libtcod 1.1 released

On 22 fév, 05:08, Joni Toivanen <jomio... RemoveThis @gmail.com> wrote:
>
> The problem with that code is that it doesn't really work with a valid
> implementation of vsnprintf. According to the C99 the vsnprintf function
> should return "the number of characters that would have been written had
> [the buffer size] been sufficiently large, not counting the terminating
> null character, or a negative value if an encoding error occurred." (I'm
> guessing that an encoding error means invalid fmt string).
>
> At least Microsoft Visual C++ libraries and older versions of glibc
> implement the vsnprintf function incorrectly and return a negative value
> if the buffer isn't big enough, and your code would work on them, but on
> recent versions of glibc and other C libraries that have C99 conforming
> vsnprintf your code will fail if the initial 512 chars isn't enough.
>
> Here's a version of the function that should (untested, sorry) also work
> with a C99 conforming vsnprintf:
>
> static char *TCOD_console_vsprint(const char *fmt, va_list ap) {
>         static char *msg=NULL;
>         static int buflen=0;
>         int len;
>         bool ok=false;
>         if (!msg) {
>                 buflen=512;
>                 msg=(char *)calloc(sizeof(char),buflen);
>         }
>         len = vsnprintf(msg,buflen,fmt,ap);
>         if (len >= buflen) {
>                 // buffer not big enough, glibc
>                 buflen = len + 1;  // null char not included in len
>                 msg=(char *)calloc(sizeof(char),buflen);
>                 vsnprintf(msg,buflen,fmt,ap);
>                 return(msg);
>         } else if (len < 0) {
>                 // buffer not big enough, MSVC
>                 do {
>                         buflen*=2;
>                         free( msg );
>                         msg=(char *)calloc(sizeof(char),buflen);
>                 } while(vsnprintf(msg,buflen,fmt,ap) < 0);
>         }
>         return msg;
>
> }
>
> (I tried to imitate your coding style, but if it looks ugly to you, just
> reformat it -- provided that it works at all
>
> Hope this helps...

Thanks for the information, I didn't read the whole man page...
Strangely, the code works on mingw32 with gcc 3.2.3. I guess it must
use a glibc < 2.1...

--
jice
 >> Stay informed about: libtcod 1.1 released 

On Feb 22, 10:47 am, jice <jice.nos... DeleteThis @gmail.com> wrote:
>
> Thanks for the information, I didn't read the whole man page...
> Strangely, the code works on mingw32 with gcc 3.2.3. I guess it must
> use a glibc < 2.1...
>

I didn't have anything else to do this morning (hah), so I tested the
two
different versions of the function on my Linux box. Your version
worked
okay if the string was less than 512 chars long, but if it was longer
than that, it clipped the end off. The code I pasted worked even for
strings longer than 512 chars, but I've only tested it on Linux, so I
don't
know how it will react to MSVC or anything other than glibc and gcc...

Cheers

(Posted from google groups, so the formatting might be off...)
 >> Stay informed about: libtcod 1.1 released 

jice wrote:
> On 21 fév, 11:46, Ido Yehieli <Ido.Yehi....RemoveThis@gmail.com> wrote:
>>
>> I can imagine you have a line of the sorts of "char buffer[256];"
>> somewhere in your code?
>> If you are using c99 you can take advantage of variable-length arrays:
>>
>> char buffer[snprintf(NULL, 0, your_string)];
>>
>> Or just dynamically allocate if you use c89.
>>
>> -Ido.
>
> To avoid calling vsnprintf twice for each printed string, I'm now
> using a resizable buffer :
>
> static char *TCOD_console_vsprint(const char *fmt, va_list ap) {
> static char *msg=NULL;
> static int buflen=0;
> bool ok=false;
> if (!msg) {
> buflen=512;
> msg=(char *)calloc(sizeof(char),buflen);
> }
> do {
> ok = (vsnprintf(msg,buflen,fmt,ap) != -1);
> if (!ok) {
> buflen*=2;
> free( msg );
> msg=(char *)calloc(sizeof(char),buflen);
> }
> } while (! ok);
> return msg;
> }

The problem with that code is that it doesn't really work with a valid
implementation of vsnprintf. According to the C99 the vsnprintf function
should return "the number of characters that would have been written had
[the buffer size] been sufficiently large, not counting the terminating
null character, or a negative value if an encoding error occurred." (I'm
guessing that an encoding error means invalid fmt string).

At least Microsoft Visual C++ libraries and older versions of glibc
implement the vsnprintf function incorrectly and return a negative value
if the buffer isn't big enough, and your code would work on them, but on
recent versions of glibc and other C libraries that have C99 conforming
vsnprintf your code will fail if the initial 512 chars isn't enough.

Here's a version of the function that should (untested, sorry) also work
with a C99 conforming vsnprintf:

static char *TCOD_console_vsprint(const char *fmt, va_list ap) {
static char *msg=NULL;
static int buflen=0;
int len;
bool ok=false;
if (!msg) {
buflen=512;
msg=(char *)calloc(sizeof(char),buflen);
}
len = vsnprintf(msg,buflen,fmt,ap);
if (len >= buflen) {
// buffer not big enough, glibc
buflen = len + 1; // null char not included in len
msg=(char *)calloc(sizeof(char),buflen);
vsnprintf(msg,buflen,fmt,ap);
return(msg);
} else if (len < 0) {
// buffer not big enough, MSVC
do {
buflen*=2;
free( msg );
msg=(char *)calloc(sizeof(char),buflen);
} while(vsnprintf(msg,buflen,fmt,ap) < 0);
}
return msg;
}

(I tried to imitate your coding style, but if it looks ugly to you, just
reformat it -- provided that it works at all

Hope this helps...
 >> Stay informed about: libtcod 1.1 released