How to extract real number from string?

Question

I saw this answer (you can use the minor example in Ideone there, it's equivalent), so I wrote the function below to do the job, but it will stop when reading the dot before the decimal digits and will treat pi like 3 and 14. How to change the code to extract the real number? I do not care about speed or memory usage, I just want to do the job.

float extract_number(char* str) {
    char* p = str;
    while (*p) {
        if (isdigit(*p)) {
            long val = strtol(p, &p, 10);
            printf("%ld\n", val);
            return val;
        } else {
            p++;
        }
    }
    return -1.0;
}

Here is the str I am working with:

Initializing data structure(s) took 149.898690000 seconds wall clock time.

You're going to loose a lot of precision converting from `long` to `float`. — PC Luddite, Jul 12 '15 at 20:43
Consider `strtod` instead of `strtol`. (It returns `double`, not `float`, not sure if it is OK in your case.) — AlexD, Jul 12 '15 at 20:45
Uh, what? `strtol` converts to a [long](http://en.cppreference.com/w/c/string/byte/strtol), not a float. So there's your problem. — Jongware, Jul 12 '15 at 20:57
Correct @Jongware, you deserve the credits, are you going to answer please? `strtof` solved the problem! — gsamaras, Jul 12 '15 at 21:00
@AlexD gave the hint (but for the wrong reason). You can use either `strtod` or `strtof`. — Jongware, Jul 12 '15 at 21:01
Oh @AlexD, sorry, missed your edit. Anyway, one from you, should answer then, you deserve the credits. AlexD has the token :P — gsamaras, Jul 12 '15 at 21:03
@Jongware Your comment is more precise, please go ahead with answering. — AlexD, Jul 12 '15 at 21:10

paulsm4 · Answer 1 · 2015-07-12T20:50:02.023

2

Personally, I like sscanf for string parsing:

double extract_number(char* str) {
  double x;
  int result = sscanf (str, "%lf", &x);
  if (result != 1)
    return -1.0;
  return x;
}

NOTE:

If you want to separate the integral and fractional parts of a floating point value, look at modf().

edited Jul 12 '15 at 20:50

answered Jul 12 '15 at 20:46

paulsm4

114,292
17
138
190

-1 Input overflow triggers undefined behavior, therefore it is unsafe to use the `scanf` functions to parse numbers, period. – zwol Jul 12 '15 at 20:50
Paul, really sorry, but my question was not clear. Updated and your code won't work now. – gsamaras Jul 12 '15 at 20:50
@gsamaras: 1) sscanf() is perfectly appropriate: you can check for "illegal string" by verifying "result == 1" (successfully read the "%lf" argument). 2) Use "modf()" to split integer/fraction (if your platform supports it). 3) Using "-1" to flag "illegal input" probably isn't the most robust solution. I would consider adding an "error flag" either as your function return, or as a reference parameter. In any case, I much prefer "sscanf()" to "strtod()". Because you don't have to mess with errno - which everybody and his brother uses, and which isn't always thread-safe). IMHO... – paulsm4 Jul 12 '15 at 21:09
@paulsm4 I'm sorry, it does not work to "check for 'illegal string' by verifying 'result == 1'", because, again, *input overflow triggers **undefined behavior**.* See the longer discussion on this answer: http://stackoverflow.com/questions/24302160/scanf-on-an-istream-object/24318630#24318630 The *only* fully reliable (that is, all errors are detectable and the behavior is fully defined regardless of input) numeric conversion functions in the C library are the `strto*` functions. – zwol Jul 13 '15 at 14:18
@paulsm4 Also, FYI, `errno` is required to be thread-safe by both POSIX.1(-2001) and C11 (which is the first C standard revision to have any concept of threads) and I haven't tripped over an implementation that got that wrong in at least 15 years. – zwol Jul 13 '15 at 14:22

score 2 · Accepted Answer · edited May 23 '17 at 11:51

strtol converts text to a long, not a float. From strtol:

Interprets an integer value in a byte string pointed to by str.

Use either strtof or strtod to convert to a float or double. The prototype uses a second argument, commonly called endptr, which is a char * that will be updated to point to the first not-converted character, but since you don't need it here you can set it to NULL.

A couple more optimizations on that routine:

you can safely modify the argument *str
no need for an intermediate value, you can return the result of strtof
no need for an else after an explicit return
if the value -1.0f is a possible valid return value, you should use NAN.

float extract_number (char* str)
{
    while (*str)
    {
        if (isdigit(*str))
        {
            return strtof(str, NULL);
        }
        str++;
    }
    return NAN;
}

To return NAN you need to include math.h, but it is a fairly recent addition (and, apparently, an omission before!) so it may not be available in your system. If it isn't, use this totally wicked hack (kudos to J.Kraftcheck's answer here):

return strtof("NaN", NULL);

How to extract real number from string?

2 Answers2