FreeTDS User Guide: A Guide to Installing, Configuring, and Running FreeTDS
Prev	Chapter 11. Programming	Next

DB-Library for the Tenderfoot

	Few things are harder to put up with than the annoyance of a good example.
	Mark Twain

Below is a complete sample working db-lib program, presented as a series of examples.
Features of sample code
Processes command-line options to select the server, database, username, and password
Remaining arguments on the command line comprise the SQL query to execute
Installs error and message handlers
Illustrates correct row-processing
Illustrates correct error detection and handling
Other sample code may be found in the distribution, in the cleverly named samples directory. A complete program, heavily commented for your perusal, is apps/bsqldb.c.

What's the big deal with errors?

Correct handling of errors is extremely important in database applications because they involve two systems most others don't: the network and the database server. Both can give rise to errors that, if not detected and reported when they occur, let the application proceed blithely on until something truly mysterious happens. In the worst case, in the absence of a properly reported error, the application may seem to have updated the data, when in fact it did not.

Every db-lib application uses the network, making it subject to network failures. Database programs also almost always have very high data integrity requirements. It is necessary to know the row was absolutely, positively committed, once and only once, without error or exception. Without taking great care to trap and handle all error conditions, no statement about the program's reliability can be made with confidence.

How to Get and Build the sample code

Run doc/grep_sample_code to extract the C code from the User Guide SGML source.
Compile
Link

Files Required to Build the Sample Code

sybfront.h
sybdb.h
libsybdb.a or libsybdb.so

Your library's extension may vary according to your operating system.

The source code may be built with commands similar to these. The precise options and paths depend on your particular system. The commands below work with the GNU compiler and linker on an ELF system with dynamic linking, common on Linux and BSD systems.

Example 11-1. Building the Sample Code

    $ ../doc/grep_sample_code ../doc/userguide.sgml > sample.c
    $ cc -I /usr/local/include -Wl,-L/usr/local/lib -Wl,-R/usr/local/lib sample.c -lsybdb -o sample

where /usr/local/include and /usr/local/lib are respectively the locations of your header files and libraries.

We now proceed to the code proper.

Header files

We need two header files to use db-lib. We need a few others to deal with I/O in C, as you know. Also declare the error and message handler functions, more about which later.

Example 11-2. Sample Code: db-lib header files

#include <stdio.h>
#include <stdlib.h>
#include <string.h>
#include <assert.h>
#include <errno.h>
#include <unistd.h>
#include <libgen.h>

#include <sqlfront.h>	/* sqlfront.h always comes first */
#include <sybdb.h>	/* sybdb.h is the only other file you need */

int err_handler(DBPROCESS*, int, int, int, char*, char*);
int msg_handler(DBPROCESS*, DBINT, int, int, char*, char*, char*, int);

Prolog

Nothing special here. Collect the command line parameters. We do this with the standard getopts(3) function. Cf. man 3 getopts for details.

Example 11-3. Sample Code: db-lib prolog

extern char *optarg;
extern int optind;

const static char syntax[] = 
		"syntax: example -S server -D db -U user -P passwd\n";

struct {
	char *appname, *servername, *dbname, *username, *password;
} options = {0,0,0,0,0};

int
main(int argc, char *argv[])
{
	int i, ch;
	LOGINREC *login;	
	DBPROCESS *dbproc;	
	RETCODE erc;		
	
	options.appname = basename(argv[0]);
	
	while ((ch = getopt(argc, argv, "U:P:S:D:")) != -1) {
		switch (ch) {
		case 'S':
			options.servername = strdup(optarg);
			break;
		case 'D':
			options.dbname = strdup(optarg);
			break;
		case 'U':
			options.username = strdup(optarg);
			break;
		case 'P':
			options.password = strdup(optarg);
			break;
		case '?':
		default:
			fprintf(stderr, syntax);
			exit(1);
		}
	}

	argc -= optind;
	argv += optind;
	
	if (! (options.servername
	    && options.username   && options.password)) {
		fprintf(stderr, syntax);
		exit(1);
	}

111Prolog Notes

: LOGINREC is a structure that describes the client. It's passed to the server at connect time.
: DBPROCESS is a structure that describes the connection. It is returned by dbopen().
: RETCODE is the most common return code type for db-lib functions.

Initialize

Initialize the library. Create and populate a LOGINREC record.

Example 11-4. Sample Code: db-lib Initialize


	if (dbinit() == FAIL) {
		fprintf(stderr, "%s:%d: dbinit() failed\n", 
				options.appname, __LINE__);
		exit(1);
	}
	

	dberrhandle(err_handler);
	dbmsghandle(msg_handler);


	if ((login = dblogin()) == NULL) {
		fprintf(stderr, "%s:%d: unable to allocate login structure\n", 
				options.appname, __LINE__);
		exit(1);
	}


	DBSETLUSER(login, options.username);
	DBSETLPWD(login, options.password);

112Initialization Notes

: Always make dbinit() the first db-lib call.
: Install the error- and mesage-handlers right away. They're explained in more detail later.
: dblogin() almost never fails. But check! No point in trying to use a null pointer.
: The LOGIN record isn't directly accessible. It's populated via macros like these. There are other fields, but these two are essential. Look for SETLsomething in the documentation.

Connect to the server

dbopen() forms a connection with the server. We pass our LOGINREC pointer (which describes the client end), and the name of the server. Then, optionally, we change to our favored database. If that step is skipped, the user lands in his default database.

Example 11-5. Sample Code: db-lib Connect to the server

	if ((dbproc = dbopen(login, options.servername)) == NULL) {
		fprintf(stderr, "%s:%d: unable to connect to %s as %s\n", 
				options.appname, __LINE__, 
				options.servername, options.username);
		exit(1);
	}

	if (options.dbname  && (erc = dbuse(dbproc, options.dbname)) == FAIL) {
		fprintf(stderr, "%s:%d: unable to use to database %s\n", 
				options.appname, __LINE__, options.dbname);
		exit(1);
	}

Send a query

db-lib maintains a command buffer to hold the SQL to be sent to the server. Two functions — dbcmd() and dbfcmd() — build up the query from strings of text. The command buffer is reset after the query is sent to the server.
We left the SQL on the command line. We fetch it now and send it to the server.

Example 11-6. Sample Code: db-lib Send a query

	for (i=0; i < argc; i++) {
		assert(argv[i]);
		printf("%s ", argv[i]);
		if ((erc = dbfcmd(dbproc, "%s ", argv[i])) == FAIL) {
			fprintf(stderr, "%s:%d: dbcmd() failed\n", options.appname, __LINE__);
			exit(1);	


		}
	}
	printf("\n");

	if ((erc = dbsqlexec(dbproc)) == FAIL) {
		fprintf(stderr, "%s:%d: dbsqlexec() failed\n", options.appname, __LINE__);
		exit(1);		

	}

113Initialization Notes

: Failure at this juncture is rare. The library is merely allocating memory to hold the SQL.
: dbsqlexec() waits for the server to execute the query. Depending on the complexity of the query, that may take a while.

dbsqlexec() will fail if something is grossly wrong with the query, e.g. incorrect syntax or a reference to nonexistent table. It's only the first of a few places where an error can crop up in processing the query, though. Just because dbsqlexec() succeeded doesn't mean you're in the clear.

Fetch Results

A query may produce zero, one, or more results. Broadly, that entails providing buffers to db-lib to fill, and iterating over the results a row (and column) at a time.

Kinds of Results

Results is a special term: it means more than rows or no rows. To process the results means to gather the data returned by the server into the application's variables.

Table 11-4. Kinds of Results

Type	Metadata	Regular Rows	Compute Rows	Return Status	Example SQL
None	None	None	None	None	Any `INSERT`, `UPDATE`, or `DELETE` statement
Empty	1 set	None	0 or more	None	`SELECT name FROM systypes WHERE 0 = 1`
Simple	1 set	0 or more	None	None	`SELECT name FROM sysobjects`
Complex	2 or more	0 or more	1 or more	None	`SELECT name FROM sysobjects COMPUTE COUNT(name)`
Stored Procedure	0 or more	0 or more	0 or more	1 or more	`EXEC sp_help sysobjects`

As the above table shows, results can comprise ordinary rows and compute rows (resulting from a COMPUTE clause). Stored procedures may of course contain multiple SQL statements, some of which may be SELECT statements and might include COMPUTE clauses. In addition, they generate a return status (with a RETURN statement or else automatically) and perhaps OUTPUT parameters.

Data and Metadata

Observe that a row is set of columns, and each column has attributes such as type and size. The column attributes of a row are collectively known as metadata. The server always returns metadata before any data (even for a a SELECT statement that produced no rows).

Table 11-5. Result-fetching functions

Function	Fetches	Returns	Comment
`dbresults()`	metadata	`SUCCEED`, `FAIL` or, `NO_MORE_RESULTS`.	`SUCCEED` indicates just that: the query executed successfully (whew!). There may be metadata (and perhaps data) and/or stored procedure outputs available.
`nextrow()`	data	`REG_ROW`, compute_id, `NO_MORE_ROWS`, `BUF_FULL`, or `FAIL`.	Places fetched data into bound columns, if any.

Binding

Each time dbresults() returns SUCCEED, there is something to retrieve. db-lib has different functions to deal with the different kinds of results. The functions are of two kinds: those that convert the data into a form desired by the application, known as binding, and those that return the data in "native" form.

To understand binding, it may be easiest to examine two primitive functions, dbdata() and dbconvert(). dbdata() returns a pointer to the column's data. The data to which it points are in "native" form, 4 bytes for an INT, 8 bytes for a DATETIME and so on. dbconvert() converts between datatypes; you can hand it an integer and get back a character array (or a C double. You might think of dbconvert() as atoi(3) on steroids). dbbind() combines these two functions. The application indicates in what form it would like to use each column, and the library converts them on the fly as each row is read.

To bind a column is to provide a buffer to db-lib to fill and indicate which datatype the buffer is meant to hold. [1]

It may be well to pause here to observe the three ways a datatype is described in a db-lib program.

db-lib Datatype Descriptors

Sever Datatype: Describes the data as an abstract type, not representing any particular kind of storage. SYBREAL, for example, doesn't imply any particular arrangement of bits; it just means "a floating-point datatype corresponding to the T-SQL REAL type on the server." These all begin with SYB, e.g. SYBINT4.
Program Variable Datatype: Defines a C variable in a machine-independent way. Because a C defines its int type according the CPU architecture, it may have 2, 4, 8, or some other number of bytes. A DBINT on the other hand, is guaranteed to be 4 bytes and, as such, assuredly will hold any value returned by the server from a T-SQL INT column. These all begin with DB, e.g. DBREAL.
Bind Type: Prescribes a conversion operation. Indicates to dbbind() the Program Variable Datatype defined by the target buffer. Sybase and Microsoft call this the "vartype". These all end with BIND, e.g. STRINGBIND.

Typically it's more convenient to have db-lib convert the data into the desired form. The function that does that is dbind(). So: after fetching the metadata, and before fetching the data, we usually prepare the bound columns.

Fetching Data

Table 11-6. Data-fetching functions

Type	Regular rows	Compute rows	Return status	`OUTPUT` parameters
Meta	`dbnumcols()`	`dbnumcompute()`, `dbnumalts()`, `dbaltop()`, `dbbylist()`	`dbhasretstatus()`	`dbnumrets()`
Binding	`dbbind()`, `dbnullbind()`	`dbaltbind()`, `dbanullbind()`	`dbretstatus()`	none
Native	`dbdatlen()`, `dbdata()`	`dbadlen()`, `dbalttype()`, `dbaltutype()`, `dbaltlen()`, `dbadata()`	none	`dbretdata()`, `dbretlen()`, `dbretname()`, `dbrettype()`

The paradigm may now perhaps be clear: Query, fetch results, bind columns, fetch regular rows, fetch compute rows, fetch stored procedure outputs. Repeat as necessary.

Table 11-7. Putting it all together

Step	Function	Once Per	Many Times Per
Query	`dbsqlexec()`	Query	Program
Fetch metadata	`dbresults()`	SQL statement	Query
Prepare variables	`dbbind()`	Column	Statement
Fetch regular data	`dbnextrow()`	Row	Statement
Fetch compute data	`dbnextrow()`	Compute column	Statement
Fetch output parameters	`dbretdata()`	output parameter	Stored procedure
Fetch return status	`dbretstatus()`	Stored procedure	Program

Fetch All Rows!

db-lib doesn't insist every column — or even any column — be bound or otherwise retrieved into the application's variables. There is, however, one absolutely crucial, inflexible, unalterable requirement: the application must process all rows produced by the query. Before the DBPROCESS can be used for another query, the application must either fetch all rows, or cancel the results and receive an acknowledgement from the server. Cancelling is beyond the scope of this document, so for now fetch all rows.

Now, at last, some sample code that fetches data. In the interest of simplicity, we don't bind anything except regular rows.

Example 11-7. Sample Code: db-lib Fetch Results

	while ((erc = dbresults(dbproc)) != NO_MORE_RESULTS) {
		struct COL 						
		{ 
			char *name; 
			char *buffer; 
			int type, size, status; 
		} *columns, *pcol;
		int ncols;
		int row_code;
		
		if (erc == FAIL) {
			fprintf(stderr, "%s:%d: dbresults failed\n", 
					options.appname, __LINE__);
			exit(1);
		}
		
		ncols = dbnumcols(dbproc);

		if ((columns = calloc(ncols, sizeof(struct COL))) == NULL) {
			perror(NULL);
			exit(1);
		}

		/* 
		 * Read metadata and bind.  
		 */
		for (pcol = columns; pcol - columns < ncols; pcol++) {
			int c = pcol - columns + 1;
			
			pcol->name = dbcolname(dbproc, c);		
			pcol->type = dbcoltype(dbproc, c);
			pcol->size = dbcollen(dbproc, c);
			
			if (SYBCHAR != pcol->type) {			
				pcol->size = dbwillconvert(pcol->type, SYBCHAR);
			}

			printf("%*s ", pcol->size, pcol->name);

			if ((pcol->buffer = calloc(1, pcol->size + 1)) == NULL){
				perror(NULL);
				exit(1);
			}

			erc = dbbind(dbproc, c, NTBSTRINGBIND,	
					     pcol->size+1, (BYTE*)pcol->buffer);
			if (erc == FAIL) {
				fprintf(stderr, "%s:%d: dbbind(%d) failed\n", 
						options.appname, __LINE__, c);
				exit(1);
			}
			erc = dbnullbind(dbproc, c, &pcol->status);	
			if (erc == FAIL) {
				fprintf(stderr, "%s:%d: dbnullbind(%d) failed\n", 
						options.appname, __LINE__, c);
				exit(1);
			}
		}
		printf("\n");
		
		/* 
		 * Print the data to stdout.  
		 */
		while ((row_code = dbnextrow(dbproc)) != NO_MORE_ROWS){	
			switch (row_code) {
			case REG_ROW:
				for (pcol=columns; pcol - columns < ncols; pcol++) {
					char *buffer = pcol->status == -1? 
						"NULL" : pcol->buffer;
					printf("%*s ", pcol->size, buffer);
				}
				printf("\n");
				break;
				
			case BUF_FULL:
				assert(row_code != BUF_FULL);
				break;
				
			case FAIL:
				fprintf(stderr, "%s:%d: dbresults failed\n", 
						options.appname, __LINE__);
				exit(1);
				break;
				
			default: 					
				printf("Data for computeid %d ignored\n", row_code);
			}


		}

		/* free metadata and data buffers */
		for (pcol=columns; pcol - columns < ncols; pcol++) {
			free(pcol->buffer);
		}
		free(columns);

		/* 
		 * Get row count, if available.   
		 */
		if (DBCOUNT(dbproc) > -1)
			fprintf(stderr, "%d rows affected\n", DBCOUNT(dbproc));

		/* 
		 * Check return status 
		 */
		if (dbhasretstat(dbproc) == TRUE) {
			printf("Procedure returned %d\n", dbretstatus(dbproc));
		}
	}

	dbclose(dbproc);
	dbexit();
	exit(0);
}

114Data-fetching Notes

: As soon as dbresults() reports SUCCESS, the row's metadata are available.
: db-lib columns start with 1.
: dbcollen() returns the sizeof the native data (e.g. 4 bytes for a T-SQL INT). We'll use dbbind() to convert everything to strings. If the column is [VAR]CHAR, we want the column's defined size, otherwise we want its maximum size when represented as a string, which FreeTDS's dbwillconvert() returns (for fixed-length datatypes). [2]
: NTBSTRINGBIND null-terminates the character array for us. "NTB" might perhaps stand for "null terminating byte".
: A zero-length string is not a NULL! dbnullbind() arranges for the passed buffer to be set to -1 whenever that column is NULL for a particular row.
: Each time dbnextrow() returns REG_ROW, it has filled the bound buffers with the converted values for the row.
: Computed rows are left as an exercise to the reader.

Messages and Errors

Errors may originate on the server or in the library itself. The former are known as messages (because they are: they arrive as messages from the server); the latter are termed errors. Their handling is a little intimidating. It requires writing and installing a callback function (whose parameters are predefined by db-lib), and thinking about how to handle different types of errors.

Kinds of Errors

Messages: Messages arise because the server has something to say. [3]. They usually describe some problem encountered executing the SQL. Perhaps the SQL refers to a nonexistent object or attempted to violate a constraint. But they can also be benign, indicating for instance merely that the default database has changed.
Errors: Errors arise either because the application has misused db-lib in some way — say, passed a NULL DBPROCESS pointer or tried to issue a query while results were pending — or because some trouble cropped up in communicating with the server (couldn't find it, say, or didn't hear back from it).

Why these two require distinct handling is lost in the mists of time. But it does help to keep them distinct in your mind, especially while reading the documentation.

To have db-lib use your handler, pass its name to the appropriate dberrhandle() or dbmsghandle() function immediately after calling dbinit().

Example 11-8. Sample Code: db-lib Error and Message handlers

int
msg_handler(DBPROCESS *dbproc, DBINT msgno, int msgstate, int severity, 
	    char *msgtext, char *srvname, char *procname, int line)
{									
	enum {changed_database = 5701, changed_language = 5703 };	
	
	if (msgno == changed_database || msgno == changed_language) 
		return 0;

	if (msgno > 0) {
		fprintf(stderr, "Msg %ld, Level %d, State %d\n", 
				(long) msgno, severity, msgstate);

		if (strlen(srvname) > 0)
			fprintf(stderr, "Server '%s', ", srvname);
		if (strlen(procname) > 0)
			fprintf(stderr, "Procedure '%s', ", procname);
		if (line > 0)
			fprintf(stderr, "Line %d", line);

		fprintf(stderr, "\n\t");
	}
	fprintf(stderr, "%s\n", msgtext);
	
	if (severity > 10) {						
		fprintf(stderr, "%s: error: severity %d >�10, exiting\n", 
				options.appname, severity);
		exit(severity);
	}

	return 0;							
}

int
err_handler(DBPROCESS * dbproc, int severity, int dberr, int oserr, 
	    char *dberrstr, char *oserrstr)
{									
	if (dberr) {							
		fprintf(stderr, "%s: Msg %d, Level %d\n", 
				options.appname, dberr, severity);
		fprintf(stderr, "%s\n\n", dberrstr);
	}

	else {
		fprintf(stderr, "%s: DB-LIBRARY error:\n\t", options.appname);
		fprintf(stderr, "%s\n", dberrstr);
	}

	return INT_CANCEL;						
}

Handlers are always called before the function that engendered them returns control to the application.

115Error Handling Notes

: When first writing a handler, pay careful attention to the precise type of each parameter. Only by carefully matching them will you convince a modern C compiler that the address of your function is of the type accepted by dbmsghandle(). [4]
: Some messages don't convey much, as though the server gets lonely sometimes. You're not obliged to print every one.
: Severities are defined in the server documentation, and can be set by the T-SQL RAISERROR statement.
: Message handlers always and only ever return zero.
: When first writing the handler, pay careful attention to the precise type of each parameter. Only by carefully matching them will you convince a modern C compiler that the address of your function is of the type accepted by dberrhandle(). [5]
: Some messages are so severe they provoke db-lib into calling the error handler, too! If you have both installed — and of course you do, right? — then you can skip those lacking an error number.
: While INT_CANCEL is the most common return code, it's not the only one. For one thing, the error handler's return code can control how long db-lib keeps retrying timeout errors. See the documentation for details.

No matter what the error handler says or does, it can't remedy the error. It's still an error and usually the best that can happen is that the function will return FAIL. The exception is timeout conditions, when the handler can stave off failure by requesting retries.

You may be asking yourself, "OK, fine, I can print the error message. But what if I want to communicate something back to line in my program where the error occurred? How to do that?" First of all, remember the calling function — that's your application — will learn of an error from the return code. If it needs more detail, though, there are two ways to pass it.

Set a global variable.
Use setuserdata() and getuserdata().

If your application is written in C++, you may be tempted to use throw().�Don't! Your handler is a C function and, more important, it's an extension of db-lib. You can put a throw() in your handler and it will compile. But when it executes, it's going to rip through db-lib's stack. Your application will be unuseable at that point, if it doesn't cause a segment fault.

Last Remarks

We've reached the end of our db-lib tour. The almost 300 lines of C above constitute program with these features:

Sample Code features

Accepts command-line parameters and SQL.
Checks for errors and server messages.
Processes any number of results..
Prints results in columns of suitable widths.

There are things it doesn't do, in the name of simplicity.

Sample Code nonfeatures

No BCP (bulk copy) mode
No RPC (remote procedure call) mode, preventing it from retrieving output parameters.

Your humble author hopes you found it worthwhile. Happy Hacking.

Prev	Home	Next
ODBC API Implementation Summary	Up	Acknowledgments

[1]	This is the sort of thing `C++`'s type system does so much better
[2]	For IMAGE data, we need to multiply by 2, because `dbbind()` will convert each byte to a hexadecimal pair. The example program will report an error with IMAGE data.
[3]	Just one more way in which databases differ from files.
[4]	Back in K&R days, that wasn't such a problem. But there were other problems, some much worse.
[5]	If that advice sounds familiar, it's because it bears repeating.