Lib: SMF: Missing execution of ancestor state_run

[Yaxit]

Describe the bug
When using the State Machine Framework (SMF) with ancestor support, the ancestor state's run function is not being executed on the first call to smf_run_state(). However, it correctly executes on subsequent calls. According to SMF documentation, when a state has an ancestor, the ancestor's run function should be called automatically after the child state's run function completes.

To Reproduce
Below is a minimal example to reproduce the issue:

#include <zephyr/smf.h>

/* Forward declaration of state table */
const struct smf_state demo_states[];

/* List of demo states */
enum demo_state { S0, S1};

/* User defined object */
struct s_object {
        /* This must be first */
        struct smf_ctx ctx;

        /* Other state specific data add here */
} s_obj;

int cnt = 0;
/* State S0 */
static void s0_entry(void *o){
        printk("S0 entry\n");
        smf_set_state(SMF_CTX(&s_obj), &demo_states[S1]);
}
static void s0_run(void *o){
        printk("S0 run\n");
        cnt++;
}

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. cnt=%d\n", cnt);
}

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


/* Populate state table */
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, &demo_states[S0], NULL),
};

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

        printk("\nIter 1\n");
        smf_run_state(SMF_CTX(&s_obj));
        printk("\nIter 2\n");
        smf_run_state(SMF_CTX(&s_obj));

}

Expected behavior
When using a state with an ancestor, the ancestor's run function should be automatically executed after the child state's run function on every call to smf_run_state().

Expected output:

S0 entry
S1 entry

Iter 1
S1 run. cnt=0
S0 run

Iter 2
S1 run. cnt=1
S0 run

Actual behavior
The ancestor's run function is not executed on the first call to smf_run_state() but works correctly on subsequent calls.

Actual output:

S0 entry
S1 entry

Iter 1
S1 run. cnt=0

Iter 2
S1 run. cnt=0
S0 run

Impact
This issue affects state machine hierarchies that depend on ancestor behavior being consistent across all iterations. It could lead to unexpected behavior in systems relying on ancestor state execution during the first iteration.

Environment:

• OS: Windows 11
• Toolchain: Zephyr SDK 0.16.1
• Zephyr Version: v3.7.99-1f8f3dc29142

[Yaxit]

Maybe related, but I think worth confirming: in the case of a hierarchical state machine, when a child state requests a transition to its parent, only the child's exit function is called. However, it seems that if the transition is called by the parent's run function (that is executing in the context of handling the child's run function) then also the parent's exit function is called.
I think this is also incorrect. The docs state that the exit and entry are always called in the case of a transition to self, but in this case the transition is not to self, but to the parent https://docs.zephyrproject.org/latest/services/smf/index.html#uml-state-machines

#include <zephyr/smf.h>

/* Forward declaration of state table */
const struct smf_state demo_states[];

/* List of demo states */
enum demo_state { S0, S1 , S2, S3};

/* User defined object */
struct s_object {
        /* This must be first */
        struct smf_ctx ctx;

        /* Other state specific data add here */
} s_obj;

static int s0_cnt = 0;
static int s1_cnt = 0;
/* State S0 */
static void s0_entry(void *o){
        printk("S0 entry\n");
        if(s1_cnt <= 0){
                smf_set_state(SMF_CTX(&s_obj), &demo_states[S1]);
        }
}
static void s0_run(void *o){
        printk("S0 run\n");
        printk("S0 run cnt = %d\n", s0_cnt);
        smf_set_state(SMF_CTX(&s_obj), &demo_states[S0]);
        s0_cnt++;
}

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");
        printk("S1 run cnt = %d\n", s1_cnt);
        s1_cnt++;
}

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


/* Populate state table */
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, &demo_states[S0], NULL),
};

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

        printk("\nIter 1\n");
        smf_run_state(SMF_CTX(&s_obj));
        printk("\nIter 2\n");
        smf_run_state(SMF_CTX(&s_obj));
        printk("\nIter 3\n");
        smf_run_state(SMF_CTX(&s_obj));
        printk("\nIter 4\n");
        smf_run_state(SMF_CTX(&s_obj));
}

*** Using Zephyr OS v3.7.99-1f8f3dc29142 ***
S0 entry
S1 entry

Iter 1
S1 run
S1 run cnt = 0

Iter 2
S1 run
S1 run cnt = 1
S0 run
S0 run cnt = 0
S1 exit
S0 exit
S0 entry

Iter 3
S0 run
S0 run cnt = 1
S0 exit
S0 entry

[jelledevleeschouwer]

@Yaxit, I think the problem lies in the fact that it was assumed the smf_set_state() is only called from the context of the smf_run_state() function, where internal->new_state is always reset in smf_execute_ancestor_run_actions(). However, in your case, you transition to the child state in the entry action of the parent state S0 which is ran from the context of smf_set_initial(), where the internal->new_state is not reset (though intended) after running the entry actions.

A workaround would be to refactor your snippet to set the initial state to S1 directly in your main().
A permanent fix would be to reset internal->new_state at the exit of the smf_set_initial, for which I created pull request #91921

[jelledevleeschouwer]

Hmm, seems it's already fixed in fa795f4.