snprintf.c - a portable implementation of snprintf
(including vsnprintf.c, asnprintf, vasnprintf, asprintf, vasprintf)
snprintf.c - a portable implementation of snprintf,
including
vsnprintf.c, asnprintf, vasnprintf, asprintf, vasprintf
snprintf is a routine to convert numeric and string arguments
to formatted strings. It is similar to sprintf(3) provided in a
system's C library, yet it requires an additional argument - the buffer
size - and it guarantees never to store anything beyond the given buffer,
regardless of the format or arguments to be formatted. Some newer
operating systems do provide snprintf in their C library,
but many do not or do provide an inadequate (slow or idiosyncratic)
version, which calls for a portable implementation of this routine.
This program is free software; you can redistribute it
and/or modify it under the terms of the
Frontier Artistic License
which comes with this Kit.
Features
careful adherence to specs regarding flags, field width and precision;
good performance for large string handling (large format, large argument
or large paddings). Performance is similar to system's sprintf
and in several cases significantly better (make sure you compile with
optimizations turned on, tell the compiler the code is strict ANSI
if necessary to give it more freedom for optimizations);
return value semantics per ISO/IEC 9899:1999 ("ISO C99");
written in standard ISO/ANSI C - requires an ANSI C compiler.
Supported conversion specifiers and data types
This snprintf only supports the following conversion specifiers:
s, c, d, o, u, x, X, p (and synonyms: i, D, U, O - see below)
with flags: '-', '+', ' ', '0' and '#'.
An asterisk is supported for field width as well as precision.
Length modifiers 'h' (short int), 'l' (long int),
and 'll' (long long int) are supported.
NOTE:
If macro SNPRINTF_LONGLONG_SUPPORT is not defined (default)
the length modifier 'll' is recognized but treated the same as 'l',
which may cause argument value truncation!
Defining SNPRINTF_LONGLONG_SUPPORT requires that your system's
sprintf also handles length modifier 'll'.
long long int is a language extension which may not be portable.
Conversion of numeric data (conversion specifiers d, o, u, x, X, p)
with length modifiers (none or h, l, ll) is left to the system
routine sprintf, but all handling of flags, field width and precision
as well as c and s conversions is done very carefully by this portable routine.
If a string precision (truncation) is specified (e.g. %.8s) it is
guaranteed the string beyond the specified precision will not be referenced.
Length modifiers h, l and ll are ignored for c and s conversions
(data types wint_t and wchar_t are not supported).
The following common synonyms for conversion characters are supported:
i is a synonym for d
D is a synonym for ld, explicit length modifiers are ignored
U is a synonym for lu, explicit length modifiers are ignored
O is a synonym for lo, explicit length modifiers are ignored
The D, O and U conversion characters are nonstandard, they are supported
for backward compatibility only, and should not be used for new code.
The following is specifically not supported:
flag ' (thousands' grouping character) is recognized but ignored
numeric conversion specifiers: f, e, E, g, G and synonym F,
as well as the new a and A conversion specifiers
length modifier 'L' (long double)
and 'q' (quad - use 'll' instead)
wide character/string conversions: lc, ls, and nonstandard
synonyms C and S
writeback of converted string length: conversion character n
the n$ specification for direct reference to n-th argument
locales
It is permitted for str_m to be zero, and it is permitted to specify NULL
pointer for resulting string argument if str_m is zero (as per ISO C99).
The return value is the number of characters which would be generated
for the given input, excluding the trailing null. If this value
is greater or equal to str_m, not all characters from the result
have been stored in str, output bytes beyond the (str_m-1) -th character
are discarded. If str_m is greater than zero it is guaranteed
the resulting string will be null-terminated.
NOTE that this matches the ISO C99, OpenBSD, and GNU C library 2.1,
but is different from some older and vendor implementations,
and is also different from XPG, XSH5, SUSv2 specifications.
For historical discussion on changes in the semantics and standards
of snprintf see printf(3) man page in the Linux programmers manual.
Routines asprintf and vasprintf return a pointer (in the ptr argument)
to a buffer sufficiently large to hold the resulting string. This pointer
should be passed to free(3) to release the allocated storage when it is
no longer needed. If sufficient space cannot be allocated, these functions
will return -1 and set ptr to be a NULL pointer. These two routines are a
GNU C library extensions (glibc).
Routines asnprintf and vasnprintf are similar to asprintf and vasprintf,
yet, like snprintf and vsnprintf counterparts, will write at most str_m-1
characters into the allocated output string, the last character in the
allocated buffer then gets the terminating null. If the formatted string
length (the return value) is greater than or equal to the str_m argument,
the resulting string was truncated and some of the formatted characters
were discarded. These routines present a handy way to limit the amount
of allocated memory to some sane value.
There is a very low-traffic mailing list snprintf-announce@ijs.si
where announcements about new versions will be posted
as well as warnings about threatening bugs if discovered.
The posting is restricted to snprintf developer(s).
You can also subscribe to the list by mailing
the command SUBSCRIBE either in the subject or in the message body
to the address snprintf-announce-request@ijs.si . You will be asked for
confirmation before subscription will be effective.
The list of members is only accessible to the list administrator,
so there is no need for concern about automatic e-mail address gatherers.
Questions about the mailing list and concerns for the attention
of a person should be sent to snprintf-announce-admin@ijs.si
There is no general discussion list about portable snprintf
at the moment. Please send comments and suggestion to the author.
Revision history
Version 1.3 fixes a runaway loop problem from 1.2. Please upgrade.
1999-06-30 V1.3 Mark Martinec <mark.martinec@ijs.si>
fixed runaway loop (eventually crashing when str_l wraps
beyond 2^31) while copying format string without
conversion specifiers to a buffer that is too short
(thanks to Edwin Young <edwiny@autonomy.com> for spotting the problem);
added macros PORTABLE_SNPRINTF_VERSION_(MAJOR|MINOR) to snprintf.h
2000-02-14 V2.0 (never released) Mark Martinec <mark.martinec@ijs.si>
relaxed license terms:
The Artistic License now applies.
You may still apply the GNU GENERAL PUBLIC LICENSE
as was distributed with previous versions, if you prefer;
added vsnprintf (patch also independently proposed by
Caolán McNamara 2000-05-04, and Keith M Willenson 2000-06-01)
2000-06-27 V2.1 Mark Martinec <mark.martinec@ijs.si>
removed POSIX check for str_m < 1; value 0 for str_m is
allowed by ISO C99 (and GNU C library 2.1) (pointed out
on 2000-05-04 by Caolán McNamara, caolan@ csn dot ul dot ie).
Besides relaxed license this change in standards adherence
is the main reason to bump up the major version number;
added nonstandard routines asnprintf, vasnprintf, asprintf,
vasprintf that dynamically allocate storage for the
resulting string; these routines are not compiled by default,
see comments where NEED_V?ASN?PRINTF macros are defined;
autoconf contributed by Caolán McNamara
2000-10-06 V2.2 Mark Martinec <mark.martinec@ijs.si>
BUG FIX: the %c conversion used a temporary variable
that was no longer in scope when referenced,
possibly causing incorrect resulting character;
BUG FIX: make precision and minimal field width unsigned
to handle huge values (2^31 <= n < 2^32) correctly;
also be more careful in the use of signed/unsigned/size_t
internal variables -- probably more careful than many
vendor implementations, but there may still be a case
where huge values of str_m, precision or minimal field
could cause incorrect behaviour;
use separate variables for signed/unsigned arguments,
and for short/int, long, and long long argument lengths
to avoid possible incompatibilities on certain
computer architectures. Also use separate variable
arg_sign to hold sign of a numeric argument,
to make code more transparent;
some fiddling with zero padding and "0x" to make it
Linux compatible;
systematically use macros fast_memcpy and fast_memset
instead of case-by-case hand optimization; determine some
breakeven string lengths for different architectures;
terminology change: format -> conversion specifier,
C9x -> ISO/IEC 9899:1999 ("ISO C99"),
alternative form -> alternate form,
data type modifier -> length modifier;
several comments rephrased and new ones added;
make compiler not complain about 'credits' defined but
not used;
Other implementations of snprintf
I am aware of some other (more or less) portable implementations
of snprintf. I do not claim they are free software - please refer
to their respective copyright and licensing terms.
If you know of other versions please let
me know.
a very thorough implementation (src/util_snprintf.c)
by the Apache Group distributed with the
Apache web server
- http://www.apache.org/ .
Does its own floating point conversions using routines
ecvt(3), fcvt(3) and gcvt(3) from the standard C library
or from the GNU libc.
This is from the code:
This software [...] was originally based
on public domain software written at the
National Center
for Supercomputing Applications, University of Illinois,
Urbana-Champaign.
[...] This code is based on, and used with the permission of,
the SIO stdio-replacement strx_* functions by Panos Tsirigotis
<panos@alumni.cs.colorado.edu> for xinetd.
QCI
Utilities use a modified version of snprintf from the Apache group.
implementations as distributed with
OpenBSD,
FreeBSD, and
NetBSD
are all wrappers to vfprintf.c, which is derived from software
contributed to Berkeley by Chris Torek.
Brandon Long's
<blong@fiction.net>
modified version
of Prof. Patrick Powell's snprintf with contributions from others.
With minimal floating point support.
implementation (src/snprintf.c) as distributed with
sendmail - http://www.sendmail.org/
is a cleaned up Prof. Patrick Powell's version
to compile properly and to support .precision and %lx.
implementation used by
newlog
(a replacement for syslog(3)) made available by
the SOS Corporation.
Enabling floating point support is a compile-time option.
implementation by Michael Richardson
<mcr@metis.milkyway.com>
is available at
http://sandelman.ottawa.on.ca/SSW/snp/snp.html.
It is based on BSD44-lite's vfprintf() call, modified to function
on SunOS. Needs internal routines from the 4.4 strtod (included),
requires GCC to compile the long long (aka quad_t) portions.
implementation from Tomi Salo
<ttsalo@ssh.fi>
distributed with
SSH 2.0
Unix Server. Not in public domain.
Floating point conversions done by system's sprintf.