dex/vendor/gopkg.in/gorp.v1/README.md

673 lines
19 KiB
Markdown
Raw Normal View History

# Go Relational Persistence
[![build status](https://secure.travis-ci.org/go-gorp/gorp.png)](http://travis-ci.org/go-gorp/gorp)
I hesitate to call gorp an ORM. Go doesn't really have objects, at least
not in the classic Smalltalk/Java sense. There goes the "O". gorp doesn't
know anything about the relationships between your structs (at least not
yet). So the "R" is questionable too (but I use it in the name because,
well, it seemed more clever).
The "M" is alive and well. Given some Go structs and a database, gorp
should remove a fair amount of boilerplate busy-work from your code.
I hope that gorp saves you time, minimizes the drudgery of getting data
in and out of your database, and helps your code focus on algorithms,
not infrastructure.
* Bind struct fields to table columns via API or tag
* Support for embedded structs
* Support for transactions
* Forward engineer db schema from structs (great for unit tests)
* Pre/post insert/update/delete hooks
* Automatically generate insert/update/delete statements for a struct
* Automatic binding of auto increment PKs back to struct after insert
* Delete by primary key(s)
* Select by primary key(s)
* Optional trace sql logging
* Bind arbitrary SQL queries to a struct
* Bind slice to SELECT query results without type assertions
* Use positional or named bind parameters in custom SELECT queries
* Optional optimistic locking using a version column (for update/deletes)
## Installation
# install the library:
go get gopkg.in/gorp.v1
// use in your .go code:
import (
"gopkg.in/gorp.v1"
)
## Versioning
This project provides a stable release (v1.x tags) and a bleeding edge codebase (master).
`gopkg.in/gorp.v1` points to the latest v1.x tag. The API's for v1 are stable and shouldn't change. Development takes place at the master branch. Althought the code in master should always compile and test successfully, it might break API's. We aim to maintain backwards compatibility, but API's and behaviour might be changed to fix a bug. Also note that API's that are new in the master branch can change until released as v2.
If you want to use bleeding edge, use `github.com/go-gorp/gorp` as import path.
## API Documentation
Full godoc output from the latest v1 release is available here:
https://godoc.org/gopkg.in/gorp.v1
For the latest code in master:
https://godoc.org/github.com/go-gorp/gorp
## Quickstart
```go
package main
import (
"database/sql"
"gopkg.in/gorp.v1"
_ "github.com/mattn/go-sqlite3"
"log"
"time"
)
func main() {
// initialize the DbMap
dbmap := initDb()
defer dbmap.Db.Close()
// delete any existing rows
err := dbmap.TruncateTables()
checkErr(err, "TruncateTables failed")
// create two posts
p1 := newPost("Go 1.1 released!", "Lorem ipsum lorem ipsum")
p2 := newPost("Go 1.2 released!", "Lorem ipsum lorem ipsum")
// insert rows - auto increment PKs will be set properly after the insert
err = dbmap.Insert(&p1, &p2)
checkErr(err, "Insert failed")
// use convenience SelectInt
count, err := dbmap.SelectInt("select count(*) from posts")
checkErr(err, "select count(*) failed")
log.Println("Rows after inserting:", count)
// update a row
p2.Title = "Go 1.2 is better than ever"
count, err = dbmap.Update(&p2)
checkErr(err, "Update failed")
log.Println("Rows updated:", count)
// fetch one row - note use of "post_id" instead of "Id" since column is aliased
//
// Postgres users should use $1 instead of ? placeholders
// See 'Known Issues' below
//
err = dbmap.SelectOne(&p2, "select * from posts where post_id=?", p2.Id)
checkErr(err, "SelectOne failed")
log.Println("p2 row:", p2)
// fetch all rows
var posts []Post
_, err = dbmap.Select(&posts, "select * from posts order by post_id")
checkErr(err, "Select failed")
log.Println("All rows:")
for x, p := range posts {
log.Printf(" %d: %v\n", x, p)
}
// delete row by PK
count, err = dbmap.Delete(&p1)
checkErr(err, "Delete failed")
log.Println("Rows deleted:", count)
// delete row manually via Exec
_, err = dbmap.Exec("delete from posts where post_id=?", p2.Id)
checkErr(err, "Exec failed")
// confirm count is zero
count, err = dbmap.SelectInt("select count(*) from posts")
checkErr(err, "select count(*) failed")
log.Println("Row count - should be zero:", count)
log.Println("Done!")
}
type Post struct {
// db tag lets you specify the column name if it differs from the struct field
Id int64 `db:"post_id"`
Created int64
Title string
Body string
}
func newPost(title, body string) Post {
return Post{
Created: time.Now().UnixNano(),
Title: title,
Body: body,
}
}
func initDb() *gorp.DbMap {
// connect to db using standard Go database/sql API
// use whatever database/sql driver you wish
db, err := sql.Open("sqlite3", "/tmp/post_db.bin")
checkErr(err, "sql.Open failed")
// construct a gorp DbMap
dbmap := &gorp.DbMap{Db: db, Dialect: gorp.SqliteDialect{}}
// add a table, setting the table name to 'posts' and
// specifying that the Id property is an auto incrementing PK
dbmap.AddTableWithName(Post{}, "posts").SetKeys(true, "Id")
// create the table. in a production system you'd generally
// use a migration tool, or create the tables via scripts
err = dbmap.CreateTablesIfNotExists()
checkErr(err, "Create tables failed")
return dbmap
}
func checkErr(err error, msg string) {
if err != nil {
log.Fatalln(msg, err)
}
}
```
## Examples
### Mapping structs to tables
First define some types:
```go
type Invoice struct {
Id int64
Created int64
Updated int64
Memo string
PersonId int64
}
type Person struct {
Id int64
Created int64
Updated int64
FName string
LName string
}
// Example of using tags to alias fields to column names
// The 'db' value is the column name
//
// A hyphen will cause gorp to skip this field, similar to the
// Go json package.
//
// This is equivalent to using the ColMap methods:
//
// table := dbmap.AddTableWithName(Product{}, "product")
// table.ColMap("Id").Rename("product_id")
// table.ColMap("Price").Rename("unit_price")
// table.ColMap("IgnoreMe").SetTransient(true)
//
type Product struct {
Id int64 `db:"product_id"`
Price int64 `db:"unit_price"`
IgnoreMe string `db:"-"`
}
```
Then create a mapper, typically you'd do this one time at app startup:
```go
// connect to db using standard Go database/sql API
// use whatever database/sql driver you wish
db, err := sql.Open("mymysql", "tcp:localhost:3306*mydb/myuser/mypassword")
// construct a gorp DbMap
dbmap := &gorp.DbMap{Db: db, Dialect: gorp.MySQLDialect{"InnoDB", "UTF8"}}
// register the structs you wish to use with gorp
// you can also use the shorter dbmap.AddTable() if you
// don't want to override the table name
//
// SetKeys(true) means we have a auto increment primary key, which
// will get automatically bound to your struct post-insert
//
t1 := dbmap.AddTableWithName(Invoice{}, "invoice_test").SetKeys(true, "Id")
t2 := dbmap.AddTableWithName(Person{}, "person_test").SetKeys(true, "Id")
t3 := dbmap.AddTableWithName(Product{}, "product_test").SetKeys(true, "Id")
```
### Struct Embedding
gorp supports embedding structs. For example:
```go
type Names struct {
FirstName string
LastName string
}
type WithEmbeddedStruct struct {
Id int64
Names
}
es := &WithEmbeddedStruct{-1, Names{FirstName: "Alice", LastName: "Smith"}}
err := dbmap.Insert(es)
```
See the `TestWithEmbeddedStruct` function in `gorp_test.go` for a full example.
### Create/Drop Tables ###
Automatically create / drop registered tables. This is useful for unit tests
but is entirely optional. You can of course use gorp with tables created manually,
or with a separate migration tool (like goose: https://bitbucket.org/liamstask/goose).
```go
// create all registered tables
dbmap.CreateTables()
// same as above, but uses "if not exists" clause to skip tables that are
// already defined
dbmap.CreateTablesIfNotExists()
// drop
dbmap.DropTables()
```
### SQL Logging
Optionally you can pass in a logger to trace all SQL statements.
I recommend enabling this initially while you're getting the feel for what
gorp is doing on your behalf.
Gorp defines a `GorpLogger` interface that Go's built in `log.Logger` satisfies.
However, you can write your own `GorpLogger` implementation, or use a package such
as `glog` if you want more control over how statements are logged.
```go
// Will log all SQL statements + args as they are run
// The first arg is a string prefix to prepend to all log messages
dbmap.TraceOn("[gorp]", log.New(os.Stdout, "myapp:", log.Lmicroseconds))
// Turn off tracing
dbmap.TraceOff()
```
### Insert
```go
// Must declare as pointers so optional callback hooks
// can operate on your data, not copies
inv1 := &Invoice{0, 100, 200, "first order", 0}
inv2 := &Invoice{0, 100, 200, "second order", 0}
// Insert your rows
err := dbmap.Insert(inv1, inv2)
// Because we called SetKeys(true) on Invoice, the Id field
// will be populated after the Insert() automatically
fmt.Printf("inv1.Id=%d inv2.Id=%d\n", inv1.Id, inv2.Id)
```
### Update
Continuing the above example, use the `Update` method to modify an Invoice:
```go
// count is the # of rows updated, which should be 1 in this example
count, err := dbmap.Update(inv1)
```
### Delete
If you have primary key(s) defined for a struct, you can use the `Delete`
method to remove rows:
```go
count, err := dbmap.Delete(inv1)
```
### Select by Key
Use the `Get` method to fetch a single row by primary key. It returns
nil if no row is found.
```go
// fetch Invoice with Id=99
obj, err := dbmap.Get(Invoice{}, 99)
inv := obj.(*Invoice)
```
### Ad Hoc SQL
#### SELECT
`Select()` and `SelectOne()` provide a simple way to bind arbitrary queries to a slice
or a single struct.
```go
// Select a slice - first return value is not needed when a slice pointer is passed to Select()
var posts []Post
_, err := dbmap.Select(&posts, "select * from post order by id")
// You can also use primitive types
var ids []string
_, err := dbmap.Select(&ids, "select id from post")
// Select a single row.
// Returns an error if no row found, or if more than one row is found
var post Post
err := dbmap.SelectOne(&post, "select * from post where id=?", id)
```
Want to do joins? Just write the SQL and the struct. gorp will bind them:
```go
// Define a type for your join
// It *must* contain all the columns in your SELECT statement
//
// The names here should match the aliased column names you specify
// in your SQL - no additional binding work required. simple.
//
type InvoicePersonView struct {
InvoiceId int64
PersonId int64
Memo string
FName string
}
// Create some rows
p1 := &Person{0, 0, 0, "bob", "smith"}
dbmap.Insert(p1)
// notice how we can wire up p1.Id to the invoice easily
inv1 := &Invoice{0, 0, 0, "xmas order", p1.Id}
dbmap.Insert(inv1)
// Run your query
query := "select i.Id InvoiceId, p.Id PersonId, i.Memo, p.FName " +
"from invoice_test i, person_test p " +
"where i.PersonId = p.Id"
// pass a slice to Select()
var list []InvoicePersonView
_, err := dbmap.Select(&list, query)
// this should test true
expected := InvoicePersonView{inv1.Id, p1.Id, inv1.Memo, p1.FName}
if reflect.DeepEqual(list[0], expected) {
fmt.Println("Woot! My join worked!")
}
```
#### SELECT string or int64
gorp provides a few convenience methods for selecting a single string or int64.
```go
// select single int64 from db (use $1 instead of ? for postgresql)
i64, err := dbmap.SelectInt("select count(*) from foo where blah=?", blahVal)
// select single string from db:
s, err := dbmap.SelectStr("select name from foo where blah=?", blahVal)
```
#### Named bind parameters
You may use a map or struct to bind parameters by name. This is currently
only supported in SELECT queries.
```go
_, err := dbm.Select(&dest, "select * from Foo where name = :name and age = :age", map[string]interface{}{
"name": "Rob",
"age": 31,
})
```
#### UPDATE / DELETE
You can execute raw SQL if you wish. Particularly good for batch operations.
```go
res, err := dbmap.Exec("delete from invoice_test where PersonId=?", 10)
```
### Transactions
You can batch operations into a transaction:
```go
func InsertInv(dbmap *DbMap, inv *Invoice, per *Person) error {
// Start a new transaction
trans, err := dbmap.Begin()
if err != nil {
return err
}
trans.Insert(per)
inv.PersonId = per.Id
trans.Insert(inv)
// if the commit is successful, a nil error is returned
return trans.Commit()
}
```
### Hooks
Use hooks to update data before/after saving to the db. Good for timestamps:
```go
// implement the PreInsert and PreUpdate hooks
func (i *Invoice) PreInsert(s gorp.SqlExecutor) error {
i.Created = time.Now().UnixNano()
i.Updated = i.Created
return nil
}
func (i *Invoice) PreUpdate(s gorp.SqlExecutor) error {
i.Updated = time.Now().UnixNano()
return nil
}
// You can use the SqlExecutor to cascade additional SQL
// Take care to avoid cycles. gorp won't prevent them.
//
// Here's an example of a cascading delete
//
func (p *Person) PreDelete(s gorp.SqlExecutor) error {
query := "delete from invoice_test where PersonId=?"
err := s.Exec(query, p.Id); if err != nil {
return err
}
return nil
}
```
Full list of hooks that you can implement:
PostGet
PreInsert
PostInsert
PreUpdate
PostUpdate
PreDelete
PostDelete
All have the same signature. for example:
func (p *MyStruct) PostUpdate(s gorp.SqlExecutor) error
### Optimistic Locking
gorp provides a simple optimistic locking feature, similar to Java's JPA, that
will raise an error if you try to update/delete a row whose `version` column
has a value different than the one in memory. This provides a safe way to do
"select then update" style operations without explicit read and write locks.
```go
// Version is an auto-incremented number, managed by gorp
// If this property is present on your struct, update
// operations will be constrained
//
// For example, say we defined Person as:
type Person struct {
Id int64
Created int64
Updated int64
FName string
LName string
// automatically used as the Version col
// use table.SetVersionCol("columnName") to map a different
// struct field as the version field
Version int64
}
p1 := &Person{0, 0, 0, "Bob", "Smith", 0}
dbmap.Insert(p1) // Version is now 1
obj, err := dbmap.Get(Person{}, p1.Id)
p2 := obj.(*Person)
p2.LName = "Edwards"
dbmap.Update(p2) // Version is now 2
p1.LName = "Howard"
// Raises error because p1.Version == 1, which is out of date
count, err := dbmap.Update(p1)
_, ok := err.(gorp.OptimisticLockError)
if ok {
// should reach this statement
// in a real app you might reload the row and retry, or
// you might propegate this to the user, depending on the desired
// semantics
fmt.Printf("Tried to update row with stale data: %v\n", err)
} else {
// some other db error occurred - log or return up the stack
fmt.Printf("Unknown db err: %v\n", err)
}
```
## Database Drivers
gorp uses the Go 1 `database/sql` package. A full list of compliant drivers is available here:
http://code.google.com/p/go-wiki/wiki/SQLDrivers
Sadly, SQL databases differ on various issues. gorp provides a Dialect interface that should be
implemented per database vendor. Dialects are provided for:
* MySQL
* PostgreSQL
* sqlite3
Each of these three databases pass the test suite. See `gorp_test.go` for example
DSNs for these three databases.
Support is also provided for:
* Oracle (contributed by @klaidliadon)
* SQL Server (contributed by @qrawl) - use driver: github.com/denisenkom/go-mssqldb
Note that these databases are not covered by CI and I (@coopernurse) have no good way to
test them locally. So please try them and send patches as needed, but expect a bit more
unpredicability.
## Known Issues
### SQL placeholder portability
Different databases use different strings to indicate variable placeholders in
prepared SQL statements. Unlike some database abstraction layers (such as JDBC),
Go's `database/sql` does not standardize this.
SQL generated by gorp in the `Insert`, `Update`, `Delete`, and `Get` methods delegates
to a Dialect implementation for each database, and will generate portable SQL.
Raw SQL strings passed to `Exec`, `Select`, `SelectOne`, `SelectInt`, etc will not be
parsed. Consequently you may have portability issues if you write a query like this:
```go
// works on MySQL and Sqlite3, but not with Postgresql
err := dbmap.SelectOne(&val, "select * from foo where id = ?", 30)
```
In `Select` and `SelectOne` you can use named parameters to work around this.
The following is portable:
```go
err := dbmap.SelectOne(&val, "select * from foo where id = :id",
map[string]interface{} { "id": 30})
```
### time.Time and time zones
gorp will pass `time.Time` fields through to the `database/sql` driver, but note that
the behavior of this type varies across database drivers.
MySQL users should be especially cautious. See: https://github.com/ziutek/mymysql/pull/77
To avoid any potential issues with timezone/DST, consider using an integer field for time
data and storing UNIX time.
## Running the tests
The included tests may be run against MySQL, Postgresql, or sqlite3.
You must set two environment variables so the test code knows which driver to
use, and how to connect to your database.
```sh
# MySQL example:
export GORP_TEST_DSN=gomysql_test/gomysql_test/abc123
export GORP_TEST_DIALECT=mysql
# run the tests
go test
# run the tests and benchmarks
go test -bench="Bench" -benchtime 10
```
Valid `GORP_TEST_DIALECT` values are: "mysql", "postgres", "sqlite3"
See the `test_all.sh` script for examples of all 3 databases. This is the script I run
locally to test the library.
## Performance
gorp uses reflection to construct SQL queries and bind parameters. See the BenchmarkNativeCrud vs BenchmarkGorpCrud in gorp_test.go for a simple perf test. On my MacBook Pro gorp is about 2-3% slower than hand written SQL.
## Help/Support
IRC: #gorp
Mailing list: gorp-dev@googlegroups.com
Bugs/Enhancements: Create a github issue
## Pull requests / Contributions
Contributions are very welcome. Please follow these guidelines:
* Fork the `master` branch and issue pull requests targeting the `master` branch
* If you are adding an enhancement, please open an issue first with your proposed change.
* Changes that break backwards compatibility in the public API are only accepted after we
discuss on a GitHub issue for a while.
Thanks!
## Contributors
* matthias-margush - column aliasing via tags
* Rob Figueiredo - @robfig
* Quinn Slack - @sqs