State Machine Framework (SMF) unexpected exit

[Yaxit]

Hi, I'm starting to play around with the State Machine Framework. Playin around with the example I discovered something that I find a bit counterintuitive. Take two states S0 and S1
When switching from s0_run to S1 with smf_set_state(SMF_CTX(&s_obj), &demo_states[S0]), I see that first s0_exit is executed, then s1_entry is executed, and lastly the rest of s0_run is executed. Somehow I expected that either s0_run would finish execution before s0_exit, or that the rest of the function would just not be executed?

In general, if this is desired behaviour, the state should really only changed as the very last thing, which I guess would be reasonable to document?

/* State S0 */
static void s0_entry(void *o){
        printk("S0 entry\n");
}
static void s0_run(void *o){
	printk("S0 run\n");
        smf_set_state(SMF_CTX(&s_obj), &demo_states[S1]);
	printk("S0 run finished\n");
}
static void s0_exit(void *o){
        printk("S0 exit\n");
}

/* State S1 */
static void s1_entry(void *o){
	printk("S1 entry\n");

}
static void s1_run(void *o){
	printk("S1 run\n");
        smf_set_state(SMF_CTX(&s_obj), &demo_states[S0]);
	printk("S1 run finished\n");

}
static void s1_exit(void *o){
        printk("S1 exit\n");
}


/* Populate state table */
static const struct smf_state demo_states[] = {
        [S0] = SMF_CREATE_STATE(s0_entry, s0_run, s0_exit, NULL, NULL),
        [S1] = SMF_CREATE_STATE(s1_entry, s1_run, s1_exit, NULL, NULL),
};

int main(void)
{
        int32_t ret;

        /* Set initial state */
        smf_set_initial(SMF_CTX(&s_obj), &demo_states[S0]);

        /* Run the state machine */
        while(1) {
                /* State machine terminates if a non-zero value is returned */
                ret = smf_run_state(SMF_CTX(&s_obj));
                if (ret) {
                        /* handle return code and terminate state machine */
                        break;
                }
                k_msleep(1000);
        }
}

S0 entry
S0 run
S0 exit
S1 entry
S0 run finished
S1 run
S1 exit

[glenn-andrews]

Sorry for the delay in getting back to you.

You are correct that as soon as smf_set_state() is run the entry and exit functions are called, then control comes back to the state run action.

There's no way to avoid this with the current architecture, and it is expected that smf_set_state() is the last action performed in any given control path.

To fix it we'd need to get smf_set_state() to set the destination, then return a value from the run action that triggered the calling of the entry and exit actions.

#83854 is a proposal to modify SMF to return a value from the run action. Once it is accepted, we could expand it with an SMF_EVENT_TRANSITION return value to trigger the transition upon return.

The downside is if anyone has put code after the smf_set_state() to clean up, ti would break their expected code flow. But hopefully not many people have done that.

[Yaxit]

Thanks, I'll follow the PR!

[glenn-andrews]

Note the current PR does NOT fix this issue. I'd make a subsequent PR to return a transition flag to run the entry/exit actions.

For now just make sure smf_set_state() is the last thing in your code path for the run action.