-- |
--   Copyright   :  (c) Sam Truzjan 2013
--   License     :  BSD3
--   Maintainer  :  pxqr.sta@gmail.com
--   Stability   :  experimental
--   Portability :  portable
--
--   Every node maintains a routing table of known good nodes. The
--   nodes in the routing table are used as starting points for
--   queries in the DHT. Nodes from the routing table are returned in
--   response to queries from other nodes.
--
--   For more info see:
--   <http://www.bittorrent.org/beps/bep_0005.html#routing-table>
--
{-# LANGUAGE RecordWildCards #-}
{-# LANGUAGE ViewPatterns    #-}
{-# LANGUAGE TypeOperators   #-}
{-# LANGUAGE DeriveGeneric   #-}
{-# LANGUAGE ScopedTypeVariables #-}
{-# OPTIONS_GHC -fno-warn-orphans #-}
module Network.BitTorrent.DHT.Routing
       ( -- * Table
         Table
       , Info(..)

         -- * Attributes
       , BucketCount
       , defaultBucketCount
       , BucketSize
       , defaultBucketSize
       , NodeCount

         -- * Query
       , Network.BitTorrent.DHT.Routing.null
       , Network.BitTorrent.DHT.Routing.full
       , thisId
       , shape
       , Network.BitTorrent.DHT.Routing.size
       , Network.BitTorrent.DHT.Routing.depth
       , compatibleNodeId

         -- * Lookup
       , K
       , defaultK
       , TableKey (..)
       , kclosest

         -- * Construction
       , Network.BitTorrent.DHT.Routing.nullTable
       , Event(..)
       , CheckPing(..)
       , Network.BitTorrent.DHT.Routing.insert

         -- * Conversion
       , Network.BitTorrent.DHT.Routing.TableEntry
       , Network.BitTorrent.DHT.Routing.toList

         -- * Routing
       , Timestamp
       , Routing
       , runRouting
       ) where

import Control.Applicative as A
import Control.Arrow
import Control.Monad
import Data.Function
import Data.Functor.Identity
import Data.List as L hiding (insert)
import Data.Maybe
import Data.Monoid
import Data.PSQueue as PSQ
import Data.Serialize as S hiding (Result, Done)
import qualified Data.Sequence as Seq
import Data.Time
import Data.Time.Clock.POSIX
import Data.Word
import GHC.Generics
import Text.PrettyPrint as PP hiding ((<>))
import Text.PrettyPrint.HughesPJClass (pPrint,Pretty)
import qualified Data.ByteString as BS
import Data.Bits

import Data.Torrent
import Network.BitTorrent.Address

{-----------------------------------------------------------------------
--  Routing monad
-----------------------------------------------------------------------}

-- | Last time the node was responding to our queries.
--
--   Not all nodes that we learn about are equal. Some are \"good\" and
--   some are not. Many nodes using the DHT are able to send queries
--   and receive responses, but are not able to respond to queries
--   from other nodes. It is important that each node's routing table
--   must contain only known good nodes. A good node is a node has
--   responded to one of our queries within the last 15 minutes. A
--   node is also good if it has ever responded to one of our queries
--   and has sent us a query within the last 15 minutes. After 15
--   minutes of inactivity, a node becomes questionable. Nodes become
--   bad when they fail to respond to multiple queries in a row. Nodes
--   that we know are good are given priority over nodes with unknown
--   status.
--
type Timestamp = POSIXTime

-- | Some routing operations might need to perform additional IO.
data Routing ip result
  = Full
  | Done      result
  | GetTime                ( Timestamp    -> Routing ip result)
  | NeedPing (NodeAddr ip) ( Bool         -> Routing ip result)
  | Refresh   NodeId       (Routing ip result)

instance Functor (Routing ip) where
  fmap _  Full             = Full
  fmap f (Done          r) = Done          (     f   r)
  fmap f (GetTime       g) = GetTime       (fmap f . g)
  fmap f (NeedPing addr g) = NeedPing addr (fmap f . g)
  fmap f (Refresh  nid  g) = Refresh  nid  (fmap f g)

instance Monad (Routing ip) where
  return = Done

  Full         >>= _ = Full
  Done       r >>= m = m r
  GetTime    f >>= m = GetTime     $ \ t -> f t >>= m
  NeedPing a f >>= m = NeedPing a  $ \ p -> f p >>= m
  Refresh  n f >>= m = Refresh  n  $ f >>= m

instance Applicative (Routing ip) where
  pure  = return
  (<*>) = ap

instance Alternative (Routing ip) where
  empty = Full

  Full         <|> m = m
  Done     a   <|> _ = Done a
  GetTime  f   <|> m = GetTime    $ \ t -> f t <|> m
  NeedPing a f <|> m = NeedPing a $ \ p -> f p <|> m
  Refresh  n f <|> m = Refresh  n (f <|> m)

-- | Run routing table operation.
runRouting :: Monad m
           => (NodeAddr ip -> m Bool) -- ^ ping the specific node;
           -> (NodeId      -> m ())   -- ^ refresh nodes;
           -> m Timestamp             -- ^ get current time;
           -> Routing ip f            -- ^ operation to run;
           -> m (Maybe f)             -- ^ operation result;
runRouting ping_node find_nodes timestamper = go
  where
    go  Full             = return (Nothing)
    go (Done          r) = return (Just  r)
    go (GetTime       f) = do
      t <- timestamper
      go (f t)

    go (NeedPing addr f) = do
      pong <- ping_node addr
      go (f pong)

    go (Refresh nid f) = do
      find_nodes nid
      go f

{-----------------------------------------------------------------------
    Bucket
-----------------------------------------------------------------------}
-- TODO: add replacement cache to the bucket
--
-- When a k-bucket is full and a new node is discovered for that
-- k-bucket, the least recently seen node in the k-bucket is
-- PINGed. If the node is found to be still alive, the new node is
-- place in a secondary list, a replacement cache. The replacement
-- cache is used only if a node in the k-bucket stops responding. In
-- other words: new nodes are used only when older nodes disappear.

-- | Timestamp - last time this node is pinged.
type NodeEntry ip = Binding (NodeInfo ip) Timestamp

instance (Serialize k, Serialize v) => Serialize (Binding k v) where
  get = (:->) <$> get <*> get
  put (k :-> v) = put k >> put v

-- TODO instance Pretty where

-- | Number of nodes in a bucket.
type BucketSize  = Int

-- | Maximum number of 'NodeInfo's stored in a bucket. Most clients
-- use this value.
defaultBucketSize :: BucketSize
defaultBucketSize = 8

data QueueMethods m elem fifo = QueueMethods
    { pushBack :: elem -> fifo -> m fifo
    , popFront :: fifo -> m (Maybe elem, fifo)
    , emptyQueue :: m fifo
    }

{-
fromQ :: Functor m =>
       ( a -> b )
       -> ( b -> a )
       -> QueueMethods m elem a
       -> QueueMethods m elem b
fromQ embed project QueueMethods{..} =
    QueueMethods { pushBack = \e -> fmap embed . pushBack e . project
                 , popFront = fmap (second embed) . popFront . project
                 , emptyQueue = fmap embed emptyQueue
                 }
-}

seqQ :: QueueMethods Identity (NodeInfo ip) (Seq.Seq (NodeInfo ip))
seqQ = QueueMethods
    { pushBack = \e fifo -> pure (fifo Seq.|> e)
    , popFront = \fifo -> case Seq.viewl fifo of
                            e Seq.:< fifo' -> pure (Just e, fifo')
                            Seq.EmptyL     -> pure (Nothing, Seq.empty)
    , emptyQueue = pure Seq.empty
    }

type BucketQueue ip = Seq.Seq (NodeInfo ip)

bucketQ :: QueueMethods Identity (NodeInfo ip) (BucketQueue ip)
bucketQ = seqQ

-- | Bucket is also limited in its length — thus it's called k-bucket.
--   When bucket becomes full, we should split it in two lists by
--   current span bit. Span bit is defined by depth in the routing
--   table tree. Size of the bucket should be choosen such that it's
--   very unlikely that all nodes in bucket fail within an hour of
--   each other.
--
data Bucket ip = Bucket { bktNodes :: PSQ (NodeInfo ip) Timestamp
                        , bktQ     :: BucketQueue ip
                        } deriving (Show,Generic)

instance (Eq ip, Serialize ip) => Serialize (Bucket ip)

instance (Serialize k, Serialize v, Ord k, Ord v)
       => Serialize (PSQ k v) where
  get = PSQ.fromList <$> get
  put = put . PSQ.toList

-- | Update interval, in seconds.
delta :: NominalDiffTime
delta = 15 * 60

-- | Should maintain a set of stable long running nodes.
--
-- Note: pings are triggerd only when a bucket is full.
insertBucket :: (Eq ip, Alternative f)  => Timestamp -> Event ip -> Bucket ip
           -> f ([CheckPing ip], Bucket ip)
insertBucket curTime (TryInsert info) bucket
  -- just update timestamp if a node is already in bucket
  | already_have
    = pure ( [], map_ns $ PSQ.insertWith max info curTime )
  -- bucket is good, but not full => we can insert a new node
  | PSQ.size (bktNodes bucket) < defaultBucketSize
    = pure ( [], map_ns $ PSQ.insert info curTime )
  -- If there are any questionable nodes in the bucket have not been
  -- seen in the last 15 minutes, the least recently seen node is
  -- pinged. If any nodes in the bucket are known to have become bad,
  -- then one is replaced by the new node in the next insertBucket
  -- iteration.
  | not (L.null stales)
    = pure ( [CheckPing stales],  map_q $ pushBack bucketQ info )
  -- When the bucket is full of good nodes, the new node is simply discarded.
  -- We must return 'A.empty' here to ensure that bucket splitting happens
  -- inside 'modifyBucket'.
  | otherwise = A.empty
 where
    stales = map key $ PSQ.atMost (curTime - delta) $ bktNodes bucket

    already_have = maybe False (const True) $ PSQ.lookup info (bktNodes bucket)

    map_ns f = bucket { bktNodes = f (bktNodes bucket) }
    map_q f = bucket { bktQ = runIdentity $ f (bktQ bucket) }

insertBucket curTime (PingResult bad_node got_response) bucket
    = pure ([], Bucket (upd $ bktNodes bucket) popped)
 where
    (top, popped) = runIdentity $ popFront bucketQ (bktQ bucket)
    upd | got_response     = id
        | Just info <- top = PSQ.insert info curTime . PSQ.delete bad_node
        | otherwise        = id

type BitIx = Word

partitionQ :: Monad f => QueueMethods f elem b -> (elem -> Bool) -> b -> f (b, b)
partitionQ imp test q0 = do
    pass0 <- emptyQueue imp
    fail0 <- emptyQueue imp
    let flipfix a b f = fix f a b
    flipfix q0 (pass0,fail0) $ \rec q qs -> do
        (mb,q') <- popFront imp q
        case mb of
            Nothing -> return qs
            Just e  -> do qs' <- select (pushBack imp e) qs
                          rec q' qs'
                where
                    select :: Functor f => (b -> f b) -> (b, b) -> f (b, b)
                    select f = if test e then \(a,b) -> flip (,) b <$> f a
                                         else \(a,b) ->      (,) a <$> f b

split :: Eq ip => BitIx -> Bucket ip -> (Bucket ip, Bucket ip)
split i b = (Bucket ns qs, Bucket ms rs)
  where
    (ns,ms) = (PSQ.fromList *** PSQ.fromList) . partition (spanBit . key) . PSQ.toList $ bktNodes b
    (qs,rs) = runIdentity $ partitionQ bucketQ spanBit $ bktQ b
    spanBit entry = testIdBit (nodeId entry) i

{-----------------------------------------------------------------------
--  Table
-----------------------------------------------------------------------}

-- | Number of buckets in a routing table.
type BucketCount = Int

defaultBucketCount :: BucketCount
defaultBucketCount = 20

data Info ip = Info
    { myBuckets :: Table ip
    , myNodeId :: NodeId
    , myAddress :: ip
    }
 deriving (Eq, Show, Generic)

instance (Eq ip, Serialize ip) => Serialize (Info ip)

-- | The routing table covers the entire 'NodeId' space from 0 to 2 ^
-- 160. The routing table is subdivided into 'Bucket's that each cover
-- a portion of the space. An empty table has one bucket with an ID
-- space range of @min = 0, max = 2 ^ 160@. When a node with ID \"N\"
-- is inserted into the table, it is placed within the bucket that has
-- @min <= N < max@. An empty table has only one bucket so any node
-- must fit within it. Each bucket can only hold 'K' nodes, currently
-- eight, before becoming 'Full'. When a bucket is full of known good
-- nodes, no more nodes may be added unless our own 'NodeId' falls
-- within the range of the 'Bucket'. In that case, the bucket is
-- replaced by two new buckets each with half the range of the old
-- bucket and the nodes from the old bucket are distributed among the
-- two new ones. For a new table with only one bucket, the full bucket
-- is always split into two new buckets covering the ranges @0..2 ^
-- 159@ and @2 ^ 159..2 ^ 160@.
--
data Table ip
  -- most nearest bucket
  = Tip  NodeId BucketCount (Bucket ip)

  -- left biased tree branch
  | Zero (Table  ip) (Bucket ip)

  -- right biased tree branch
  | One  (Bucket ip) (Table  ip)
    deriving (Show, Generic)

instance Eq ip => Eq (Table ip) where
  (==) = (==) `on` Network.BitTorrent.DHT.Routing.toList

instance Serialize NominalDiffTime where
  put = putWord32be . fromIntegral   . fromEnum
  get = (toEnum     . fromIntegral) <$> getWord32be

-- | Normally, routing table should be saved between invocations of
-- the client software. Note that you don't need to store /this/
-- 'NodeId' since it is already included in routing table.
instance (Eq ip, Serialize ip) => Serialize (Table ip)

-- | Shape of the table.
instance Pretty (Table ip) where
  pPrint t
    | bucketCount < 6 = hcat $ punctuate ", " $ L.map PP.int ss
    |    otherwise    = brackets $
      PP.int (L.sum    ss) <> " nodes, " <>
      PP.int bucketCount   <> " buckets"
    where
      bucketCount = L.length ss
      ss = shape t

-- | Empty table with specified /spine/ node id.
nullTable :: Eq ip => NodeId -> BucketCount -> Table ip
nullTable nid n = Tip nid (bucketCount (pred n)) (Bucket PSQ.empty (runIdentity $ emptyQueue bucketQ))
  where
    bucketCount x = max 0 (min 159 x)

-- | Test if table is empty. In this case DHT should start
-- bootstrapping process until table becomes 'full'.
null :: Table ip -> Bool
null (Tip _ _ b) = PSQ.null $ bktNodes b
null  _          = False

-- | Test if table have maximum number of nodes. No more nodes can be
-- 'insert'ed, except old ones becomes bad.
full :: Table ip -> Bool
full (Tip  _ n _) = n == 0
full (Zero   t b) = PSQ.size (bktNodes b) == defaultBucketSize && full t
full (One    b t) = PSQ.size (bktNodes b) == defaultBucketSize && full t

-- | Get the /spine/ node id.
thisId :: Table ip -> NodeId
thisId (Tip  nid _ _) = nid
thisId (Zero table _) = thisId table
thisId (One _  table) = thisId table

-- | Number of nodes in a bucket or a table.
type NodeCount   = Int

-- | Internally, routing table is similar to list of buckets or a
-- /matrix/ of nodes. This function returns the shape of the matrix.
shape :: Table ip -> [BucketSize]
shape = map (PSQ.size . bktNodes) . toBucketList

-- | Get number of nodes in the table.
size :: Table ip -> NodeCount
size = L.sum . shape

-- | Get number of buckets in the table.
depth :: Table ip -> BucketCount
depth = L.length . shape

lookupBucket :: NodeId -> Table ip -> Maybe (Bucket ip)
lookupBucket nid = go 0
  where
    go i (Zero table bucket)
      |  testIdBit nid i  = pure bucket
      |     otherwise     = go (succ i) table
    go i (One  bucket table)
      |  testIdBit nid i  = go (succ i) table
      |     otherwise     = pure bucket
    go _ (Tip _ _ bucket) = pure bucket

compatibleNodeId :: Table ip -> IO NodeId
compatibleNodeId tbl = genBucketSample prefix br
 where
    br = bucketRange (L.length (shape tbl) - 1) True
    bs = BS.pack $ take nodeIdSize $ tablePrefix tbl ++ repeat 0
    prefix = NodeId bs

tablePrefix :: Table ip -> [Word8]
tablePrefix = map (packByte . take 8 . (++repeat False))
              . chunksOf 8
              . tableBits
 where
    packByte = foldl1' (.|.) . zipWith bitmask [7,6 .. 0]
    bitmask ix True = bit ix
    bitmask _  _    = 0

tableBits :: Table ip -> [Bool]
tableBits (One _ tbl)  = True : tableBits tbl
tableBits (Zero tbl _) = False : tableBits tbl
tableBits (Tip _ _ _)  = []

chunksOf :: Int -> [e] -> [[e]]
chunksOf i ls = map (take i) (build (splitter ls)) where
  splitter :: [e] -> ([e] -> a -> a) -> a -> a
  splitter [] _ n = n
  splitter l c n  = l `c` splitter (drop i l) c n

build :: ((a -> [a] -> [a]) -> [a] -> [a]) -> [a]
build g = g (:) []

-- | Count of closest nodes in find_node request.
type K = Int

-- | Default 'K' is equal to 'defaultBucketSize'.
defaultK :: K
defaultK = 8

class TableKey k where
  toNodeId :: k -> NodeId

instance TableKey NodeId where
  toNodeId = id

instance TableKey InfoHash where
  toNodeId = either (error msg) id . S.decode . S.encode
    where -- TODO unsafe coerse?
      msg = "tableKey: impossible"

-- | Get a list of /K/ closest nodes using XOR metric. Used in
-- 'find_node' and 'get_peers' queries.
kclosest :: Eq ip => TableKey a => K -> a -> Table ip -> [NodeInfo ip]
kclosest k (toNodeId -> nid)
  = L.take k . rank nodeId nid
  . L.map PSQ.key . PSQ.toList . fromMaybe PSQ.empty
  . fmap bktNodes
  . lookupBucket nid

{-----------------------------------------------------------------------
--  Routing
-----------------------------------------------------------------------}

splitTip :: Eq ip => NodeId -> BucketCount -> BitIx -> Bucket ip -> Table ip
splitTip nid n i bucket
  | testIdBit nid i = (One  zeros (Tip nid (pred n) ones))
  |    otherwise    = (Zero (Tip nid (pred n) zeros) ones)
  where
    (ones, zeros) = split i bucket

-- | Used in each query.
--
-- TODO: Kademlia non-empty subtrees should should split if they have less than
-- k nodes in them.  Which subtrees I mean is illustrated in Fig 1. of Kademlia
-- paper.  The rule requiring additional splits is in section 2.4.
modifyBucket
  :: forall f ip xs. (Alternative f, Eq ip) =>
     NodeId -> (Bucket ip -> f (xs, Bucket ip)) -> Table ip -> f (xs,Table ip)
modifyBucket nodeId f = go (0 :: BitIx)
  where
    go :: BitIx -> Table ip -> f (xs, Table ip)
    go i (Zero table  bucket)
      | testIdBit nodeId i  = second (Zero  table) <$> f bucket
      |     otherwise       = second (`Zero` bucket) <$> go (succ i) table
    go i (One  bucket table )
      | testIdBit nodeId i  = second (One  bucket)   <$> go (succ i) table
      |     otherwise       = second (`One` table)   <$> f bucket
    go i (Tip nid n bucket)
      |        n == 0       = second (Tip nid n)    <$> f bucket
      |     otherwise       = second (Tip nid n)    <$> f bucket
                           <|>  go i (splitTip nid n i bucket)

-- | Triggering event for atomic table update
data Event ip = TryInsert { foreignNode :: NodeInfo ip }
              | PingResult { foreignNode :: NodeInfo ip
                           , ponged :: Bool
                           }
    deriving (Eq,Ord,Show)

eventId :: Event ip -> NodeId
eventId (TryInsert NodeInfo{..})    = nodeId
eventId (PingResult NodeInfo{..} _) = nodeId

-- | Actions requested by atomic table update
data CheckPing ip = CheckPing [NodeInfo ip]
    deriving (Eq,Ord,Show)


-- | Atomic 'Table' update
insert :: (Alternative m, Eq ip) => Timestamp -> Event ip -> Table ip -> m ([CheckPing ip], Table ip)
insert tm event = modifyBucket (eventId event) (insertBucket tm event)


{-----------------------------------------------------------------------
--  Conversion
-----------------------------------------------------------------------}

type TableEntry ip = (NodeInfo ip, Timestamp)

tableEntry :: NodeEntry ip -> TableEntry ip
tableEntry (a :-> b) = (a, b)

-- | Non-empty list of buckets.
toBucketList :: Table ip -> [Bucket ip]
toBucketList (Tip _ _ b) = [b]
toBucketList (Zero  t b) = b : toBucketList t
toBucketList (One   b t) = b : toBucketList t

toList :: Eq ip => Table ip -> [[TableEntry ip]]
toList = L.map (L.map tableEntry . PSQ.toList . bktNodes) . toBucketList