.circumflex-orm.2.5.source-code.vendor.scala Maven / Gradle / Ivy
The newest version!
package ru.circumflex
package orm
import java.sql._
/*!# Vendor-specific SQL dialects
Following vendors are currently supported by Circumflex ORM:
* PostgreSQL 8.3+;
* MySQL 5.7+;
* H2 database.
We also provide limited support for `Oracle`, `MS SQL Server` and `DB2`.
We would appreciate any commits for better vendors support.
*/
class H2Dialect extends Dialect {
override def driverClass = "org.h2.Driver"
override def textType = "VARCHAR"
override def createIndex(idx: Index): String = {
var result = "CREATE "
if (idx.isUnique) result += "UNIQUE "
result += "INDEX " + idx.name + " ON " + idx.relation.qualifiedName +
" (" + idx.expression + ")"
if (idx.whereClause != EmptyPredicate)
result += " WHERE " + idx.whereClause.toInlineSql
result
}
override def dropSchema(schema: Schema) = "DROP SCHEMA " + schema.name
}
class PostgreSQLDialect extends Dialect {
override def driverClass = "org.postgresql.Driver"
override def timestampType = "TIMESTAMPTZ"
override def parameterizedIn(ex: String, params: Iterable[String]) = {
if (params.size == 0) "FALSE"
else ex + " IN (" + params.mkString(", ") + ")"
}
}
class MySQLDialect extends Dialect {
override def supportsSchema = false
override def driverClass = "com.mysql.jdbc.Driver"
override def initializeField[R <: Record[_, R]](field: Field[_, R]) {
// do nothing -- for MySQL you don't need to create manually a sequence for auto-incrementable fields
}
override def defaultExpression[R <: Record[_, R]](field: Field[_, R]): String =
field match {
case a: AutoIncrementable[_, _] if (a.isAutoIncrement) =>" AUTO_INCREMENT"
case _ => field.defaultExpression.map(" DEFAULT " + _).getOrElse("")
}
override def identityLastIdPredicate[PK, R <: Record[PK, R]](node: RelationNode[PK, R]): Predicate =
new SimpleExpression(node.alias + "." + node.relation.PRIMARY_KEY.name + " = LAST_INSERT_ID()", Nil)
override def identityLastIdQuery[PK, R <: Record[PK, R]](node: RelationNode[PK, R]): SQLQuery[PK] =
new Select(expr[PK]("LAST_INSERT_ID()"))
override def sequenceNextValQuery[PK, R <: Record[PK, R]](node: RelationNode[PK, R]): SQLQuery[PK] =
throw new UnsupportedOperationException("This operation is unsupported in MySQL dialect.")
override def createIndex(idx: Index): String = {
var result = "CREATE "
if (idx.isUnique) result += "UNIQUE "
result += "INDEX " + idx.name + " USING " + idx.usingClause +
" ON " + idx.relation.qualifiedName + " (" + idx.expression + ")"
if (idx.whereClause != EmptyPredicate)
ORM_LOG.warn("Ignoring WHERE clause of INDEX " + idx.name +
": predicates are not supported.")
result
}
override def delete[PK, R <: Record[PK, R]](dml: Delete[PK, R]): String = {
var result = "DELETE " + dml.node.alias + " FROM " + dml.node.toSql
if (dml.whereClause != EmptyPredicate) result += " WHERE " + dml.whereClause.toSql
result
}
}
class OracleDialect extends Dialect {
override def driverClass = "oracle.jdbc.driver.OracleDriver"
override def fkNoAction = "SET_NULL"
override def fkRestrict = "SET NULL"
override def fkSetDefault = "SET NULL"
override def numericType(precision: Int, scale: Int): String =
"NUMBER" + (if (precision == -1) "" else "(" + precision + "," + scale + ")")
override def textType = "VARCHAR2(4000)"
override def varcharType(length: Int): String =
"VARCHAR2" + (if (length == -1) "" else "(" + length + ")")
override def booleanType = "NUMBER(1)"
override def supportsSchema = false
override def createIndex(idx: Index): String = {
var result = "CREATE "
if (idx.isUnique) result += "UNIQUE "
result += "INDEX " + idx.name +
" ON " + idx.relation.qualifiedName + " (" + idx.expression + ")"
if (idx.whereClause != EmptyPredicate)
ORM_LOG.warn("Ignoring WHERE clause of INDEX " + idx.name +
": predicates are not supported.")
result
}
override def defaultExpression[R <: Record[_, R]](field: Field[_, R]): String =
field.defaultExpression.map(" DEFAULT " + _).getOrElse("")
override def sequenceNextValQuery[PK, R <: Record[PK, R]](node: RelationNode[PK, R]): SQLQuery[PK] =
new Select(expr[PK](sequenceName(node.relation.PRIMARY_KEY) + ".nextval FROM dual"))
override def identityLastIdPredicate[PK, R <: Record[PK, R]](node: RelationNode[PK, R]): Predicate =
throw new UnsupportedOperationException("This operation is unsupported in Oracle dialect.")
override def identityLastIdQuery[PK, R <: Record[PK, R]](node: RelationNode[PK, R]): SQLQuery[PK] =
throw new UnsupportedOperationException("This operation is unsupported in Oracle dialect.")
}
class DB2Dialect extends Dialect {
override def driverClass = "com.ibm.db2.jcc.DB2Driver"
override def prepareStatement(conn: Connection, sql: String): PreparedStatement =
conn.prepareStatement(sql, ResultSet.TYPE_SCROLL_INSENSITIVE)
override def textType = "VARCHAR(4000)"
override def booleanType = "SMALLINT"
override def defaultExpression[R <: Record[_, R]](field: Field[_, R]): String =
field match {
case a: AutoIncrementable[_, _] if (a.isAutoIncrement) => " GENERATED BY DEFAULT AS IDENTITY"
case _ => field.defaultExpression.map(" DEFAULT " + _).getOrElse("")
}
override def sequenceNextValQuery[PK, R <: Record[PK, R]](node: RelationNode[PK, R]): SQLQuery[PK] =
new NativeSQLQuery[PK](expr[PK]("NEXTVAL FOR " + sequenceName(node.relation.PRIMARY_KEY)),
prepareExpr("SELECT {*} FROM SYSIBM.SYSDUMMY1"))
override def identityLastIdPredicate[PK, R <: Record[PK, R]](node: RelationNode[PK, R]): Predicate =
new SimpleExpression(node.alias + "." + node.relation.PRIMARY_KEY.name + " = IDENTITY_VAL_LOCAL()", Nil)
override def identityLastIdQuery[PK, R <: Record[PK, R]](node: RelationNode[PK, R]): SQLQuery[PK] =
new NativeSQLQuery[PK](expr[PK]("IDENTITY_VAL_LOCAL()"),
prepareExpr("SELECT {*} FROM SYSIBM.SYSDUMMY1"))
override def createIndex(idx: Index): String = {
var result = "CREATE "
if (idx.isUnique) result += "UNIQUE "
result += "INDEX " + idx.name +
" ON " + idx.relation.qualifiedName + " (" + idx.expression + ")"
if (idx.whereClause != EmptyPredicate)
ORM_LOG.warn("Ignoring WHERE clause of INDEX " + idx.name +
": predicates are not supported.")
result
}
override def dropSchema(schema: Schema) = "DROP SCHEMA " + schema.name + " RESTRICT"
override def insert[PK, R <: Record[PK, R]](dml: Insert[PK, R]): String = {
var result = "INSERT INTO " + dml.relation.qualifiedName
if (dml.fields.size > 0)
result += " (" + dml.fields.map(_.name).mkString(", ") +
") VALUES (" + dml.fields.map(_.placeholder).mkString(", ") + ")"
else result += " VALUES DEFAULT"
result
}
}
class MSSQLDialect extends Dialect {
override def driverClass = "com.microsoft.sqlserver.jdbc.SQLServerDriver"
override def booleanType = "BIT"
override def supportsSchema = false
override def defaultExpression[R <: Record[_, R]](field: Field[_, R]): String =
field match {
case a: AutoIncrementable[_, _] if (a.isAutoIncrement) =>" IDENTITY"
case _ => field.defaultExpression.map(" DEFAULT " + _).getOrElse("")
}
override def columnDefinition[R <: Record[_, R]](field: Field[_, R]): String = {
var result = field.name + " " + field.sqlType
result += defaultExpression(field)
if (field.isNotNull) result += " NOT NULL"
result
}
override def createIndex(idx: Index): String = {
var result = "CREATE "
if (idx.isUnique) result += "UNIQUE "
result += "INDEX " + idx.name + " ON " + idx.relation.qualifiedName +
" (" + idx.expression + ")"
if (idx.whereClause != EmptyPredicate)
result += " WHERE " + idx.whereClause.toInlineSql
result
}
override def sequenceNextValQuery[PK, R <: Record[PK, R]](node: RelationNode[PK, R]): SQLQuery[PK] =
throw new UnsupportedOperationException("This operation is unsupported in MS SQL Server dialect.")
override def identityLastIdPredicate[PK, R <: Record[PK, R]](node: RelationNode[PK, R]): Predicate =
new SimpleExpression(node.alias + "." + node.relation.PRIMARY_KEY.name + " = @@IDENTITY", Nil)
override def identityLastIdQuery[PK, R <: Record[PK, R]](node: RelationNode[PK, R]): SQLQuery[PK] =
new Select(expr[PK]("@@IDENTITY"))
}