The previous chapter introduced the major top-level mechanisms in MySQL++. Now we’ll dig down a little deeper and get into real examples. We start off with the basics that every MySQL++ program will have to deal with, then work up to more complex topics that are still widely interesting. You can stop reading the manual after this chapter and still get a lot out of MySQL++, ignoring the more advanced parts we present in later chapters. All of the examples are complete running programs. If you built the library from source, the examples should have been built as well. If you use RPMs instead, the example programs’ source code and a simplified Makefile are in the mysql++-devel package. They are typically installed in /usr/share/doc/mysql++-devel-*/examples, but it can vary on different Linuxes. Before you get started, please read through any of the README*.txt files included with the MySQL++ distribution that are relevant to your platform. We won’t repeat all of that here. Most of the examples require a test database, created by resetdb. You can run it like so: resetdb [-s server_addr] [-u user] [-p password] Actually, there’s a problem with that. It assumes that the MySQL++ library is already installed in a directory that the operating system’s dynamic linker can find. (MySQL++ is almost never built statically.) Unless you’re installing from RPMs, you’ve had to build the library from source, and you should run at least a few of the examples before installing the library to be sure it’s working correctly. Since your operating system’s dynamic linkage system can’t find the MySQL++ libraries without help until they’re installed, we’ve created a few helper scripts to help run the examples. MySQL++ comes with the exrun shell script for Unixy systems, and the exrun.bat batch file for Windows. You pass the example program and its arguments to the exrun helper, which sets up the library search path so that it will use the as-yet uninstalled version of the MySQL++ library in preference to any other on your system: ./exrun resetdb [-s server_addr] [-u user] [-p password] That’s the typical form for a Unixy system. You leave off the ./ bit on Windows. You can leave it off on a Unixy system, too, if you have . in your PATH. (Not a recommendation, just an observation.) All of the program arguments are optional. If you don’t give -s, the underlying MySQL C API assumes the server is on the local machine. It chooses one of several different IPC options based on the platform configuration. There are many different forms you can give as server_addr with -s to override this default behavior: localhost — this is the default; it doesn’t buy you anything On Windows, a simple period tells the underlying MySQL C API to use named pipes, if it’s available. 172.20.0.252:12345 — this would connect to IP address 172.20.0.252 on TCP port 12345. my.server.name:svc_name — this would first look up TCP service name svc_name in your system’s network services database (/etc/services on Unixy systems, and something like c:\windows\system32\drivers\etc\services on modern Windows variants). If it finds an entry for the service, it then tries to connect to that port on the domain name given. For the TCP forms, you can mix names and numbers for the host and port/service parts in any combination. If the server name doesn’t contain a colon, it uses the default port, 3306. If you don’t give -u, it assumes your user name on the database server is the same as your login name on the local machine. If you don’t give -p, it will assume the MySQL user doesn’t have a password. (One hopes this isn’t the case...) When running resetdb, the user name needs to be for an account with permission to create the test database. Once the database is created, you can use any account when running the other examples that has DELETE, INSERT, SELECT and UPDATE permissions for the test database. The MySQL root user can do all this, of course, but you might want to set up a separate user, having only the permissions necessary to work with the test database: CREATE USER mysqlpp_test@'%' IDENTIFIED BY ’nunyabinness'; GRANT ALL PRIVILEGES ON mysql_cpp_data.* TO mysqlpp_test@'%'; You could then create the sample database with the following command: ./exrun resetdb -u mysqlpp_test -p nunyabinness (Again, leave off the ./ bit on Windows.) You may have to re-run resetdb after running some of the other examples, as they change the database. See README-examples.txt for more details on running the examples. The following example demonstrates how to open a connection, execute a simple query, and display the results. This is examples/simple1.cpp: This example simply gets the entire "item" column from the example table, and prints those values out. Notice that MySQL++’s derives from std::vector, and provides an interface that makes it a vector work-alike. This means you can access elements with subscript notation, walk through them with iterators, run STL algorithms on them, etc. Row provides a little more in this area than a plain old vector: you can also access fields by name using subscript notation. The only thing that isn’t explicit in the code above is that we delegate command line argument parsing to parse_command_line() in the excommon module. This function exists to give the examples a consistent interface, not to hide important details. You can treat it like a black box: it takes argc and argv as inputs and sends back database connection parameters. The simple1 example above was pretty trivial. Let’s get a little deeper. Here is examples/simple2.cpp: The main point of this example is that we’re accessing fields in the row objects by name, instead of index. This is slower, but obviously clearer. We’re also printing out the entire table, not just one column. By default, MySQL++ uses exceptions to signal errors. We’ve been suppressing this in all the examples so far by passing false to Connection’s constructor. This kept these early examples simple at the cost of some flexibility and power in error handling. In a real program, we recommend that you leave exceptions enabled. You do this by either using the default Connection constructor, or by using the create-and-connect constructor. All of MySQL++’s custom exceptions derive from a common base class, . That in turn derives from Standard C++’s std::exception class. Since the library can indirectly cause exceptions to come from the Standard C++ Library, it’s possible to catch all exceptions from MySQL++ by just catching std::exception. However, it’s better to have individual catch blocks for each of the concrete exception types that you expect, and add a handler for either Exception or std::exception to act as a “catch-all” for unexpected exceptions. When exceptions are suppressed, MySQL++ signals errors by returning either an error code or an object that tests as false, or by setting an error flag on the object. Classes that allow you to suppress exceptions derive from the interface. When an OptionalExceptions derivative creates another object that also derives from this interface, it passes on its exception flag. Since everything flows from the object, disabling exceptions on it at the start of the program disables all optional exceptions. This is why passing false for the Connection constructor’s “throw exceptions” parameter suppresses all optional exceptions in the simple[1-3] examples. It keeps them, well, simple. This exception suppression mechanism is quite granular. It’s possible to leave exceptions enabled most of the time, but suppress them in sections of the code where they aren’t helpful. To do this, put the section of code that you want to not throw exceptions inside a block, and create a object at the top of that block. When created, it saves the exception flag of the OptionalExceptions derivative you pass to it, and then disables exceptions on it. When the NoExceptions object goes out of scope at the end of the block, it restores the exceptions flag to its previous state: mysqlpp::Connection con; // default ctor, so exceptions enabled { mysqlpp::NoExceptions ne(con); if (!con.select_db("a_db_that_might_not_exist_yet")) { // Our DB doesn’t exist yet, so create and select it here; no need // to push handling of this case way off in an exception handler. } } When one OptionalExceptions derivative passes its exceptions flag to another such object, it is only passing a copy; the two objects’ flags operate independently. There’s no way to globally enable or disable this flag on existing objects in a single call. If you’re using the NoExceptions feature and you’re still seeing optional exceptions thrown, you disabled exceptions on the wrong object. The exception thrower could be unrelated to the object you disabled exceptions on, it could be its parent, or it could be a child created before you disabled optional exceptions. MySQL++ throws some exceptions unconditionally: MySQL++ checks array indices, always. For instance, if your code said “row[21]” on a row containing only 5 fields, you’d get a BadIndex exception. If you say “row["fred"]” on a row without a “fred” field, you get a BadFieldName exception. In the past, MySQL++ delegated some of its index checking to the STL containers underpinning it, so you could get std::range_error instead. As of MySQL++ v3.0.7, this should no longer happen, but there may be instances where it still does. will always throw when you ask it to do an improper type conversion. For example, you’ll get an exception if you try to convert “1.25” to int, but not when you convert “1.00” to int. In the latter case, MySQL++ knows that it can safely throw away the fractional part. If you use template queries and don’t pass enough parameters when instantiating the template, Query will throw a exception. If you use a C++ data type in a query that MySQL++ doesn’t know to convert to SQL, MySQL++ will throw a exception. It typically happens with , especially when using data types other than the ones defined in lib/sql_types.h. It’s educational to modify the examples to force exceptions. For instance, misspell a field name, use an out-of-range index, or change a type to force a String conversion error. SQL syntax often requires certain data to be quoted. Consider this query: SELECT * FROM stock WHERE item = 'Hotdog Buns' Because the string “Hotdog Buns” contains a space, it must be quoted. With MySQL++, you don’t have to add these quote marks manually: string s = "Hotdog Buns"; query << "SELECT * FROM stock WHERE item = " << quote_only << s; That code produces the same query string as in the previous example. We used the MySQL++ quote_only manipulator, which causes single quotes to be added around the next item inserted into the stream. This works for any type of data that can be converted to MySQL++’s SQLTypeAdapter type, plus the template. SSQLS also uses these manipulators internally. Quoting is pretty simple, but SQL syntax also often requires that certain characters be “escaped”. Imagine if the string in the previous example was “Frank’s Brand Hotdog Buns” instead. The resulting query would be: SELECT * FROM stock WHERE item = 'Frank’s Brand Hotdog Buns' That’s not valid SQL syntax. The correct syntax is: SELECT * FROM stock WHERE item = 'Frank'’s Brand Hotdog Buns' As you might expect, MySQL++ provides that feature, too, through its escape manipulator. But here, we want both quoting and escaping. That brings us to the most widely useful manipulator: string s = "Frank’s Brand Hotdog Buns"; query << "SELECT * FROM stock WHERE item = " << quote << s; The quote manipulator both quotes strings and escapes any characters that are special in SQL. MySQL++ provides other manipulators as well. See the manip.h page in the reference manual. It’s important to realize that MySQL++’s quoting and escaping mechanism is type-aware. Manipulators have no effect unless you insert the manipulator into a Query or SQLQueryParms stream. SQLQueryParms is used as a stream only as an implementation detail within the library. End user code simply sees it as a std::vector derivative. Also, values are only quoted and/or escaped if they are of a data type that may need it. For example, Date must be quoted but never needs to be escaped, and integer types need neither quoting nor escaping. Manipulators are suggestions to the library, not commands: MySQL++ will ignore these suggestions if it knows it won’t result in syntactically-incorrect SQL. It’s also important to realize that quoting and escaping in Query streams and template queries is never implicit.By contrast, the Query methods that take an SSQLS do add quotes and escape strings implicitly. It can do this because SSQLS knows all the SQL code and data types, so it never has to guess whether quoting or escaping is appropriate. You must use manipulators and template query flags as necessary to tell MySQL++ where quoting and escaping is necessary. It would be nice if MySQL++ could do quoting and escaping implicitly based on data type, but this isn’t possible in all cases.Unless you’re smarter than I am, you don’t immediately see why explicit manipulators are necessary. We can tell when quoting and escaping is not appropriate based on type, so doesn’t that mean we know when it is appropriate? Alas, no. For most data types, it is possible to know, or at least make an awfully good guess, but it’s a complete toss-up for C strings, const char*. A C string could be either a literal string of SQL code, or it can be a value used in a query. Since there’s no easy way to know and it would damage the library’s usability to mandate that C strings only be used for one purpose or the other, the library requires you to be explicit. Since MySQL++ can’t reliably guess when quoting and escaping is appropriate, and the programmer doesn’t need toOne hopes the programmer knows., MySQL++ makes you tell it. The C++ and SQL data type systems have several differences that can cause problems when using MySQL++, or any other SQL based system, for that matter. Most of the data types you can store in a SQL database are either numbers or text strings. If you’re only looking at the data going between the database server and your application, there aren’t even numbers: SQL is a textual language, so numbers and everything else gets transferred between the client and the database server in text string form.Yes, we’re aware that there is a feature in MySQL that lets you transfer row data in a binary form, but we don’t support this yet. We may, someday, probably as an extension to SSQLS. The only real reason to do so is to shave off some of the data translation overhead, which is typically neglibible in practice, swamped by the far greater disk and network I/O overheads inherent in use of a client-server database system like MySQL. Consequently, MySQL++ has a lot of special support for text strings, and can translate to several C++ numeric data types transparently. Some people worry that this translation via an intermediate string form will cause data loss. Obviously the text string data types are immune from problems in this regard. We’re also confident that MySQL++ translates BLOB and integer data types losslessly. The biggest worry is with floating-point numbers. (The FLOAT and DOUBLE SQL data types.) We did have a problem with this in older versions of MySQL++, but we believe we fixed it completely in v3.0.2. No one has since proven data loss via this path. There is still a known problem SQL’s DECIMAL data type is a configurable-precision fixed-point number format. MySQL++ currently translates these to double, a floating-point data format, the closest thing available in the C++ type system. Since the main reason to use DECIMAL is to get away from the weird roundoff behavior of floating-point numbers, this could be viewed as a serious problem. The thing is, though, in all the years MySQL++ has been around, I don’t remember anyone actually complaining about it. Apparently there’s either no one using DECIMAL with MySQL++, or they’re ignoring any roundoff errors they get as a result. Until this wheel squeaks, it’s not likely to be greased. To fix this, we’ll have to create a new custom data type to hold such column values, which will be a lot of work for apparently little return. with the SQL DECIMAL type, which is somewhat related to the floating-point issue, but it’s apparently rarely encountered, which is why it hasn’t been fixed yet. The best way to avoid problems with data translation is to always use the special MySQL++ data types defined in lib/sql_types.h corresponding to your SQL schema. These typedefs begin with sql_ and end with a lowercase version of the standard SQL type name, with spaces replaced by underscores. For instance, the SQL type TINYINT UNSIGNED is represented in MySQL++ by mysqlpp::sql_tinyint_unsigned. (See the next section for extensions to this feature for handling SQL nulls.) MySQL++ doesn’t force you to use these typedefs. It tries to be flexible with regard to data conversions, so you could probably use int anywhere you use mysqlpp::sql_tinyint_unsigned, for example. That said, the MySQL++ typedefs give several advantages: Space efficiency: the MySQL++ types are no larger than necessary to hold the MySQL data. Portability: if your program has to run on multiple different system types (even just 32- and 64-bit versions of the same operating system and processor type) using the MySQL++ typedefs insulates your code from platform changes. Clarity: using C++ types named similarly to the SQL types reduces the risk of confusion when working with code in both languages at the same time. Compatibility: using the MySQL++ types ensures that data conversions between SQL and C++ forms are compatible. Naïve use of plain old C++ types can result in data truncation, exceptions, and worse. Type compatibility is important not just at the time you write your program, it also helps forward compatibility: we occasionally change the definitions of the MySQL++ typedefs to reduce the differences between the C++ and SQL type systems. We’ll be fixing the DECIMAL issue brought up above this way, for instance; if your program uses sql_decimal instead of the current underlying type, double, your program will pick up this improvement automatically with just a recompile. Most of these typedefs use standard C++ data types, but a few are aliases for a MySQL++ specific type. For instance, the SQL type DATETIME is mirrored in MySQL++ by mysqlpp::DateTime. For consistency, sql_types.h includes a typedef alias for DateTime called mysqlpp::sql_datetime. MySQL++ doesn’t have typedefs for the most exotic data types, like those for the geospatial types. Patches to correct this will be thoughtfully considered. Both C++ and SQL have things in them called NULL, but they differ in several ways. Consequently, MySQL++ has to provide special support for this, rather than just wrap native C++ facilities as it can with most data type issues. The primary distinction is one of type. In SQL, “NULL” is a type modifier, which affects whether you can legally store a null value in that column. There’s simply nothing like it in C++. To emulate SQL NULL, MySQL++ provides the Null template to allow the creation of distinct “nullable” versions of existing C++ types. So for example, if you have a TINYINT UNSIGNED column that can have nulls, the proper declaration for MySQL++ would be: mysqlpp::Null<mysqlpp::sql_tinyint_unsigned> myfield; As of MySQL++ 3.1, we also provide shorter aliases for such types: mysqlpp::sql_tinyint_unsigned_null myfield; These types are declared in lib/sql_types.h. You might want to scan through that to see what all is available. Template instantiations are first-class types in the C++ language, so there’s no possible confusion between this feature of MySQL++ and C++’s native NULL concept. There’s a secondary distinction between SQL null and anything available in the standard C++ type system: SQL null is a distinct value, equal to nothing else. We can’t use C++’s NULL for this because it is ambiguous, being equal to 0 in integer context. MySQL++ provides the global null object, which you can assign to a Null template instance to make it equal to SQL null: myfield = mysqlpp::null; If you insert a MySQL++ field holding a SQL null into a C++ IOstream, you get “(NULL)”, something fairly unlikely to be in a normal output string, thus reasonably preserving the uniqueness of the SQL null value. MySQL++ also tries to enforce the uniqueness of the SQL null value at compile time in assignments and data conversions. If you try to store a SQL null in a field type that isn’t wrapped by Null or try to assign a Null-wrapped field value to a variable of the inner non-wrapped type, the compiler will emit some ugly error message, yelling about CannotConvertNullToAnyOtherDataType. (The exact message is compiler-dependent.) If you don’t like these behaviors, you can change them by passing a different value for the second parameter to template Null. By default, this parameter is , meaning that we should enforce the uniqueness of SQL null. To relax the distinctions, you can instantiate the Null template with a different behavior type: or . Consider this code: mysqlpp::Null<unsigned char, mysqlpp::NullIsZero> myfield(mysqlpp::null); cout << myfield << endl; cout << int(myfield) << endl; This will print “0” twice. If you had used the default for the second Null template parameter, the first output statement would have printed “(NULL)”, and the second wouldn’t even compile. MySQL++ has two classes that work like std::string to some degree: and . These classes exist to provide functionality that std::string doesn’t provide, but they are neither derivatives of nor complete supersets of std::string. As a result, end-user code generally doesn’t deal with these classes directly, because std::string is a better general-purpose string type. In fact, MySQL++ itself uses std::string most of the time, too. But, the places these specialized stringish types do get used are so important to the way MySQL++ works that it’s well worth taking the time to understand them. The simpler of the two is SQLTypeAdapter, or STA for short.In version 2 of MySQL++ and earlier, SQLTypeAdapter was called SQLString, but it was confusing because its name and the fact that it derived from std::string suggested that it was a general-purpose string type. MySQL++ even used it this way in a few places internally. In v3, we made it a simple base class and renamed it to reflect its proper limited function. As its name suggests, its only purpose is to adapt other data types to be used with SQL. It has a whole bunch of conversion constructors, one for all data types we expect to be used with MySQL++ for values in queries. SQL queries are strings, so constructors that take stringish types just make a copy of that string, and all the others “stringize” the value in the format needed by SQL.SQLTypeAdapter doesn’t do itself. That happens elsewhere, right at the point that the STA gets used to build a query. The conversion constructors preserve type information, so this stringization process doesn’t throw away any essential information. STA is used anywhere MySQL++ needs to be able to accept any of several data types for use in a SQL query. Major users are Query’s template query mechanism and the Query stream quoting and escaping mechanism. You care about STA because any time you pass a data value to MySQL++ to be used in building a SQL query, it goes through STA. STA is one of the key pieces in MySQL++ that makes it easy to generate syntactically-correct SQL queries. If MySQL++ can be said to have its own generic string type, it’s String, but it’s not really functional enough for general use. It’s possible that in future versions of MySQL++ we’ll expand its interface to include everything std::string does, so that’s why it’s called that.If you used MySQL++ before v3, String used to be called ColData. It was renamed because starting in v2.3, we began using it for holding more than just column data. I considered renaming it SQLString instead, but that would have confused old MySQL++ users to no end. Instead, I followed the example of Set, MySQL++’s specialized std::set variant. The key thing String provides over std::string is conversion of strings in SQL value formats to their plain old C++ data types. For example, if you initialize it with the string “2007-11-19”, you can assign the String to a Date, not because Date knows how to initialize itself from String, but the reverse: String has a bunch of implicit conversion operators defined for it, so you can use it in any type context that makes sense in your application. Because Row::operator[] returns String, you can say things like this: int x = row["x"]; In a very real sense, String is the inverse of STA: String converts SQL value strings to C++ data types, and STA converts C++ data types to SQL value strings.During the development of MySQL++ v3.0, I tried merging SQLTypeAdapter and String into a single class to take advantage of this. The resulting class gave the C++ compiler the freedom to tie itself up in knots, because it was then allowed to convert almost any data type to almost any other. You’d get a tangle of ambiguous data type conversion errors from the most innocent code. String has two main uses. By far the most common use is as the field value type of Row, as exemplified above. It’s not just the return type of Row::operator[], though: it’s actually the value type used within Row’s internal array. As a result, any time MySQL++ pulls data from the database, it goes through String when converting it from the string form used in SQL result sets to the C++ data type you actually want the data in. It’s the core of the structure population mechanism in the SSQLS feature, for example. Because String is the last pristine form of data in a result set before it gets out of MySQL++’s internals where end-user code can see it, MySQL++’s sql_blob and related typedefs are aliases for String. Using anything else would require copies; while the whole “networked database server” thing means most of MySQL++ can be quite inefficient and still not affect benchmark results meaningfully, BLOBs tend to be big, so making unnecessary copies can really make a difference. Which brings us to... To avoid unnecessary buffer copies, both STA and String are implemented in terms of a reference-counted copy-on-write buffer scheme. Both classes share the same underlying mechanism, and so are interoperable. This means that if you construct one of these objects from another, it doesn’t actually copy the string data, it only copies a pointer to the data buffer, and increments its reference count. If the object has new data assigned to it or it’s otherwise modified, it decrements its reference count and creates its own copy of the buffer. This has a lot of practical import, such as the fact that even though Row::operator[] returns Strings by value, it’s still efficient. Historically, there was no way to hold arbitrary-sized blocks of raw binary data in an SQL database. There was resistance to adding such a feature to SQL for a long time because it’s better, where possible, to decompose blocks of raw binary data into a series of numbers and text strings that can be stored in the database. This lets you query, address and manipulate elements of the data block individually. A classic SQL newbie mistake is trying to treat the database server as a file system. Some embedded platforms use a database engine as a file system, but MySQL doesn’t typically live in that world. When your platform already has a perfectly good file system, you should use it for big, nondecomposable blocks of binary data in most cases. A common example people use when discussing this is images in database-backed web applications. If you store the image in the database, you have to write code to retrieve the image from the database and send it to the client; there’s more overhead, and less efficient use of the system’s I/O caching system. If you store the image in the filesystem, all you have to do is point the web server to the directory where the images live, and put a URL for that image in your generated HTML. Because you’re giving the web server a direct path to a file on disk, operation is far more efficient. Web servers are very good at slurping whole files off of disk and sending them out to the network, and operating systems are very good at caching file accesses. Plus, you avoid the overhead of pushing the data through the high-level language your web app is written in, which is typically an interpreted language, not C++. Some people still hold out on this, claiming that database engines have superior security features, but I call bunk on that, too. Operating systems and web servers are capable of building access control systems every bit as granular and secure as a database system. Occasionally you really do need to store a nondecomposable block of binary data in the database. For such cases, modern SQL database servers support BLOB data types, for Binary Large OBject. This is often just called binary data, though of course all data in a modern computer is binary at some level. The tricky part about dealing with binary data in MySQL++ is to ensure that you don’t ever treat the data as a C string, which is really easy to do accidentally. C strings treat zero bytes as special end-of-string characters, but they’re not special at all in binary data. We’ve made a lot of improvements to the way MySQL++ handles string data to avoid this problem, but it’s still possible to bypass these features, wrecking your BLOBs. These examples demonstrate correct techniques. Above, I opined that it’s usually incorrect to store image data in a database, particularly with web apps, of which CGI is a primitive form. Still, it makes a nice, simple example. Instead of a single example program, we have here a matched pair. The first example takes the name of a JPEG file on the command line along with all the other common example program parameters, loads that file into memory, and stores it in a BLOB column in the database. This example also demonstrates how to retrieve the value assigned to an auto-increment column in the previous insertion. This example uses that feature in the typical way, to create unique IDs for rows as they’re inserted. Here is examples/load_jpeg.cpp: Notice that we used the escape manipulator when building the INSERT query above. This is because mysqlpp::sql_blob is just an alias for one of the special MySQL++ string types, which don’t do automatic quoting and escaping. They can’t, because MySQL++ also uses these data types to hold raw SQL query strings, which would break due to doubled quoting and/or escaping if it were automatic. The other example in this pair is rather short, considering how much it does. It parses a CGI query string giving the image ID, uses that to retreive data loaded into the database by load_jpeg, and writes it out in the form a web server wants when processing a CGI call, all with adequate real-world error handling. This is examples/cgi_jpeg.cpp: While you can run it by hand, it’s best to install this in a web server’s CGI program directory, then call it with a URL like http://my.server.com/cgi-bin/cgi_jpeg?id=1. That retrieves the JPEG with ID 1 from the database and returns it to the web server, which will send it on to the browser. We’ve included an image with MySQL++ that you can use with this example pair, examples/logo.jpg. The class makes it easier to use SQL transactions in an exception-safe manner. Normally you create the Transaction object on the stack before you issue the queries in your transaction set. Then, when all the queries in the transaction set have been issued, you call Transaction::commit(), which commits the transaction set. If the Transaction object goes out of scope before you call commit(), the transaction set is rolled back. This ensures that if some code throws an exception after the transaction is started but before it is committed, the transaction isn’t left unresolved. examples/transaction.cpp illustrates this: One of the downsides of transactions is that the locking it requires in the database server is prone to deadlocks. The classic case where this happens is when two programs both want access to the same two rows within a single transaction each, but they modify them in opposite orders. If the timing is such that the programs interleave their lock acquisitions, the two come to an impasse: neither can get access to the other row they want to modify until the other program commits its transaction and thus release the row locks, but neither can finish the transaction because they’re waiting on row locks the database server is holding on behalf of the other program. The MySQL server is smart enough to detect this condition, but the best it can do is abort the second transaction. This breaks the impasse, allowing the first program to complete its transaction. The second program now has to deal with the fact that its transaction just got aborted. There’s a subtlety in detecting this situation when using MySQL++. By default, MySQL++ signals errors like these with exceptions. In the exception handler, you might expect to get ER_LOCK_DEADLOCK from Query::errnum() (or Connection::errnum(), same thing), but what you’ll almost certainly get instead is 0, meaning “no error.” Why? It’s because you’re probably using a Transaction object to get automatic roll-backs in the face of exceptions. In this case, the roll-back happens before your exception handler is called by issuing a ROLLBACK query to the database server. Thus, Query::errnum() returns the error code associated with this roll-back query, not the deadlocked transaction that caused the exception. To avoid this problem, a few of the exception objects as of MySQL++ v3.0 include this last error number in the exception object itself. It’s populated at the point of the exception, so it can differ from the value you would get from Query::errnum() later on when the exception handler runs. The example examples/deadlock.cpp demonstrates the problem: This example works a little differently than the others. You run one copy of the example, then when it pauses waiting for you to press Enter, you run another copy. Then, depending on which one you press Enter in, one of the two will abort with the deadlock exception. You can see from the error message you get that it matters which method you call to get the error number. What you do about it is up to you as it depends on your program’s design and system architecture. There are three major ways to execute a query in MySQL++: Query::execute(), Query::store(), and Query::use(). Which should you use, and why? execute() is for queries that do not return data per se. For instance, CREATE INDEX. You do get back some information from the MySQL server, which execute() returns to its caller in a object. In addition to the obvious — a flag stating whether the query succeeded or not — this object also contains things like the number of rows that the query affected. If you only need the success status, it’s a little more efficient to call Query::exec() instead, as it simply returns bool. If your query does pull data from the database, the simplest option is store(). (All of the examples up to this point have used this method.) This returns a object, which contains the entire result set. It’s especially convenient because StoreQueryResult derives from std::vector<mysqlpp::Row>, so it opens the whole panoply of STL operations for accessing the rows in the result set. Access rows randomly with subscript notation, iterate forwards and backwards over the result set, run STL algorithms on the set...it all works naturally. If you like the idea of storing your results in an STL container but don’t want to use std::vector, you can call Query::storein() instead. It lets you store the results in any standard STL container (yes, both sequential and set-associative types) instead of using StoreQueryResult. You do miss out on some of the additional database information held by StoreQueryResult’s other base class, , however. store*() queries are convenient, but the cost of keeping the entire result set in main memory can sometimes be too high. It can be surprisingly costly, in fact. A MySQL database server stores data compactly on disk, but it returns query data to the client in a textual form. This results in a kind of data bloat that affects numeric and BLOB types the most. MySQL++ and the underlying C API library also have their own memory overheads in addition to this. So, if you happen to know that the database server stores every record of a particular table in 1 KB, pulling a million records from that table could easily take several GB of memory with a store() query, depending on what’s actually stored in that table. For these large result sets, the superior option is a use() query. This returns a object, which is similar to StoreQueryResult, but without all of the random-access features. This is because a “use” query tells the database server to send the results back one row at a time, to be processed linearly. It’s analogous to a C++ stream’s input iterator, as opposed to a random-access iterator that a container like vector offers. By accepting this limitation, you can process arbitrarily large result sets. This technique is demonstrated in examples/simple3.cpp: This example does the same thing as simple2, only with a “use” query instead of a “store” query. Valuable as use() queries are, they should not be the first resort in solving problems of excessive memory use. It’s better if you can find a way to simply not pull as much data from the database in the first place. Maybe you’re saying SELECT * even though you don’t immedidately need all the columns from the table. Or, maybe you’re filtering the result set with C++ code after you get it from the database server. If you can do that filtering with a more restrictive WHERE clause on the SELECT, it’ll not only save memory, it’ll save bandwidth between the database server and client, and can even save CPU time. If the filtering criteria can’t be expressed in a WHERE clause, however, read on to the next section. Sometimes you must pull more data from the database server than you actually need and filter it in memory. SQL’s WHERE clause is powerful, but not as powerful as C++. Instead of storing the full result set and then picking over it to find the rows you want to keep, use Query::store_if(). This is examples/store_if.cpp: I doubt anyone really needs to select rows from a table that have a prime number in a given field. This example is meant to be just barely more complex than SQL can manage, to avoid obscuring the point. That point being, the Query::store_if() call here gives you a container full of results meeting a criterion that you probably can’t express in SQL. You will no doubt have much more useful criteria in your own programs. If you need a more complex query than the one store_if() knows how to build when given an SSQLS examplar, there are two overloads that let you use your own query string. One overload takes the query string directly, and the other uses the query string built with Query’s stream interface. SQL is more than just a database query language. Modern database engines can actually do some calculations on the data on the server side. But, this isn’t always the best way to get something done. When you need to mix code and a query, MySQL++’s Query::for_each() facility might be just what you need. This is examples/for_each.cpp: You only need to read the main() function to get a good idea of what the program does. The key line of code passes an SSQLS examplar and a functor to Query::for_each(). for_each() uses the SSQLS instance to build a select * from TABLE query, stock in this case. It runs that query internally, calling gather_stock_stats on each row. This is a pretty contrived example; you could actually do this in SQL, but we’re trying to prevent the complexity of the code from getting in the way of the demonstration here. Just as with store_if(), described above, there are two other overloads for for_each() that let you use your own query string. MySQL has a large number of options that control how it makes the connection to the database server, and how that connection behaves. The defaults are sufficient for most programs, so only one of the MySQL++ example programs make any connection option changes. Here is examples/multiquery.cpp: This is a fairly complex example demonstrating the multi-query and stored procedure features in newer versions of MySQL. Because these are new features, and they change the communication between the client and server, you have to enable these features in a connection option. The key line is right up at the top of main(), where it creates a object and passes it to Connection::set_option(). That method will take a pointer to any derivative of : you just create such an object on the heap and pass it in, which gives Connection the data values it needs to set the option. You don’t need to worry about releasing the memory used by the Option objects; it’s done automatically. The only tricky thing about setting options is that only a few of them can be set after the connection is up. Most need to be set just as shown in the example above: create an unconnected Connection object, set your connection options, and only then establish the connection. The option setting mechanism takes care of applying the options at the correct time in the connection establishment sequence. If you’re familiar with setting connection options in the MySQL C API, you’ll have to get your head around the fact that MySQL++’s connection option mechanism is a much simpler, higher-level design that doesn’t resemble the C API in any way. The C API has something like half a dozen different mechanisms for setting options that control the connection. The flexibility of the C++ type system allows us to wrap all of these up into a single high-level mechanism while actually getting greater type safety than the C API allows. By default, current MySQL servers have an 8 hour idle timeout on connections. This is not a problem if your program never has to run for more than 8 hours or reliably queries the database more often than that. And, it’s a good thing for the database server, because even an idle connection takes up server resources. Many programs must run continually, however, and may experience long idle periods, such as nights and weekends when no one is around to make the program issue database queries. It’s therefore common for people writing such programs to get a bug report from the field complaining that the program died overnight or over a long weekend, usually with some error message about the database server going away. They then check the DB server, find that it’s still running and never did restart and scratch their heads wondering what happened. What happened is that the server’s connection idle timeout expired, so it closed the connection to the client. You cannot detect this condition by calling Connection::connected(). When that returns true, it just means that either the connect-on-create constructor or the connect() call succeeded and that we haven’t observed the connection to be down since then. When the database server closes an idle connection, you won’t know it until after you try to issue a query. This is simply due to the nature of network programming. One way around this problem is to configure MySQL to have a longer idle timeout. This timeout is in seconds, so the default of 8 hours is 28,800 seconds. You would want to figure out the longest possible time that your program could be left idle, then pick a value somewhat longer than that. For instance, you might decide that the longest reasonable idle time is a long 4-day weekend — 345,600 seconds — which you could round up to 350,000 or 400,000 to allow for a little bit of additional idle time on either end of that period. Alternately, you can change this parameter on a per-connection basis by using MySQL++’s connection option. A completely different way to tackle this, if your program doesn’t block forever waiting on I/O while idle, is to periodically call Connection::ping(). Don’t ping the server too often! It takes a tiny amount of processing capability to handle a ping, which can add up to a significant amount if done often enough by a client, or even just rarely by enough clients. Also, a lower ping frequency can let your program ride through some types of network faults — a switch reboot, for instance — without needing a reconnect. I like to ping the DB server no more often than half the connection timeout. With the default of 8 hours, then, I’d ping between every 4 and 7 hours. This sends the smallest possible amount of data to the database server, which will reset its idle timer and cause it to respond, so ping() returns true. If it returns false instead, you know you need to reconnect to the server. Periodic pinging is easiest to do if your program uses asynchronous I/O, threads, or some kind of event loop to ensure that you can call something periodically even while the rest of the program has nothing to do. An interesting variant on this strategy is to ping the server before each query, or, better, before each group of queries within a larger operation. It has an advantage over pinging during idle time in that the client is about to use far more server resources to handle the query than it will take to handle the ping, so the ping time gets lost in the overhead. On the other hand, if the client issues queries frequently when not idle, it can result in a lot more pings than would happen if you just pinged every N hours while idle. Finally, some programmers prefer to wrap the querying mechanism in an error handler that catches the “server has gone away” error and tries to reestablish the connection and reissue the query. This adds some complexity, but it makes your program more robust without taking up unnecessary resources. If you did this, you could even change the server to drop idle connections more often, thus tying up fewer TCP/IP stack resources. An important limitation of the MySQL C API library — which MySQL++ is built atop, so it shares this limitation — is that you can only have one query in progress on each connection to the database server. If you try to issue a second query while one is still in progress, you get an obscure error message about “Commands out of sync” from the underlying C API library. (You normally get this message in a MySQL++ exception unless you have exceptions disabled, in which case you get a failure code and Connection::error() returns this message.) There are lots of ways to run into this limitation: The easiest way is to try to use a single object in a multithreaded program, with more than one thread attempting to use it to issue queries. Unless you put in a lot of work to synchronize access, this is almost guaranteed to fail at some point, giving the dread “Commands out of sync” error. You might then think to give each thread that issues queries its own Connection object. You can still run into trouble if you pass the data you get from queries around to other threads. What can happen is that one of these child objects indirectly calls back to the Connection at a time where it’s involved with another query. This is properly covered elsewhere, in .) One way to run into this problem without using threads is with “use” queries, discussed above. If you don’t consume all rows from a query before you issue another on that connection, you are effectively trying to have multiple concurrent queries on a single connection. Here’s a recipie for this particular disaster: UseQueryResult r1 = query.use("select garbage from plink where foobie='tamagotchi'"); UseQueryResult r2 = query.use("select blah from bonk where bletch='smurf'"); The second use() call fails because the first result set hasn’t been consumed yet. Still another way to run into this limitation is if you use MySQL’s multi-query feature. This lets you give multiple queries in a single call, separated by semicolons, and get back the results for each query separately. If you issue three queries using Query::store(), you only get back the first query’s results with that call, and then have to call store_next() to get the subsequent query results. MySQL++ provides Query::more_results() so you know whether you’re done, or need to call store_next() again. Until you reach the last result set, you can’t issue another query on that connection. Finally, there’s a way to run into this that surprises almost everyone sooner or later: stored procedures. MySQL normally returns at least two result sets for a stored procedure call. The simple case is that the stored procedure contains a single SQL query, and it succeeds: you get two results, first the results of the embedded SQL query, and then the result of the call itself. If there are multiple SQL queries within the stored procedure, you get more than two result sets. Until you consume them all, you can’t start a new query on the connection. As above, you want to have a loop calling more_results() and store_next() to work your way through all of the result sets produced by the stored procedure call. The following example demonstrates how to get information about the fields in a result set, such as the name of the field and the SQL type. This is examples/fieldinf.cpp: