#ifndef USE_PTHREAD

#include "thread.h"
#include "debug.h"
#include "pthread.h"
#include <bits/pthreadtypes.h>
#include <stdlib.h>
#include <string.h>
#include <sys/queue.h>
#include <ucontext.h>
#include <valgrind/valgrind.h>

#define FINISHED 0x1
#define IS_FINISHED(entry) (entry->status & FINISHED)
#define ALLOCATED 0x2
#define WAS_ALLOCATED(entry) (entry->status & ALLOCATED)
#define WAITING 0x4
#define IS_WAITING(entry) (entry->status & WAITING)
#define GET_WAITING_THREAD(entry) ((struct context_entry*)entry->waiting)
#define IS_WAITING_THREAD_FINISHED(entry) (GET_WAITING_THREAD(entry)->status & FINISHED)
#define WAITED 0x8
#define IS_WAITED(entry) (entry->status & WAITED)

#ifndef STACK_SIZE
#define STACK_SIZE 4096
#endif

// Variables used to clean up everything at the end of the processus
static char stack_for_freeing[STACK_SIZE] = {0};
static int stack_valgrind_id = 0;
static ucontext_t context_for_freeing;

struct context_entry {
    TAILQ_ENTRY(context_entry)
    link; // Use to navigate inside the list
    ucontext_t context;
    thread_t id;
    void *waiting;
    void* retvalue;
    int valgrind_id;
    char status;
    char stack[STACK_SIZE];
};

// Use TailQ from queue BSD
static TAILQ_HEAD(context_head, context_entry) head = TAILQ_HEAD_INITIALIZER(head);
// Current running thread
static struct context_entry* running = NULL;

int thread_yield(void)
{
    TRACE("thread_yield");
    if (TAILQ_EMPTY(&head)) {
        return 0;
    }

    /* Current strategy :
     * if we have checked the number of threads then keep the running one
     * otherwise, take the first element of the list should not be null
     * remove it from the head and put it at the end to take it in the next round
     * check if the thread is not finished and is not waiting for a non finished thread
     * check if the thread is not the running one.
     */
    struct context_entry* first = TAILQ_FIRST(&head);
    if (!first) {
        return -1;
    }
    TAILQ_REMOVE(&head, first, link);
    if (!IS_FINISHED(running) && !(IS_WAITING(running) && !IS_WAITING_THREAD_FINISHED(running))) {
        TAILQ_INSERT_TAIL(&head, running, link);
    }
    TRACE("PICKING %p (previous was %p)", first->id, running->id);
    // Switch to the new thread.
    struct context_entry* old_runner = running;
    running = first;
    swapcontext(&old_runner->context, &running->context);
    return 0;
}

thread_t thread_self(void)
{
    // This condition should not be true at any moment after main call
    if (running == NULL) {
        return 0;
    }
    return running->id;
}

/**
 * Wrap the function used by the thread to handle `return` statement
 * without using thread_exit.
 */
void thread_function_wrapper(void* (*func)(void*), void* funcarg)
{
    TRACE("Wrapper for %p\n", func);
    thread_exit(func(funcarg));
}

/**
 * Create an entry and put it at the end of the FIFO
 */
int thread_create(thread_t* newthread, void* (*func)(void*), void* funcarg)
{
    TRACE("Create a new thread that execute function %p", func);
    struct context_entry* new_entry = malloc(sizeof(*new_entry));
    memset(new_entry->stack, 0, STACK_SIZE);

    getcontext(&new_entry->context);

    new_entry->context.uc_stack.ss_sp = new_entry->stack;
    new_entry->context.uc_stack.ss_size = STACK_SIZE;
    new_entry->context.uc_stack.ss_flags = 0;

    // Tell Valgrind that the memory area of the future stack is a stack
    new_entry->valgrind_id = VALGRIND_STACK_REGISTER(
        new_entry->context.uc_stack.ss_sp,
        new_entry->context.uc_stack.ss_sp + new_entry->context.uc_stack.ss_size);

    // Use the entry's memory address as an id.
    new_entry->id = new_entry;
    TRACE("ALLOCATED %p", new_entry);
    new_entry->status = ALLOCATED;
    new_entry->retvalue = NULL;

    *newthread = new_entry->id;

    makecontext(&new_entry->context, (void (*)(void))thread_function_wrapper, 2, func, funcarg);

    TAILQ_INSERT_TAIL(&head, new_entry, link);
    return 0;
}

void print_entry(struct context_entry* entry)
{
    TRACE("CONTEXT (%p, %p, %d);", entry, entry->id, WAS_ALLOCATED(entry));
}

int thread_join(thread_t thread, void** retval)
{
    TRACE("Join thread %p", thread);
    struct context_entry* entry = thread;
    // Check if the target is not already waited by another
    if (IS_WAITED(entry) || IS_WAITING(entry) && GET_WAITING_THREAD(entry) == running) {
        return -1;
    }

    if (!IS_FINISHED(entry)) {
        // Use status to be in waiting state
        running->status |= WAITING;
        // Use retvalue to share which thread we are currently waiting for
        running->waiting = entry;
        // Mark the waited thread as waited to not be waited by any other thread.
        entry->status |= WAITED;
        entry->retvalue = running;
        do {
            thread_yield();
        } while (!IS_FINISHED(entry));
        // Exit from waiting state
        running->status &= ~WAITING;
    }

    // Save returned value if needed
    TRACE("RETURNING %p IN %p", entry->retvalue, retval);
    if (retval)
        *retval = entry->retvalue;

    // Clean up
    DBG("(entry, was_alloacted) : %p,%d", entry, WAS_ALLOCATED(entry));
    if (WAS_ALLOCATED(entry)) {
        VALGRIND_STACK_DEREGISTER(entry->valgrind_id);
    }
    free(entry);

    return 0;
}

void thread_exit(void* retval)
{
    TRACE("Exit thread %p", running);
    print_entry(running);
    if (running == NULL) {
        exit(0);
    }

    running->status |= FINISHED;
    if (IS_WAITED(running)) {
        // If the thread was waited by another thread, we need to wake it up.
        struct context_entry* waited = running->retvalue;
        TAILQ_INSERT_TAIL(&head, waited, link);
    }
    running->retvalue = retval;

    if (!TAILQ_EMPTY(&head)) {
        thread_yield();
    }
    exit(0);
}

void clear_context(void)
{
    TRACE("INSIDE CLEAR");
    struct context_entry* last = NULL;
    // Loop over remaining threads to clean them from the heap.
    while (!TAILQ_EMPTY(&head)) {
        last = TAILQ_FIRST(&head);
        TAILQ_REMOVE(&head, last, link);
        if (WAS_ALLOCATED(last)) {
            VALGRIND_STACK_DEREGISTER(last->valgrind_id);
        }
        if (IS_WAITED(last)) {
            struct context_entry* waited = last->retvalue;
            TAILQ_INSERT_TAIL(&head, waited, link);
        }
        free(last);
    }
    VALGRIND_STACK_DEREGISTER(stack_valgrind_id);
    exit(0);
}

void __attribute__((constructor)) setup_main_thread()
{
    TRACE("premain");
    // Create an entry for the main thread.
    struct context_entry* main = malloc(sizeof(*main));
    // memset(main, 0, sizeof(*main));
    getcontext(&main->context);
    main->id = main;
    main->status = 0;
    main->retvalue = NULL;
    running = main;

    // Create a context with static stack to clean everything at the end.
    getcontext(&context_for_freeing);
    stack_valgrind_id = VALGRIND_STACK_REGISTER(stack_for_freeing, stack_for_freeing + STACK_SIZE);
    context_for_freeing.uc_stack.ss_sp = stack_for_freeing;
    context_for_freeing.uc_stack.ss_size = STACK_SIZE;
    makecontext(&context_for_freeing, (void (*)(void)) clear_context, 0);
}

void __attribute__((destructor)) clear_last_thread()
{
    TRACE("POST");
    // Running is the initial main thread. No need to switch to a static stack.
    TAILQ_INSERT_HEAD(&head, running, link);
    if (!WAS_ALLOCATED(running)) {
        clear_context();
        exit(0);
    }

    // Running's stack was allocated by us, lets switch to a static stack first.
    swapcontext(&running->context, &context_for_freeing);
    exit(0);
}

int thread_mutex_init(thread_mutex_t* mutex)
{
    return pthread_mutex_init((pthread_mutex_t*)mutex, NULL);
}

int thread_mutex_destroy(thread_mutex_t* mutex)
{
    return pthread_mutex_destroy((pthread_mutex_t*)mutex);
}

int thread_mutex_lock(thread_mutex_t* mutex)
{
    return pthread_mutex_lock((pthread_mutex_t*)mutex);
}

int thread_mutex_unlock(thread_mutex_t* mutex)
{
    return pthread_mutex_unlock((pthread_mutex_t*)mutex);
}

#endif