                    `proq:"update Product set name = :$1.Name:, cost = :$1.Cost: where id = :$1.Id:"`
	FindByNameAndCost    func(ctx context.Context, q proteus.ContextQuerier, name string, cost float64) ([]Product, error) `proq:"select * from Product where name=:$1: and cost=:$2:"`
}

If you want to map the output of a DAO with a proq tag to a struct, then create a struct and put the following struct tag on each field that you want to map to a value in the output:

prof - The fields on the dto that are mapped to select parameters in a query

Storing queries outside of struct tags

Struct tags are cumbersome for all but the shortest queries. In order to allow a more natural way to store longer queries, one or more instances of the proteus.QueryMapper interface can be passed into the proteus.Build function. In order to reference a query stored in an proteus.QueryMapper, you should put q:name as the value of the proq struct tag, where name is the name for the query.

For example:

	m := proteus.MapMapper{
		"q1": "select * from foo where id = :id:",
		"q2": "update foo set x=:x: where id = :id:",
	}

	type s struct {
		GetF   func(ctx context.Context, q proteus.ContextQuerier, id string) (f, error)                `proq:"q:q1" prop:"id"`
		Update func(ctx context.Context, e proteus.ContextExecutor, id string, x string) (int64, error) `proq:"q:q2" prop:"id,x"`
	}
	
	sImpl := s{}
	err := proteus.ShouldBuild(context.Background(), &sImpl, proteus.Sqlite, m)

Out of the box, you can use either a map[string]string or a unix properties file to store your queries. In order to use a map[string]string, cast your map to proteus.MapMapper (or just declare your variable to be of type proteus.MapMapper). To use a properties file, call the method proteus.PropFileToQueryMapper with the name of the property file that contains your queries.

Generating function variables

Some people don't want to use structs and struct tags to implement their SQL mapping layer. Starting with version 0.11.0, Proteus can also generate functions that aren't fields in a struct.

First, create an instance of a proteus.Builder. The factory function takes a proteus.Adapter and zero or more proteus.QueryMapper instances:

    b := NewBuilder(Postgres)

Next, declare a function variable with the signature you want. The parameters for the function variables follow the same rules as the function fields:

    var f func(c context.Context, e ContextExecutor, name string, age int) (int64, error)
    var g func(c context.Context, q ContextQuerier, id int) (*Person, error)

Then call the BuildFunction method on your proteus.Builder instance, passing in a pointer to your function variable, the SQL query (or a query mapper reference), and the parameter names as a string slice:

    err := b.BuildFunction(ctx, &f, "INSERT INTO PERSON(name, age) VALUES(:name:, :age:)", []string{"name", "age"})
    if err != nil {
        t.Fatalf("build function failed: %v", err)
    }
    
    err = b.BuildFunction(ctx, &g, "SELECT * FROM PERSON WHERE id = :id:", []string{"id"})
    if err != nil {
        t.Fatalf("build function 2 failed: %v", err)
    }

Finally, call your functions, to run your SQL queries:

    db := setupDbPostgres()
    defer db.Close()
    ctx := context.Background()
    
    rows, err := f(ctx, db, "Fred", 20)
    if err != nil {
        t.Fatalf("create failed: %v", err)
    }
    fmt.Println(rows) // prints 1
    
    p, err := g(ctx, db, 1)
    if err != nil {
        t.Fatalf("get failed: %v", err)
    }
    fmt.Println(p) // prints {1, Fred, 20}

Ad-hoc database queries

While Proteus is focused on type safety, sometimes you just want to run a query without associating it with a function. Starting with version 0.11.0, Proteus allows you to run ad-hoc database queries.

First, create an instance of a proteus.Builder. The factory function takes a proteus.Adapter and zero or more proteus.QueryMapper instances:

    b := NewBuilder(Postgres)

Next, run your query by passing it to the Exec, ExecResult, or Query methods on proteus.Builder.

Exec expects a context.Context, a proteus.ContextExecutor, the query, and a map of string to interface{}, where the keys are the parameter names and the values are the parameter values. It returns an int64 with the number of rows modified and and error.

ExecResult expects a context.Context, a proteus.ContextExecutor, the query, and a map of string to interface{}, where the keys are the parameter names and the values are the parameter values. It returns a sql.Result with the number of rows modified and and error.

Query expected a context.Context, a proteus.ContextQuerier, the query, a map of string to interface{}, where the keys are the parameter names and the values are the parameter values, and a pointer to the value that should be populated by the query. The method returns an error.

    db := setupDbPostgres()
    defer db.Close()
    ctx := context.Background()

    rows, err := b.Exec(c, db, "INSERT INTO PERSON(name, age) VALUES(:name:, :age:)", map[string]interface{}{"name": "Fred", "age": 20})
    if err != nil {
        t.Fatalf("create failed: %v", err)
    } 
    fmt.Println(rows) // prints 1

    var p *Person
    err = b.Query(c, db, "SELECT * FROM PERSON WHERE id = :id:", map[string]interface{}{"id": 1}, &p)
    if err != nil {
        t.Fatalf("get failed: %v", err)
    }
    fmt.Println(*p) // prints {1, Fred, 20}

Ad-hoc queries support all of the functionality of Proteus except for type safety. You can reference queries in proteus.QueryMapper instances, build out dynamic in clauses, extract values from struct instances, and map to a struct with prof tags on its fields.

Valid function signatures

API

FAQ

1. Why doesn't Proteus generate a struct that meets an interface?

Go has some surprising limitations. One of them is that there is no way to use reflection to build a implementation of a method; you can only build implementations of functions. The difference is subtle, but the net result is that you cannot supply an interface to the reflection API and get back something that implements that interface.

Another go limitation is that there is also no way to attach metadata to an interface's method. only struct fields can have associated metadata.

A third limitation is that a go interface is not satisfied by a struct that has fields of function type, even if the names of the functions and the types of the function parameters match an interface.

Given these limitations, Proteus uses structs to hold the generated functions. If you want an interface that describes the functionality provided by the struct, you can do something like this:

type ProductDaoS struct {
	FindById             func(ctx context.Context, q proteus.ContextQuerier, id int) (Product, error)                       `proq:"select * from Product where id = :id:" prop:"id"`
	Update               func(ctx context.Context, e proteus.ContextExecutor, p Product) (int64, error)                     `proq:"update Product set name = :p.Name:, cost = :p.Cost: where id = :p.Id:" prop:"p"`
	FindByNameAndCost    func(ctx context.Context, q proteus.ContextQuerier, name string, cost float64) ([]Product, error)  `proq:"select * from Product where name=:name: and cost=:cost:" prop:"name,cost"`
}

type ProductDao interface {
    FindById(context.Context, proteus.ContextQuerier, int) (Product, error)
    Update(context.Context, proteus.ContextExecutor, Product) (int64, error)
    FindByNameAndCost(context.Context, proteus.ContextQuerier, string, float64) ([]Product, error)
}

type productWrapper struct {
    ProductDaoS
}

func (pw productWrapper) FindById(ctx context.Context, q proteus.ContextQuerier, id int) (Product, error) {
    return pw.ProductDaoS.FindById(ctx, q, id)
}

func (pw productWrapper) Update(ctx context.Context, e proteus.ContextExecutor, p Product) (int64, error) {
    return pw.ProductDaoS.Update(ctx, e, p)
}

func (pw productWrapper) FindByNameAndCost(ctx context.Context, q proteus.ContextQuerier, n string, c float64) ([]Product, error) {
    return pw.ProductDaoS.FindByNameAndCost(ctx, q, n, c)
}

func NewProductDao(ctx context.Context) ProductDao {
    p := ProductDaoS{}
    err := proteus.ShouldBuild(ctx, &p,proteus.Sqlite)
    if err != nil {
        panic(err)
    }
    return productWrapper{p}
}

A future version of Proteus may include a tool that can be used with go generate to automatically create the wrapper and interface.

2. Why do I have to specify the parameter names with a struct tag?

This is another limitation of go's reflection API. The names of parameters are not available at runtime to be inspected, and must be supplied by another way in order to be referenced in a query. If you do not want to use a prop struct tag, you can use positional parameters ($1, $2, etc.) instead.

Logging

In order to avoid tying the client code to one particular implementation, Proteus includes its own logger that can be bridged to any other Go logger.

By default, Proteus will log nothing. The simplest way to change the amount of information logged by Proteus is by calling the function proteus.SetLogLevel. This function takes in a value of type logger.Level.

If you are using the proteus.ShouldBuild function to generate your DAOs, you can supply a logging level by passing in the context returned by the function logger.WithLevel . This value is overridden by the logging level set by proteus.SetLogLevel.

You can also include additional values in the logs by passing in the context returned by the function logger.WithValues. This function adds one or more logger.Pair values to the context.

If you are using the context.Context support in Proteus 0.10.0 and later, the logger level can be supplied in the context passed into the function. This will override any logger level specified when proteus.ShouldBuild was invoked. You can also supply additional logging fields for the function call by using a context returned by the logger.WithValues function.

All Proteus logging output include 3 default fields (in order):

time (value is a time.Time in UTC)
level (the level of the log as a logger.Level)
message (the message supplied with the log

The default logger provided with the Proteus logging package outputs to os.Stdout in a JSON format. The default logger implementation is configurable; you can specify a different io.Writer by using the call:

    logger.Config(logger.DefaultLogger{
        Writer: myWriterImpl,
    })

The format of the output of the DefaultLogger can be configured by supplying a logger.Formatter:

    logger.Config(logger.DefaultLogger{
        Writer: myWriterImpl,
        Formatter: myFormatter,
    })

A helper type logger.FormatterFunc will turn any function with the signature func(vals ...interface{}) string into a logger.Formatter.

The first 6 values passed to the Format method are:

position	value
0	"time"
1	a `time.Time` in UTC
2	"level"
3	a `logger.Level`
4	"message"
5	the message for the log

If you want to supply your own logger implementation, pass an implementation of logger.Logger into logger.Config. This interface matches the definition used by go-kit. The default parameters are in the order specified above. There is a logger.LoggerFunc helper type to convert any function with the signature of func(vals ...interface{}) error into a logger.Logger.

Feel free to use this logger within your own code. If this logger proves to be useful, it might be broken into its own top-level package.

Future Directions

There are more interesting features coming to Proteus. They are (in likely order of implementation):

Build in clauses using a field from a slice of struct or map
Generate batch values clauses using a slice of struct or map
more expansive performance measurement support and per-request logging control
go generate tool to create a wrapper struct and interface for a Proteus DAO
API implementations for nonSQL data stores and HTTP requests
Additional struct tags to generate basic CRUD queries automatically. This is about as far as I'd go in implementing ORM-like features. The goal for Proteus is to be an adapter between a data source and business logic, not to define your entire programming model.

faruihuihui/proteus