When converting a double to a string, what can the first string character be?

Question

When converting a double to a string like so:

static void strfromdouble(char *buf, size_t n, double d) {
    snprintf(buf, n, "%.14g", d);
}

Which characters could end up in buf[0] after the call to snprintf?

Obviously, it could be one of the following: 0-9, -, . but since we are using floating point math here, there are probably more characters that could end up in buf[0] like I for Inf or N for NaN and maybe even more?

Is the behaviour of snprintf standardized in this situation across different compilers and platforms? If it is, where is the specification, i.e. where can I look up which characters my code would have to be able to deal with in buf[0]?

Answer 1

When converting a double to a string, what can the first string character be?

Easy to generate "0123456789-+ iInN". But other char are possible, depending on how far we want to stretch it.

Obviously, it could be one of the following: 0-9, -, . but since we are using floating point math here, there are probably more characters that could end up in buf[0] like I for Inf or N for NaN and maybe even more?

snprintf(buf, n, "%.14g", d);

When n = 1, expect buf[0] == '\0'

When n = 0, expect buf[0] to be unchanged so any char value is possible there.

With "%.14g" I would not expect 'I' nor 'N', but 'i' and 'n'.
"%.14G" would produce 'I' and 'N'.
A leading '.' does not happen with this format.

With other locales, (research setlocale()), the '-' may be something else. I know of no locale that does that though.

Other formats

With other format "% something [gefaGefa]" strings:

The first character could be a space ' ' or '+'. Like '-', that can vary with locale.

If a leading '.' is possible (I am confident this is not possible with standard formats), others locales may use an alternate such as ','.

---

With implementation specific formats, any lead character is possible.

With invalid formats, any lead character is possible due to undefined behavior.

---

#include <stdio.h>

int main(void) {
  char s[1000];
  double inf = strtod("inf", 0);
  double nan = strtod("nan", 0);
  snprintf(s, sizeof s, "%g\n", 0.0);   printf("%3d %.1s\n", s[0], s);
  snprintf(s, sizeof s, "%g\n", 9.0);   printf("%3d %.1s\n", s[0], s);
  snprintf(s, sizeof s, "%g\n", -1.0);  printf("%3d %.1s\n", s[0], s);
  snprintf(s, sizeof s, "%+g\n", +1.0); printf("%3d %.1s\n", s[0], s);
  snprintf(s, sizeof s, "%g\n", inf);   printf("%3d %.1s\n", s[0], s);
  snprintf(s, sizeof s, "%G\n", inf);   printf("%3d %.1s\n", s[0], s);
  snprintf(s, sizeof s, "%g\n", nan);   printf("%3d %.1s\n", s[0], s);
  snprintf(s, sizeof s, "%G\n", nan);   printf("%3d %.1s\n", s[0], s);
  snprintf(s, sizeof s, "%5g\n", 1.0);  printf("%3d %.1s\n", s[0], s);
  snprintf(s, 1, "%g\n", 1.0);          printf("%3d %.1s\n", s[0], s);
  s[0] = 'z';
  snprintf(s, 0, "%g\n", 1.0);          printf("%3d %.1s\n", s[0], s);
  return 0;
}

Output

 48 0
 57 9
 45 -
 43 +
105 i
 73 I
110 n
 78 N
 32  
  0 
122 z

---

Is the behaviour of snprintf standardized in this situation across different compilers and platforms?

Yes. Compliant possibilities include "0123456789-+ iInN". Some compilers may add additional implementation defined ones.

---

where is the specification (?)

Look for the C spec. Latest is C17

Where do I find the current C or C++ standard documents?

---

Details:

infinity --> [-]inf or [-]infinity (efga) or uppercase version (EFGA)
nan --> [-]nan or [-]nan(n-char-sequence) or uppercase prefix version

%f --> [−]ddd.ddd,if a decimal-point character appears,at least one digit appears before it.
%F --> like %f
%e --> [−]d.ddd e±dd
%e --> like %e, but with E
%g --> converted in style f or e
%G --> converted in style F or E
%a --> [−]0xh.hhhh p±d
%A --> like %a, but with X,P