setjmp/longjmp

[Ravi Uday]

Hi,

Can anyone explain me why do we use setjmp and longjump functions.
I read through the manual pages/doc but wasnt able to get a clear
picture of the same.
Any small example illustrating these would be real helping.
Also, what happens when one jumps from one process/task to other, even though
it is OT can anybody clear this concept.
Thanks,
- Ravi

[Gordon Burditt]

>Can anyone explain me why do we use setjmp and longjump functions.

I read through the manual pages/doc but wasnt able to get a clear
picture of the same.
setjmp and longjmp are a long-distance version of goto, and
therefore they are greatly frowned upon by the Structured Programming Police.
jmp_buf buf;
some_type_t *p;

if (setjmp(buf) == 0) {
/* try something that might get an error */
...
p = malloc(sizeof(*p));
if (p == NULL) {
longjmp(buf, 1); /* we got an error */
}
...
} else {
/* we got an error, (someone called longjmp) handle it */
/* e.g. use the slow, file-based method instead of */
/* the in-memory method */
}

The tricky thing you can do with longjmp, though, is that it
can be called from functions called from the /* try something that
might get an error */ code. The functions signalling the error
need not have any idea where the error handler is (although they
do need to know where buf is, either passed as a parameter or a
static or global.)

Another tricky thing that can be done, and isn't portable, is to
catch signals and have the signal handler call longjmp(). This
way, the /* we got an error, handle it */ part might correct for
division by zero, overflow, or dereferencing NULL. Or it might
just note that it crashed and log the error.

void sigsegv_handler(int sig) {
/* you're pretty much stuck with buf being a global or */
/* static variable here, as you can't change the parameters */

longjmp(buf, 1);
}

Yet another tricky thing involves a stack of jmp_bufs and nested
error handlers.
Any small example illustrating these would be real helping. Also, what happens when one jumps from one process/task to other, even though
it is OT can anybody clear this concept.

In general, you DON'T jump from one process/task to another. Other
programs aren't in your address space. And if you try, expect a
real mess.

Gordon L. Burditt

[Dave Vandervies]

In article <ce********@library1.airnews.net>,
Gordon Burditt <go***********@burditt.org> wrote:

Can anyone explain me why do we use setjmp and longjump functions.
I read through the manual pages/doc but wasnt able to get a clear
picture of the same.

setjmp and longjmp are a long-distance version of goto, and
therefore they are greatly frowned upon by the Structured Programming Police.

setjmp can also behave as a COME FROM, which is occasionally useful,
especially if something like this needs to be added to an existing
program:
--------
static jmp_buf env;

void go_back_to_sleep(void)
{
longjmp(env,1);
}

int main(void)
{
/*Do one-time initialization*/

/*This means "COME FROM longjmp in go_back_to_sleep()"*/
setjmp(env);

/*Wait for wake-up call*/

/*Do per-run initialization*/

while(1)
{
/*Do stuff, including calling functions that may call
go_back_to_sleep() under appropriate conditions
*/
}

/*never reached*/
}
--------
This is probably not the best way to do this in new code, but it's
often the easiest way to change an existing program's handling of "We're
done doing stuff" from "exit" to "go back to sleep and wait for another
wake-up call".
dave

--
Dave Vandervies dj******@csclub.uwaterloo.ca

It's like trying to give yourself a voluntary lobotomy.
--Andy Glew in comp.arch