esp-idf/components/esp_event/test/test_event.c
2019-11-03 16:19:30 +08:00

1286 lines
48 KiB
C

#include <stdbool.h>
#include <string.h>
#include "esp_event.h"
#include "sdkconfig.h"
#include "freertos/FreeRTOS.h"
#include "freertos/task.h"
#include "freertos/portmacro.h"
#include "esp_log.h"
#include "driver/periph_ctrl.h"
#include "driver/timer.h"
#include "esp_event.h"
#include "esp_event_private.h"
#include "esp_event_internal.h"
#include "esp_heap_caps.h"
#include "sdkconfig.h"
#include "unity.h"
#include "test_utils.h"
static const char* TAG = "test_event";
#define TEST_CONFIG_ITEMS_TO_REGISTER 5
#define TEST_CONFIG_TASKS_TO_SPAWN 2
#define TEST_CONFIG_WAIT_MULTIPLIER 5
// The initial logging "initializing test" is to ensure mutex allocation is not counted against memory not being freed
// during teardown.
#define TEST_SETUP() \
ESP_LOGI(TAG, "initializing test"); \
size_t free_mem_before = heap_caps_get_free_size(MALLOC_CAP_DEFAULT); \
test_setup(); \
s_test_core_id = xPortGetCoreID(); \
s_test_priority = uxTaskPriorityGet(NULL);
#define TEST_TEARDOWN() \
test_teardown(); \
vTaskDelay(pdMS_TO_TICKS(CONFIG_ESP_INT_WDT_TIMEOUT_MS * TEST_CONFIG_WAIT_MULTIPLIER)); \
TEST_ASSERT_EQUAL(free_mem_before, heap_caps_get_free_size(MALLOC_CAP_DEFAULT));
typedef struct {
void* data;
SemaphoreHandle_t start;
SemaphoreHandle_t done;
} task_arg_t;
typedef struct {
esp_event_base_t base;
int32_t id;
esp_event_handler_t* handles;
int32_t num;
esp_event_loop_handle_t loop;
bool is_registration;
} handler_registration_data_t;
typedef struct {
esp_event_base_t base;
int32_t id;
esp_event_loop_handle_t loop;
int32_t num;
} post_event_data_t;
typedef struct {
int performed;
int expected;
SemaphoreHandle_t done;
} performance_data_t;
typedef struct {
void* data;
SemaphoreHandle_t mutex;
} simple_arg_t;
typedef struct {
int *arr;
int index;
} ordered_data_t;
static BaseType_t s_test_core_id;
static UBaseType_t s_test_priority;
ESP_EVENT_DECLARE_BASE(s_test_base1);
ESP_EVENT_DECLARE_BASE(s_test_base2);
ESP_EVENT_DEFINE_BASE(s_test_base1);
ESP_EVENT_DEFINE_BASE(s_test_base2);
enum {
TEST_EVENT_BASE1_EV1,
TEST_EVENT_BASE1_EV2,
TEST_EVENT_BASE1_MAX
};
enum {
TEST_EVENT_BASE2_EV1,
TEST_EVENT_BASE2_EV2,
TEST_EVENT_BASE2_MAX
};
static BaseType_t test_event_get_core(void)
{
static int calls = 0;
if (portNUM_PROCESSORS > 1) {
return (s_test_core_id + calls++) % portNUM_PROCESSORS;
} else {
return s_test_core_id;
}
}
static esp_event_loop_args_t test_event_get_default_loop_args(void)
{
esp_event_loop_args_t loop_config = {
.queue_size = CONFIG_ESP_SYSTEM_EVENT_QUEUE_SIZE,
.task_name = "loop",
.task_priority = s_test_priority,
.task_stack_size = 2048,
.task_core_id = test_event_get_core()
};
return loop_config;
}
static void test_event_simple_handler(void* event_handler_arg, esp_event_base_t event_base, int32_t event_id, void* event_data)
{
if (!event_handler_arg) {
return;
}
simple_arg_t* arg = (simple_arg_t*) event_handler_arg;
xSemaphoreTake(arg->mutex, portMAX_DELAY);
int* count = (int*) arg->data;
if (event_data == NULL) {
(*count)++;
} else {
(*count) += *((int*) event_data);
}
xSemaphoreGive(arg->mutex);
}
static void test_event_ordered_dispatch(void* event_handler_arg, esp_event_base_t event_base, int32_t event_id, void* event_data)
{
int *arg = (int*) event_handler_arg;
ordered_data_t *data = *((ordered_data_t**) (event_data));
data->arr[data->index++] = *arg;
}
static void test_event_performance_handler(void* event_handler_arg, esp_event_base_t event_base, int32_t event_id, void* event_data)
{
performance_data_t* data = (performance_data_t*) event_handler_arg;
data->performed++;
if (data->performed >= data->expected) {
xSemaphoreGive(data->done);
}
}
static void test_event_post_task(void* args)
{
task_arg_t* arg = (task_arg_t*) args;
post_event_data_t* data = arg->data;
xSemaphoreTake(arg->start, portMAX_DELAY);
for (int i = 0; i < data->num; i++) {
TEST_ASSERT_EQUAL(ESP_OK, esp_event_post_to(data->loop, data->base, data->id, NULL, 0, portMAX_DELAY));
vTaskDelay(1);
}
xSemaphoreGive(arg->done);
vTaskDelete(NULL);
}
static void test_event_simple_handler_registration_task(void* args)
{
task_arg_t* arg = (task_arg_t*) args;
handler_registration_data_t* data = (handler_registration_data_t*) arg->data;
xSemaphoreTake(arg->start, portMAX_DELAY);
for(int i = 0; i < data->num; i++) {
if (data->is_registration) {
TEST_ASSERT_EQUAL(ESP_OK, esp_event_handler_register_with(data->loop, data->base, data->id, data->handles[i], NULL));
} else {
TEST_ASSERT_EQUAL(ESP_OK, esp_event_handler_unregister_with(data->loop, data->base, data->id, data->handles[i]));
}
vTaskDelay(1);
}
xSemaphoreGive(arg->done);
vTaskDelete(NULL);
}
static void test_handler_post_w_task(void* event_handler_arg, esp_event_base_t event_base, int32_t event_id, void* event_data)
{
simple_arg_t* arg = (simple_arg_t*) event_handler_arg;
esp_event_loop_handle_t* loop = (esp_event_loop_handle_t*) event_data;
int* count = (int*) arg->data;
(*count)++;
if (*count <= 2) {
if (event_base == s_test_base1 && event_id == TEST_EVENT_BASE1_EV1) {
TEST_ASSERT_EQUAL(ESP_OK, esp_event_post_to(*loop, s_test_base1, TEST_EVENT_BASE1_EV2, NULL, 0, portMAX_DELAY));
} else{
xSemaphoreGive((SemaphoreHandle_t) arg->mutex);
}
} else {
// Test that once the queue is full and the handler attempts to post to the same loop,
// posting does not block indefinitely.
if (event_base == s_test_base1 && event_id == TEST_EVENT_BASE1_EV1) {
xSemaphoreTake((SemaphoreHandle_t) arg->mutex, portMAX_DELAY);
TEST_ASSERT_EQUAL(ESP_ERR_TIMEOUT, esp_event_post_to(*loop, s_test_base1, TEST_EVENT_BASE1_EV2, NULL, 0, portMAX_DELAY));
}
}
}
static void test_handler_post_wo_task(void* event_handler_arg, esp_event_base_t event_base, int32_t event_id, void* event_data)
{
simple_arg_t* arg = (simple_arg_t*) event_handler_arg;
esp_event_loop_handle_t* loop = (esp_event_loop_handle_t*) event_data;
int* count = (int*) arg->data;
(*count)++;
if (*count <= 2) {
if (event_base == s_test_base1 && event_id == TEST_EVENT_BASE1_EV1) {
TEST_ASSERT_EQUAL(ESP_OK, esp_event_post_to(*loop, s_test_base1, TEST_EVENT_BASE1_EV2, NULL, 0, portMAX_DELAY));
} else{
xSemaphoreGive((SemaphoreHandle_t) arg->mutex);
}
} else {
// Test that once the queue is full and the handler attempts to post to the same loop,
// posting does not block indefinitely.
if (event_base == s_test_base1 && event_id == TEST_EVENT_BASE1_EV1) {
xSemaphoreTake((SemaphoreHandle_t) arg->mutex, portMAX_DELAY);
TEST_ASSERT_EQUAL(ESP_OK, esp_event_post_to(*loop, s_test_base1, TEST_EVENT_BASE1_EV2, NULL, 0, portMAX_DELAY));
TEST_ASSERT_EQUAL(ESP_ERR_TIMEOUT, esp_event_post_to(*loop, s_test_base1, TEST_EVENT_BASE1_EV2, NULL, 0, portMAX_DELAY));
}
}
}
static void test_handler_unregister_itself(void* event_handler_arg, esp_event_base_t event_base, int32_t event_id, void* event_data)
{
esp_event_loop_handle_t* loop = (esp_event_loop_handle_t*) event_data;
int* unregistered = (int*) event_handler_arg;
(*unregistered) += (event_base == s_test_base1 ? 0 : 10) + event_id + 1;
// Unregister this handler for this event
TEST_ASSERT_EQUAL(ESP_OK, esp_event_handler_unregister_with(*loop, event_base, event_id, test_handler_unregister_itself));
}
static void test_post_from_handler_loop_task(void* args)
{
esp_event_loop_handle_t event_loop = (esp_event_loop_handle_t) args;
while(1) {
TEST_ASSERT_EQUAL(ESP_OK, esp_event_loop_run(event_loop, portMAX_DELAY));
}
}
static void test_setup(void)
{
TEST_ASSERT_TRUE(TEST_CONFIG_TASKS_TO_SPAWN >= 2);
TEST_ASSERT_EQUAL(ESP_OK, esp_event_loop_create_default());
}
static void test_teardown(void)
{
TEST_ASSERT_EQUAL(ESP_OK, esp_event_loop_delete_default());
}
#define TIMER_DIVIDER 16 // Hardware timer clock divider
#define TIMER_SCALE (TIMER_BASE_CLK / TIMER_DIVIDER) // convert counter value to seconds
#define TIMER_INTERVAL0_SEC (2.0) // sample test interval for the first timer
#if CONFIG_ESP_EVENT_POST_FROM_ISR
static void test_handler_post_from_isr(void* event_handler_arg, esp_event_base_t event_base, int32_t event_id, void* event_data)
{
SemaphoreHandle_t *sem = (SemaphoreHandle_t*) event_handler_arg;
// Event data is just the address value (maybe have been truncated due to casting).
int *data = (int*) event_data;
TEST_ASSERT_EQUAL(*data, (int) (*sem));
xSemaphoreGive(*sem);
}
void IRAM_ATTR test_event_on_timer_alarm(void* para)
{
/* Retrieve the interrupt status and the counter value
from the timer that reported the interrupt */
uint64_t timer_counter_value =
timer_group_get_counter_value_in_isr(TIMER_GROUP_0, TIMER_0);
timer_group_intr_clr_in_isr(TIMER_GROUP_0, TIMER_0);
timer_counter_value += (uint64_t) (TIMER_INTERVAL0_SEC * TIMER_SCALE);
timer_group_set_alarm_value_in_isr(TIMER_GROUP_0, TIMER_0, timer_counter_value);
int data = (int) para;
// Posting events with data more than 4 bytes should fail.
TEST_ASSERT_EQUAL(ESP_ERR_INVALID_ARG, esp_event_isr_post(s_test_base1, TEST_EVENT_BASE1_EV1, &data, 5, NULL));
// This should succeedd, as data is int-sized. The handler for the event checks that the passed event data
// is correct.
BaseType_t task_unblocked;
TEST_ASSERT_EQUAL(ESP_OK, esp_event_isr_post(s_test_base1, TEST_EVENT_BASE1_EV1, &data, sizeof(data), &task_unblocked));
if (task_unblocked == pdTRUE) {
portYIELD_FROM_ISR();
}
}
#endif //CONFIG_ESP_EVENT_POST_FROM_ISR
TEST_CASE("can create and delete event loops", "[event]")
{
/* this test aims to verify that:
* - creating loops with and without a task succeeds
* - event queue can accomodate the set queue size, and drops the post when exceeded
* - deleting loops with unconsumed posts and unregistered handlers (when unregistration is enabled) does not leak memory */
TEST_SETUP();
esp_event_loop_handle_t loop1; // with dedicated task
esp_event_loop_handle_t loop2; // without dedicated task
esp_event_loop_handle_t loop3; // with leftover post and handlers
esp_event_loop_args_t loop_args = test_event_get_default_loop_args();
TEST_ASSERT_EQUAL(ESP_OK, esp_event_loop_create(&loop_args, &loop1));
loop_args.task_name = NULL;
TEST_ASSERT_EQUAL(ESP_OK, esp_event_loop_create(&loop_args, &loop2));
TEST_ASSERT_EQUAL(ESP_OK, esp_event_loop_create(&loop_args, &loop3));
TEST_ASSERT_EQUAL(ESP_OK, esp_event_handler_register_with(loop3, s_test_base1, TEST_EVENT_BASE1_EV1, (void*) 0x00000001, NULL));
TEST_ASSERT_EQUAL(ESP_OK, esp_event_handler_register_with(loop3, s_test_base1, TEST_EVENT_BASE1_EV2, (void*) 0x00000002, NULL));
TEST_ASSERT_EQUAL(ESP_OK, esp_event_handler_register_with(loop3, s_test_base2, TEST_EVENT_BASE1_EV1, (void*) 0x00000003, NULL));
for (int i = 0; i < loop_args.queue_size; i++) {
int mod = i % 4;
switch(mod) {
case 0:
TEST_ASSERT_EQUAL(ESP_OK, esp_event_post_to(loop3, s_test_base1, TEST_EVENT_BASE1_EV1, NULL, 0, portMAX_DELAY));
break;
case 1:
TEST_ASSERT_EQUAL(ESP_OK, esp_event_post_to(loop3, s_test_base2, TEST_EVENT_BASE1_EV1, NULL, 0, portMAX_DELAY));
break;
case 2:
TEST_ASSERT_EQUAL(ESP_OK, esp_event_post_to(loop3, s_test_base1, TEST_EVENT_BASE1_EV2, NULL, 0, portMAX_DELAY));
break;
case 3:
TEST_ASSERT_EQUAL(ESP_OK, esp_event_post_to(loop3, s_test_base2, TEST_EVENT_BASE1_EV2, NULL, 0, portMAX_DELAY));
break;
default:
break;
}
}
TEST_ASSERT_EQUAL(ESP_ERR_TIMEOUT, esp_event_post_to(loop3, s_test_base1, TEST_EVENT_BASE1_EV1, NULL, 0, pdMS_TO_TICKS(10)));
TEST_ASSERT_EQUAL(ESP_OK, esp_event_loop_delete(loop1));
TEST_ASSERT_EQUAL(ESP_OK, esp_event_loop_delete(loop2));
TEST_ASSERT_EQUAL(ESP_OK, esp_event_loop_delete(loop3));
TEST_TEARDOWN();
}
TEST_CASE("can register/unregister handlers for all events/all events for a specific base", "[event]")
{
/* this test aims to verify that handlers can be registered to be called on all events
* or for all events with specific bases */
TEST_SETUP();
esp_event_loop_handle_t loop;
int count = 0;
simple_arg_t arg = {
.data = &count,
.mutex = xSemaphoreCreateMutex()
};
esp_event_loop_args_t loop_args = test_event_get_default_loop_args();
loop_args.task_name = NULL;
TEST_ASSERT_EQUAL(ESP_OK, esp_event_loop_create(&loop_args, &loop));
/* Register the handler twice to the same base and id but with a different argument (expects to return ESP_OK and log a warning)
* This aims to verify: 1) Handler's argument to be updated
* 2) Registration not to leak memory
*/
TEST_ASSERT_EQUAL(ESP_OK, esp_event_handler_register_with(loop, ESP_EVENT_ANY_BASE, ESP_EVENT_ANY_ID, test_event_simple_handler, NULL));
TEST_ASSERT_EQUAL(ESP_OK, esp_event_handler_register_with(loop, ESP_EVENT_ANY_BASE, ESP_EVENT_ANY_ID, test_event_simple_handler, &arg));
TEST_ASSERT_EQUAL(ESP_OK, esp_event_handler_register_with(loop, s_test_base1, ESP_EVENT_ANY_ID, test_event_simple_handler, &arg));
TEST_ASSERT_EQUAL(ESP_ERR_INVALID_ARG, esp_event_handler_register_with(loop, ESP_EVENT_ANY_BASE, TEST_EVENT_BASE1_EV1, test_event_simple_handler, &arg));
TEST_ASSERT_EQUAL(ESP_OK, esp_event_handler_register_with(loop, s_test_base1, TEST_EVENT_BASE1_EV1, test_event_simple_handler, &arg));
TEST_ASSERT_EQUAL(ESP_OK, esp_event_handler_register_with(loop, s_test_base1, TEST_EVENT_BASE1_EV2, test_event_simple_handler, &arg));
TEST_ASSERT_EQUAL(ESP_OK, esp_event_handler_register_with(loop, s_test_base2, TEST_EVENT_BASE1_EV1, test_event_simple_handler, &arg));
TEST_ASSERT_EQUAL(ESP_ERR_INVALID_ARG, esp_event_post_to(loop, ESP_EVENT_ANY_BASE, ESP_EVENT_ANY_ID, NULL, 0, portMAX_DELAY));
TEST_ASSERT_EQUAL(ESP_ERR_INVALID_ARG, esp_event_post_to(loop, s_test_base1, ESP_EVENT_ANY_ID, NULL, 0, portMAX_DELAY));
TEST_ASSERT_EQUAL(ESP_ERR_INVALID_ARG, esp_event_post_to(loop, ESP_EVENT_ANY_BASE, TEST_EVENT_BASE1_EV1, NULL, 0, portMAX_DELAY));
TEST_ASSERT_EQUAL(ESP_OK, esp_event_post_to(loop, s_test_base1, TEST_EVENT_BASE1_EV1, NULL, 0, portMAX_DELAY)); // exec loop, base and id level (+3)
TEST_ASSERT_EQUAL(ESP_OK, esp_event_post_to(loop, s_test_base1, TEST_EVENT_BASE1_EV2, NULL, 0, portMAX_DELAY)); // exec loop, base and id level (+3)
// Post unknown events. Respective loop level and base level handlers should still execute.
TEST_ASSERT_EQUAL(ESP_OK, esp_event_post_to(loop, s_test_base1, TEST_EVENT_BASE1_MAX, NULL, 0, portMAX_DELAY)); // exec loop and base level (+2)
TEST_ASSERT_EQUAL(ESP_OK, esp_event_post_to(loop, s_test_base2, TEST_EVENT_BASE2_MAX, NULL, 0, portMAX_DELAY)); // exec loop level (+1)
TEST_ASSERT_EQUAL(ESP_OK, esp_event_loop_run(loop, pdMS_TO_TICKS(10)));
TEST_ASSERT_EQUAL(9, count); // 3 + 3 + 2 + 1
TEST_ASSERT_EQUAL(ESP_OK, esp_event_loop_delete(loop));
vSemaphoreDelete(arg.mutex);
TEST_TEARDOWN();
}
TEST_CASE("can unregister handler", "[event]")
{
/* this test aims to verify that unregistered handlers no longer execute when events are raised */
TEST_SETUP();
esp_event_loop_handle_t loop;
esp_event_loop_args_t loop_args = test_event_get_default_loop_args();
loop_args.task_name = NULL;
TEST_ASSERT_EQUAL(ESP_OK, esp_event_loop_create(&loop_args, &loop));
int count = 0;
simple_arg_t arg = {
.data = &count,
.mutex = xSemaphoreCreateMutex()
};
TEST_ASSERT_EQUAL(ESP_OK, esp_event_handler_register_with(loop, s_test_base1, TEST_EVENT_BASE1_EV1, test_event_simple_handler, &arg));
TEST_ASSERT_EQUAL(ESP_OK, esp_event_handler_register_with(loop, s_test_base2, TEST_EVENT_BASE1_EV1, test_event_simple_handler, &arg));
TEST_ASSERT_EQUAL(ESP_OK, esp_event_post_to(loop, s_test_base1, TEST_EVENT_BASE1_EV1, NULL, 0, portMAX_DELAY));
TEST_ASSERT_EQUAL(ESP_OK, esp_event_post_to(loop, s_test_base2, TEST_EVENT_BASE1_EV1, NULL, 0, portMAX_DELAY));
TEST_ASSERT_EQUAL(ESP_OK, esp_event_loop_run(loop, pdMS_TO_TICKS(10)));
TEST_ASSERT_EQUAL(2, count);
TEST_ASSERT_EQUAL(ESP_OK, esp_event_handler_unregister_with(loop, s_test_base1, TEST_EVENT_BASE1_EV1, test_event_simple_handler));
TEST_ASSERT_EQUAL(ESP_OK, esp_event_post_to(loop, s_test_base1, TEST_EVENT_BASE1_EV1, NULL, 0, portMAX_DELAY));
TEST_ASSERT_EQUAL(ESP_OK, esp_event_post_to(loop, s_test_base2, TEST_EVENT_BASE1_EV1, NULL, 0, portMAX_DELAY));
TEST_ASSERT_EQUAL(ESP_OK, esp_event_loop_run(loop, pdMS_TO_TICKS(10)));
TEST_ASSERT_EQUAL(3, count);
TEST_ASSERT_EQUAL(ESP_OK, esp_event_loop_delete(loop));
vSemaphoreDelete(arg.mutex);
TEST_TEARDOWN();
}
TEST_CASE("handler can unregister itself", "[event]")
{
/* this test aims to verify that handlers can unregister themselves */
TEST_SETUP();
esp_event_loop_handle_t loop;
esp_event_loop_args_t loop_args = test_event_get_default_loop_args();
loop_args.task_name = NULL;
TEST_ASSERT_EQUAL(ESP_OK, esp_event_loop_create(&loop_args, &loop));
int unregistered = 0;
/*
* s_test_base1, ev1 = 1
* s_test_base1, ev2 = 2
* s_test_base2, ev1 = 11
* s_test_base2, ev2 = 12
*/
int expected_unregistered = 0;
TEST_ASSERT_EQUAL(ESP_OK, esp_event_handler_register_with(loop, s_test_base1, TEST_EVENT_BASE1_EV1, test_handler_unregister_itself, &unregistered));
TEST_ASSERT_EQUAL(ESP_OK, esp_event_handler_register_with(loop, s_test_base1, TEST_EVENT_BASE1_EV2, test_handler_unregister_itself, &unregistered));
TEST_ASSERT_EQUAL(ESP_OK, esp_event_handler_register_with(loop, s_test_base2, TEST_EVENT_BASE2_EV1, test_handler_unregister_itself, &unregistered));
TEST_ASSERT_EQUAL(ESP_OK, esp_event_handler_register_with(loop, s_test_base2, TEST_EVENT_BASE2_EV2, test_handler_unregister_itself, &unregistered));
TEST_ASSERT_EQUAL(ESP_OK, esp_event_post_to(loop, s_test_base1, TEST_EVENT_BASE1_EV2, &loop, sizeof(loop), portMAX_DELAY));
TEST_ASSERT_EQUAL(ESP_OK, esp_event_loop_run(loop, pdMS_TO_TICKS(10)));
expected_unregistered = 2; // base1, ev2
TEST_ASSERT_EQUAL(expected_unregistered, unregistered);
TEST_ASSERT_EQUAL(ESP_OK, esp_event_post_to(loop, s_test_base1, TEST_EVENT_BASE1_EV1, &loop, sizeof(loop), portMAX_DELAY));
TEST_ASSERT_EQUAL(ESP_OK, esp_event_post_to(loop, s_test_base2, TEST_EVENT_BASE2_EV1, &loop, sizeof(loop), portMAX_DELAY));
TEST_ASSERT_EQUAL(ESP_OK, esp_event_loop_run(loop, pdMS_TO_TICKS(10)));
expected_unregistered += 1 + 11; // base1, ev1 + base2, ev1
TEST_ASSERT_EQUAL(expected_unregistered, unregistered);
TEST_ASSERT_EQUAL(ESP_OK, esp_event_post_to(loop, s_test_base2, TEST_EVENT_BASE2_EV2, &loop, sizeof(loop), portMAX_DELAY));
TEST_ASSERT_EQUAL(ESP_OK, esp_event_loop_run(loop, pdMS_TO_TICKS(10)));
expected_unregistered += 12; // base2, ev2
TEST_ASSERT_EQUAL(expected_unregistered, unregistered);
TEST_ASSERT_EQUAL(ESP_OK, esp_event_post_to(loop, s_test_base1, TEST_EVENT_BASE1_EV1, &loop, sizeof(loop), portMAX_DELAY));
TEST_ASSERT_EQUAL(ESP_OK, esp_event_post_to(loop, s_test_base1, TEST_EVENT_BASE1_EV2, &loop, sizeof(loop), portMAX_DELAY));
TEST_ASSERT_EQUAL(ESP_OK, esp_event_post_to(loop, s_test_base2, TEST_EVENT_BASE2_EV1, &loop, sizeof(loop), portMAX_DELAY));
TEST_ASSERT_EQUAL(ESP_OK, esp_event_post_to(loop, s_test_base2, TEST_EVENT_BASE2_EV2, &loop, sizeof(loop), portMAX_DELAY));
TEST_ASSERT_EQUAL(ESP_OK, esp_event_loop_run(loop, pdMS_TO_TICKS(10)));
TEST_ASSERT_EQUAL(expected_unregistered, unregistered); // all handlers unregistered
TEST_ASSERT_EQUAL(ESP_OK, esp_event_loop_delete(loop));
TEST_TEARDOWN();
}
TEST_CASE("can exit running loop at approximately the set amount of time", "[event]")
{
/* this test aims to verify that running loop does not block indefinitely in cases where
* events are posted frequently */
TEST_SETUP();
esp_event_loop_handle_t loop;
esp_event_loop_args_t loop_args = test_event_get_default_loop_args();
loop_args.task_name = NULL;
TEST_ASSERT_EQUAL(ESP_OK, esp_event_loop_create(&loop_args, &loop));
performance_data_t handler_data = {
.performed = 0,
.expected = INT32_MAX,
.done = xSemaphoreCreateBinary()
};
TEST_ASSERT_EQUAL(ESP_OK, esp_event_handler_register_with(loop, s_test_base1, TEST_EVENT_BASE1_EV1, test_event_performance_handler, &handler_data));
post_event_data_t post_event_data = {
.base = s_test_base1,
.id = TEST_EVENT_BASE1_EV1,
.loop = loop,
.num = INT32_MAX
};
task_arg_t post_event_arg = {
.data = &post_event_data,
.done = xSemaphoreCreateBinary(),
.start = xSemaphoreCreateBinary()
};
TaskHandle_t post_task;
xTaskCreatePinnedToCore(test_event_post_task, "post", 2048, &post_event_arg, s_test_priority, &post_task, test_event_get_core());
int runtime_ms = 10;
int runtime_us = runtime_ms * 1000;
int64_t start, diff;
start = esp_timer_get_time();
xSemaphoreGive(post_event_arg.start);
// Run the loop for the runtime_ms set amount of time, regardless of whether events
// are still being posted to the loop.
TEST_ASSERT_EQUAL(ESP_OK, esp_event_loop_run(loop, pdMS_TO_TICKS(runtime_ms)));
diff = (esp_timer_get_time() - start);
// Threshold is 25 percent.
TEST_ASSERT(diff < runtime_us * 1.25f);
// Verify that the post task still continues
TEST_ASSERT_NOT_EQUAL(pdTRUE, xSemaphoreTake(post_event_arg.done, pdMS_TO_TICKS(10)));
vSemaphoreDelete(post_event_arg.done);
vSemaphoreDelete(post_event_arg.start);
vSemaphoreDelete(handler_data.done);
vTaskDelete(post_task);
TEST_ASSERT_EQUAL(ESP_OK, esp_event_loop_delete(loop));
TEST_TEARDOWN();
}
TEST_CASE("can register/unregister handlers simultaneously", "[event]")
{
/* this test aims to verify that the event handlers list remains consistent despite
* simultaneous access by differenct tasks */
TEST_SETUP();
const char* base = "base";
int32_t id = 0;
esp_event_loop_handle_t loop;
esp_event_loop_args_t loop_args = test_event_get_default_loop_args();
TEST_ASSERT_EQUAL(ESP_OK, esp_event_loop_create(&loop_args, &loop));
ESP_LOGI(TAG, "registering handlers");
handler_registration_data_t* registration_data = calloc(TEST_CONFIG_TASKS_TO_SPAWN, sizeof(*registration_data));
task_arg_t* registration_arg = calloc(TEST_CONFIG_TASKS_TO_SPAWN, sizeof(*registration_arg));
for (int i = 0; i < TEST_CONFIG_TASKS_TO_SPAWN; i++) {
registration_data[i].base = base;
registration_data[i].id = id;
registration_data[i].loop = loop;
registration_data[i].handles = calloc(TEST_CONFIG_ITEMS_TO_REGISTER, sizeof(esp_event_handler_t));
registration_data[i].num = TEST_CONFIG_ITEMS_TO_REGISTER;
registration_data[i].is_registration = true;
for (int j = 0; j < TEST_CONFIG_ITEMS_TO_REGISTER; j++) {
registration_data[i].handles[j] = (void*) (i * TEST_CONFIG_ITEMS_TO_REGISTER) + (j + TEST_CONFIG_ITEMS_TO_REGISTER);
}
registration_arg[i].start = xSemaphoreCreateBinary();
registration_arg[i].done = xSemaphoreCreateBinary();
registration_arg[i].data = &registration_data[i];
xTaskCreatePinnedToCore(test_event_simple_handler_registration_task, "register", 2048, &registration_arg[i], s_test_priority, NULL, test_event_get_core());
}
// Give the semaphores to the spawned registration task
for (int i = 0; i < TEST_CONFIG_TASKS_TO_SPAWN; i++) {
xSemaphoreGive(registration_arg[i].start);
}
// Take the same semaphores in order to proceed
for (int i = 0; i < TEST_CONFIG_TASKS_TO_SPAWN; i++) {
xSemaphoreTake(registration_arg[i].done, portMAX_DELAY);
}
ESP_LOGI(TAG, "checking consistency of handlers list");
// Check consistency of events list
for (int i = 0; i < TEST_CONFIG_TASKS_TO_SPAWN; i++) {
for (int j = 0; j < TEST_CONFIG_ITEMS_TO_REGISTER; j++) {
TEST_ASSERT_TRUE(esp_event_is_handler_registered(loop, base, id, registration_data[i].handles[j]));
}
}
ESP_LOGI(TAG, "unregistering handlers");
/* Test if tasks can unregister simultaneously */
// Unregister registered events
handler_registration_data_t* unregistration_data = calloc(TEST_CONFIG_TASKS_TO_SPAWN, sizeof(*unregistration_data));
task_arg_t* unregistration_arg = calloc(TEST_CONFIG_TASKS_TO_SPAWN, sizeof(*unregistration_arg));
for (int i = 0; i < TEST_CONFIG_TASKS_TO_SPAWN; i++) {
unregistration_data[i].base = base;
unregistration_data[i].id = id;
unregistration_data[i].loop = loop;
unregistration_data[i].handles = calloc(TEST_CONFIG_ITEMS_TO_REGISTER, sizeof(esp_event_handler_t));
unregistration_data[i].num = TEST_CONFIG_ITEMS_TO_REGISTER;
unregistration_data[i].is_registration = false;
memcpy(unregistration_data[i].handles, registration_data[i].handles, TEST_CONFIG_ITEMS_TO_REGISTER * sizeof(esp_event_handler_t));
unregistration_arg[i].data = &unregistration_data[i];
unregistration_arg[i].start = xSemaphoreCreateBinary();
unregistration_arg[i].done = xSemaphoreCreateBinary();
xTaskCreatePinnedToCore(test_event_simple_handler_registration_task, "unregister", 2048, &unregistration_arg[i], s_test_priority, NULL, test_event_get_core());
}
// Give the semaphores to the spawned unregistration task
for (int i = 0; i < TEST_CONFIG_TASKS_TO_SPAWN; i++) {
xSemaphoreGive(unregistration_arg[i].start);
}
// Take the same semaphores in order to proceed
for (int i = 0; i < TEST_CONFIG_TASKS_TO_SPAWN; i++) {
xSemaphoreTake(unregistration_arg[i].done, portMAX_DELAY);
}
ESP_LOGI(TAG, "checking consistency of handlers list");
// Check consistency of events list
for (int i = 0; i < TEST_CONFIG_TASKS_TO_SPAWN; i++) {
for (int j = 0; j < TEST_CONFIG_ITEMS_TO_REGISTER; j++) {
TEST_ASSERT_FALSE(esp_event_is_handler_registered(loop, base, id, registration_data[i].handles[j]));
}
}
// Do cleanup
for (int i = 0; i < TEST_CONFIG_TASKS_TO_SPAWN; i++) {
free(registration_data[i].handles);
vSemaphoreDelete(registration_arg[i].start);
vSemaphoreDelete(registration_arg[i].done);
free(unregistration_data[i].handles);
vSemaphoreDelete(unregistration_arg[i].start);
vSemaphoreDelete(unregistration_arg[i].done);
}
free(registration_data);
free(unregistration_data);
free(registration_arg);
free(unregistration_arg);
TEST_ASSERT_EQUAL(ESP_OK, esp_event_loop_delete(loop));
TEST_TEARDOWN();
}
TEST_CASE("can post and run events", "[event]")
{
/* this test aims to verify that:
* - multiple tasks can post to the queue simultaneously
* - handlers recieve the appropriate handler arg and associated event data */
TEST_SETUP();
esp_event_loop_handle_t loop;
esp_event_loop_args_t loop_args = test_event_get_default_loop_args();
loop_args.task_name = NULL;
loop_args.queue_size = TEST_CONFIG_TASKS_TO_SPAWN * TEST_CONFIG_ITEMS_TO_REGISTER;
TEST_ASSERT_EQUAL(ESP_OK, esp_event_loop_create(&loop_args, &loop));
int count = 0;
simple_arg_t arg = {
.data = &count,
.mutex = xSemaphoreCreateMutex()
};
TEST_ASSERT_EQUAL(ESP_OK, esp_event_handler_register_with(loop, s_test_base1, TEST_EVENT_BASE1_EV1, test_event_simple_handler, &arg));
post_event_data_t* post_event_data = calloc(TEST_CONFIG_TASKS_TO_SPAWN, sizeof(*post_event_data));
task_arg_t* post_event_arg = calloc(TEST_CONFIG_TASKS_TO_SPAWN, sizeof(*post_event_arg));
for (int i = 0; i < TEST_CONFIG_TASKS_TO_SPAWN; i++)
{
post_event_data[i].base = s_test_base1;
post_event_data[i].id = TEST_EVENT_BASE1_EV1;
post_event_data[i].loop = loop;
post_event_data[i].num = TEST_CONFIG_ITEMS_TO_REGISTER;
post_event_arg[i].data = &post_event_data[i];
post_event_arg[i].start = xSemaphoreCreateBinary();
post_event_arg[i].done = xSemaphoreCreateBinary();
xTaskCreatePinnedToCore(test_event_post_task, "post", 2048, &post_event_arg[i], s_test_priority, NULL, test_event_get_core());
}
for (int i = 0; i < TEST_CONFIG_TASKS_TO_SPAWN; i++) {
xSemaphoreGive(post_event_arg[i].start);
}
// Execute some events as they are posted
for (int i = 0; i < (TEST_CONFIG_TASKS_TO_SPAWN * TEST_CONFIG_ITEMS_TO_REGISTER) / 2; i++) {
TEST_ASSERT_EQUAL(ESP_OK, esp_event_loop_run(loop, pdMS_TO_TICKS(10)));
}
for (int i = 0; i < TEST_CONFIG_TASKS_TO_SPAWN; i++) {
xSemaphoreTake(post_event_arg[i].done, portMAX_DELAY);
}
// Execute the rest
TEST_ASSERT_EQUAL(ESP_OK, esp_event_loop_run(loop, pdMS_TO_TICKS(10)));
TEST_ASSERT_EQUAL(TEST_CONFIG_TASKS_TO_SPAWN * TEST_CONFIG_ITEMS_TO_REGISTER, count);
// Cleanup
for (int i = 0; i < TEST_CONFIG_TASKS_TO_SPAWN; i++) {
vSemaphoreDelete(post_event_arg[i].start);
vSemaphoreDelete(post_event_arg[i].done);
}
free(post_event_data);
free(post_event_arg);
TEST_ASSERT_EQUAL(ESP_OK, esp_event_loop_delete(loop));
vSemaphoreDelete(arg.mutex);
TEST_TEARDOWN();
}
static void loop_run_task(void* args)
{
esp_event_loop_handle_t event_loop = (esp_event_loop_handle_t) args;
while(1) {
esp_event_loop_run(event_loop, portMAX_DELAY);
}
}
static void performance_test(bool dedicated_task)
{
// rand() seems to do a one-time allocation. Call it here so that the memory it allocates
// is not counted as a leak.
unsigned int _rand __attribute__((unused)) = rand();
TEST_SETUP();
const char test_base[] = "qwertyuiopasdfghjklzxvbnmmnbvcxzqwertyuiopasdfghjklzxvbnmmnbvcxz";
#define TEST_CONFIG_BASES (sizeof(test_base) - 1)
#define TEST_CONFIG_IDS (TEST_CONFIG_BASES / 2)
// Create loop
esp_event_loop_args_t loop_args = test_event_get_default_loop_args();
esp_event_loop_handle_t loop;
if (!dedicated_task) {
loop_args.task_name = NULL;
}
TEST_ASSERT_EQUAL(ESP_OK, esp_event_loop_create(&loop_args, &loop));
performance_data_t data;
// Register the handlers
for (int base = 0; base < TEST_CONFIG_BASES; base++) {
for (int id = 0; id < TEST_CONFIG_IDS; id++) {
TEST_ASSERT_EQUAL(ESP_OK, esp_event_handler_register_with(loop, test_base + base, id, test_event_performance_handler, &data));
}
}
TaskHandle_t mtask = NULL;
if (!dedicated_task) {
xTaskCreate(loop_run_task, "loop_run", loop_args.task_stack_size, (void*) loop, loop_args.task_priority, &mtask);
}
// Perform performance test
float running_sum = 0;
float running_count = 0;
for (int bases = 1; bases <= TEST_CONFIG_BASES; bases *= 2) {
for (int ids = 1; ids <= TEST_CONFIG_IDS; ids *= 2) {
data.performed = 0;
data.expected = bases * ids;
data.done = xSemaphoreCreateBinary();
// Generate randomized list of posts
int post_bases[TEST_CONFIG_BASES];
int post_ids[TEST_CONFIG_IDS];
for (int i = 0; i < bases; i++) {
post_bases[i] = i;
}
for (int i = 0; i < ids; i++) {
post_ids[i] = i;
}
for (int i = 0; i < bases; i++) {
int rand_a = rand() % bases;
int rand_b = rand() % bases;
int temp = post_bases[rand_a];
post_bases[rand_a]= post_bases[rand_b];
post_bases[rand_b] = temp;
}
for (int i = 0; i < ids; i++) {
int rand_a = rand() % ids;
int rand_b = rand() % ids;
int temp = post_ids[rand_a];
post_ids[rand_a]= post_ids[rand_b];
post_ids[rand_b] = temp;
}
// Post the events
int64_t start = esp_timer_get_time();
for (int base = 0; base < bases; base++) {
for (int id = 0; id < ids; id++) {
TEST_ASSERT_EQUAL(ESP_OK, esp_event_post_to(loop, test_base + post_bases[base], post_ids[id], NULL, 0, portMAX_DELAY));
}
}
xSemaphoreTake(data.done, portMAX_DELAY);
int64_t elapsed = esp_timer_get_time() - start;
// Record data
TEST_ASSERT_EQUAL(data.expected, data.performed);
running_count++;
running_sum += data.performed / (elapsed / (1000000.0));
vSemaphoreDelete(data.done);
}
}
int average = (int) (running_sum / (running_count));
if (!dedicated_task) {
((esp_event_loop_instance_t*) loop)->task = mtask;
}
TEST_ASSERT_EQUAL(ESP_OK, esp_event_loop_delete(loop));
TEST_TEARDOWN();
#ifdef CONFIG_ESP_EVENT_LOOP_PROFILING
ESP_LOGI(TAG, "events dispatched/second with profiling enabled: %d", average);
// Enabling profiling will slow down event dispatch, so the set threshold
// is not valid when it is enabled.
#else
#ifndef CONFIG_SPIRAM
TEST_PERFORMANCE_GREATER_THAN(EVENT_DISPATCH, "%d", average);
#else
TEST_PERFORMANCE_GREATER_THAN(EVENT_DISPATCH_PSRAM, "%d", average);
#endif // CONFIG_SPIRAM
#endif // CONFIG_ESP_EVENT_LOOP_PROFILING
}
TEST_CASE("performance test - dedicated task", "[event]")
{
performance_test(true);
}
TEST_CASE("performance test - no dedicated task", "[event]")
{
performance_test(false);
}
TEST_CASE("can post to loop from handler - dedicated task", "[event]")
{
TEST_SETUP();
esp_event_loop_handle_t loop_w_task;
esp_event_loop_args_t loop_args = test_event_get_default_loop_args();
int count;
simple_arg_t arg = {
.data = &count,
.mutex = xSemaphoreCreateBinary()
};
TEST_ASSERT_EQUAL(ESP_OK, esp_event_loop_create(&loop_args, &loop_w_task));
count = 0;
// Test that a handler can post to a different loop while there is still slots on the queue
TEST_ASSERT_EQUAL(ESP_OK, esp_event_handler_register_with(loop_w_task, s_test_base1, TEST_EVENT_BASE1_EV1, test_handler_post_w_task, &arg));
TEST_ASSERT_EQUAL(ESP_OK, esp_event_handler_register_with(loop_w_task, s_test_base1, TEST_EVENT_BASE1_EV2, test_handler_post_w_task, &arg));
TEST_ASSERT_EQUAL(ESP_OK, esp_event_post_to(loop_w_task, s_test_base1, TEST_EVENT_BASE1_EV1, &loop_w_task, sizeof(&loop_w_task), portMAX_DELAY));
xSemaphoreTake(arg.mutex, portMAX_DELAY);
TEST_ASSERT_EQUAL(2, count);
// Test that other tasks can still post while there is still slots in the queue, while handler is executing
count = 100;
TEST_ASSERT_EQUAL(ESP_OK, esp_event_post_to(loop_w_task, s_test_base1, TEST_EVENT_BASE1_EV1, &loop_w_task, sizeof(&loop_w_task), portMAX_DELAY));
for (int i = 0; i < loop_args.queue_size; i++) {
TEST_ASSERT_EQUAL(ESP_OK, esp_event_post_to(loop_w_task, s_test_base1, TEST_EVENT_BASE1_EV2, NULL, 0, portMAX_DELAY));
}
TEST_ASSERT_EQUAL(ESP_ERR_TIMEOUT, esp_event_post_to(loop_w_task, s_test_base1, TEST_EVENT_BASE1_EV2, NULL, 0,
pdMS_TO_TICKS(CONFIG_ESP_INT_WDT_TIMEOUT_MS * TEST_CONFIG_WAIT_MULTIPLIER)));
xSemaphoreGive(arg.mutex);
vTaskDelay(pdMS_TO_TICKS(CONFIG_ESP_INT_WDT_TIMEOUT_MS * TEST_CONFIG_WAIT_MULTIPLIER));
TEST_ASSERT_EQUAL(ESP_OK, esp_event_loop_delete(loop_w_task));
vSemaphoreDelete(arg.mutex);
TEST_TEARDOWN();
}
TEST_CASE("can post to loop from handler - no dedicated task", "[event]")
{
TEST_SETUP();
esp_event_loop_handle_t loop_wo_task;
esp_event_loop_args_t loop_args = test_event_get_default_loop_args();
int count;
simple_arg_t arg = {
.data = &count,
.mutex = xSemaphoreCreateBinary()
};
count = 0;
loop_args.queue_size = 1;
loop_args.task_name = NULL;
TEST_ASSERT_EQUAL(ESP_OK, esp_event_loop_create(&loop_args, &loop_wo_task));
TaskHandle_t mtask;
xTaskCreate(test_post_from_handler_loop_task, "task", 2584, (void*) loop_wo_task, s_test_priority, &mtask);
// Test that a handler can post to a different loop while there is still slots on the queue
TEST_ASSERT_EQUAL(ESP_OK, esp_event_handler_register_with(loop_wo_task, s_test_base1, TEST_EVENT_BASE1_EV1, test_handler_post_wo_task, &arg));
TEST_ASSERT_EQUAL(ESP_OK, esp_event_handler_register_with(loop_wo_task, s_test_base1, TEST_EVENT_BASE1_EV2, test_handler_post_wo_task, &arg));
TEST_ASSERT_EQUAL(ESP_OK, esp_event_post_to(loop_wo_task, s_test_base1, TEST_EVENT_BASE1_EV1, &loop_wo_task, sizeof(&loop_wo_task), portMAX_DELAY));
xSemaphoreTake(arg.mutex, portMAX_DELAY);
TEST_ASSERT_EQUAL(2, count);
count = 100;
TEST_ASSERT_EQUAL(ESP_OK, esp_event_post_to(loop_wo_task, s_test_base1, TEST_EVENT_BASE1_EV1, &loop_wo_task, sizeof(&loop_wo_task), portMAX_DELAY));
vTaskDelay(pdMS_TO_TICKS(CONFIG_ESP_INT_WDT_TIMEOUT_MS * TEST_CONFIG_WAIT_MULTIPLIER));
// For loop without tasks, posting is more restrictive. Posting should wait until execution of handler finishes
TEST_ASSERT_EQUAL(ESP_ERR_TIMEOUT, esp_event_post_to(loop_wo_task, s_test_base1, TEST_EVENT_BASE1_EV2, NULL, 0,
pdMS_TO_TICKS(CONFIG_ESP_INT_WDT_TIMEOUT_MS * TEST_CONFIG_WAIT_MULTIPLIER)));
xSemaphoreGive(arg.mutex);
vTaskDelay(pdMS_TO_TICKS(CONFIG_ESP_INT_WDT_TIMEOUT_MS * TEST_CONFIG_WAIT_MULTIPLIER));
vTaskDelete(mtask);
TEST_ASSERT_EQUAL(ESP_OK, esp_event_loop_delete(loop_wo_task));
vSemaphoreDelete(arg.mutex);
TEST_TEARDOWN();
}
static void test_event_simple_handler_template(void* handler_arg, esp_event_base_t base, int32_t id, void* event_arg)
{
int* count = (int*) handler_arg;
(*count)++;
}
static void test_event_simple_handler_1(void* handler_arg, esp_event_base_t base, int32_t id, void* event_arg)
{
test_event_simple_handler_template(handler_arg, base, id, event_arg);
}
static void test_event_simple_handler_3(void* handler_arg, esp_event_base_t base, int32_t id, void* event_arg)
{
test_event_simple_handler_template(handler_arg, base, id, event_arg);
}
static void test_event_simple_handler_2(void* handler_arg, esp_event_base_t base, int32_t id, void* event_arg)
{
test_event_simple_handler_template(handler_arg, base, id, event_arg);
}
static void test_registration_from_handler_hdlr(void* handler_arg, esp_event_base_t base, int32_t id, void* event_arg)
{
esp_event_loop_handle_t* loop = (esp_event_loop_handle_t*) event_arg;
TEST_ASSERT_EQUAL(ESP_OK, esp_event_handler_register_with(*loop, s_test_base1, TEST_EVENT_BASE1_EV2, test_event_simple_handler_1, handler_arg));
TEST_ASSERT_EQUAL(ESP_OK, esp_event_handler_register_with(*loop, s_test_base1, TEST_EVENT_BASE1_EV2, test_event_simple_handler_2, handler_arg));
TEST_ASSERT_EQUAL(ESP_OK, esp_event_handler_register_with(*loop, s_test_base1, TEST_EVENT_BASE1_EV2, test_event_simple_handler_3, handler_arg));
}
static void test_unregistration_from_handler_hdlr(void* handler_arg, esp_event_base_t base, int32_t id, void* event_arg)
{
esp_event_loop_handle_t* loop = (esp_event_loop_handle_t*) event_arg;
TEST_ASSERT_EQUAL(ESP_OK, esp_event_handler_unregister_with(*loop, s_test_base1, TEST_EVENT_BASE1_EV2, test_event_simple_handler_1));
TEST_ASSERT_EQUAL(ESP_OK, esp_event_handler_unregister_with(*loop, s_test_base1, TEST_EVENT_BASE1_EV2, test_event_simple_handler_2));
TEST_ASSERT_EQUAL(ESP_OK, esp_event_handler_unregister_with(*loop, s_test_base1, TEST_EVENT_BASE1_EV2, test_event_simple_handler_3));
}
TEST_CASE("can register from handler", "[event]")
{
TEST_SETUP();
esp_event_loop_handle_t loop;
esp_event_loop_args_t loop_args = test_event_get_default_loop_args();
loop_args.task_name = NULL;
TEST_ASSERT_EQUAL(ESP_OK, esp_event_loop_create(&loop_args, &loop));
int count = 0;
TEST_ASSERT_EQUAL(ESP_OK, esp_event_handler_register_with(loop, s_test_base1, TEST_EVENT_BASE1_EV1, test_registration_from_handler_hdlr, &count));
TEST_ASSERT_EQUAL(ESP_OK, esp_event_handler_register_with(loop, s_test_base2, TEST_EVENT_BASE2_EV1, test_unregistration_from_handler_hdlr, &count));
TEST_ASSERT_EQUAL(ESP_OK, esp_event_post_to(loop, s_test_base1, TEST_EVENT_BASE1_EV1, &loop, sizeof(&loop), portMAX_DELAY));
TEST_ASSERT_EQUAL(ESP_OK, esp_event_loop_run(loop, pdMS_TO_TICKS(10)));
TEST_ASSERT_EQUAL(ESP_OK, esp_event_post_to(loop, s_test_base1, TEST_EVENT_BASE1_EV2, NULL, 0, portMAX_DELAY));
TEST_ASSERT_EQUAL(ESP_OK, esp_event_loop_run(loop, pdMS_TO_TICKS(10)));
TEST_ASSERT_EQUAL(3, count);
TEST_ASSERT_EQUAL(ESP_OK, esp_event_post_to(loop, s_test_base2, TEST_EVENT_BASE2_EV1, &loop, sizeof(&loop), portMAX_DELAY));
TEST_ASSERT_EQUAL(ESP_OK, esp_event_loop_run(loop, pdMS_TO_TICKS(10)));
TEST_ASSERT_EQUAL(ESP_OK, esp_event_post_to(loop, s_test_base1, TEST_EVENT_BASE1_EV2, NULL, 0, portMAX_DELAY));
TEST_ASSERT_EQUAL(ESP_OK, esp_event_loop_run(loop, pdMS_TO_TICKS(10)));
TEST_ASSERT_EQUAL(3, count);
TEST_ASSERT_EQUAL(ESP_OK, esp_event_loop_delete(loop));
TEST_TEARDOWN();
}
static void test_create_loop_handler(void* handler_args, esp_event_base_t base, int32_t id, void* event_data)
{
esp_event_loop_args_t loop_args = test_event_get_default_loop_args();
if (id == TEST_EVENT_BASE1_EV1) {
TEST_ASSERT_EQUAL(ESP_OK, esp_event_loop_create(&loop_args, (esp_event_loop_handle_t*) handler_args));
} else {
TEST_ASSERT_EQUAL(ESP_OK, esp_event_loop_delete(*((esp_event_loop_handle_t*) handler_args)));
}
}
TEST_CASE("can create and delete loop from handler", "[event]")
{
TEST_SETUP();
esp_event_loop_handle_t loop;
esp_event_loop_handle_t test_loop;
esp_event_loop_args_t loop_args = test_event_get_default_loop_args();
loop_args.task_name = NULL;
TEST_ASSERT_EQUAL(ESP_OK, esp_event_loop_create(&loop_args, &loop));
TEST_ASSERT_EQUAL(ESP_OK, esp_event_handler_register_with(loop, s_test_base1, TEST_EVENT_BASE1_EV1, test_create_loop_handler, &test_loop));
TEST_ASSERT_EQUAL(ESP_OK, esp_event_handler_register_with(loop, s_test_base1, TEST_EVENT_BASE1_EV2, test_create_loop_handler, &test_loop));
TEST_ASSERT_EQUAL(ESP_OK, esp_event_post_to(loop, s_test_base1, TEST_EVENT_BASE1_EV1, NULL, 0, portMAX_DELAY));
TEST_ASSERT_EQUAL(ESP_OK, esp_event_loop_run(loop, pdMS_TO_TICKS(10)));
TEST_ASSERT_EQUAL(ESP_OK, esp_event_post_to(loop, s_test_base1, TEST_EVENT_BASE1_EV2, NULL, 0, portMAX_DELAY));
TEST_ASSERT_EQUAL(ESP_OK, esp_event_loop_run(loop, pdMS_TO_TICKS(10)));
TEST_ASSERT_EQUAL(ESP_OK, esp_event_loop_delete(loop));
TEST_TEARDOWN();
}
TEST_CASE("events are dispatched in the order they are registered", "[event]")
{
TEST_SETUP();
esp_event_loop_handle_t loop;
esp_event_loop_args_t loop_args = test_event_get_default_loop_args();
loop_args.task_name = NULL;
TEST_ASSERT_EQUAL(ESP_OK, esp_event_loop_create(&loop_args, &loop));
int id_arr[7];
for (int i = 0; i < 7; i++) {
id_arr[i] = i;
}
int data_arr[12] = {0};
TEST_ASSERT_EQUAL(ESP_OK, esp_event_handler_register_with(loop, s_test_base2, TEST_EVENT_BASE2_EV1, test_event_ordered_dispatch, id_arr + 0));
TEST_ASSERT_EQUAL(ESP_OK, esp_event_handler_register_with(loop, ESP_EVENT_ANY_BASE, ESP_EVENT_ANY_ID, test_event_ordered_dispatch, id_arr + 1));
TEST_ASSERT_EQUAL(ESP_OK, esp_event_handler_register_with(loop, s_test_base1, ESP_EVENT_ANY_ID, test_event_ordered_dispatch, id_arr + 2));
TEST_ASSERT_EQUAL(ESP_OK, esp_event_handler_register_with(loop, s_test_base2, TEST_EVENT_BASE2_EV2, test_event_ordered_dispatch, id_arr + 3));
TEST_ASSERT_EQUAL(ESP_OK, esp_event_handler_register_with(loop, s_test_base1, TEST_EVENT_BASE1_EV1, test_event_ordered_dispatch, id_arr + 4));
TEST_ASSERT_EQUAL(ESP_OK, esp_event_handler_register_with(loop, s_test_base2, ESP_EVENT_ANY_ID, test_event_ordered_dispatch, id_arr + 5));
TEST_ASSERT_EQUAL(ESP_OK, esp_event_handler_register_with(loop, s_test_base1, TEST_EVENT_BASE1_EV2, test_event_ordered_dispatch, id_arr + 6));
esp_event_dump(stdout);
ordered_data_t data = {
.arr = data_arr,
.index = 0
};
ordered_data_t* dptr = &data;
TEST_ASSERT_EQUAL(ESP_OK, esp_event_post_to(loop, s_test_base2, TEST_EVENT_BASE1_EV2, &dptr, sizeof(dptr), portMAX_DELAY));
TEST_ASSERT_EQUAL(ESP_OK, esp_event_post_to(loop, s_test_base1, TEST_EVENT_BASE1_EV1, &dptr, sizeof(dptr), portMAX_DELAY));
TEST_ASSERT_EQUAL(ESP_OK, esp_event_post_to(loop, s_test_base1, TEST_EVENT_BASE1_EV2, &dptr, sizeof(dptr), portMAX_DELAY));
TEST_ASSERT_EQUAL(ESP_OK, esp_event_post_to(loop, s_test_base2, TEST_EVENT_BASE1_EV1, &dptr, sizeof(dptr), portMAX_DELAY));
TEST_ASSERT_EQUAL(ESP_OK, esp_event_loop_run(loop, pdMS_TO_TICKS(10)));
// Expected data executing the posts above
int ref_arr[12] = {1, 3, 5, 1, 2, 4, 1, 2, 6, 0, 1, 5};
for (int i = 0; i < 12; i++) {
TEST_ASSERT_EQUAL(ref_arr[i], data_arr[i]);
}
TEST_ASSERT_EQUAL(ESP_OK, esp_event_loop_delete(loop));
TEST_TEARDOWN();
}
#if CONFIG_ESP_EVENT_POST_FROM_ISR
TEST_CASE("can properly prepare event data posted to loop", "[event]")
{
TEST_SETUP();
esp_event_loop_handle_t loop;
esp_event_loop_args_t loop_args = test_event_get_default_loop_args();
loop_args.task_name = NULL;
TEST_ASSERT_EQUAL(ESP_OK, esp_event_loop_create(&loop_args, &loop));
esp_event_post_instance_t post;
esp_event_loop_instance_t* loop_def = (esp_event_loop_instance_t*) loop;
TEST_ASSERT_EQUAL(ESP_OK, esp_event_post_to(loop, s_test_base1, TEST_EVENT_BASE1_EV1, NULL, 0, portMAX_DELAY));
TEST_ASSERT_EQUAL(pdTRUE, xQueueReceive(loop_def->queue, &post, portMAX_DELAY));
TEST_ASSERT_EQUAL(false, post.data_set);
TEST_ASSERT_EQUAL(false, post.data_allocated);
TEST_ASSERT_EQUAL(NULL, post.data.ptr);
int sample = 0;
TEST_ASSERT_EQUAL(ESP_OK, esp_event_isr_post_to(loop, s_test_base1, TEST_EVENT_BASE1_EV1, &sample, sizeof(sample), NULL));
TEST_ASSERT_EQUAL(pdTRUE, xQueueReceive(loop_def->queue, &post, portMAX_DELAY));
TEST_ASSERT_EQUAL(true, post.data_set);
TEST_ASSERT_EQUAL(false, post.data_allocated);
TEST_ASSERT_EQUAL(false, post.data.val);
TEST_ASSERT_EQUAL(ESP_OK, esp_event_loop_delete(loop));
TEST_TEARDOWN();
}
TEST_CASE("can post events from interrupt handler", "[event]")
{
SemaphoreHandle_t sem = xSemaphoreCreateBinary();
/* Select and initialize basic parameters of the timer */
timer_config_t config;
config.divider = TIMER_DIVIDER;
config.counter_dir = TIMER_COUNT_UP;
config.counter_en = TIMER_PAUSE;
config.alarm_en = TIMER_ALARM_EN;
config.intr_type = TIMER_INTR_LEVEL;
config.auto_reload = false;
timer_init(TIMER_GROUP_0, TIMER_0, &config);
/* Timer's counter will initially start from value below.
Also, if auto_reload is set, this value will be automatically reload on alarm */
timer_set_counter_value(TIMER_GROUP_0, TIMER_0, 0x00000000ULL);
/* Configure the alarm value and the interrupt on alarm. */
timer_set_alarm_value(TIMER_GROUP_0, TIMER_0, TIMER_INTERVAL0_SEC * TIMER_SCALE);
timer_enable_intr(TIMER_GROUP_0, TIMER_0);
timer_isr_register(TIMER_GROUP_0, TIMER_0, test_event_on_timer_alarm,
(void *) sem, ESP_INTR_FLAG_IRAM, NULL);
timer_start(TIMER_GROUP_0, TIMER_0);
TEST_SETUP();
TEST_ASSERT_EQUAL(ESP_OK, esp_event_handler_register(s_test_base1, TEST_EVENT_BASE1_EV1,
test_handler_post_from_isr, &sem));
xSemaphoreTake(sem, portMAX_DELAY);
TEST_TEARDOWN();
}
#endif // CONFIG_ESP_EVENT_POST_FROM_ISR