10.†Incompatible Library Changes

This chapter documents those library changes since the epochal 1.7.9 release that break end-user programs. You can dig this stuff out of the ChangeLog.md file, but the change log focuses more on explaining and justifying the facets of each change, while this section focuses on how to migrate your code between these library versions.

Since pure additions do not break programs, those changes are still documented only in the change log.

10.1.†API Changes

This section documents files, functions, methods and classes that were removed or changed in an incompatible way. If your program uses the changed item, you will have to change something in your program to get it to compile after upgrading to each of these versions.


Removed Row::operator[]() overloads except the one for size_type, and added Row::lookup_by_name() to provide the “subscript by string” functionality. In practical terms, this change means that the row["field"] syntax no longer works; you must use the new lookup_by_name method instead.

Renamed the generated library on POSIX systems from libsqlplus to libmysqlpp.


Removed SQLQuery::operator=(), and the same for its Query subclass. Use the copy constructor instead, if you need to copy one query to another query object.


The library used to have two names for many core classes: a short one, such as Row and a longer one, MysqlRow. The library now uses the shorter names exclusively.

All symbols within MySQL++ are in the mysqlpp namespace now if you use the new mysql++.h header. If you use the older sqlplus.hh or mysql++.hh headers, these symbols are hoist up into the global namespace. The older headers cause the compiler to emit warnings if you use them, and they will go away someday.


Connection class changes
  • Connection::create_db() and drop_db() return true on success. They returned false in v1.7.x! This change will only affect your code if you have exceptions disabled.

  • Renamed Connection::real_connect() to connect(), made several more of its parameters default, and removed the old connect() method, as it’s now a strict subset of the new one. The only practical consequence is that if your program was using real_connect(), you will have to change it to connect().

  • Replaced Connection::read_option() with new set_option() mechanism. In addition to changing the name, programs using this function will have to use the new Connection::Option enumerated values, accept a true return value as meaning success instead of 0, and use the proper argument type. Regarding the latter, read_option() took a const char* argument, but because it was just a thin wrapper over the MySQL C API function mysql_options(), the actual value being pointed to could be any of several types. This new mechanism is properly type-safe.

Exception-related changes
  • Classes Connection, Query, Result, ResUse, and Row now derive from OptionalExceptions which gives these classes a common interface for disabling exceptions. In addition, almost all of the per-method exception-disabling flags were removed. The preferred method for disabling exceptions on these objects is to create an instance of the new NoExceptions class on the stack, which disables exceptions on an OptionalExceptions subclass as long as the NoExceptions instance is in scope. You can instead call disable_exceptions() on any of these objects, but if you only want them disabled temporarily, it’s easy to forget to re-enable them later.

  • In the previous version of MySQL++, those classes that supported optional exceptions that could create instances of other such classes were supposed to pass this flag on to their children. That is, if you created a Connection object with exceptions enabled, and then asked it to create a Query object, the Query object also had exceptions disabled. The problem is, this didn’t happen in all cases where it should have in v1.7. This bug is fixed in v2.0. If your program begins crashing due to uncaught exceptions after upgrading to v2.0, this is the most likely cause. The most expeditious fix in this situation is to use the new NoExceptions feature to return these code paths to the v1.7 behavior. A better fix is to rework your program to avoid or deal with the new exceptions.

  • All custom MySQL++ exceptions now derive from the new Exception interface. The practical upshot of this is that the variability between the various exception types has been eliminated. For instance, to get the error string, the BadQuery exception had a string member called error plus a method called what(). Both did the same thing, and the what() method is more common, so the error string was dropped from the interface. None of the example programs had to be changed to work with the new exceptions, so if your program handles MySQL++ exceptions the same way they do, your program won’t need to change, either.

  • Renamed SQLQueryNEParams exception to BadParamCount to match style of other exception names.

  • Added BadOption, ConnectionFailed, DBSelectionFailed, EndOfResults, EndOfResultSets, LockFailed, and ObjectNotInitialized exception types, to fix overuse of BadQuery. Now the latter is used only for errors on query execution. If your program has a “catch-all” block taking a std::exception for each try block containing MySQL++ statements, you probably won’t need to change your program. Otherwise, the new exceptions will likely show up as program crashes due to unhandled exceptions.

Query class changes
  • In previous versions, Connection had a querying interface similar to class Query’s. These methods were intended only for Query’s use; no example ever used this interface directly, so no end-user code is likely to be affected by this change.

  • A more likely problem arising from the above change is code that tests for query success by calling the Connection object’s success() method or by casting it to bool. This will now give misleading results, because queries no longer go through the Connection object. Class Query has the same success-testing interface, so use it instead.

  • Query now derives from std::ostream instead of std::stringstream.

Result/ResUse class changes
  • Renamed ResUse::mysql_result() to raw_result() so it’s database server neutral.

  • Removed ResUse::eof(), as it wrapped the deprecated and unnecessary MySQL C API function mysql_eof(). See the simple3 and usequery examples to see the proper way to test for the end of a result set.

Row class changes
  • Removed “field name” form of Row::field_list(). It was pointless.

  • Row subscripting works more like v1.7.9: one can subscript a Row with a string (e.g. row["myfield"]), or with an integer (e.g. row[5]). lookup_by_name() was removed. Because row[0] is ambiguous (0 could mean the first field, or be a null pointer to const char*), there is now Row::at(), which can look up any field by index.

Miscellaneous changes
  • Where possible, all distributed Makefiles only build dynamic libraries. (Shared objects on most Unices, DLLs on Windows, etc.) Unless your program is licensed under the GPL or LGPL, you shouldn’t have been using the static libraries from previous versions anyway.

  • Removed the backwards-compatibility headers sqlplus.hh and mysql++.hh. If you were still using these, you will have to change to mysql++.h, which will put all symbols in namespace mysqlpp.

  • Can no longer use arrow operator (->) on the iterators into the Fields, Result and Row containers.


Code like this will have to change:

query << "delete from mytable where myfield=%0:myvalue";
query.def["myvalue"] = some_value;

...to something more like this:

query << "delete from mytable where myfield=%0";

The first code snippet abuses the default template query parameter mechanism (Query::def) to fill out the template instead of using one of the overloaded forms of execute(), store() or use() taking one or more SQLString parameters. The purpose of Query::def is to allow for default template parameters over multiple queries. In the first snippet above, there is only one parameter, so in order to justify the use of template queries in the first place, it must be changing with each query. Therefore, it isn’t really a “default” parameter at all. We did not make this change maliciously, but you can understand why we are not in any hurry to restore this “feature”.

(Incidentally, this change was made to allow better support for BLOB columns.)


Connection::set_option() calls now set the connection option immediately, instead of waiting until just before the connnection is actually established. Code that relied on the old behavior could see unhandled exceptions, since option setting errors are now thrown from a different part of the code. You want to wrap the actual set_option() call now, not Connection::connect()

FieldNames and FieldTypes are no longer exported from the library. If you are using these classes directly from Visual C++ or MinGW, your code won’t be able to dynamically link to a DLL version of the library any more. These are internal classes, however, so no one should be using them directly.


Class name changes

Several classes changed names in this release:

  • ColData is now String.

  • NullisBlank is now NullIsBlank. (Note the capital I.) Similar changes for NullisNull and NullisZero.

  • ResNSel is now SimpleResult.

  • Result is now StoreQueryResult.

  • ResUse is now UseQueryResult.

  • SQLString is now SQLTypeAdapter.

When first building existing code against this version, you may find it helpful to define the macro MYSQLPP_OLD_CLASS_NAMES in your program’s build options. This will turn on some macros that set up aliases for the new class names matching their corresponding old names. Then, when you’ve fixed up any other issues that may prevent your program from building with the new MySQL++, you can turn it back off and fix up any class name differences.

If you were only using ColData in a BLOB context, you should use sql_blob or one of the related typedefs defined in lib/sql_types.h instead, to insulate your code from changes like these.

The SQLString change shouldn’t affect you, as this class was not designed to be used by end user code. But, due to the old name and the fact that it used to derive from std::string, some might have been tempted to use it as an enhanced std::string. Such code will undoubtedly break, but can probably be fixed by just changing it to use std::string instead.

Connection class changes

The option setting mechanism has been redesigned. (Yes, again.) There used to be an enum in Connection with a value for each option we understood, and an overload of Connection::set_option() for each argument type we understood. It was possible to pass any option value to any set_option() overload, and the problem would only be detected at run time. Now each option is represented by a class derived from the new Option abstract base class, and set_option() simply takes a pointer to one of these objects. See examples/multiquery.cpp for the syntax. Since each Option subclass takes only the parameter types it actually understands, it’s now completely type-safe at compile time.

The new option setting mechanism also has the virtue of being more powerful so it let us replace several existing things within Connection with new options:

  • Replaced enable_ssl() with SslOption.

  • Replaced the compress parameter to the Connection create-and-connect constructor and Connection::connect() method with CompressOption.

  • Replaced the connect_timeout parameter with ConnectTimeoutOption.

  • Defined Option subclasses for each of the flags you would previously set using the client_flag parameter. There are about a dozen of these, so instead of listing them, look in lib/options.h for something with a similar name.

Collapsed Connection’s host, port, and socket_name parameters down into a new combined server parameter which is parsed to determine what kind of connection you mean. These interfaces are still compatible with v2.3 and earlier up through the port parameter.

Moved Connection::affected_rows(), info() and insert_id() methods to class Query, as they relate to the most recently-executed query.

Changed the return type of Connection::ping() from int to bool. If you were calling ping() in bool context or using its return value in bool context, you will need to reverse the sense of the test because the previous return code used zero to mean success. Now it returns true to indicate success.

Renamed several methods:

  • Use client_version() instead of api_version() or client_info().

  • Use ipc_version() instead of host_info().

  • Use protocol_version() instead of proto_info().

  • Use server_version() instead of server_info().

  • Use status() instead of stat().

Also, removed close() in favor of disconnect(), which has always done the same thing.

Date and Time class changes

The sql_timestamp typedef is now an alias for DateTime, not Time.

There used to be implicit conversion constructors from ColData (now String), std::string and const char* for the Date, DateTime, and Time classes. It’s still possible to do these conversions, but only explicitly. (This had to be done to make Null<T> work in SSQLSes.)

The most likely place to run into problems as a result of this change is in code like this:

void some_function(const mysqlpp::DateTime& dt);


The function call needs to be changed to:

Exception changes

If an error occurs during the processing of a “use” query (as opposed to the initial execution) we throw the new UseQueryError exception instead of BadQuery.

If you pass bad values to the Row ctor so that it can’t initialize itself properly, it throws the ObjectNotInitialized exception instead of BadQuery.

Together, these two changes mean that BadQuery is now used solely to indicate a problem executing the actual SQL query statement.

Field and Fields class changes

Field is now a real C++ class, not just a typedef for the corresponding C API class. Major portability impacts are:

  • It has no public data members. Where sensible, there is a public accessor function of the same name as the corresponding field in the C API structure.

  • The main exception to this is the flags data member. This is a bitfield in the C API data structure and you had to use MySQL-specific constants to break values out of it. MySQL++’s new Field class provides a public member function returning bool for each of these flags.

  • The new class doesn’t include all of the data members from the C API version. We left out those that aren’t used within MySQL++ or its examples, or whose function we couldn’t understand. Basically, if we couldn’t document a reason to use it, we left it out.

Fields used to be a std::vector work-alike which worked with the C API to access fields and return them as though they were simply contained directly within the Fields object. Now that we have a real MySQL++ class to hold information about each field without reference to the C API, we were able to replace the Fields class with:

typedef std::vector<Field> Fields;

If anything, this should give a pure superset of the old functionality, but it’s possible it could break end user code.

Query class changes

If you were using char as an 8-bit integer in query building, there are several places in MySQL++ v3 where it will now be treated as a single-character string. MySQL++ has had the tiny_int class for many years now specifically to provide a true 8-bit integer without the semantic confusion surrounding the old C char type. Either use tiny_int, or use the SQL type aliases sql_tinyint and sql_tinyint_unsigned instead.

The ‘r’ and ‘R’ template query parameter modifiers were removed. They made the library do quoting and both quoting and escaping (respectively) regardless of the data type of the parameter. There are no corresponding Query stream manipulators, so for symmetery we had to decide whether to add such manipulators or remove the tquery modifiers. There should never be a reason to force quoting or escaping other than to work around a MySQL++ bug, and it’s better to just fix the bug than work around it, so removed the tquery modifiers.

Query::store_next() and Result::fetch_row() no longer throw the EndOfResults and EndOfResultSets exceptions; these are not exceptional conditions! These methods simply return false when you hit the end of the result set now.

Renamed Query::def to Query::template_defaults to make its purpose clearer.

Removed Query::preview(). The most direct replacement for this set of overloaded methods is the parallel set of str() methods, which were just aliases before. (Chose str() over preview() because it’s standard C++ nomenclature.) But if you’re just looking to get a copy of a built query string and you aren’t using template queries, you can now insert the Query into a stream and get the same result.

For example, a lot of code in the examples that used to say things like:

cout << query.preview() << endl;

now looks like this:

cout << query << endl;
Result, ResUse, and ResNSel class changes

In addition to the class name changes described above, UseQueryResult is no longer StoreQueryResult’s base class. There is a new abstract class called ResultBase containing much of what used to be in ResUse, and it is the base of both of these concrete result set types. This should only affect your code if you were using ResUse references to refer to Result objects.

Removed a bunch of duplicate methods:

  • Use num_fields() instead of columns().

  • Use field_names() instead of names().

  • Use num_rows() instead of rows().

  • Use field_types() instead of types().

Renamed several methods for “grammar” reasons. For example, some methods returned a single object but had a “plural” name, implying that it returned a container of objects. In cases like this, we changed the name to agree with the return value. Some of these also fall into the duplicate method category above:

  • Use field(unsigned int) instead of fields(unsigned int).

  • Use field_num(const std::string&) instead of names(const std::string&).

  • Use field_name(int) instead of names(int).

  • Use field_type(int) instead of types(int).

Removed several “smelly” methods:

  • purge(): was an internal implementation detail, not something for end user code to call

  • raw_result(): end user code shouldn’t be digging down to the C API data structures, but if you really need something like this, look at the implementation of Query::storein(). Its workings will probably be educational.

  • reset_names(): no reason to call this, especially now that the field name list is initialized once at startup and then never changed

  • reset_field_names(): just an alias for previous

  • reset_types(): same argument as for reset_names()

  • reset_field_types(): just an alias for previous

ResUse::field_num() would unconditionally throw a BadFieldName exception when you asked for a field that doesn’t exist. Now, if exceptions are disabled on the object, it just returns -1.

SimpleResult’s member variables are all now private, and have read-only accessor functions of the same name.

Code like this used to work:

mysqlpp::Row row;
mysqlpp::Result::size_type i;
for (i = 0; row = res[i]; ++i) {
  // Do something with row here

That is, indexing past the end of a “store” result set would just return an empty row object, which tests as false in bool context, so it ends the loop. Now that StoreQueryResult is a std::vector derivative, this either crashes your program or causes the standard library to throw an exception, depending on what debugging features your version of STL has. The proper technique is:

mysqlpp::Row row;
mysqlpp::StoreQueryResult::size_type i;
for (i = 0; i < res.num_rows(); ++i) {
  row = res[i];
  // Do something with row here

...or, in a more C++ish idiom:

mysqlpp::Row row;
mysqlpp::StoreQueryResult::const_iterator it;
for (it = res.begin(); it != res.end(); ++it) {
  row = *it;
  // Do something with row here
Row class changes

Removed Row::raw_data(), raw_size() and raw_string(). These were useful with BLOB data back when MySQL++ didn’t handle embedded null characters very well, and when copies of ColData objects were expensive. Neither is true now, so they have no value any more. Equivalent calls are:

mysqlpp::String s = row[0];
s.data();                          // raw_data() equivalent
s.length();                        // raw_size() equivalent
std::string(s.data(), s.length()); // raw_string() equivalent

Row::operator[](const char*) would unconditionally throw a BadFieldName exception when you asked for a field that doesn’t exist. Now, if exceptions are disabled on the Row object, it just returns a reference to an empty String object. You can tell when this happens because such an object tests as false in bool context.

Specialized SQL Structure (SSQLS) changes

Renamed custom* to ssqls*. There is a backwards-compatibility header custom.h which includes ssqls.h for you, but it will go away in a future version of MySQL++.

SSQLSes get populated by field name now, not by field order. In v2, it was absolutely required that your SSQLS had its fields declared in exactly the same order as the fields in the database server, and there could be no gaps. An ALTER TABLE command would almost always necessitate redefining the corresponding SSQLS and rebuilding your program. Some alterations actually made using SSQLS impossible. For the most part, this change just gives your program additional flexibility in the face of future changes. However, code that was taking advantage of this low-level fact will break when moving to v3. Before I explain how, let’s go over the high-level functional changes you’ll find in v3’s SSQLS mechanism.

Because MySQL++ no longer needs the SSQLS field order to match the SQL field order, the sql_create_c_order_* SSQLS creation macro was dropped in v3. We were also able to drop the ordering parameters from sql_create_complete_*. That in turn means there is no longer a difference between the way it and sql_create_c_names_* work, so the latter was also dropped. Thus, there are now only two groups of SSQLS creation macros left: sql_create_*, which works pretty much as it always has, and sql_create_complete_*, which is the same except for the lack of ordering parameters.

In general, you should be using sql_create_* for all SSQLSes unless you need to use different names for data members in C++ than you use for the corresponding columns in SQL. In that case, use sql_create_complete_* instead.

In v2, it was possible to have different SQL column names than SSQLS data member names while still using sql_create_* if you only used SSQLS for data retrieval.[25] In v3, you must use sql_create_complete_* for absolutely all uses of SSQLS when you want the C++ field names to differ from the SQL column names.

The new Null<T> support in SSQLSes causes an internal compiler error in Visual C++ 2003. (VC++ 2005 and newer have no trobule with it.) A poll on the mailing list says there aren’t many people still stuck on this version, so we just ifdef’d out the SSQLS mechanism and all the examples that use it when built with VC++ 2003. If this affects you, see Section†5.15, “SSQLS and Visual C++ 2003” for suggestions on ways to cope.

If you are using types other than MySQL++’s sql_* ones [26] in your SSQLSes, code that previously worked may now see TypeLookupFailed exceptions. (This can be thrown even if exceptions are otherwise disabled in MySQL++.) This version of MySQL++ is stricter about mapping SQL to C++ type information, and vice versa. If the library can’t find a suitable mapping from one type system to the other, it throws this exception, because its only other option would be to crash or raise an assertion. This typically happens when building SQL queries, so you can probably handle it the same way as if the subsequent query excecution failed. If you’re catching the generic mysqlpp::Exception, your error handling code might not need to change. If you see this exception, it does mean you need to look into your use of data types, though. The table that controls this is mysql_type_info::types, defined at the top of lib/type_info.cpp. Every data type in lib/sql_types.h has a corresponding record in this table, so if you stick to those types, you’ll be fine. It’s also okay to use types your C++ compiler can convert directly to these predefined types.

The _table static member variable for each SSQLS is now private. The recommended way to access this remains unchanged: the table() static member function.

table() used to return a modifiable reference to the table name. Now there are two overloads, one which returns an unmodifiable pointer to the table name, and the other which takes const char* so you can override the default table name. So, the code we used to recommend for changing the SSQLS’s table name:

my_ssqls_type::table() = "MyTableName";

now needs to be:

Miscellaneous changes

MySQL++ does quoting and escaping much more selectively now. Basically, if the library can tell you’re not building a SQL query using one of the standard methods, it assumes you’re outputting values for human consumption, so it disables quoting and SQL escaping. If you need to build your own mechanism to replace this, quoting is easy to do, and Query::escape_string() can do SQL escaping for you.

Removed success() in Connection, Query and SimpleResult (neť ResNSel) and simply made these classes testable in bool context to get the same information. An additional change in Connection is that it used to be considered “unsuccessful” when the connection was down. Since the sense of this test is now whether the object is in a good state, it only returns false when the connection attempt fails. Call Connection::is_connected() if you just want to test whether the connection is up.

The debug mode build of the library now has a "_d" suffix for Visual C++, and Xcode. This lets you have both versions installed without conflict. The release build uses the current naming scheme. If you have an existing program building against MySQL++ on these platforms, you’ll need to change your build options to use the new name in debug mode.

Renamed NO_LONG_LONGS to MYSQLPP_NO_LONG_LONGS to avoid a risk of collision in the global macro namespace.


Most MySQL++ classes with at() or operator []() methods now throw the new BadIndex exception when you pass an out-of-range index. These methods variously either did not check their indices, or threw std::out_of_range when passed a bad index.

I say “most” because there is at least one MySQL++ class that doesn’t follow this rule. Fields is just a typedef for a specialization of std::vector, and the Standard has its own rules for index checking.

10.2.†ABI Changes

This section documents those library changes that require you to rebuild your program so that it will link with the new library. Most of the items in the previous section are also ABI changes, but this section is only for those items that shouldn’t require any code changes in your program.

If you were going to rebuild your program after installing the new library anyway, you can probably ignore this section.


The Query classes now subclass from stringstream instead of the deprecated strstream.


Fixed several const-incorrectnesses in the Query classes.


Removed “reset query” parameters from several Query class members. This is not an API change, because the parameters were given default values, and the library would ignore any value other than the default. So, any program that tried to make them take another value wouldn’t have worked anyway.


Some freestanding functions didn’t get moved into namespace mysqlpp when that namespace was created. This release fixed that. It doesn’t affect the API if your program’s C++ source files say using namespace mysqlpp within them.


Removed Connection::infoo(). (I’d call this an API change if I thought there were any programs out there actually using this...)

Collapsed the Connection constructor taking a bool (for setting the throw_exceptions flag) and the default constructor into a single constructor using a default for the parameter.

Classes Connection and Query are now derived from the Lockable interface, instead of implementing their own lock/unlock functions.

In several instances, functions that took objects by value now take them by const reference, for efficiency.

Merged SQLQuery class’s members into class Query.

Merged RowTemplate class’s members into class Row.

Reordered member variable declarations in some classes. The most common instance is when the private section was declared before the public section; it is now the opposite way. This can change the object’s layout in memory, so a program linking to the library must be rebuilt.

Simplified the date and time class hierarchy. Date used to derive from mysql_date, Time used to derive from mysql_time, and DateTime used to derive from both of those. All three of these classes used to derive from mysql_dt_base. All of the mysql_* classes’ functionality and data has been folded into the leaf classes, and now the only thing shared between them is their dependence on the DTbase template. Since the leaf classes’ interface has not changed and end-user code shouldn’t have been using the other classes, this shouldn’t affect the API in any practical way.

mysql_type_info now always initializes its private num member. Previously, this would go uninitialized if you used the default constructor. Now there is no default ctor, but the ctor taking one argument (which sets num) has a default.


Removed reset_query parameters from Query member functions. None of these have been honored at least going back to v1.7.9, so this is not an API change. As of this version, Query now automatically detects when it can safely reset itself after executing a query, so it’s not necessary to ask for a reset except when using template queries.

Removed overloads of Query::execute(), store(), and use() that take only a const char*. This is not an API change because there was an equivalent call chain for this already. This change just snaps a layer of indirection.

Query::error() is now const and returns const char* instead of a std::string by value.

Removed Lockable mechanism as it was conceptually flawed. Connection and Query consequently no longer derive from Lockable. Since it was basically useless in prior versions, it can’t be construed as an API change.


Connection::thread_aware(), thread_start() and thread_end() are now static methods, so a program can call them before creating a connection. Ditto for DBDriver methods of the same name.

ConnectionPool::release() is now virtual, so a subclass can override it.


ConnectionPool::grab() is now virtual; same reason as above.

Query can now be tested in bool context, as was intended for v3.0.0. Had to change the “safe bool” method signature to make it happen, so technically it’s an API change, but it’s still used the same way.


The addition of a few new virtual methods to ConnectionPool inadvertently changed the library ABI. I knew adding fields changed the ABI, but erroneously assumed that the inverse of that truth — that adding methods was always safe — was also true. Adding normal methods is safe, but adding virtual methods breaks the ABI because it changes the class’s vtable size.

That left us with two bad choices: either we could come out with a 3.1.1 that removed these methods to restore the prior ABI, or we could just declare this the “new ABI” and move on, resolving not to fall into this trap again. We’ve chosen the latter path.

[25] In MySQL++ v2, data retreival (Query::storein(), SSQLS(const Row& other), etc.) worked fine regardless of whether your SSQLS field names matched those in the corresponding SQL table, because the SSQLS was populated by position, not by field name. Thus, if all you used SSQLS for was data retrieval, you could define your structures with sql_create_* in v2. This was never recommended, because such an SSQLS wouldn’t work with other features of MySQL++ like Query::insert() because they depend on being able to map names from C++ to SQL and back. You needed to use sql_create_c_names_* to make these features work in v2 in the face of a naming scheme difference between C++ and SQL.

[26] These typedefs have been available since MySQL++ v2.1.