sscanf rejecting leading zeros in integer reads

Question

I want to create a simple "date" read using sscanf which accepts inputs as:

"dd/mm/yyyy"

Both "dd" and "mm" fields can be, as much, 2 digits longs (eg. 0, 6 or 11, but not 123). The "years" field can be either 0 or a four digits field. A value of 0 in any of these three fields means the day, month or year of the system have to be taken instead.

That format must be strict, so, if the format of the input doesn't fit the pattern, the user must be notified.

My attempt is:

int d, m, y;
char const* input = "23/7/1990";

int n = sscanf(input, "%2u/%2u/%4u", &d, &m, &y);

if (n != 3) throw InvalidDate("Invalid format");

// Fill 0 values with system date.
// Check date correctness with `mktime` and `localtime`.

The problem is this sscanf format accepts non-allowed inputs as:

char const* invalid1 = "23/ 12/ 1990";
char const* invalid2 = "23/12/1990/123whatever......."

So, is there any tricks/modifiers to reject leading zeros before integers, to mark the end of string, or to cause a detectable failure if more input is parsed?

For the last case (invalid2; detectable failure at the end of string), a possible solution would be:

int d, m, y;
char trick;

char const* input = "23/7/1990";

int n = sscanf(input, "%2u/%2u/%4u%c", &d, &m, &y, &trick);

// If it fills four fields, means the input was too long.
if (fields != 3) throw InvalidDate("Invalid format");

// Fill 0 values with system date.

But I don't know if there is a better way to detect the end-of-string. Even more, this "format" (with the leading '%c') causes sscanf believes the input was wrong with valid dates (for example, "23/6/1990" provokes the last char is not fill; if scanf were used instead of sscanf, ferror would be set). I have even tried with "%2u/%2u/%4u\0" but the compiler warns me with "embedded \0 in format".

So, what is the best solution without using regular expressions or stringstream?

By the way, is there other ways to "cheat" sscanf?

Answer 1

You may use boost regex library which can do a lot of these stuffs. Check the code below:

#include <boost/regex.hpp>
#include <iostream>
#include <string>

int main()
{
    // Expression to match
    boost::regex e("(^\\d{1,2})/(\\d{1,2})/(\\d{4})$");

    // Results are here
    boost::match_results<std::string::const_iterator>  results;

    std::string val_to_match = "1/11/1990";
    if (boost::regex_search(val_to_match, results, e) && results.size() == 4) {
        std::cout << "Matched "  << results[0] << std::endl; 
        int i = 1;
        while (i < 4) {
            std::cout << "Value: " << i <<  "  "<< results[i] << std::endl;
            i++;
        }
    } else {
        std::cout << "Couldn't match \n";
    }

    return 0;
}

Answer 2

Modified your code, and got this working:

 void parseDate(const char *date) {

      char trick;
      int d, m, y, n = sscanf(date, "%2u/%2u/%4u%c", &d, &m, &y, &trick);

      (n != 3 || y < 999)) ? 
           puts("Invalid format!") : printf("%u %u %u\n", d, m, y);
 }

You mentioned that "year" can be either zero or a four-digit number, so I modified your code to accept 1000 to 9999 only. Or else, 23/7/1a990 case will have a year of 1.

Tested this one and put the output to a file.

Results:

Sample date: 23/7/1990
Output: 23 7 1990

Sample date: 23/12/1990/123whatever.......
Output: Invalid format!

Sample date: 23/ 12/ 1990
Output: 23 12 1990

Sample date: 23/12/19a90
Output: Invalid format!

Sample date: 2a/1
Output: Invalid format!

Sample date: a23/12/1990
Output: Invalid format!

Sample date: 23/12/199000
Output: Invalid format!

You can refer to this thread: How to parse and validate a date in std::string in C++?. One answer there suggests using strptime.

Answer 3

How about this? You can use %[^0-9] conversion specification to read characters between two numbers.

#include <stdio.h>
#include <string.h>

void process_date(const char* input){
  int d, m, y;
  char sep1[3], sep2[3], trick;
  int n;

  n = sscanf(
    input, "%2u%2[^0-9]%2u%2[^0-9]%4u%c",
    &d, sep1, &m, sep2, &y, &trick);

  if(!(n == 5 && strcmp(sep1, "/") == 0 && strcmp(sep2, "/") == 0)){
    fprintf(stderr, "Invalid format (input = %s).\n", input);
    return;
  }

  printf("d = %d, m = %d, y = %d.\n", d, m, y);
}

int main(){
  process_date("23/7/1990");
  process_date("23/12/1990");
  process_date("23/7/0");
  process_date("23/0/1990");
  process_date("0/7/1990");

  process_date("23/ 12/ 1990");
  process_date("23/12/1990/123whatever.......");
  process_date("123/7/1990");
  process_date("23/12/19a90");
  process_date("2a/1");
  process_date("a23/12/1990");
  process_date("23/12/199000");

  return 0;
}

Outputs:

d = 23, m = 7, y = 1990.
d = 23, m = 12, y = 1990.
d = 23, m = 7, y = 0.
d = 23, m = 0, y = 1990.
d = 0, m = 7, y = 1990.
Invalid format (input = 23/ 12/ 1990).
Invalid format (input = 23/12/1990/123whatever.......).
Invalid format (input = 123/7/1990).
Invalid format (input = 23/12/19a90).
Invalid format (input = 2a/1).
Invalid format (input = a23/12/1990).
Invalid format (input = 23/12/199000).

Answer 4

How about something like this? It doesn't use sscanf, but as has been said in the comments, it'd be hard to make that function work as you want:

int d, m, y;

int date[3];        //holds day/month/year in its cells
int tokenCount = 0;
char* pc;
int result = 0;
char* pch = strtok(input, "/");

while (pch != NULL)
{
    if (strlen(pch) == 0)
    {
        throw InvalidDate("Invalid format");
    }

    //atoi is stupid, there's no way to tell whether the string didn't contain a valid integer or if it contained a zero
    result = strtol(pch, &pc, 10);
    if (*pc != 0)
    {
        throw InvalidDate("Invalid format");
    }

    if (tokenCount > 2)     //we got too many tokens
    {
        throw InvalidDate("Invalid format");
    }

    date[tokenCount] = result;
    tokenCount++;

    pch = strtok(NULL, "/");
}

if (tokenCount != 3)
{
    //not enough tokens were supplied
    throw InvalidDate("Invalid format");
}


d = date[0];
m = date[1];
y = date[2];

You can then do some more checking, such as whether the month is between 1-12.

One thing to bear in mind is that strtok modifies the string it receives, so make sure to make a copy.

Answer 5

So. Since it seems everybody agrees there's no way to make sscanf fits better to this pattern, I think the best solution is:

char const* input = "23/7/1990";

int d, m, y;

{ // Search blanks due to `sscanf` limitations.
    for (unsigned i = 0; i < 10 and input[i] != '\0'; ++i)
        if (isspace(input[i]))
           throw InvalidDate("Invalid format");

} { // Check format (with extra input detection).
    char trick;
    int n = sscanf(input, "%2u/%2u/%4u%c", &d, &m, &y, &trick);

    if (n != 3 or (y != 0 and y < 1000))
        throw InvalidDate("Invalid format");
}

// Fill 0 values with system date.
// Check date correctness with `mktime` and `localtime`.

EDIT: Before, I used strpbrk to detect blanks (sscanf ignores it before numbers). The problem with that solution was strpbrk parses the complete input until it finds something. If input is too long, but without blanks, the execution would be very slow. Since I know the max allowable size of input, I've change it by a 10-loop for which uses isspace.

Performance can be of course improved by throwing if '\0' is found too soon, but determining "too soon" within for would be too verbose . So, I left this work for sscanf, making the first for better-defined.

Any other "complains" to that solution are strongly wellcome.