liquibase.change.AbstractSQLChange Maven / Gradle / Ivy
Go to download
Show more of this group Show more artifacts with this name
Show all versions of liquibase-core Show documentation
Show all versions of liquibase-core Show documentation
Liquibase is a tool for managing and executing database changes.
package liquibase.change;
import liquibase.change.core.RawSQLChange;
import liquibase.Scope;
import liquibase.GlobalConfiguration;
import liquibase.database.Database;
import liquibase.database.core.MSSQLDatabase;
import liquibase.database.core.PostgresDatabase;
import liquibase.exception.DatabaseException;
import liquibase.exception.UnexpectedLiquibaseException;
import liquibase.exception.ValidationErrors;
import liquibase.exception.Warnings;
import liquibase.statement.SqlStatement;
import liquibase.statement.core.RawSqlStatement;
import liquibase.util.StringUtil;
import java.io.*;
import java.util.ArrayList;
import java.util.List;
/**
* A common parent for all raw SQL related changes regardless of where the sql was sourced from.
*
* Implements the necessary logic to choose how the SQL string should be parsed to generate the statements.
*
*/
@SuppressWarnings("java:S5998")
public abstract class AbstractSQLChange extends AbstractChange implements DbmsTargetedChange {
private boolean stripComments;
private boolean splitStatements;
/**
*
* @deprecated To be removed when splitStatements is changed to be type Boolean
*
*/
private boolean splitStatementsSet;
private String endDelimiter;
private String sql;
private String dbms;
protected String encoding;
protected AbstractSQLChange() {
setStripComments(null);
setSplitStatements(null);
}
public InputStream openSqlStream() throws IOException {
return null;
}
@Override
@DatabaseChangeProperty(since = "3.0", exampleValue = "h2, oracle")
public String getDbms() {
return dbms;
}
@Override
public void setDbms(final String dbms) {
this.dbms = dbms;
}
/**
* {@inheritDoc}
* @param database
* @return always true (in AbstractSQLChange)
*/
@Override
public boolean supports(Database database) {
return true;
}
@Override
public Warnings warn(Database database) {
return new Warnings();
}
@Override
public ValidationErrors validate(Database database) {
ValidationErrors validationErrors = new ValidationErrors();
if (StringUtil.trimToNull(sql) == null) {
validationErrors.addError("'sql' is required");
}
return validationErrors;
}
/**
* Return if comments should be stripped from the SQL before passing it to the database.
*
* This will always return a non-null value and should be a boolean rather than a Boolean, but that breaks the Bean Standard.
*/
@DatabaseChangeProperty(description = "Set to true to remove any comments in the SQL before executing, otherwise false. Defaults to false if not set")
public Boolean isStripComments() {
return stripComments;
}
/**
* Return true if comments should be stripped from the SQL before passing it to the database.
* Passing null sets stripComments to the default value (false).
*/
public void setStripComments(Boolean stripComments) {
if (stripComments == null) {
this.stripComments = false;
} else {
this.stripComments = stripComments;
}
}
/**
* Return if the SQL should be split into multiple statements before passing it to the database.
* By default, statements are split around ";" and "go" delimiters.
*
* This will always return a non-null value and should be a boolean rather than a Boolean, but that breaks the Bean Standard.
*/
@DatabaseChangeProperty(description = "Set to false to not have liquibase split statements on ;'s and GO's. Defaults to true if not set")
public Boolean isSplitStatements() {
return splitStatements;
}
/**
* Set whether SQL should be split into multiple statements.
* Passing null sets stripComments to the default value (true).
*/
public void setSplitStatements(Boolean splitStatements) {
if (splitStatements == null) {
this.splitStatements = true;
} else {
this.splitStatements = splitStatements;
splitStatementsSet = true;
}
}
/**
* @deprecated To be removed when splitStatements is changed to be Boolean type
* @return
*/
@Deprecated
public boolean isSplitStatementsSet() {
return splitStatementsSet;
}
/**
* Return the raw SQL managed by this Change
*/
@DatabaseChangeProperty(serializationType = SerializationType.DIRECT_VALUE)
public String getSql() {
return sql;
}
/**
* Set the raw SQL managed by this Change. The passed sql is trimmed and set to null if an empty string is passed.
*/
public void setSql(String sql) {
this.sql = StringUtil.trimToNull(sql);
}
/**
* Set the end delimiter used to split statements. Will return null if the default delimiter should be used.
*
* @see #splitStatements
*/
@DatabaseChangeProperty(description = "Delimiter to apply to the end of the statement. Defaults to ';', may be set to ''.", exampleValue = "\\nGO")
public String getEndDelimiter() {
return endDelimiter;
}
/**
* Set the end delimiter for splitting SQL statements. Set to null to use the default delimiter.
* @param endDelimiter
*/
public void setEndDelimiter(String endDelimiter) {
this.endDelimiter = endDelimiter;
}
/**
* Calculates the checksum based on the contained SQL.
*
* @see liquibase.change.AbstractChange#generateCheckSum()
*/
@Override
public CheckSum generateCheckSum() {
InputStream stream = null;
try {
stream = openSqlStream();
String sql = this.sql;
if ((stream == null) && (sql == null)) {
sql = "";
}
if (sql != null) {
stream = new ByteArrayInputStream(sql.getBytes(GlobalConfiguration.OUTPUT_FILE_ENCODING.getCurrentValue())
);
}
return CheckSum.compute(new NormalizingStream(this.getEndDelimiter(), this.isSplitStatements(), this.isStripComments(), stream), false);
} catch (IOException e) {
throw new UnexpectedLiquibaseException(e);
} finally {
if (stream != null) {
try {
stream.close();
} catch (IOException e) {
Scope.getCurrentScope().getLog(getClass()).fine("Error closing stream", e);
}
}
}
}
/**
* Generates one or more SqlStatements depending on how the SQL should be parsed.
* If split statements is set to true then the SQL is split and each command is made into a separate SqlStatement.
*
* If stripping comments is true then any comments are removed before the splitting is executed.
* The set SQL is passed through the {@link java.sql.Connection#nativeSQL} method if a connection is available.
*/
@Override
public SqlStatement[] generateStatements(Database database) {
List returnStatements = new ArrayList<>();
String sql = StringUtil.trimToNull(getSql());
if (sql == null) {
return new SqlStatement[0];
}
String processedSQL = normalizeLineEndings(sql);
if (this instanceof RawSQLChange && ((RawSQLChange) this).isRerunnable()) {
returnStatements.add(new RawSqlStatement(processedSQL, getEndDelimiter()));
return returnStatements.toArray(new SqlStatement[returnStatements.size()]);
}
for (String statement : StringUtil.processMultiLineSQL(processedSQL, isStripComments(), isSplitStatements(), getEndDelimiter())) {
if (database instanceof MSSQLDatabase) {
statement = statement.replaceAll("\\n", "\r\n");
}
String escapedStatement = statement;
try {
if (database.getConnection() != null) {
escapedStatement = database.getConnection().nativeSQL(statement);
}
} catch (DatabaseException e) {
escapedStatement = statement;
}
returnStatements.add(new RawSqlStatement(escapedStatement, getEndDelimiter()));
}
return returnStatements.toArray(new SqlStatement[returnStatements.size()]);
}
@Override
public boolean generateStatementsVolatile(Database database) {
return false;
}
@Override
public boolean generateRollbackStatementsVolatile(Database database) {
return false;
}
@Override
public ChangeStatus checkStatus(Database database) {
return new ChangeStatus().unknown("Cannot check raw sql status");
}
protected String normalizeLineEndings(String string) {
return string.replace("\r", "");
}
public static class NormalizingStream extends InputStream {
private ByteArrayInputStream headerStream;
private PushbackInputStream stream;
private byte[] quickBuffer = new byte[100];
private List resizingBuffer = new ArrayList<>();
private int lastChar = 'X';
private boolean seenNonSpace;
public NormalizingStream(String endDelimiter, Boolean splitStatements, Boolean stripComments, InputStream stream) {
this.stream = new PushbackInputStream(stream, 2048);
try {
this.headerStream = new ByteArrayInputStream((endDelimiter+":"+splitStatements+":"+stripComments+":").getBytes(GlobalConfiguration.OUTPUT_FILE_ENCODING.getCurrentValue()));
} catch (UnsupportedEncodingException e) {
throw new UnexpectedLiquibaseException(e);
}
}
@Override
public int read() throws IOException {
if (headerStream != null) {
int returnChar = headerStream.read();
if (returnChar != -1) {
return returnChar;
}
headerStream = null;
}
int returnChar = stream.read();
if (isWhiteSpace(returnChar)) {
returnChar = ' ';
}
while ((returnChar == ' ') && (!seenNonSpace || (lastChar == ' '))) {
returnChar = stream.read();
if (isWhiteSpace(returnChar)) {
returnChar = ' ';
}
}
seenNonSpace = true;
lastChar = returnChar;
if ((lastChar == ' ') && isOnlyWhitespaceRemaining()) {
return -1;
}
return returnChar;
}
@Override
public int available() throws IOException {
return stream.available();
}
@Override
public boolean markSupported() {
return stream.markSupported();
}
@Override
public synchronized void mark(int readLimit) {
stream.mark(readLimit);
}
@Override
public synchronized void reset() throws IOException {
stream.reset();
}
private boolean isOnlyWhitespaceRemaining() throws IOException {
try {
int quickBufferUsed = 0;
while (true) {
byte read = (byte) stream.read();
if (quickBufferUsed >= quickBuffer.length) {
resizingBuffer.add(read);
} else {
quickBuffer[quickBufferUsed++] = read;
}
if (read == -1) {
return true;
}
if (!isWhiteSpace(read)) {
if (!resizingBuffer.isEmpty()) {
byte[] buf = new byte[resizingBuffer.size()];
for (int i=0; i< resizingBuffer.size(); i++) {
buf[i] = resizingBuffer.get(i);
}
stream.unread(buf);
}
stream.unread(quickBuffer, 0, quickBufferUsed);
return false;
}
}
} finally {
resizingBuffer.clear();
}
}
private boolean isWhiteSpace(int read) {
return (read == ' ') || (read == '\n') || (read == '\r') || (read == '\t');
}
@Override
public void close() throws IOException {
stream.close();
}
}
}