Multiple calls to vsnprintf fail with SEGV under AMD64

[Paul Price]

In a few places in our code base, we do something like the following:

void printFunc(const char *format, va_list ap)
{

int size = vsnprintf(NULL, 0, format, ap);
psString string = psStringAlloc(size);
vsnprintf(string, size + 1, format, ap);
now do something with string

}

Please note that, while this code works under x86 and x86_64, it produces a SEGV
on AMD64, with a trace similar to the following:

#0 0x00002aaaac9ad270 in strlen () from /lib/tls/libc.so.6
#1 0x00002aaaac981cd8 in vfprintf () from /lib/tls/libc.so.6
#2 0x00002aaaac9a0aa4 in vsnprintf () from /lib/tls/libc.so.6
#3 0x00002aaaaaebf189 in printFunc (.......)

The reason for the problem is that one cannot *technically* traverse a va_list
twice (the traversal is done in the depths of vsnprintf). While glibc for some
architectures must reset the va_list pointer to the start of the list after a
traversal, this is not guaranteed behaviour, so one must either copy the va_list
(va_copy), or start it again (va_start).

References:

• ​http://lists.debian.org/debian-ia64/2005/05/msg00004.html
• Harbison & Steele, p330.

Josh is writing a function, psStringAppendV, that will take care of this issue.

Then we will clean up all instances we can find.

[jhoblitt]

psDB has been fixed by elimating all calls to vsnprintf() by using the new
psStringAppendV() instead.

[Paul Price]

Fixed psTrace, psLogMsg as well.