$Id: why_not_scanf.txt,v 1.1 2013/04/28 17:58:02 db Exp db $

Why scanf sucks

Occasionally the beginning C programmer is tempted to use scanf for
decoding an input stream. 

Here's a typical naive attempt.

scanf("%s %d",string, &i);

Typically the beginner is mystified when the terminal appears to hang
without returning the expected string and integer. The format string
has forgotten to account for the newline still in the buffer, so the sscanf
gets confused. Moreover, different systems have different ideas about what
should be in the buffer on a carriage return as well. If your incompetent
teacher insists you have to use sscanf() you could try a getchar(); or two
to flush the input before trying another scanf, but by far the safest bet 
is to use an fgets(); then parse the string yourself. The fgets will smoothly
pull the newline out for you. No fuss, no muss.

Now that you have replaced that scanf() with a fgets()
The second major pitfall is assuming that you can now safely use a sscanf()

fgets(input_buf, sizeof(input_buf) - 1, stdin);
sscanf(input_buf,"%s %d", string, &i);

This is certainly better but still error prone. What if your input
consists of two spaces not just one? What if the input is a string
and a float?

You are far better off doing something like

/*
 * Sample program to illustrate how to parse a simple
 * input from user
 *
 * Diane Bruce 2012
 */

#include <stdio.h>
#include <stddef.h>
#include <err.h>
#include <string.h>
#include <assert.h>
#include <stdlib.h>
#include <limits.h>
#include <err.h>
#include <sysexits.h>

#define MAXLINE 	256

int	valid_int(char *s);

int 
main(void)
{
char	input_buf[MAXLINE];
char	*string;
char	*token;  
char	*p;
char	*q;
int	i;

	fgets(input_buf, sizeof(input_buf) - 1, stdin);

	p = input_buf;
	if ((q = strchr(p,'\n')) != NULL)	/* strip any newline */
		*q = '\0';
	token = strsep(&p, " ");
	if (token == NULL || *token == '\0')
		errx(EX_DATAERR,"Missing string");
	string = strdup(token);
	assert(string != NULL);

	/* Now deal with integer */
	token = strsep(&p, " ");
	if (token == NULL || *token == '\0')
		errx(EX_DATAERR,"Missing integer");

	if (valid_int(token))
		i = strtol(token, NULL, 10);
	else
		errx(EX_DATAERR,"Garbled integer");

	printf("string = %s integer = %d\n", string, i);
}

/*
 * valid_int
 *
 * inputs	- string
 * output	- 1 if valid int , 0 if not
 * side effects	- none
 */

int
valid_int(char *s)
{

	while (*s != '\0') {
		if (*s != '-' && !isdigit(*s)) 
			return(0);

		s++;	/* N.B. isdigit is often a macro */
	}
	return (1);
}

...

The slightly more advanced C coder may be tempted to use regular expressions
in their sscanf(), after all it works in perl ever so well.
Don't be tempted.

It is far safer to painfully parse as I have shown than risk an
error due to an unexpected match in a R.E. you have missed.
This author has seen this happen in one large IRC network, where the
entire network was KLINED (every single user kicked off) due to an
unexpected match in a sscanf(). The network will remain unnamed, the coders
faces are red enough.

Note that sometimes you are forced to use sscanf(), floats are a large
exception.

sscanf("%f", float_string); 

All I can say here is make very sure float_string is sanitized so
the only thing it can contain is a float.