Files
ent/dialect/sql/schema/sqlite.go
Jannik Clausen 7017cbc898 dialect/sql/schema: file based type store (#2644)
* dialect/sql/schema: file based type store

This PR adds support for a file based type storage when using versioned migrations. The file called `.ent_types` is written to the migration directory alongside the migration files and will be kept in sync for every migration file generation run.

In order to not break existing code, where the type storage might differ for different deployment, global unique ID mut be enabled by using a new option. This will also be raised as an error to the user when attempting to use versioned migrations and global unique ID.

Documentation will be added to this PR once feedback on the code is gathered.

* apply CR

* fix tests

* change format of types file to exclude it from atlas.sum file

* docs and drift test

* apply CR
2022-06-15 16:10:15 +02:00

459 lines
14 KiB
Go

// Copyright 2019-present Facebook Inc. All rights reserved.
// This source code is licensed under the Apache 2.0 license found
// in the LICENSE file in the root directory of this source tree.
package schema
import (
"context"
"fmt"
"strconv"
"strings"
"entgo.io/ent/dialect"
"entgo.io/ent/dialect/sql"
"entgo.io/ent/schema/field"
"ariga.io/atlas/sql/migrate"
"ariga.io/atlas/sql/schema"
"ariga.io/atlas/sql/sqlite"
)
// SQLite is an SQLite migration driver.
type SQLite struct {
dialect.Driver
WithForeignKeys bool
}
// init makes sure that foreign_keys support is enabled.
func (d *SQLite) init(ctx context.Context, tx dialect.ExecQuerier) error {
on, err := exist(ctx, tx, "PRAGMA foreign_keys")
if err != nil {
return fmt.Errorf("sqlite: check foreign_keys pragma: %w", err)
}
if !on {
// foreign_keys pragma is off, either enable it by execute "PRAGMA foreign_keys=ON"
// or add the following parameter in the connection string "_fk=1".
return fmt.Errorf("sqlite: foreign_keys pragma is off: missing %q in the connection string", "_fk=1")
}
return nil
}
func (d *SQLite) tableExist(ctx context.Context, conn dialect.ExecQuerier, name string) (bool, error) {
query, args := sql.Select().Count().
From(sql.Table("sqlite_master")).
Where(sql.And(
sql.EQ("type", "table"),
sql.EQ("name", name),
)).
Query()
return exist(ctx, conn, query, args...)
}
// setRange sets the start value of table PK.
// SQLite tracks the AUTOINCREMENT in the "sqlite_sequence" table that is created and initialized automatically
// whenever a table that contains an AUTOINCREMENT column is created. However, it populates to it a rows (for tables)
// only after the first insertion. Therefore, we check. If a record (for the given table) already exists in the "sqlite_sequence"
// table, we updated it. Otherwise, we insert a new value.
func (d *SQLite) setRange(ctx context.Context, conn dialect.ExecQuerier, t *Table, value int64) error {
query, args := sql.Select().Count().
From(sql.Table("sqlite_sequence")).
Where(sql.EQ("name", t.Name)).
Query()
exists, err := exist(ctx, conn, query, args...)
switch {
case err != nil:
return err
case exists:
query, args = sql.Update("sqlite_sequence").Set("seq", value).Where(sql.EQ("name", t.Name)).Query()
default: // !exists
query, args = sql.Insert("sqlite_sequence").Columns("name", "seq").Values(t.Name, value).Query()
}
return conn.Exec(ctx, query, args, nil)
}
func (d *SQLite) tBuilder(t *Table) *sql.TableBuilder {
b := sql.CreateTable(t.Name)
for _, c := range t.Columns {
b.Column(d.addColumn(c))
}
if t.Annotation != nil {
addChecks(b, t.Annotation)
}
// Unlike in MySQL, we're not able to add foreign-key constraints to table
// after it was created, and adding them to the `CREATE TABLE` statement is
// not always valid (because circular foreign-keys situation is possible).
// We stay consistent by not using constraints at all, and just defining the
// foreign keys in the `CREATE TABLE` statement.
if d.WithForeignKeys {
for _, fk := range t.ForeignKeys {
b.ForeignKeys(fk.DSL())
}
}
// If it's an ID based primary key with autoincrement, we add
// the `PRIMARY KEY` clause to the column declaration. Otherwise,
// we append it to the constraint clause.
if len(t.PrimaryKey) == 1 && t.PrimaryKey[0].Increment {
return b
}
for _, pk := range t.PrimaryKey {
b.PrimaryKey(pk.Name)
}
return b
}
// cType returns the SQLite string type for the given column.
func (*SQLite) cType(c *Column) (t string) {
if c.SchemaType != nil && c.SchemaType[dialect.SQLite] != "" {
return c.SchemaType[dialect.SQLite]
}
switch c.Type {
case field.TypeBool:
t = "bool"
case field.TypeInt8, field.TypeUint8, field.TypeInt16, field.TypeUint16, field.TypeInt32,
field.TypeUint32, field.TypeUint, field.TypeInt, field.TypeInt64, field.TypeUint64:
t = "integer"
case field.TypeBytes:
t = "blob"
case field.TypeString, field.TypeEnum:
// SQLite does not impose any length restrictions on
// the length of strings, BLOBs or numeric values.
t = fmt.Sprintf("varchar(%d)", DefaultStringLen)
case field.TypeFloat32, field.TypeFloat64:
t = "real"
case field.TypeTime:
t = "datetime"
case field.TypeJSON:
t = "json"
case field.TypeUUID:
t = "uuid"
case field.TypeOther:
t = c.typ
default:
panic(fmt.Sprintf("unsupported type %q for column %q", c.Type, c.Name))
}
return t
}
// addColumn returns the DSL query for adding the given column to a table.
func (d *SQLite) addColumn(c *Column) *sql.ColumnBuilder {
b := sql.Column(c.Name).Type(d.cType(c)).Attr(c.Attr)
c.unique(b)
if c.PrimaryKey() && c.Increment {
b.Attr("PRIMARY KEY AUTOINCREMENT")
}
c.nullable(b)
c.defaultValue(b)
return b
}
// addIndex returns the query for adding an index to SQLite.
func (d *SQLite) addIndex(i *Index, table string) *sql.IndexBuilder {
return i.Builder(table).IfNotExists()
}
// dropIndex drops a SQLite index.
func (d *SQLite) dropIndex(ctx context.Context, tx dialect.Tx, idx *Index, table string) error {
query, args := idx.DropBuilder("").Query()
return tx.Exec(ctx, query, args, nil)
}
// fkExist returns always true to disable foreign-keys creation after the table was created.
func (d *SQLite) fkExist(context.Context, dialect.Tx, string) (bool, error) { return true, nil }
// table returns always error to indicate that SQLite dialect doesn't support incremental migration.
func (d *SQLite) table(ctx context.Context, tx dialect.Tx, name string) (*Table, error) {
rows := &sql.Rows{}
query, args := sql.Select("name", "type", "notnull", "dflt_value", "pk").
From(sql.Table(fmt.Sprintf("pragma_table_info('%s')", name)).Unquote()).
OrderBy("pk").
Query()
if err := tx.Query(ctx, query, args, rows); err != nil {
return nil, fmt.Errorf("sqlite: reading table description %w", err)
}
// Call Close in cases of failures (Close is idempotent).
defer rows.Close()
t := NewTable(name)
for rows.Next() {
c := &Column{}
if err := d.scanColumn(c, rows); err != nil {
return nil, fmt.Errorf("sqlite: %w", err)
}
if c.PrimaryKey() {
t.PrimaryKey = append(t.PrimaryKey, c)
}
t.AddColumn(c)
}
if err := rows.Err(); err != nil {
return nil, err
}
if err := rows.Close(); err != nil {
return nil, fmt.Errorf("sqlite: closing rows %w", err)
}
indexes, err := d.indexes(ctx, tx, name)
if err != nil {
return nil, err
}
// Add and link indexes to table columns.
for _, idx := range indexes {
switch {
case idx.primary:
case idx.Unique && len(idx.columns) == 1:
name := idx.columns[0]
c, ok := t.column(name)
if !ok {
return nil, fmt.Errorf("index %q column %q was not found in table %q", idx.Name, name, t.Name)
}
c.Key = UniqueKey
c.Unique = true
fallthrough
default:
t.addIndex(idx)
}
}
return t, nil
}
// table loads the table indexes from the database.
func (d *SQLite) indexes(ctx context.Context, tx dialect.Tx, name string) (Indexes, error) {
rows := &sql.Rows{}
query, args := sql.Select("name", "unique", "origin").
From(sql.Table(fmt.Sprintf("pragma_index_list('%s')", name)).Unquote()).
Query()
if err := tx.Query(ctx, query, args, rows); err != nil {
return nil, fmt.Errorf("reading table indexes %w", err)
}
defer rows.Close()
var idx Indexes
for rows.Next() {
i := &Index{}
origin := sql.NullString{}
if err := rows.Scan(&i.Name, &i.Unique, &origin); err != nil {
return nil, fmt.Errorf("scanning index description %w", err)
}
i.primary = origin.String == "pk"
idx = append(idx, i)
}
if err := rows.Err(); err != nil {
return nil, err
}
if err := rows.Close(); err != nil {
return nil, fmt.Errorf("closing rows %w", err)
}
for i := range idx {
columns, err := d.indexColumns(ctx, tx, idx[i].Name)
if err != nil {
return nil, err
}
idx[i].columns = columns
// Normalize implicit index names to ent naming convention. See:
// https://github.com/sqlite/sqlite/blob/e937df8/src/build.c#L3583
if len(columns) == 1 && strings.HasPrefix(idx[i].Name, "sqlite_autoindex_"+name) {
idx[i].Name = columns[0]
}
}
return idx, nil
}
// indexColumns loads index columns from index info.
func (d *SQLite) indexColumns(ctx context.Context, tx dialect.Tx, name string) ([]string, error) {
rows := &sql.Rows{}
query, args := sql.Select("name").
From(sql.Table(fmt.Sprintf("pragma_index_info('%s')", name)).Unquote()).
OrderBy("seqno").
Query()
if err := tx.Query(ctx, query, args, rows); err != nil {
return nil, fmt.Errorf("reading table indexes %w", err)
}
defer rows.Close()
var names []string
if err := sql.ScanSlice(rows, &names); err != nil {
return nil, err
}
return names, nil
}
// scanColumn scans the column information from SQLite column description.
func (d *SQLite) scanColumn(c *Column, rows *sql.Rows) error {
var (
pk sql.NullInt64
notnull sql.NullInt64
defaults sql.NullString
)
if err := rows.Scan(&c.Name, &c.typ, &notnull, &defaults, &pk); err != nil {
return fmt.Errorf("scanning column description: %w", err)
}
c.Nullable = notnull.Int64 == 0
if pk.Int64 > 0 {
c.Key = PrimaryKey
}
if c.typ == "" {
return fmt.Errorf("missing type information for column %q", c.Name)
}
parts, size, _, err := parseColumn(c.typ)
if err != nil {
return err
}
switch strings.ToLower(parts[0]) {
case "bool", "boolean":
c.Type = field.TypeBool
case "blob":
c.Type = field.TypeBytes
case "integer":
// All integer types have the same "type affinity".
c.Type = field.TypeInt
case "real", "float", "double":
c.Type = field.TypeFloat64
case "datetime":
c.Type = field.TypeTime
case "json":
c.Type = field.TypeJSON
case "uuid":
c.Type = field.TypeUUID
case "varchar", "char", "text":
c.Size = size
c.Type = field.TypeString
case "decimal", "numeric":
c.Type = field.TypeOther
}
if defaults.Valid {
return c.ScanDefault(defaults.String)
}
return nil
}
// alterColumns returns the queries for applying the columns change-set.
func (d *SQLite) alterColumns(table string, add, _, _ []*Column) sql.Queries {
queries := make(sql.Queries, 0, len(add))
for i := range add {
c := d.addColumn(add[i])
if fk := add[i].foreign; fk != nil {
c.Constraint(fk.DSL())
}
queries = append(queries, sql.Dialect(dialect.SQLite).AlterTable(table).AddColumn(c))
}
// Modifying and dropping columns is not supported and disabled until we
// will support https://www.sqlite.org/lang_altertable.html#otheralter
return queries
}
// tables returns the query for getting the in the schema.
func (d *SQLite) tables() sql.Querier {
return sql.Select("name").
From(sql.Table("sqlite_schema")).
Where(sql.EQ("type", "table"))
}
// needsConversion reports if column "old" needs to be converted
// (by table altering) to column "new".
func (d *SQLite) needsConversion(old, new *Column) bool {
c1, c2 := d.cType(old), d.cType(new)
return c1 != c2 && old.typ != c2
}
// Atlas integration.
func (d *SQLite) atOpen(conn dialect.ExecQuerier) (migrate.Driver, error) {
return sqlite.Open(&db{ExecQuerier: conn})
}
func (d *SQLite) atTable(t1 *Table, t2 *schema.Table) {
if t1.Annotation != nil {
setAtChecks(t1, t2)
}
}
func (d *SQLite) atTypeC(c1 *Column, c2 *schema.Column) error {
if c1.SchemaType != nil && c1.SchemaType[dialect.SQLite] != "" {
t, err := sqlite.ParseType(strings.ToLower(c1.SchemaType[dialect.SQLite]))
if err != nil {
return err
}
c2.Type.Type = t
return nil
}
var t schema.Type
switch c1.Type {
case field.TypeBool:
t = &schema.BoolType{T: "bool"}
case field.TypeInt8, field.TypeUint8, field.TypeInt16, field.TypeUint16, field.TypeInt32,
field.TypeUint32, field.TypeUint, field.TypeInt, field.TypeInt64, field.TypeUint64:
t = &schema.IntegerType{T: sqlite.TypeInteger}
case field.TypeBytes:
t = &schema.BinaryType{T: sqlite.TypeBlob}
case field.TypeString, field.TypeEnum:
// SQLite does not impose any length restrictions on
// the length of strings, BLOBs or numeric values.
t = &schema.StringType{T: sqlite.TypeText}
case field.TypeFloat32, field.TypeFloat64:
t = &schema.FloatType{T: sqlite.TypeReal}
case field.TypeTime:
t = &schema.TimeType{T: "datetime"}
case field.TypeJSON:
t = &schema.JSONType{T: "json"}
case field.TypeUUID:
t = &sqlite.UUIDType{T: "uuid"}
case field.TypeOther:
t = &schema.UnsupportedType{T: c1.typ}
default:
t, err := sqlite.ParseType(strings.ToLower(c1.typ))
if err != nil {
return err
}
c2.Type.Type = t
}
c2.Type.Type = t
return nil
}
func (d *SQLite) atUniqueC(t1 *Table, c1 *Column, t2 *schema.Table, c2 *schema.Column) {
// For UNIQUE columns, SQLite create an implicit index named
// "sqlite_autoindex_<table>_<i>". Ent uses the PostgreSQL approach
// in its migration, and name these indexes as "<table>_<column>_key".
for _, idx := range t1.Indexes {
// Index also defined explicitly, and will be add in atIndexes.
if idx.Unique && d.atImplicitIndexName(idx, t1, c1) {
return
}
}
t2.AddIndexes(schema.NewUniqueIndex(fmt.Sprintf("%s_%s_key", t2.Name, c1.Name)).AddColumns(c2))
}
func (d *SQLite) atImplicitIndexName(idx *Index, t1 *Table, c1 *Column) bool {
if idx.Name == c1.Name {
return true
}
p := fmt.Sprintf("sqlite_autoindex_%s_", t1.Name)
if !strings.HasPrefix(idx.Name, p) {
return false
}
i, err := strconv.ParseInt(strings.TrimPrefix(idx.Name, p), 10, 64)
return err == nil && i > 0
}
func (d *SQLite) atIncrementC(_ *schema.Table, c *schema.Column) {
c.AddAttrs(&sqlite.AutoIncrement{})
}
func (d *SQLite) atIncrementT(t *schema.Table, v int64) {
t.AddAttrs(&sqlite.AutoIncrement{Seq: v})
}
func (d *SQLite) atIndex(idx1 *Index, t2 *schema.Table, idx2 *schema.Index) error {
for _, c1 := range idx1.Columns {
c2, ok := t2.Column(c1.Name)
if !ok {
return fmt.Errorf("unexpected index %q column: %q", idx1.Name, c1.Name)
}
idx2.AddParts(&schema.IndexPart{C: c2})
}
return nil
}
func (SQLite) atTypeRangeSQL(ts ...string) string {
for i := range ts {
ts[i] = fmt.Sprintf("('%s')", ts[i])
}
return fmt.Sprintf("INSERT INTO `%s` (`type`) VALUES %s", TypeTable, strings.Join(ts, ", "))
}