Many resources are needed to download a project. Please understand that we have to compensate our server costs. Thank you in advance. Project price only 1 $
You can buy this project and download/modify it how often you want.
package liquibase.change.core;
import liquibase.CatalogAndSchema;
import liquibase.change.*;
import liquibase.changelog.ChangeSet;
import liquibase.database.AbstractJdbcDatabase;
import liquibase.database.Database;
import liquibase.database.core.MSSQLDatabase;
import liquibase.database.core.MySQLDatabase;
import liquibase.database.core.PostgresDatabase;
import liquibase.datatype.DataTypeFactory;
import liquibase.datatype.LiquibaseDataType;
import liquibase.exception.DatabaseException;
import liquibase.exception.UnexpectedLiquibaseException;
import liquibase.exception.Warnings;
import liquibase.executor.ExecutorService;
import liquibase.executor.LoggingExecutor;
import liquibase.io.EmptyLineAndCommentSkippingInputStream;
import liquibase.logging.LogService;
import liquibase.logging.LogType;
import liquibase.logging.Logger;
import liquibase.resource.ResourceAccessor;
import liquibase.resource.UtfBomAwareReader;
import liquibase.snapshot.InvalidExampleException;
import liquibase.snapshot.SnapshotGeneratorFactory;
import liquibase.statement.BatchDmlExecutablePreparedStatement;
import liquibase.statement.ExecutablePreparedStatementBase;
import liquibase.statement.InsertExecutablePreparedStatement;
import liquibase.statement.SqlStatement;
import liquibase.statement.core.InsertOrUpdateStatement;
import liquibase.statement.core.InsertSetStatement;
import liquibase.statement.core.InsertStatement;
import liquibase.structure.core.Column;
import liquibase.structure.core.DataType;
import liquibase.structure.core.Table;
import liquibase.util.BooleanParser;
import liquibase.util.StreamUtil;
import liquibase.util.StringUtils;
import liquibase.util.csv.CSVReader;
import java.io.IOException;
import java.io.InputStream;
import java.io.Reader;
import java.util.*;
import static java.util.ResourceBundle.getBundle;
@DatabaseChange(name = "loadData",
description = "Loads data from a CSV file into an existing table. A value of NULL in a cell will be " +
"converted to a database NULL rather than the string 'NULL'.\n" +
"Lines starting with # (hash) sign are treated as comments. You can change comment pattern by " +
"specifying 'commentLineStartsWith' property in loadData tag." +
"To disable comments set 'commentLineStartsWith' to empty value'\n" +
"\n" +
"If the data type for a load column is set to NUMERIC, numbers are parsed in US locale (e.g. 123.45)." +
"\n" +
"Date/Time values included in the CSV file should be in ISO format " +
"http://en.wikipedia.org/wiki/ISO_8601 in order to be parsed correctly by Liquibase. Liquibase will " +
"initially set the date format to be 'yyyy-MM-dd'T'HH:mm:ss' and then it checks for two special " +
"cases which will override the data format string.\n" +
"\n" +
"If the string representing the date/time includes a '.', then the date format is changed to " +
"'yyyy-MM-dd'T'HH:mm:ss.SSS'\n" +
"If the string representing the date/time includes a space, then the date format is changed " +
"to 'yyyy-MM-dd HH:mm:ss'\n" +
"Once the date format string is set, Liquibase will then call the SimpleDateFormat.parse() method " +
"attempting to parse the input string so that it can return a Date/Time. If problems occur, " +
"then a ParseException is thrown and the input string is treated as a String for the INSERT command " +
"to be generated.\n" +
"If UUID type is used UUID value is stored as string and NULL in cell is supported. Column config should be used in xml.",
priority = ChangeMetaData.PRIORITY_DEFAULT, appliesTo = "table",
since = "1.7")
public class LoadDataChange extends AbstractChange implements ChangeWithColumns {
/**
* CSV Lines starting with that sign(s) will be treated as comments by default
*/
public static final String DEFAULT_COMMENT_PATTERN = "#";
private static final Logger LOG = LogService.getLog(LoadDataChange.class);
private static ResourceBundle coreBundle = getBundle("liquibase/i18n/liquibase-core");
private String catalogName;
private String schemaName;
private String tableName;
private String file;
private String commentLineStartsWith = DEFAULT_COMMENT_PATTERN;
private Boolean relativeToChangelogFile;
private String encoding;
private String separator = liquibase.util.csv.CSVReader.DEFAULT_SEPARATOR + "";
private String quotchar = liquibase.util.csv.CSVReader.DEFAULT_QUOTE_CHARACTER + "";
private List columns = new ArrayList<>();
private Boolean usePreparedStatements;
/**
* Transform a value read from a CSV file into a string to be written into the database if the column type
* is not known.
*
* @param value the value to transform
* @return if the value is empty or the string "NULL" (case-insensitive), return the empty string.
* If not, the value "toString()" representation (trimmed of spaces left and right) is returned.
*/
protected static String getValueToWrite(Object value) {
if ((value == null) || "NULL".equalsIgnoreCase(value.toString())) {
return "";
} else {
return value.toString().trim();
}
}
// TODO: We can currently do this for INSERT operations, but not yet for UPDATE operations, so loadUpdateDataChange
// will overwrite this flag for now.
protected boolean hasPreparedStatementsImplemented() {
return true;
}
@Override
public boolean supports(Database database) {
return true;
}
@Override
public boolean generateRollbackStatementsVolatile(Database database) {
return true;
}
@DatabaseChangeProperty(
since = "3.0",
mustEqualExisting = "table.catalog"
)
public String getCatalogName() {
return catalogName;
}
public void setCatalogName(String catalogName) {
this.catalogName = catalogName;
}
@DatabaseChangeProperty(mustEqualExisting = "table.schema")
public String getSchemaName() {
return schemaName;
}
public void setSchemaName(String schemaName) {
this.schemaName = schemaName;
}
@DatabaseChangeProperty(
description = "Name of the table to insert data into",
requiredForDatabase = "all",
mustEqualExisting = "table"
)
public String getTableName() {
return tableName;
}
public void setTableName(String tableName) {
this.tableName = tableName;
}
@DatabaseChangeProperty(
description = "CSV file to load",
exampleValue = "com/example/users.csv",
requiredForDatabase = "all"
)
public String getFile() {
return file;
}
public void setFile(String file) {
this.file = file;
}
public Boolean getUsePreparedStatements() {
return usePreparedStatements;
}
public void setUsePreparedStatements(Boolean usePreparedStatements) {
this.usePreparedStatements = usePreparedStatements;
}
public String getCommentLineStartsWith() {
return commentLineStartsWith;
}
public void setCommentLineStartsWith(String commentLineStartsWith) {
//if the value is null (not provided) we want to use default value
if (commentLineStartsWith == null) {
this.commentLineStartsWith = DEFAULT_COMMENT_PATTERN;
} else if ("".equals(commentLineStartsWith)) {
this.commentLineStartsWith = null;
} else {
this.commentLineStartsWith = commentLineStartsWith;
}
}
public Boolean isRelativeToChangelogFile() {
return relativeToChangelogFile;
}
public void setRelativeToChangelogFile(Boolean relativeToChangelogFile) {
this.relativeToChangelogFile = relativeToChangelogFile;
}
@DatabaseChangeProperty(
description = "Encoding of the CSV file (defaults to UTF-8)",
exampleValue = "UTF-8"
)
public String getEncoding() {
return encoding;
}
public void setEncoding(String encoding) {
this.encoding = encoding;
}
@DatabaseChangeProperty(exampleValue = ",")
public String getSeparator() {
return separator;
}
public void setSeparator(String separator) {
if ((separator != null) && "\\t".equals(separator)) {
separator = "\t";
}
this.separator = separator;
}
@DatabaseChangeProperty(exampleValue = "'")
public String getQuotchar() {
return quotchar;
}
public void setQuotchar(String quotchar) {
this.quotchar = quotchar;
}
@Override
public void addColumn(LoadDataColumnConfig column) {
columns.add(column);
}
@Override
@DatabaseChangeProperty(description = "Defines how the data should be loaded.", requiredForDatabase = "all")
public List getColumns() {
return columns;
}
@Override
public void setColumns(List columns) {
this.columns = columns;
}
@Override
public SqlStatement[] generateStatements(Database database) {
boolean databaseSupportsBatchUpdates = false;
try {
databaseSupportsBatchUpdates = database.supportsBatchUpdates();
} catch (DatabaseException e) {
throw new UnexpectedLiquibaseException(e);
}
CSVReader reader = null;
try {
reader = getCSVReader();
if (reader == null) {
throw new UnexpectedLiquibaseException("Unable to read file " + this.getFile());
}
String[] headers = reader.readNext();
if (headers == null) {
throw new UnexpectedLiquibaseException("Data file " + getFile() + " was empty");
}
// If we do not have a column list yet, take the column list we interpolated from the CSV headers
// earlier.
if (columns.isEmpty()) {
columns.addAll(getColumnsFromHeaders(headers));
}
// If we have an real JDBC connection to the database, ask the database for any missing column types.
try {
retrieveMissingColumnLoadTypes(columns, database);
} catch (DatabaseException e) {
throw new UnexpectedLiquibaseException(e);
}
List batchedStatements = new ArrayList<>();
boolean anyPreparedStatements = false;
String[] line;
// Start at '1' to take into account the header (already processed):
int lineNumber = 1;
boolean isCommentingEnabled = StringUtils.isNotEmpty(commentLineStartsWith);
List statements = new ArrayList<>();
while ((line = reader.readNext()) != null) {
lineNumber++;
if
((line.length == 0) || ((line.length == 1) && (StringUtils.trimToNull(line[0]) == null)) ||
(isCommentingEnabled && isLineCommented(line))
) {
//nothing interesting on this line
continue;
}
// Ensure each line has the same number of columns defined as does the header.
// (Failure could indicate unquoted strings with commas, for example).
if (line.length != headers.length) {
throw new UnexpectedLiquibaseException(
"CSV file " + getFile() + " Line " + lineNumber + " has " + line.length +
" values defined, Header has " + headers.length +
". Numbers MUST be equal (check for unquoted string with embedded commas)"
);
}
boolean needsPreparedStatement = false;
if (usePreparedStatements != null && usePreparedStatements) {
needsPreparedStatement = true;
}
List columnsFromCsv = new ArrayList<>();
for (int i = 0; i < headers.length; i++) {
Object value = line[i];
String columnName = headers[i].trim();
ColumnConfig valueConfig = new ColumnConfig();
ColumnConfig columnConfig = getColumnConfig(i, headers[i].trim());
if (columnConfig != null) {
if ("skip".equalsIgnoreCase(columnConfig.getType())) {
continue;
}
// don't overwrite header name unless there is actually a value to override it with
if (columnConfig.getName() != null) {
columnName = columnConfig.getName();
}
valueConfig.setName(columnName);
if (columnConfig.getType() != null) {
if (columnConfig.getType().equalsIgnoreCase(LOAD_DATA_TYPE.BOOLEAN.toString())) {
if ("NULL".equalsIgnoreCase(value.toString())) {
valueConfig.setValue(null);
} else {
valueConfig.setValueBoolean(
BooleanParser.parseBoolean(value.toString().toLowerCase())
);
}
} else if (columnConfig.getType().equalsIgnoreCase(LOAD_DATA_TYPE.NUMERIC.toString())) {
if ("NULL".equalsIgnoreCase(value.toString())) {
valueConfig.setValue(null);
} else {
valueConfig.setValueNumeric(value.toString());
}
} else if
(
columnConfig.getType().toLowerCase().contains("date")
|| columnConfig.getType().toLowerCase().contains("time")
) {
if ("NULL".equalsIgnoreCase(value.toString())) {
valueConfig.setValue(null);
} else {
valueConfig.setValueDate(value.toString());
}
} else if (columnConfig.getType().equalsIgnoreCase(LOAD_DATA_TYPE.STRING.toString())) {
if ("NULL".equalsIgnoreCase(value.toString())) {
valueConfig.setValue(null);
} else {
valueConfig.setValue(value.toString());
}
} else if (columnConfig.getType().equalsIgnoreCase(LOAD_DATA_TYPE.COMPUTED.toString())) {
if ("NULL".equalsIgnoreCase(value.toString())) {
valueConfig.setValue(null);
} else {
liquibase.statement.DatabaseFunction function =
new liquibase.statement.DatabaseFunction(value.toString());
valueConfig.setValueComputed(function);
}
} else if (columnConfig.getType().equalsIgnoreCase(LOAD_DATA_TYPE.SEQUENCE.toString())) {
String sequenceName;
if ("NULL".equalsIgnoreCase(value.toString())) {
sequenceName = columnConfig.getDefaultValue();
if (sequenceName == null) {
throw new UnexpectedLiquibaseException(
"Must set a sequence name in the loadData column defaultValue attribute"
);
}
} else {
sequenceName = value.toString();
}
liquibase.statement.SequenceNextValueFunction function =
new liquibase.statement.SequenceNextValueFunction(sequenceName);
valueConfig.setValueComputed(function);
} else if (columnConfig.getType().equalsIgnoreCase(LOAD_DATA_TYPE.BLOB.toString())) {
if ("NULL".equalsIgnoreCase(value.toString())) {
valueConfig.setValue(null);
} else {
valueConfig.setValueBlobFile(value.toString());
needsPreparedStatement = true;
}
} else if (columnConfig.getType().equalsIgnoreCase(LOAD_DATA_TYPE.CLOB.toString())) {
if ("NULL".equalsIgnoreCase(value.toString())) {
valueConfig.setValue(null);
} else {
valueConfig.setValueClobFile(value.toString());
needsPreparedStatement = true;
}
} else if (columnConfig.getType().equalsIgnoreCase(LOAD_DATA_TYPE.UUID.toString())) {
valueConfig.setType(columnConfig.getType());
if ("NULL".equalsIgnoreCase(value.toString())) {
valueConfig.setValue(null);
} else {
valueConfig.setValue(value.toString());
}
} else {
throw new UnexpectedLiquibaseException(
String.format(coreBundle.getString("loaddata.type.is.not.supported"),
columnConfig.getType()
)
);
}
} else {
// columnConfig did not specify a type
valueConfig.setValue(getValueToWrite(value));
}
} else {
// No columnConfig found. Assume header column name to be the table column name.
if (columnName.contains("(") || (columnName.contains(")") && (database instanceof
AbstractJdbcDatabase))) {
columnName = ((AbstractJdbcDatabase) database).quoteObject(columnName, Column.class);
}
valueConfig.setName(columnName);
valueConfig.setValue(getValueToWrite(value));
}
columnsFromCsv.add(valueConfig);
}
// end of: iterate through all the columns of a CSV line
// Try to use prepared statements if any of the two following conditions apply:
// 1. There is no other option than using a prepared statement (e.g. in cases of LOBs)
// 2. The database supports batched statements (for improved performance) AND we are not in an
// "SQL" mode (i.e. we generate an SQL file instead of actually modifying the database).
if
(
(needsPreparedStatement ||
(databaseSupportsBatchUpdates &&
!(ExecutorService.getInstance().getExecutor(database) instanceof LoggingExecutor)
)
)
&& hasPreparedStatementsImplemented()
) {
anyPreparedStatements = true;
ExecutablePreparedStatementBase stmt =
this.createPreparedStatement(
database, getCatalogName(), getSchemaName(), getTableName(), columnsFromCsv,
getChangeSet(), getResourceAccessor()
);
batchedStatements.add(stmt);
} else {
InsertStatement insertStatement =
this.createStatement(getCatalogName(), getSchemaName(), getTableName());
for (ColumnConfig column : columnsFromCsv) {
String columnName = column.getName();
Object value = column.getValueObject();
if (value == null) {
value = "NULL";
}
insertStatement.addColumnValue(columnName, value);
}
statements.add(insertStatement);
}
// end of: will we use a PreparedStatement?
}
// end of: loop for every input line from the CSV file
if (anyPreparedStatements) {
// If we have only prepared statements and the database supports batching, let's roll
if (databaseSupportsBatchUpdates && statements.isEmpty() && (!batchedStatements.isEmpty())) {
return new SqlStatement[] {
new BatchDmlExecutablePreparedStatement(
database, getCatalogName(), getSchemaName(),
getTableName(), columns,
getChangeSet(), getResourceAccessor(),
batchedStatements)
};
} else {
return statements.toArray(new SqlStatement[statements.size()]);
}
} else {
if (statements.isEmpty()) {
// avoid returning unnecessary dummy statement
return new SqlStatement[0];
}
InsertSetStatement statementSet = this.createStatementSet(
getCatalogName(), getSchemaName(), getTableName()
);
for (SqlStatement stmt : statements) {
statementSet.addInsertStatement((InsertStatement) stmt);
}
if ((database instanceof MSSQLDatabase) || (database instanceof MySQLDatabase) || (database
instanceof PostgresDatabase)) {
List innerStatements = statementSet.getStatements();
if ((innerStatements != null) && (!innerStatements.isEmpty()) && (innerStatements.get(0)
instanceof InsertOrUpdateStatement)) {
//cannot do insert or update in a single statement
return statementSet.getStatementsArray();
}
// we only return a single "statement" - it's capable of emitting multiple sub-statements,
// should the need arise, on generation.
return new SqlStatement[]{statementSet};
} else {
return statementSet.getStatementsArray();
}
}
} catch (IOException e) {
throw new RuntimeException(e);
} catch (UnexpectedLiquibaseException ule) {
if ((getChangeSet() != null) && (getChangeSet().getFailOnError() != null) && !getChangeSet()
.getFailOnError()) {
LOG.info(LogType.LOG, "Change set " + getChangeSet().toString(false) +
" failed, but failOnError was false. Error: " + ule.getMessage());
return new SqlStatement[0];
} else {
throw ule;
}
} finally {
if (null != reader) {
try {
reader.close();
} catch (IOException e) {
// Do nothing
}
}
}
}
/**
* Iterate through the List of LoadDataColumnConfig and ask the database for any column types that we have
* no data type of.
* @param columns a list of LoadDataColumnConfigs to process
*/
@SuppressWarnings("CommentedOutCodeLine")
private void retrieveMissingColumnLoadTypes(List columns, Database database) throws
DatabaseException {
boolean matched = false;
// If no column is missing type information, we are already done.
for (LoadDataColumnConfig c : columns) {
if (c.getType() == null) {
matched = true;
}
}
if (!matched) {
return;
}
/* The above is the JDK7 version of:
if (columns.stream().noneMatch(c -> c.getType() == null)) {
return;
}
*/
// Snapshot the database table
CatalogAndSchema catalogAndSchema = new CatalogAndSchema(getCatalogName(), getSchemaName());
catalogAndSchema = catalogAndSchema.standardize(database);
Table targetTable = new Table(catalogAndSchema.getCatalogName(), catalogAndSchema.getSchemaName(),
database.correctObjectName(getTableName(), Table.class));
Table snapshotOfTable;
try {
snapshotOfTable = SnapshotGeneratorFactory.getInstance().createSnapshot(
targetTable,
database);
} catch (InvalidExampleException e) {
throw new DatabaseException(e);
}
if (snapshotOfTable == null) {
LOG.warning(LogType.LOG, String.format(
coreBundle.getString("could.not.snapshot.table.to.get.the.missing.column.type.information"),
database.escapeTableName(
targetTable.getSchema().getCatalogName(),
targetTable.getSchema().getName(),
targetTable.getName())
));
return;
}
// Save the columns of the database table in a lookup table
Map tableColumns = new HashMap<>();
for (Column c : snapshotOfTable.getColumns()) {
// Normalise the LoadDataColumnConfig column names to the database
tableColumns.put(database.correctObjectName(c.getName(), Column.class), c);
}
/* The above is the JDK7 version of:
snapshotOfTable.getColumns().forEach(c -> tableColumns.put(c.getName(), c));
*/
// Normalise the LoadDataColumnConfig column names to the database
Map columnConfigs = new HashMap<>();
for (LoadDataColumnConfig c : columns) {
columnConfigs.put(
database.correctObjectName(c.getName(), Column.class),
c
);
}
/* The above is the JDK7 version of:
columns.forEach(c -> columnConfigs.put(
database.correctObjectName(c.getName(), Column.class),
c
));
*/
for (Map.Entry entry : columnConfigs.entrySet()) {
if (!(entry.getValue().getType() == null)) {
continue;
}
LoadDataColumnConfig columnConfig = entry.getValue();
DataType dataType = tableColumns.get(entry.getKey()).getType();
if (dataType == null) {
LOG.warning(LogType.LOG, String.format(coreBundle.getString("unable.to.find.load.data.type"),
columnConfig.toString(), snapshotOfTable.toString()));
columnConfig.setType(LOAD_DATA_TYPE.STRING.toString());
} else {
LiquibaseDataType liquibaseDataType = DataTypeFactory.getInstance()
.fromDescription(dataType.toString(), database);
if (liquibaseDataType != null) {
columnConfig.setType(liquibaseDataType.getLoadTypeName().toString());
} else {
LOG.warning(LogType.LOG, String.format(coreBundle.getString("unable.to.convert.load.data.type"),
columnConfig.toString(), snapshotOfTable.toString(), dataType.toString()));
}
}
}
/* The above is the JDK7 version of:
columnConfigs.entrySet().stream()
.filter(entry -> entry.getValue().getType() == null)
.forEach(entry -> {
LoadDataColumnConfig columnConfig = entry.getValue();
DataType dataType = tableColumns.get(entry.getKey()).getType();
if (dataType == null) {
LOG.warning(String.format(coreBundle.getString("unable.to.find.load.data.type"),
columnConfig.toString(), snapshotOfTable.toString() ));
columnConfig.setType(LOAD_DATA_TYPE.STRING.toString());
} else {
LiquibaseDataType liquibaseDataType = DataTypeFactory.getInstance()
.fromDescription(dataType.toString(), database);
if (liquibaseDataType != null) {
columnConfig.setType(liquibaseDataType.getLoadTypeName().toString());
} else {
LOG.warning(String.format(coreBundle.getString("unable.to.convert.load.data.type"),
columnConfig.toString(), snapshotOfTable.toString(), liquibaseDataType.toString()));
}
}
}
);
*/
}
/**
* If no columns (and their data types) are specified in the loadData change, we interpolate their names from
* the header columns of the CSV file.
* @param headers the headers of the CSV file
* @return a List of LoadDataColumnConfigs
*/
private List getColumnsFromHeaders(String[] headers) {
ArrayList result = new ArrayList<>();
int i=0;
for (String columnNameFromHeader : headers) {
LoadDataColumnConfig loadDataColumnConfig = new LoadDataColumnConfig();
loadDataColumnConfig.setIndex(i);
loadDataColumnConfig.setHeader(columnNameFromHeader);
loadDataColumnConfig.setName(columnNameFromHeader);
result.add(loadDataColumnConfig);
i++;
}
return result;
}
private boolean isLineCommented(String[] line) {
return StringUtils.startsWith(line[0], commentLineStartsWith);
}
@Override
public boolean generateStatementsVolatile(Database database) {
return true;
}
public CSVReader getCSVReader() throws IOException {
ResourceAccessor resourceAccessor = getResourceAccessor();
if (resourceAccessor == null) {
throw new UnexpectedLiquibaseException("No file resourceAccessor specified for " + getFile());
}
InputStream stream = StreamUtil.openStream(file, isRelativeToChangelogFile(), getChangeSet(), resourceAccessor);
if (stream == null) {
return null;
}
Reader streamReader;
if (getEncoding() == null) {
streamReader = new UtfBomAwareReader(stream);
} else {
streamReader = new UtfBomAwareReader(stream, getEncoding());
}
char quotchar;
if (StringUtils.trimToEmpty(this.quotchar).isEmpty()) {
// hope this is impossible to have a field surrounded with non ascii char 0x01
quotchar = '\1';
} else {
quotchar = this.quotchar.charAt(0);
}
if (separator == null) {
separator = liquibase.util.csv.CSVReader.DEFAULT_SEPARATOR + "";
}
return new CSVReader(streamReader, separator.charAt(0), quotchar);
}
protected ExecutablePreparedStatementBase createPreparedStatement(
Database database, String catalogName, String schemaName, String tableName,
List columns, ChangeSet changeSet, ResourceAccessor resourceAccessor) {
return new InsertExecutablePreparedStatement(database, catalogName, schemaName, tableName, columns,
changeSet, resourceAccessor);
}
protected InsertStatement createStatement(String catalogName, String schemaName, String tableName) {
return new InsertStatement(catalogName, schemaName, tableName);
}
protected InsertSetStatement createStatementSet(String catalogName, String schemaName, String tableName) {
return new InsertSetStatement(catalogName, schemaName, tableName);
}
protected ColumnConfig getColumnConfig(int index, String header) {
for (LoadDataColumnConfig config : columns) {
if ((config.getIndex() != null) && config.getIndex().equals(index)) {
return config;
}
if ((config.getHeader() != null) && config.getHeader().equalsIgnoreCase(header)) {
return config;
}
if ((config.getName() != null) && config.getName().equalsIgnoreCase(header)) {
return config;
}
}
return null;
}
@Override
public ChangeStatus checkStatus(Database database) {
return new ChangeStatus().unknown("Cannot check loadData status");
}
@Override
public String getConfirmationMessage() {
return "Data loaded from " + getFile() + " into " + getTableName();
}
@Override
public CheckSum generateCheckSum() {
InputStream stream = null;
try {
stream = StreamUtil.openStream(file, isRelativeToChangelogFile(), getChangeSet(), getResourceAccessor());
if (stream == null) {
throw new UnexpectedLiquibaseException(getFile() + " could not be found");
}
stream = new EmptyLineAndCommentSkippingInputStream(stream, commentLineStartsWith);
return CheckSum.compute(getTableName() + ":" + CheckSum.compute(stream, /*standardizeLineEndings*/ true));
} catch (IOException e) {
throw new UnexpectedLiquibaseException(e);
} finally {
if (stream != null) {
try {
stream.close();
} catch (IOException ignore) {
// Do nothing
}
}
}
}
@Override
public Warnings warn(Database database) {
return null;
}
@Override
public String getSerializedObjectNamespace() {
return STANDARD_CHANGELOG_NAMESPACE;
}
@SuppressWarnings("HardCodedStringLiteral")
public enum LOAD_DATA_TYPE {
BOOLEAN, NUMERIC, DATE, STRING, COMPUTED, SEQUENCE, BLOB, CLOB, SKIP,UUID
}
}