173

itoa() is a really handy function to convert a number to a string. Linux does not seem to have itoa(), is there an equivalent function or do I have to use sprintf(str, "%d", num)?

5
  • 4
    any reason not to use sprintf(str, "%d", num)? is it much slower than itoa? Commented Sep 15, 2014 at 18:53
  • 5
    @javapowered, for one, itoa allows arbitrary base conversion, printf specifiers don't. Commented Aug 26, 2016 at 15:35
  • @javapowered sprintf() is not signal safe Commented Apr 3, 2019 at 15:28
  • Any reason not to use gcvt() from standard library? Commented May 6, 2019 at 10:13
  • @C--, for one, gcvt is not part of C standard and POSIX.1-2008 removed it from POSIX too. Commented Apr 6, 2022 at 5:39

20 Answers 20

128

EDIT: Sorry, I should have remembered that this machine is decidedly non-standard, having plugged in various non-standard libc implementations for academic purposes ;-)

As itoa() is indeed non-standard, as mentioned by several helpful commenters, it is best to use sprintf(target_string,"%d",source_int) or (better yet, because it's safe from buffer overflows) snprintf(target_string, size_of_target_string_in_bytes, "%d", source_int). I know it's not quite as concise or cool as itoa(), but at least you can Write Once, Run Everywhere (tm) ;-)

Here's the old (edited) answer

You are correct in stating that the default gcc libc does not include itoa(), like several other platforms, due to it not technically being a part of the standard. See here for a little more info. Note that you have to

#include <stdlib.h> 

Of course you already know this, because you wanted to use itoa() on Linux after presumably using it on another platform, but... the code (stolen from the link above) would look like:

Example

/* itoa example */ #include <stdio.h> #include <stdlib.h> int main () { int i; char buffer [33]; printf ("Enter a number: "); scanf ("%d",&i); itoa (i,buffer,10); printf ("decimal: %s\n",buffer); itoa (i,buffer,16); printf ("hexadecimal: %s\n",buffer); itoa (i,buffer,2); printf ("binary: %s\n",buffer); return 0; } 

Output:

Enter a number: 1750 decimal: 1750 hexadecimal: 6d6 binary: 11011010110 
Sign up to request clarification or add additional context in comments.

9 Comments

Hmmm, compiling that on Debian gives me "undefined reference to `itoa'". Maybe something is wrong with my system.
I get the same on Ubuntu 8.04. I can find no reference to itoa in stdio.h or stdlib.h either (not suprising since it is not part of the standard)
edited for correctness, thanks guys! sorry, I always forget that this isn't a vanilla Linux box ;-)
I have edited the answer to include the buffer size argument; I believe everything is as it should be now, I don't see a problem with the order of the arguments per se. Am I missing something?
It doesn't work for Linux? what is the outcome of the question/answer (Non standard seems to be all linuxes?)
|
31

itoa is not a standard C function. You can implement your own. It appeared in the first edition of Kernighan and Ritchie's The C Programming Language, on page 60. The second edition of The C Programming Language ("K&R2") contains the following implementation of itoa, on page 64. The book notes several issues with this implementation, including the fact that it does not correctly handle the most negative number

 /* itoa: convert n to characters in s */ void itoa(int n, char s[]) { int i, sign; if ((sign = n) < 0) /* record sign */ n = -n; /* make n positive */ i = 0; do { /* generate digits in reverse order */ s[i++] = n % 10 + '0'; /* get next digit */ } while ((n /= 10) > 0); /* delete it */ if (sign < 0) s[i++] = '-'; s[i] = '\0'; reverse(s); } 

The function reverse used above is implemented two pages earlier:

 #include <string.h> /* reverse: reverse string s in place */ void reverse(char s[]) { int i, j; char c; for (i = 0, j = strlen(s)-1; i<j; i++, j--) { c = s[i]; s[i] = s[j]; s[j] = c; } } 

5 Comments

well, maybe let's avoid calling strlen() within the loop - it should be called once, before the loop starts... second thing: do we need two counters (i and j) ? wouldn't it be enough to iterate i from 0 to < mid (len/2) and swap elements under indices i and len - 1 - i ?
@moremover; This implementation is from "The C Programming Language" by K&R, not mine.
yeah... strlen() is called before the loop actually... didn't notice it at first...
Yes. It is called only once to initialize j.
Handling the most-negative number is easily solved by taking the unsigned absolute value. unsigned abs = n < 0 ? (0U - n) : n. Also, instead of reversing as a 2nd step, you can store backwards starting from the end of the buffer, or into a small scratch buffer, so you only need memmove or memcpy. (The upper limit on buffer size is about log10(INT_MAX), e.g. the longest signed 64-bit number is 19 decimal digits plus a '-'. But I don't know a good way to make a compile-time constant expression out of INT_MAX or INT_MIN, and a hard-coded char buf[24] is ugly.)
12

If you are calling it a lot, the advice of "just use snprintf" can be annoying. So here's what you probably want:

const char *my_itoa_buf(char *buf, size_t len, int num) { static char loc_buf[sizeof(int) * CHAR_BITS]; /* not thread safe */ if (!buf) { buf = loc_buf; len = sizeof(loc_buf); } if (snprintf(buf, len, "%d", num) == -1) return ""; /* or whatever */ return buf; } const char *my_itoa(int num) { return my_itoa_buf(NULL, 0, num); } 

6 Comments

It's not just non-thread safe, it's not very safe at all :- void some_func(char* a, char* b); some_func(itoa(123), itoa(456)); Care to guess what the function recieves?
Also, const qualifiers do nothing on function return types -- you would know this if you turned on compiler warnings :)
@cat But there aren’t any const-qualified return types here. const char * is a non-const pointer to const, which makes a lot of sense and is correct.
@Chortos-2 That's interesting, you are, of course, completely correct -- I didn't realise the semantic difference in meaning of const between const int f (void) { ... and const int* f (void) { ..., but now having tried it with a compiler, it makes sense.
The sizing of the static buffer makes no sense at all, since the string put into it is not binary. On a typical machine it's going to be 32 characters, but (assuming 32-bit int) the longest string is "-2147483648" which needs only 12 characters. Maybe it's a safe, generous, upper bound, though?
|
10

Edit: I just found out about std::to_string which is identical in operation to my own function below. It was introduced in C++11 and is available in recent versions of gcc, at least as early as 4.5 if you enable the c++0x extensions.


Not only is itoa missing from gcc, it's not the handiest function to use since you need to feed it a buffer. I needed something that could be used in an expression so I came up with this:

std::string itos(int n) { const int max_size = std::numeric_limits<int>::digits10 + 1 /*sign*/ + 1 /*0-terminator*/; char buffer[max_size] = {0}; sprintf(buffer, "%d", n); return std::string(buffer); } 

Ordinarily it would be safer to use snprintf instead of sprintf but the buffer is carefully sized to be immune to overrun.

See an example: http://ideone.com/mKmZVE

1 Comment

The question seems to be about C, which has no std:: stuff etc.
6

As Matt J wrote, there is itoa, but it's not standard. Your code will be more portable if you use snprintf.

Comments

5

Reading the code of guys who do it for a living will get you a LONG WAY.

Check out how guys from MySQL did it. The source is VERY WELL COMMENTED and will teach you much more than hacked up solutions found all over the place.

MySQL's implementation of int2str

I provide the mentioned implementation here; the link is here for reference and should be used to read the full implementation.

char * int2str(long int val, char *dst, int radix, int upcase) { char buffer[65]; char *p; long int new_val; char *dig_vec= upcase ? _dig_vec_upper : _dig_vec_lower; ulong uval= (ulong) val; if (radix < 0) { if (radix < -36 || radix > -2) return NullS; if (val < 0) { *dst++ = '-'; /* Avoid integer overflow in (-val) for LLONG_MIN (BUG#31799). */ uval = (ulong)0 - uval; } radix = -radix; } else if (radix > 36 || radix < 2) return NullS; /* The slightly contorted code which follows is due to the fact that few machines directly support unsigned long / and %. Certainly the VAX C compiler generates a subroutine call. In the interests of efficiency (hollow laugh) I let this happen for the first digit only; after that "val" will be in range so that signed integer division will do. Sorry 'bout that. CHECK THE CODE PRODUCED BY YOUR C COMPILER. The first % and / should be unsigned, the second % and / signed, but C compilers tend to be extraordinarily sensitive to minor details of style. This works on a VAX, that's all I claim for it. */ p = &buffer[sizeof(buffer)-1]; *p = '\0'; new_val= uval / (ulong) radix; *--p = dig_vec[(uchar) (uval- (ulong) new_val*(ulong) radix)]; val = new_val; while (val != 0) { ldiv_t res; res=ldiv(val,radix); *--p = dig_vec[res.rem]; val= res.quot; } while ((*dst++ = *p++) != 0) ; return dst-1; } 

3 Comments

A link to a potential solution is always welcome, but please add context around the link so your fellow users will have some idea what it is and why it’s there. Always quote the most relevant part of an important link, in case the target site is unreachable or goes permanently offline. Take into account that being barely more than a link to an external site is a possible reason as to Why and how are some answers deleted?.
So what's so good about the snippet you posted here? What should future readers look out for?
Well, the elephant in the room is, of course, buffer overflow. How big is the supplied buffer (dst)? And, will 65 always be large enough for a long int? Sure, that's a uint64_t represented in binary, but we will inevitably have a uint128_t.
5

Where is the itoa function in Linux?

There is no such function in Linux. I use this code instead.

/* ============= itoa Convert integer to string PARAMS: - value A 64-bit number to convert - str Destination buffer; should be 66 characters long for radix2, 24 - radix8, 22 - radix10, 18 - radix16. - radix Radix must be in range -36 .. 36. Negative values used for signed numbers. ============= */ char* itoa (unsigned long long value, char str[], int radix) { char buf [66]; char* dest = buf + sizeof(buf); boolean sign = false; if (value == 0) { memcpy (str, "0", 2); return str; } if (radix < 0) { radix = -radix; if ( (long long) value < 0) { value = -value; sign = true; } } *--dest = '\0'; switch (radix) { case 16: while (value) { * --dest = '0' + (value & 0xF); if (*dest > '9') *dest += 'A' - '9' - 1; value >>= 4; } break; case 10: while (value) { *--dest = '0' + (value % 10); value /= 10; } break; case 8: while (value) { *--dest = '0' + (value & 7); value >>= 3; } break; case 2: while (value) { *--dest = '0' + (value & 1); value >>= 1; } break; default: // The slow version, but universal while (value) { *--dest = '0' + (value % radix); if (*dest > '9') *dest += 'A' - '9' - 1; value /= radix; } break; } if (sign) *--dest = '-'; memcpy (str, dest, buf +sizeof(buf) - dest); return str; } 

6 Comments

You should edit your answer to explain how this code answers the question.
Code bugged... negative values not working
calandoa, can you give precise values (value,radix) that do not work ?
@rick-rick-rick (don't forget the @ so the user is notified!): postive radix and negative value. BTW do not use negative radix, or do it properly: en.wikipedia.org/wiki/Negative_base
Shouldn't 65 be enough for the buffer? 2^64 - 1 is 64 binary digits plus 1 for the null terminator is 65.
|
4

Following function allocates just enough memory to keep string representation of the given number and then writes the string representation into this area using standard sprintf method.

char *itoa(long n) { int len = n==0 ? 1 : floor(log10l(labs(n)))+1; if (n<0) len++; // room for negative sign '-' char *buf = calloc(sizeof(char), len+1); // +1 for null snprintf(buf, len+1, "%ld", n); return buf; } 

Don't forget to free up allocated memory when out of need:

char *num_str = itoa(123456789L); // ... free(num_str); 

N.B. As snprintf copies n-1 bytes, we have to call snprintf(buf, len+1, "%ld", n) (not just snprintf(buf, len, "%ld", n))

3 Comments

It's not a good idea to call your function itoa but give it different behaviour to what common implementations of itoa actually have. This function is an OK idea but call it something else :) I'd also suggest using snprintf to calculate the buffer length instead of the floating point string; floating point can have corner case inaccuracies. And don't cast calloc
This should use labs if it's going to take a long integer. Else it might truncate.
snprintf into a fixed-size tmp buffer like char buf[64] to get the length, then malloc and copy into that. You're not getting any benefit out of calloc over malloc, since you write all the bytes. Extra copying of a very short string is less bad than having to call floating point log10. A fast approximation with an integer log2 could be useful, though, if you have a bit-scan function that will reliably inline to something efficient (like bsr on x86). (Alternative: malloc a 64 byte buffer and then realloc after you know the final length.)
3

i tried my own implementation of itoa(), it seem's work in binary, octal, decimal and hex

#define INT_LEN (10) #define HEX_LEN (8) #define BIN_LEN (32) #define OCT_LEN (11) static char * my_itoa ( int value, char * str, int base ) { int i,n =2,tmp; char buf[BIN_LEN+1]; switch(base) { case 16: for(i = 0;i<HEX_LEN;++i) { if(value/base>0) { n++; } } snprintf(str, n, "%x" ,value); break; case 10: for(i = 0;i<INT_LEN;++i) { if(value/base>0) { n++; } } snprintf(str, n, "%d" ,value); break; case 8: for(i = 0;i<OCT_LEN;++i) { if(value/base>0) { n++; } } snprintf(str, n, "%o" ,value); break; case 2: for(i = 0,tmp = value;i<BIN_LEN;++i) { if(tmp/base>0) { n++; } tmp/=base; } for(i = 1 ,tmp = value; i<n;++i) { if(tmp%2 != 0) { buf[n-i-1] ='1'; } else { buf[n-i-1] ='0'; } tmp/=base; } buf[n-1] = '\0'; strcpy(str,buf); break; default: return NULL; } return str; } 

Comments

2

direct copy to buffer : 64 bit integer itoa hex :

 char* itoah(long num, char* s, int len) { long n, m = 16; int i = 16+2; int shift = 'a'- ('9'+1); if(!s || len < 1) return 0; n = num < 0 ? -1 : 1; n = n * num; len = len > i ? i : len; i = len < i ? len : i; s[i-1] = 0; i--; if(!num) { if(len < 2) return &s[i]; s[i-1]='0'; return &s[i-1]; } while(i && n) { s[i-1] = n % m + '0'; if (s[i-1] > '9') s[i-1] += shift ; n = n/m; i--; } if(num < 0) { if(i) { s[i-1] = '-'; i--; } } return &s[i]; } 

note: change long to long long for 32 bit machine. long to int in case for 32 bit integer. m is the radix. When decreasing radix, increase number of characters (variable i). When increasing radix, decrease number of characters (better). In case of unsigned data type, i just becomes 16 + 1.

Comments

2

Here is a much improved version of Archana's solution. It works for any radix 1-16, and numbers <= 0, and it shouldn't clobber memory.

static char _numberSystem[] = "0123456789ABCDEF"; static char _twosComp[] = "FEDCBA9876543210"; static void safestrrev(char *buffer, const int bufferSize, const int strlen) { int len = strlen; if (len > bufferSize) { len = bufferSize; } for (int index = 0; index < (len / 2); index++) { char ch = buffer[index]; buffer[index] = buffer[len - index - 1]; buffer[len - index - 1] = ch; } } static int negateBuffer(char *buffer, const int bufferSize, const int strlen, const int radix) { int len = strlen; if (len > bufferSize) { len = bufferSize; } if (radix == 10) { if (len < (bufferSize - 1)) { buffer[len++] = '-'; buffer[len] = '\0'; } } else { int twosCompIndex = 0; for (int index = 0; index < len; index++) { if ((buffer[index] >= '0') && (buffer[index] <= '9')) { twosCompIndex = buffer[index] - '0'; } else if ((buffer[index] >= 'A') && (buffer[index] <= 'F')) { twosCompIndex = buffer[index] - 'A' + 10; } else if ((buffer[index] >= 'a') && (buffer[index] <= 'f')) { twosCompIndex = buffer[index] - 'a' + 10; } twosCompIndex += (16 - radix); buffer[index] = _twosComp[twosCompIndex]; } if (len < (bufferSize - 1)) { buffer[len++] = _numberSystem[radix - 1]; buffer[len] = 0; } } return len; } static int twosNegation(const int x, const int radix) { int n = x; if (x < 0) { if (radix == 10) { n = -x; } else { n = ~x; } } return n; } static char *safeitoa(const int x, char *buffer, const int bufferSize, const int radix) { int strlen = 0; int n = twosNegation(x, radix); int nuberSystemIndex = 0; if (radix <= 16) { do { if (strlen < (bufferSize - 1)) { nuberSystemIndex = (n % radix); buffer[strlen++] = _numberSystem[nuberSystemIndex]; buffer[strlen] = '\0'; n = n / radix; } else { break; } } while (n != 0); if (x < 0) { strlen = negateBuffer(buffer, bufferSize, strlen, radix); } safestrrev(buffer, bufferSize, strlen); return buffer; } return NULL; } 

Comments

2

Where is the itoa function in Linux?

As itoa() is not standard in C, various versions with various function signatures exists.
char *itoa(int value, char *str, int base); is common in *nix.

Should it be missing from Linux or if code does not want to limit portability, code could make it own.

Below is a version that does not have trouble with INT_MIN and handles problem buffers: NULL or an insufficient buffer returns NULL.

#include <stdlib.h> #include <limits.h> #include <string.h> // Buffer sized for a decimal string of a `signed int`, 28/93 > log10(2) #define SIGNED_PRINT_SIZE(object) ((sizeof(object) * CHAR_BIT - 1)* 28 / 93 + 3) char *itoa_x(int number, char *dest, size_t dest_size) { if (dest == NULL) { return NULL; } char buf[SIGNED_PRINT_SIZE(number)]; char *p = &buf[sizeof buf - 1]; // Work with negative absolute value int neg_num = number < 0 ? number : -number; // Form string *p = '\0'; do { *--p = (char) ('0' - neg_num % 10); neg_num /= 10; } while (neg_num); if (number < 0) { *--p = '-'; } // Copy string size_t src_size = (size_t) (&buf[sizeof buf] - p); if (src_size > dest_size) { // Not enough room return NULL; } return memcpy(dest, p, src_size); } 

Below is a C99 or later version that handles any base [2...36]

char *itoa_x(int number, char *dest, size_t dest_size, int base) { if (dest == NULL || base < 2 || base > 36) { return NULL; } char buf[sizeof number * CHAR_BIT + 2]; // worst case: itoa(INT_MIN,,,2) char *p = &buf[sizeof buf - 1]; // Work with negative absolute value to avoid UB of `abs(INT_MIN)` int neg_num = number < 0 ? number : -number; // Form string *p = '\0'; do { *--p = "0123456789ABCDEFGHIJKLMNOPQRSTUVWXYZ"[-(neg_num % base)]; neg_num /= base; } while (neg_num); if (number < 0) { *--p = '-'; } // Copy string size_t src_size = (size_t) (&buf[sizeof buf] - p); if (src_size > dest_size) { // Not enough room return NULL; } return memcpy(dest, p, src_size); } 

For a C89 and onward compliant code, replace inner loop with

 div_t qr; do { qr = div(neg_num, base); *--p = "0123456789ABCDEFGHIJKLMNOPQRSTUVWXYZ"[-qr.rem]; neg_num = qr.quot; } while (neg_num); 

Comments

2

glibc internal implementation

glibc 2.28 has an internal implementation:

which is used in several places internally, but I could not find if it can be exposed or how.

At least that should be a robust implementation if you are willing to extract it.

This question asks how to roll your own: How to convert an int to string in C?

Comments

2

I would prefer this: https://github.com/wsq003/itoa_for_linux

It should be the fastest itoa() ever. We use itoa() instead of sprintf() for performance reason, so a fastest itoa() with limited feature is reasonable and worthwhile.

Comments

1

If you just want to print them:

void binary(unsigned int n) { for(int shift=sizeof(int)*8-1;shift>=0;shift--) { if (n >> shift & 1) printf("1"); else printf("0"); } printf("\n"); } 

Comments

1

The replacement with snprintf is NOT complete!

It covers only bases: 2, 8, 10, 16, whereas itoa works for bases between 2 and 36.

Since I was searching a replacement for base 32, I guess I'll have to code my own!

Comments

0

I have used _itoa(...) on RedHat 6 and GCC compiler. It works.

Comments

0

If you're using SDL, there are cross-platform implementation of many functions, including itoa(). Use SDL_itoa() as defined in <SDL2/SDL_stdinc.h> .

Comments

0

Effectively, all current C compilers provide excellent support for compound literals and for inline functions, which make it possible to write thread-safe, memory-safe implementation that does not rely on static buffers (that can lead to unexpected bugs).

Basic idea is to combined a function that uses caller provided buffer, with a macro to abstract the creation of this temporary buffer. Negligent performance/memory impact for most applications.

#include <stdio.h> const char *itoa_wrapper(char *buffer, int buffer_sz, int val) { int result = snprintf(buffer, buffer_sz, "%d", val) ; // Assuming formatting always works return buffer ; } #define itoa(val) itoa_wrapper( (char[20]) {}, 20, val) int main(int argc, char **argv) { printf("ARGC=%s, 1+2+3+4=%s\n", itoa(argc), itoa(1+2+3+4)) ; } 

Short explanation: The (char [20] ) {} construct create a compound literal - of type 'char[20]' - effectively automatic temporary variables in the caller program, and passes that address as work area for the itoa_wrapper to use for storing the output.

Worth noting that there is a performance impact for creating the compound literal. It is fully initialized (with zero bytes). Relatively minor penalty unless used in a tight loop. For most logging use case - should be OK, as the overhead of logging the string to a stream, etc, much higher than the cost of zero-ing the work buffer.

Possible to improve performance by making itoa_wrapper static inline. Using custom formatter instead of snprintf (which is overkill for this task), can reduce make improve performance further.

On my desktop, 10M calls to itoa takes 0.4 seconds. Minor performance issue.

Comments

-6

You can use this program instead of sprintf.

void itochar(int x, char *buffer, int radix); int main() { char buffer[10]; itochar(725, buffer, 10); printf ("\n %s \n", buffer); return 0; } void itochar(int x, char *buffer, int radix) { int i = 0 , n,s; n = s; while (n > 0) { s = n%radix; n = n/radix; buffer[i++] = '0' + s; } buffer[i] = '\0'; strrev(buffer); } 

1 Comment

There are so many bugs in this code: 1) Doesn't actually convert hex correctly. 2) Doesn't convert 0 at all. 3) Doesn't work with negative numbers. 4) No checking for buffer overrun. I will post an improved version of this code shortly.

Start asking to get answers

Find the answer to your question by asking.

Ask question

Explore related questions

See similar questions with these tags.