On Fri, 16 Jul 2004, jacob navia wrote:
jn>
jn>"Peter Kirk" <peter> a écrit dans le message de
jn>news:40f79255$1@news.wineasy.se...
jn>> Hi there
jn>>
jn>> I have a question about static variables. There is a function which
jn>returns
jn>> a pointer to a structure "AStructure". The function has a static local
jn>> variable of type AStructure, and it is the address of this variable which
jn>is
jn>> returned. Is this good practice?
jn>>
jn>> See this pseudo code:
jn>>
jn>> AStructure *my_function()
jn>> {
jn>> static AStructure result;
jn>>
jn>> // "result" is filled with data...
jn>>
jn>> return ( &result );
jn>> }
jn>>
jn>> Thanks,
jn>> Peter
jn>>
jn>
jn>It is 100% correct as far as I see.
jn>
jn>Problems *could* arise in a
jn>multi-threaded context because of simultaneous access to a single
jn>memory location. If one thread enters this function during the time when
jn>another thread is updating the data structure problems would arise.
jn>
jn>This problem can be solved with a semaphore or a similar construct that
jn>protects the access to the static variable and allows only one thread
jn>to work on it at a time.
jn>
jn>In a single-thread environment there are no problems
There may be. If the "filled with data" is different for different
calls to the function, the following will not do what you think it should:
foo(my_function(), my_function());
A typical instance of this problem are formatting functions for user
defined types that use an internal buffer:
const char *
fmt_complex(complex c)
{
static char buf[100];
sprintf(buf, "%g+%gI", creal(c), cimag(c));
return (buf);
}
then call
void
foo(complex a, complex b)
{
printf("a=%s b=%s\n", fmt_complex(a), fmt_complex(b));
}
harti
