Go database/sql defines interfaces for SQL databases. Actual driver must be implemented in own package. And dependency injection is done as a part of import and build system. Lets go deeper to see how it is actually implemented.

This exercise will result in a simple driver backed up by csv files. Who do not want to SELECT from csv file?

Disclaimer: I wrote it in order to learn database/sql and to learn more about design of APIs for go. I regularly use deprecated methods, which have advanced variants in recent standard library. Consult documentation if you want to know more.

I have had three goals in mind

  1. Learning
  2. Simplicity
  3. 100 lines of code

database/sql intro

Here is small snippet of Go program using Postgres https://github.com/lib/pq/ driver to talk to database. It use generic methods of database/sql interfaces, so switching the driver and fixing SQL dialects, we can talk to any database system without a need to change much of the code.

import ( "database/sql" _ "github.com/lib/pq"
) ... db, err := sql.Open("postgres", dbURI) if err != nil { panic(err) } defer db.Close() _, err := db.Exec("CREATE TABLE IF NOT EXISTS TBL2(id BIGSERIAL)") if err != nil { panic(err) }

Magic inside

There is a magic inside

import ( "fmt" "database/sql" _ "github.com/lib/pq" _ "github.com/go-sql-driver/mysql"
) func main() { fmt.Printf("Drivers=%#v\n", sql.Drivers())
Drivers=[]string{"mysql", "postgres"}

Driver is magically registered during module import. It sounds like a magic, isn't? And it turns out it is actually very simple. Function init() is package initializer, which can run arbitrary code before all other code. In case of SQL driver, it contains the code to register itself.

func init() { sql.Register("postgres", &Driver{})

Similary for mysql

func init() { sql.Register("mysql", &MySQLDriver{})

Registering SQL Driver

With all those knowledge one can implement dummy SQL driver

package main import ( "fmt" "database/sql"
) type DummyDriver struct {
} // implements sql.Driver interface
func (d *DummyDriver) Open(name string) (driver.Conn, error) { fmt.Printf("[dummy]: open(%s)", name) return nil, nil
} func main() { // manually register dummy one sql.Register("dummy", &DummyDriver{}) // List registered drivers, now with dummy fmt.Printf("Drivers=%#v\n", sql.Drivers())

And to test it has been registered ...


Of course it is really dummy driver as it does not offer ANY functionality.


As we see, Driver interface returns driver.Conn, which is yet another interface. This defines three methods, which can be skipped for the csv case.

First make Open real function, so let is open the file

type DummyConn struct { name string
} // implements sql.Driver interface
// Open - open file for reading, return the connection
func (d *DummyDriver) Open(name string) (driver.Conn, error) { file, err := os.Open(name) if err != nil { return nil, err } defer file.Close() return &DummyConn{name}, nil

Here one can see the beauty of Go error reporting. File operation errors the same way as database server errors.

This is a bit uncommon for real SQL driver, however we do not have server to connect to. So Open tries to open the file at least to return file access errors as early as possible. All the heavy lifting is done below.

Here is bunch of methods not interesting for the use case ... all those returns not implemented error.

// implements driver.Conn interface
// Prepare - not implemented ...
func (c *DummyConn) Prepare(query string) (driver.Stmt, error) { return nil, fmt.Errorf("Prepare method not implemented")
// implements driver.Tx interface
// Rollback - not implemented ...
func (c *DummyConn) Rollback() error { return fmt.Errorf("Rollback method not implemented")

So for read-only file access the connection is really dumb method, there is only one file path, which needs to be passed.

Do the query

SQL queries are base of the system. Here is a constraint by supporting only one query string. However I do prefer brevity over anything else, so here we are. The driver.Queryer interface

// Queryer interface
func (c *DummyConn) Query(query string, args []driver.Value) (driver.Rows, error) { if query != "SELECT * FROM csv" { return nil, fmt.Errorf("Only `SELECT * FROM csv` string is implemented!") } r := csv.NewReader(c.file) r.FieldsPerRecord = 0 // enforce the same number of columns columns, err := r.Read() // assume first line gives you column names if err != nil { return nil, err } res := &results{r, columns} return res, nil

Again, nothing really complicated. One needs to handle errors and then return correct structure with proper implemented interface methods.

Query methods returns an object, which allows read of the result. Real database systems deal well with the read and write locks, concurrent access and more. You have maybe heard about database cursor. In a case of read only csv file, the situation is way more simpler.

As each Query can pass independent results, opened file is a part of results. So each results will have independent file handle and file position, which is an equivalent of database cursor.

Read the results

One needs to implement next interface. driver.Querier returns object with driver.Rows interface. Here is structure results used to read data through Query and Scan methods. It provides enough to implement Rows interface

type results struct { reader *csv.Reader columns []string

And implementation is fairly straightforward. We do assume that first line of csv file contains names of columns. So this is initialized as a part of query. It simplify Columns[] method a lot!

// driver.Rows interface
func (r *results) Columns() []string { return r.columns

And to not leak file handles, close them once results are read

func (r *results) Close() error { return r.file.Close()

And the most important method. Under the hood it reads data from csv file and put them to internal buffer dest[], which is then managed by database/sql code. This is the way how data ends up in Scan method later on.

func (r *results) Next(dest []driver.Value) error { d, err := r.reader.Read() if err != nil { return err } for i := 0; i != len(r.columns); i++ { dest[i] = driver.Value(d[i]) } return nil

Running together

 // manually register dummy driver sql.Register("dummy", &DummyDriver{}) ... // prepare two return objects rows := make([]*sql.Rows, 2) for i := 0; i != 2; i++ { rows[i], err = db.Query("SELECT * FROM csv") if err != nil { panic(err) } } // read results for _, r := range rows { for r.Next() { var f1, f2, f3 string err := r.Scan(&f1, &f2, &f3) if err != nil { panic(err) } fmt.Printf("first_name=%s, last_name=%s, username=%s\n", f1, f2, f3) } }

And a proof, result of running program

$ go run main.go first_name=Rob, last_name=Pike, username=rob
first_name=Ken, last_name=Thompson, username=ken
first_name=Robert, last_name=Griesemer, username=gri
first_name=Rob, last_name=Pike, username=rob
first_name=Ken, last_name=Thompson, username=ken
first_name=Robert, last_name=Griesemer, username=gri


Complete program is available at github.com/vyskocilm/gazpacho/dbmagic. And as last method, Next ends at line 100. The goal to implement dummy sql driver in 100 lines of Go code was achieved.

After going through the article reader shall be able

  1. To understand the automagic registration of SQL drivers
  2. To understand the design of database/sql better
  3. Hopefully enjoy the simplicity, but expressiveness of Go interfaces
  4. Learn that design of interfaces is hard and database/sql offer V2 interfaces in some cases.

Reddit thread