sscanf variable length string parsing

Question

I have a problem that is probably fairly common and likely has a beautiful hack that I am not aware of. I would greatly appreciate it if someone would enlighten me!

I am using C's sscanf() function to parse input and the format is "%d %d %d %s %d %s %d ..." where the first two %d are random ID integers (insignificant) for the string and the third is a count of the number of %d %s combinations to follow.

For instance, "12 34 2 3 yes 2 no" could be a string, where 12 and 34 random are ID's (unimportant to the problem) and 2 specifies the two combinations following of '3 yes' and '2 no'. The 3 preceding 'yes' specifies the length of the string following, and the same is true for the 'no' with a 2 before it. Where we can have a variable number of these combinations following and we want to catch them all with the sscanf.

Does anyone know of any way to do this with sscanf?

Thanks a lot!

If the C++ tag is accurate, you probably want to use a stringstream instead. It'll make this rather easier. — Jerry Coffin
– Jerry Coffin, Commented Mar 31, 2013 at 2:03
@TimHaggard: Is this a C question or a C++ question? They are different languages. — nneonneo
– nneonneo, Commented Mar 31, 2013 at 2:57
@nneonneo: The scanf function is part of the C Standard and also part of the C++ Standard. It is appropriate to tag scanf() questions with either or both languages. — Ben Voigt
– Ben Voigt, Commented Mar 31, 2013 at 3:25
@Ben But depending on which language is used, a good answer would be rather different, which would warrant two different questions/answers. Combining the two doesn't really help imo. — Voo
– Voo, Commented Mar 31, 2013 at 4:45

nneonneo · Accepted Answer · 2019-01-28 02:16:09Z

4

Just parse the string in two (or more) passes. This uses the %n format specifier to write the number of bytes processed, so we know where to pick up in subsequent passes:

int a, b, n, pos;
const char *buf = "12 34 2 3 yes 2 no";

assert(sscanf(buf, "%d %d %d %n", &a, &b, &n, &pos) == 3);
for(int i=0; i<n; i++) {
    int cur;

    int x;
    char y[20];

    assert(sscanf(buf+pos, "%d %19s %n", &x, y, &cur) == 2);
    printf("%d %s\n", x, y);
    pos += cur;
}

outputs

3 yes
2 no

edited Jan 28, 2019 at 2:16

answered Mar 31, 2013 at 2:10

nneonneo

181k37 gold badges331 silver badges412 bronze badges

Sign up to request clarification or add additional context in comments.

9 Comments

autistic Over a year ago

I'd wrap the sscanf's around assert, to make sure I know when the first returns something other than 3 and the second returns something other than 2 respectively. Aside from that, +1 because this is my favourite answer in the realm of C programming.

nneonneo Over a year ago

Apparently the effect of %n on the return value is somewhat undefined (ew). So I will add asserts for >=3/>=2.

nneonneo Over a year ago

As an aside, this may in fact be the only time I've ever found %n useful (outside of exploiting format string vulnerabilities to write arbitrary data to arbitrary stack addresses).

autistic Over a year ago

No, it's well defined. The C standard says "Execution of a %n directive does not increment the assignment count returned at the completion of execution of the fscanf function." The %n directive will always execute providing the previous directives succeed, which is why I suggested assert. Ideally, one would report an error instead of raising an assertion failure. However, including error reporting code in such an example takes the focus away from the direction of the solution. assert will hopefully bring that focus back, when errors do occur.

autistic Over a year ago

%n is also useful when you need the number of decimals in an integer. Many people will use code such as assert(scanf("%d", &val) == 1); do { decimal_count++; val /= 10; } while (val > 0); (erroneously, for negative values) when all that is necessary is assert(scanf("%d%n", &val, &decimal_count) == 1);.

|

Charles Salvia · Accepted Answer · 2013-03-31 02:09:19Z

1

There's no convenient way to do this with just sscanf. You'd need to dynamically generate the format string itself, before passing it to sscanf.

You might want to consider writing a specialized parsing routine for this where you call sscanf in a loop, or more preferably (since you specify the C++ tag) using an std::istringstream in a loop.

edited Mar 31, 2013 at 2:09

answered Mar 31, 2013 at 2:03

Charles Salvia

53.7k15 gold badges133 silver badges144 bronze badges

1 Comment

Tim Haggard Over a year ago

I meant sscanf. Sorry about the confusion!

Sheng · Accepted Answer · 2013-03-31 02:40:31Z

0

First of all, ssprintf() is used to generate string, not parse it. You should use sscanf. I do not know hwo to finish it in one sscanf(). But you can do it in a loop as:

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

int main(){
    int choice_id, count;
    char choice[20],id_1[20],id_2[20], count_str[20], choice_id_str[20];
    int index;
    char *input = "12 34 2 3 yes 2 no";

    sscanf(input, "%s %s %s", id_1, id_2, count_str);
    input += strlen(id_1)+strlen(id_2)+strlen(count_str)+2;
    count = atoi(count_str);
    for(index = 0; index< count; ++index){
        sscanf(input, " %s %s", choice_id_str, choice );
        choice_id = atoi(choice_id_str);
        // Process or store the record
        printf("%d: %s\n",choice_id, choice);
        input += strlen(choice_id_str) + strlen(choice) + 2;
    }
    return 0;
}

Compile with gcc (GCC) 4.1.2, and run with Linux. The output is:

-bash-3.2$ ./a.out
3: yes
2: no

edited Mar 31, 2013 at 2:40

answered Mar 31, 2013 at 2:10

Sheng

3,5651 gold badge20 silver badges21 bronze badges

4 Comments

Sheng Over a year ago

the input here is the to-parse string, such as "12 34 2 3 yes 2 no"

nneonneo Over a year ago

It is not clear how to move the pointer. scanf may consume an arbitrary amount of the input string.

nneonneo Over a year ago

That's incorrect...a space in sscanf could match an arbitrary amount of whitespace (not just one character).

Sheng Over a year ago

I did not notice that. But I think it still works well. I updated my code into full exampel.

user1896333 · Accepted Answer · 2013-03-31 05:05:58Z

Is there a maximum value for the number of %d %s pairs that follow the initial "%d %d %d" header? There must be since you're putting the values somewhere, perhaps in a struct { int i; char s[j]; } a[n]; with appropriate j and n sizes.

(BTW, your example uses "%d %d %d %s %d %s %d ..." when the description indicates it should be "%d %d %d %d %s %d %s %d %s ...", 3 decimals, followed by decimal/string pairs)

If there is a maximum just create a maximized template and test that the return code from sscanf, which should indicate the number of input items successfully matched and assigned, is your 3rd value minus the 3 header items. If the return code isn't correct when compared to the 3rd int, report a malformed line.

I once made something like this a function so the a[n] was automatic ion the stack and then the function allocated a linked list from the heap for the items sscanf-ed and returned a pointer to the first item.

Collectives™ on Stack Overflow

sscanf variable length string parsing

4 Answers 4

9 Comments

1 Comment

4 Comments

Comments

Your Answer

Hot Network Questions

Collectives™ on Stack Overflow

4 Answers 4

9 Comments

1 Comment

4 Comments

Comments

Your Answer

Sign up or log in

Post as a guest

Related