#pragma once

#include <locks.hfa>
#include <list.hfa>
#include <mutex_stmt.hfa>

// link field used for threads waiting on channel
struct wait_link {
    // used to put wait_link on a dl queue
    inline dlink(wait_link);

    // waiting thread
    struct thread$ * t;

    // shadow field
    void * elem;
};
P9_EMBEDDED( wait_link, dlink(wait_link) )

static inline void ?{}( wait_link & this, thread$ * t, void * elem ) {
    this.t = t;
    this.elem = elem;
}

// wake one thread from the list
static inline void wake_one( dlist( wait_link ) & queue ) {
    wait_link & popped = try_pop_front( queue );
    unpark( popped.t );
}

// returns true if woken due to shutdown
// blocks thread on list and releases passed lock
static inline bool block( dlist( wait_link ) & queue, void * elem_ptr, go_mutex & lock ) {
    wait_link w{ active_thread(), elem_ptr };
    insert_last( queue, w );
    unlock( lock );
    park();
    return w.elem == 0p;
}

// void * used for some fields since exceptions don't work with parametric polymorphism currently
exception channel_closed {
    // on failed insert elem is a ptr to the element attempting to be inserted
    // on failed remove elem ptr is 0p
    // on resumption of a failed insert this elem will be inserted
    // so a user may modify it in the resumption handler
    void * elem;

    // pointer to chan that is closed
    void * closed_chan;
};
vtable(channel_closed) channel_closed_vt;

// #define CHAN_STATS // define this to get channel stats printed in dtor

forall( T ) {

struct __attribute__((aligned(128))) channel {
    size_t size, front, back, count;
    T * buffer;
    dlist( wait_link ) prods, cons; // lists of blocked threads
    go_mutex mutex_lock;            // MX lock
    bool closed;                    // indicates channel close/open
    #ifdef CHAN_STATS
    size_t blocks, operations;      // counts total ops and ops resulting in a blocked thd
    #endif
};

static inline void ?{}( channel(T) &c, size_t _size ) with(c) {
    size = _size;
    front = back = count = 0;
    buffer = aalloc( size );
    prods{};
    cons{};
    mutex_lock{};
    closed = false;
    #ifdef CHAN_STATS
    blocks = 0;
    operations = 0;
    #endif
}

static inline void ?{}( channel(T) &c ){ ((channel(T) &)c){ 0 }; }
static inline void ^?{}( channel(T) &c ) with(c) {
    #ifdef CHAN_STATS
    printf("Channel %p Blocks: %lu, Operations: %lu, %.2f%% of ops blocked\n", &c, blocks, operations, ((double)blocks)/operations * 100);
    #endif
    verifyf( cons`isEmpty && prods`isEmpty, "Attempted to delete channel with waiting threads (Deadlock).\n" );
    delete( buffer );
}
static inline size_t get_count( channel(T) & chan ) with(chan) { return count; }
static inline size_t get_size( channel(T) & chan ) with(chan) { return size; }
static inline bool has_waiters( channel(T) & chan ) with(chan) { return !cons`isEmpty || !prods`isEmpty; }
static inline bool has_waiting_consumers( channel(T) & chan ) with(chan) { return !cons`isEmpty; }
static inline bool has_waiting_producers( channel(T) & chan ) with(chan) { return !prods`isEmpty; }

// closes the channel and notifies all blocked threads
static inline void close( channel(T) & chan ) with(chan) {
    lock( mutex_lock );
    closed = true;

    // flush waiting consumers and producers
    while ( has_waiting_consumers( chan ) ) {
        cons`first.elem = 0p;
        wake_one( cons );
    }
    while ( has_waiting_producers( chan ) ) {
        prods`first.elem = 0p;
        wake_one( prods );
    }
    unlock(mutex_lock);
}

static inline void is_closed( channel(T) & chan ) with(chan) { return closed; }

static inline void flush( channel(T) & chan, T elem ) with(chan) {
    lock( mutex_lock );
    while ( count == 0 && !cons`isEmpty ) {
        memcpy(cons`first.elem, (void *)&elem, sizeof(T)); // do waiting consumer work
        wake_one( cons );
    }
    unlock( mutex_lock );
}

// handles buffer insert
static inline void __buf_insert( channel(T) & chan, T & elem ) with(chan) {
    memcpy((void *)&buffer[back], (void *)&elem, sizeof(T));
    count += 1;
    back++;
    if ( back == size ) back = 0;
}

// does the buffer insert or hands elem directly to consumer if one is waiting
static inline void __do_insert( channel(T) & chan, T & elem ) with(chan) {
    if ( count == 0 && !cons`isEmpty ) {
        memcpy(cons`first.elem, (void *)&elem, sizeof(T)); // do waiting consumer work
        wake_one( cons );
    } else __buf_insert( chan, elem );
}

// needed to avoid an extra copy in closed case
static inline bool __internal_try_insert( channel(T) & chan, T & elem ) with(chan) {
    lock( mutex_lock );
    #ifdef CHAN_STATS
    operations++;
    #endif
    if ( count == size ) { unlock( mutex_lock ); return false; }
    __do_insert( chan, elem );
    unlock( mutex_lock );
    return true;
}

// attempts a nonblocking insert
// returns true if insert was successful, false otherwise
static inline bool try_insert( channel(T) & chan, T elem ) { return __internal_try_insert( chan, elem ); }

// handles closed case of insert routine
static inline void __closed_insert( channel(T) & chan, T & elem ) with(chan) {
    channel_closed except{&channel_closed_vt, &elem, &chan };
    throwResume except; // throw closed resumption
    if ( !__internal_try_insert( chan, elem ) ) throw except; // if try to insert fails (would block), throw termination
}

static inline void insert( channel(T) & chan, T elem ) with(chan) {
    // check for close before acquire mx
    if ( unlikely(closed) ) {
        __closed_insert( chan, elem );
        return;
    } 

    lock( mutex_lock );

    #ifdef CHAN_STATS
    if ( !closed ) operations++;
    #endif

    // if closed handle
    if ( unlikely(closed) ) {
        unlock( mutex_lock );
        __closed_insert( chan, elem );
        return;
    }

    // have to check for the zero size channel case
    if ( size == 0 && !cons`isEmpty ) {
        memcpy(cons`first.elem, (void *)&elem, sizeof(T));
        wake_one( cons );
        unlock( mutex_lock );
        return true;
    }

    // wait if buffer is full, work will be completed by someone else
    if ( count == size ) {
        #ifdef CHAN_STATS
        blocks++;
        #endif

        // check for if woken due to close
        if ( unlikely( block( prods, &elem, mutex_lock ) ) )
            __closed_insert( chan, elem );
        return;
    } // if

    if ( count == 0 && !cons`isEmpty ) {
        memcpy(cons`first.elem, (void *)&elem, sizeof(T)); // do waiting consumer work
        wake_one( cons );
    } else __buf_insert( chan, elem );
    
    unlock( mutex_lock );
    return;
}

// handles buffer remove
static inline void __buf_remove( channel(T) & chan, T & retval ) with(chan) {
    memcpy((void *)&retval, (void *)&buffer[front], sizeof(T));
    count -= 1;
    front = (front + 1) % size;
}

// does the buffer remove and potentially does waiting producer work
static inline void __do_remove( channel(T) & chan, T & retval ) with(chan) {
    __buf_remove( chan, retval );
    if (count == size - 1 && !prods`isEmpty ) {
        __buf_insert( chan, *(T *)prods`first.elem );  // do waiting producer work
        wake_one( prods );
    }
}

// needed to avoid an extra copy in closed case and single return val case
static inline bool __internal_try_remove( channel(T) & chan, T & retval ) with(chan) {
    lock( mutex_lock );
    #ifdef CHAN_STATS
    operations++;
    #endif
    if ( count == 0 ) { unlock( mutex_lock ); return false; }
    __do_remove( chan, retval );
    unlock( mutex_lock );
    return true;
}

// attempts a nonblocking remove
// returns [T, true] if insert was successful
// returns [T, false] if insert was successful (T uninit)
static inline [T, bool] try_remove( channel(T) & chan ) {
    T retval;
    return [ retval, __internal_try_remove( chan, retval ) ];
}

static inline T try_remove( channel(T) & chan, T elem ) {
    T retval;
    __internal_try_remove( chan, retval );
    return retval;
}

// handles closed case of insert routine
static inline void __closed_remove( channel(T) & chan, T & retval ) with(chan) {
    channel_closed except{&channel_closed_vt, 0p, &chan };
    throwResume except; // throw resumption
    if ( !__internal_try_remove( chan, retval ) ) throw except; // if try to remove fails (would block), throw termination
}

static inline T remove( channel(T) & chan ) with(chan) {
    T retval;
    if ( unlikely(closed) ) {
        __closed_remove( chan, retval );
        return retval;
    } 
    lock( mutex_lock );

    #ifdef CHAN_STATS
    if ( !closed ) operations++;
    #endif

    if ( unlikely(closed) ) {
        unlock( mutex_lock );
        __closed_remove( chan, retval );
        return retval;
    } 

    // have to check for the zero size channel case
    if ( size == 0 && !prods`isEmpty ) {
        memcpy((void *)&retval, (void *)prods`first.elem, sizeof(T));
        wake_one( prods );
        unlock( mutex_lock );
        return retval;
    }

    // wait if buffer is empty, work will be completed by someone else
    if (count == 0) {
        #ifdef CHAN_STATS
        blocks++;
        #endif
        // check for if woken due to close
        if ( unlikely( block( cons, &retval, mutex_lock ) ) )
            __closed_remove( chan, retval );
        return retval;
    }

    // Remove from buffer
    __do_remove( chan, retval );

    unlock( mutex_lock );
    return retval;
}
} // forall( T )