Merge branch 'request-strategy-rewrite'

bad_storage.go

@@ -15,11 +15,9 @@ type badStorage struct{}
 var _ storage.ClientImpl = badStorage{}
 
 func (bs badStorage) OpenTorrent(*metainfo.Info, metainfo.Hash) (storage.TorrentImpl, error) {
-	return bs, nil
+	return storage.TorrentImpl{
-}
+		Piece: bs.Piece,
-
+	}, nil
-func (bs badStorage) Close() error {
-	return nil
 }
 
 func (bs badStorage) Piece(p metainfo.Piece) storage.PieceImpl {

bencode/decode.go

@@ -205,68 +205,70 @@ func (d *Decoder) parseString(v reflect.Value) error {
 
 // Info for parsing a dict value.
 type dictField struct {
-	Value reflect.Value // Storage for the parsed value.
+	Type reflect.Type
-	// True if field value should be parsed into Value. If false, the value
+	Get  func(value reflect.Value) func(reflect.Value)
-	// should be parsed and discarded.
+	Tags tag
-	Ok                       bool
-	Set                      func() // Call this after parsing into Value.
-	IgnoreUnmarshalTypeError bool
 }
 
 // Returns specifics for parsing a dict field value.
-func getDictField(dict reflect.Value, key string) dictField {
+func getDictField(dict reflect.Type, key string) dictField {
 	// get valuev as a map value or as a struct field
 	switch dict.Kind() {
 	case reflect.Map:
-		value := reflect.New(dict.Type().Elem()).Elem()
 		return dictField{
-			Value: value,
+			Type: dict.Elem(),
-			Ok:    true,
+			Get: func(mapValue reflect.Value) func(reflect.Value) {
-			Set: func() {
+				return func(value reflect.Value) {
-				if dict.IsNil() {
+					if mapValue.IsNil() {
-					dict.Set(reflect.MakeMap(dict.Type()))
+						mapValue.Set(reflect.MakeMap(dict))
 					}
 					// Assigns the value into the map.
-				dict.SetMapIndex(reflect.ValueOf(key).Convert(dict.Type().Key()), value)
+					//log.Printf("map type: %v", mapValue.Type())
+					mapValue.SetMapIndex(reflect.ValueOf(key).Convert(dict.Key()), value)
+				}
 			},
 		}
 	case reflect.Struct:
-		sf, ok := getStructFieldForKey(dict.Type(), key)
+		return getStructFieldForKey(dict, key)
-		if !ok {
+		//if sf.r.PkgPath != "" {
-			return dictField{}
+		//	panic(&UnmarshalFieldError{
-		}
+		//		Key:   key,
-		if sf.r.PkgPath != "" {
+		//		Type:  dict.Type(),
-			panic(&UnmarshalFieldError{
+		//		Field: sf.r,
-				Key:   key,
+		//	})
-				Type:  dict.Type(),
+		//}
-				Field: sf.r,
-			})
-		}
-		return dictField{
-			Value:                    dict.FieldByIndex(sf.r.Index),
-			Ok:                       true,
-			Set:                      func() {},
-			IgnoreUnmarshalTypeError: sf.tag.IgnoreUnmarshalTypeError(),
-		}
 	default:
+		panic("unimplemented")
 		return dictField{}
 	}
 }
 
-type structField struct {
-	r   reflect.StructField
-	tag tag
-}
-
 var (
 	structFieldsMu sync.Mutex
-	structFields   = map[reflect.Type]map[string]structField{}
+	structFields   = map[reflect.Type]map[string]dictField{}
 )
 
-func parseStructFields(struct_ reflect.Type, each func(string, structField)) {
+func parseStructFields(struct_ reflect.Type, each func(key string, df dictField)) {
-	for i, n := 0, struct_.NumField(); i < n; i++ {
+	for _i, n := 0, struct_.NumField(); _i < n; _i++ {
+		i := _i
 		f := struct_.Field(i)
 		if f.Anonymous {
+			t := f.Type
+			if t.Kind() == reflect.Ptr {
+				t = t.Elem()
+			}
+			parseStructFields(t, func(key string, df dictField) {
+				innerGet := df.Get
+				df.Get = func(value reflect.Value) func(reflect.Value) {
+					anonPtr := value.Field(i)
+					if anonPtr.Kind() == reflect.Ptr && anonPtr.IsNil() {
+						anonPtr.Set(reflect.New(f.Type.Elem()))
+						anonPtr = anonPtr.Elem()
+					}
+					return innerGet(anonPtr)
+				}
+				each(key, df)
+			})
 			continue
 		}
 		tagStr := f.Tag.Get("bencode")
@@ -278,25 +280,35 @@ func parseStructFields(struct_ reflect.Type, each func(string, structField)) {
 		if key == "" {
 			key = f.Name
 		}
-		each(key, structField{f, tag})
+		each(key, dictField{f.Type, func(value reflect.Value) func(reflect.Value) {
+			return value.Field(i).Set
+		}, tag})
 	}
 }
 
 func saveStructFields(struct_ reflect.Type) {
-	m := make(map[string]structField)
+	m := make(map[string]dictField)
-	parseStructFields(struct_, func(key string, sf structField) {
+	parseStructFields(struct_, func(key string, sf dictField) {
 		m[key] = sf
 	})
 	structFields[struct_] = m
 }
 
-func getStructFieldForKey(struct_ reflect.Type, key string) (f structField, ok bool) {
+func getStructFieldForKey(struct_ reflect.Type, key string) (f dictField) {
 	structFieldsMu.Lock()
 	if _, ok := structFields[struct_]; !ok {
 		saveStructFields(struct_)
 	}
-	f, ok = structFields[struct_][key]
+	f, ok := structFields[struct_][key]
 	structFieldsMu.Unlock()
+	if !ok {
+		var discard interface{}
+		return dictField{
+			Type: reflect.TypeOf(discard),
+			Get:  func(reflect.Value) func(reflect.Value) { return func(reflect.Value) {} },
+			Tags: nil,
+		}
+	}
 	return
 }
 
@@ -314,31 +326,33 @@ func (d *Decoder) parseDict(v reflect.Value) error {
 			return nil
 		}
 
-		df := getDictField(v, keyStr)
+		df := getDictField(v.Type(), keyStr)
 
 		// now we need to actually parse it
-		if df.Ok {
+		if df.Type == nil {
-			// log.Printf("parsing ok struct field for key %q", keyStr)
-			ok, err = d.parseValue(df.Value)
-		} else {
 			// Discard the value, there's nowhere to put it.
 			var if_ interface{}
 			if_, ok = d.parseValueInterface()
 			if if_ == nil {
-				err = fmt.Errorf("error parsing value for key %q", keyStr)
+				return fmt.Errorf("error parsing value for key %q", keyStr)
 			}
+			if !ok {
+				return fmt.Errorf("missing value for key %q", keyStr)
 			}
+			continue
+		}
+		setValue := reflect.New(df.Type).Elem()
+		//log.Printf("parsing into %v", setValue.Type())
+		ok, err = d.parseValue(setValue)
 		if err != nil {
-			if _, ok := err.(*UnmarshalTypeError); !ok || !df.IgnoreUnmarshalTypeError {
+			if _, ok := err.(*UnmarshalTypeError); !ok || !df.Tags.IgnoreUnmarshalTypeError() {
 				return fmt.Errorf("parsing value for key %q: %s", keyStr, err)
 			}
 		}
 		if !ok {
 			return fmt.Errorf("missing value for key %q", keyStr)
 		}
-		if df.Ok {
+		df.Get(v)(setValue)
-			df.Set()
-		}
 	}
 }
 

bencode/decode_test.go

@@ -7,6 +7,7 @@ import (
 	"reflect"
 	"testing"
 
+	qt "github.com/frankban/quicktest"
 	"github.com/stretchr/testify/assert"
 	"github.com/stretchr/testify/require"
 )
@@ -144,7 +145,7 @@ func TestIgnoreUnmarshalTypeError(t *testing.T) {
 	}{}
 	require.Error(t, Unmarshal([]byte("d6:Normal5:helloe"), &s))
 	assert.NoError(t, Unmarshal([]byte("d6:Ignore5:helloe"), &s))
-	require.Nil(t, Unmarshal([]byte("d6:Ignorei42ee"), &s))
+	qt.Assert(t, Unmarshal([]byte("d6:Ignorei42ee"), &s), qt.IsNil)
 	assert.EqualValues(t, 42, s.Ignore)
 }
 

bencode/encode.go

@@ -133,13 +133,16 @@ func (e *Encoder) reflectValue(v reflect.Value) {
 		e.reflectString(v.String())
 	case reflect.Struct:
 		e.writeString("d")
-		for _, ef := range encodeFields(v.Type()) {
+		for _, ef := range getEncodeFields(v.Type()) {
-			field_value := v.Field(ef.i)
+			fieldValue := ef.i(v)
-			if ef.omit_empty && isEmptyValue(field_value) {
+			if !fieldValue.IsValid() {
+				continue
+			}
+			if ef.omitEmpty && isEmptyValue(fieldValue) {
 				continue
 			}
 			e.reflectString(ef.tag)
-			e.reflectValue(field_value)
+			e.reflectValue(fieldValue)
 		}
 		e.writeString("e")
 	case reflect.Map:
@@ -190,9 +193,9 @@ func (e *Encoder) reflectValue(v reflect.Value) {
 }
 
 type encodeField struct {
-	i          int
+	i         func(v reflect.Value) reflect.Value
 	tag       string
-	omit_empty bool
+	omitEmpty bool
 }
 
 type encodeFieldsSortType []encodeField
@@ -206,31 +209,55 @@ var (
 	encodeFieldsCache = make(map[reflect.Type][]encodeField)
 )
 
-func encodeFields(t reflect.Type) []encodeField {
+func getEncodeFields(t reflect.Type) []encodeField {
 	typeCacheLock.RLock()
 	fs, ok := encodeFieldsCache[t]
 	typeCacheLock.RUnlock()
 	if ok {
 		return fs
 	}
-
+	fs = makeEncodeFields(t)
 	typeCacheLock.Lock()
 	defer typeCacheLock.Unlock()
-	fs, ok = encodeFieldsCache[t]
+	encodeFieldsCache[t] = fs
-	if ok {
 	return fs
 }
 
-	for i, n := 0, t.NumField(); i < n; i++ {
+func makeEncodeFields(t reflect.Type) (fs []encodeField) {
+	for _i, n := 0, t.NumField(); _i < n; _i++ {
+		i := _i
 		f := t.Field(i)
 		if f.PkgPath != "" {
 			continue
 		}
 		if f.Anonymous {
+			t := f.Type
+			if t.Kind() == reflect.Ptr {
+				t = t.Elem()
+			}
+			anonEFs := makeEncodeFields(t)
+			for aefi := range anonEFs {
+				anonEF := anonEFs[aefi]
+				bottomField := anonEF
+				bottomField.i = func(v reflect.Value) reflect.Value {
+					v = v.Field(i)
+					if v.Kind() == reflect.Ptr {
+						if v.IsNil() {
+							// This will skip serializing this value.
+							return reflect.Value{}
+						}
+						v = v.Elem()
+					}
+					return anonEF.i(v)
+				}
+				fs = append(fs, bottomField)
+			}
 			continue
 		}
 		var ef encodeField
-		ef.i = i
+		ef.i = func(v reflect.Value) reflect.Value {
+			return v.Field(i)
+		}
 		ef.tag = f.Name
 
 		tv := getTag(f.Tag)
@@ -240,11 +267,10 @@ func encodeFields(t reflect.Type) []encodeField {
 		if tv.Key() != "" {
 			ef.tag = tv.Key()
 		}
-		ef.omit_empty = tv.OmitEmpty()
+		ef.omitEmpty = tv.OmitEmpty()
 		fs = append(fs, ef)
 	}
 	fss := encodeFieldsSortType(fs)
 	sort.Sort(fss)
-	encodeFieldsCache[t] = fs
 	return fs
 }

bencode/tags.go

@@ -24,6 +24,9 @@ func (me tag) Key() string {
 }
 
 func (me tag) HasOpt(opt string) bool {
+	if len(me) < 1 {
+		return false
+	}
 	for _, s := range me[1:] {
 		if s == opt {
 			return true

client.go

@@ -2,7 +2,6 @@ package torrent
 
 import (
 	"bufio"
-	"bytes"
 	"context"
 	"crypto/rand"
 	"encoding/binary"
@@ -18,15 +17,14 @@ import (
 	"github.com/anacrolix/dht/v2"
 	"github.com/anacrolix/dht/v2/krpc"
 	"github.com/anacrolix/log"
-	"github.com/anacrolix/missinggo/bitmap"
 	"github.com/anacrolix/missinggo/perf"
 	"github.com/anacrolix/missinggo/pubsub"
 	"github.com/anacrolix/missinggo/slices"
+	"github.com/anacrolix/missinggo/v2"
+	"github.com/anacrolix/missinggo/v2/bitmap"
+	"github.com/anacrolix/missinggo/v2/conntrack"
 	"github.com/anacrolix/missinggo/v2/pproffd"
 	"github.com/anacrolix/sync"
-	"github.com/anacrolix/torrent/internal/limiter"
-	"github.com/anacrolix/torrent/tracker"
-	"github.com/anacrolix/torrent/webtorrent"
 	"github.com/davecgh/go-spew/spew"
 	"github.com/dustin/go-humanize"
 	"github.com/google/btree"
@@ -34,15 +32,17 @@ import (
 	"golang.org/x/time/rate"
 	"golang.org/x/xerrors"
 
-	"github.com/anacrolix/missinggo/v2"
+	"github.com/anacrolix/chansync"
-	"github.com/anacrolix/missinggo/v2/conntrack"
 
 	"github.com/anacrolix/torrent/bencode"
+	"github.com/anacrolix/torrent/internal/limiter"
 	"github.com/anacrolix/torrent/iplist"
 	"github.com/anacrolix/torrent/metainfo"
 	"github.com/anacrolix/torrent/mse"
 	pp "github.com/anacrolix/torrent/peer_protocol"
 	"github.com/anacrolix/torrent/storage"
+	"github.com/anacrolix/torrent/tracker"
+	"github.com/anacrolix/torrent/webtorrent"
 )
 
 // Clients contain zero or more Torrents. A Client manages a blocklist, the
@@ -81,6 +81,8 @@ type Client struct {
 	websocketTrackers websocketTrackers
 
 	activeAnnounceLimiter limiter.Instance
+
+	updateRequests chansync.BroadcastCond
 }
 
 type ipStr string
@@ -259,7 +261,7 @@ func NewClient(cfg *ClientConfig) (cl *Client, err error) {
 				if err != nil {
 					panic(err)
 				}
-				cl.dhtServers = append(cl.dhtServers, anacrolixDhtServerWrapper{ds})
+				cl.dhtServers = append(cl.dhtServers, AnacrolixDhtServerWrapper{ds})
 				cl.onClose = append(cl.onClose, func() { ds.Close() })
 			}
 		}
@@ -293,6 +295,8 @@ func NewClient(cfg *ClientConfig) (cl *Client, err error) {
 		},
 	}
 
+	go cl.requester()
+
 	return
 }
 
@@ -311,6 +315,10 @@ func (cl *Client) AddDialer(d Dialer) {
 	}
 }
 
+func (cl *Client) Listeners() []Listener {
+	return cl.listeners
+}
+
 // Registers a Listener, and starts Accepting on it. You must Close Listeners provided this way
 // yourself.
 func (cl *Client) AddListener(l Listener) {
@@ -955,7 +963,7 @@ func (cl *Client) runHandshookConn(c *PeerConn, t *Torrent) error {
 		return fmt.Errorf("adding connection: %w", err)
 	}
 	defer t.dropConnection(c)
-	go c.writer(time.Minute)
+	c.startWriter()
 	cl.sendInitialMessages(c, t)
 	err := c.mainReadLoop()
 	if err != nil {
@@ -964,10 +972,15 @@ func (cl *Client) runHandshookConn(c *PeerConn, t *Torrent) error {
 	return nil
 }
 
+// Maximum pending requests we allow peers to send us. If peer requests are buffered on read, this
+// instructs the amount of memory that might be used to cache pending writes. Assuming 512KiB
+// (1<<19) cached for sending, for 16KiB (1<<14) chunks.
+const localClientReqq = 1 << 5
+
 // See the order given in Transmission's tr_peerMsgsNew.
 func (cl *Client) sendInitialMessages(conn *PeerConn, torrent *Torrent) {
 	if conn.PeerExtensionBytes.SupportsExtended() && cl.config.Extensions.SupportsExtended() {
-		conn.post(pp.Message{
+		conn.write(pp.Message{
 			Type:       pp.Extended,
 			ExtendedID: pp.HandshakeExtendedID,
 			ExtendedPayload: func() []byte {
@@ -976,10 +989,7 @@ func (cl *Client) sendInitialMessages(conn *PeerConn, torrent *Torrent) {
 						pp.ExtensionNameMetadata: metadataExtendedId,
 					},
 					V:            cl.config.ExtendedHandshakeClientVersion,
-					// If peer requests are buffered on read, this instructs the amount of memory
+					Reqq:         localClientReqq,
-					// that might be used to cache pending writes. Assuming 512KiB cached for
-					// sending, for 16KiB chunks.
-					Reqq:         1 << 5,
 					YourIp:       pp.CompactIp(conn.remoteIp()),
 					Encryption:   cl.config.HeaderObfuscationPolicy.Preferred || !cl.config.HeaderObfuscationPolicy.RequirePreferred,
 					Port:         cl.incomingPeerPort(),
@@ -999,11 +1009,11 @@ func (cl *Client) sendInitialMessages(conn *PeerConn, torrent *Torrent) {
 	func() {
 		if conn.fastEnabled() {
 			if torrent.haveAllPieces() {
-				conn.post(pp.Message{Type: pp.HaveAll})
+				conn.write(pp.Message{Type: pp.HaveAll})
-				conn.sentHaves.AddRange(0, bitmap.BitIndex(conn.t.NumPieces()))
+				conn.sentHaves.AddRange(0, bitmap.BitRange(conn.t.NumPieces()))
 				return
 			} else if !torrent.haveAnyPieces() {
-				conn.post(pp.Message{Type: pp.HaveNone})
+				conn.write(pp.Message{Type: pp.HaveNone})
 				conn.sentHaves.Clear()
 				return
 			}
@@ -1011,7 +1021,7 @@ func (cl *Client) sendInitialMessages(conn *PeerConn, torrent *Torrent) {
 		conn.postBitfield()
 	}()
 	if conn.PeerExtensionBytes.SupportsDHT() && cl.config.Extensions.SupportsDHT() && cl.haveDhtServer() {
-		conn.post(pp.Message{
+		conn.write(pp.Message{
 			Type: pp.Port,
 			Port: cl.dhtPort(),
 		})
@@ -1069,12 +1079,12 @@ func (cl *Client) gotMetadataExtensionMsg(payload []byte, t *Torrent, c *PeerCon
 		return err
 	case pp.RequestMetadataExtensionMsgType:
 		if !t.haveMetadataPiece(piece) {
-			c.post(t.newMetadataExtensionMessage(c, pp.RejectMetadataExtensionMsgType, d["piece"], nil))
+			c.write(t.newMetadataExtensionMessage(c, pp.RejectMetadataExtensionMsgType, d["piece"], nil))
 			return nil
 		}
 		start := (1 << 14) * piece
 		c.logger.WithDefaultLevel(log.Debug).Printf("sending metadata piece %d", piece)
-		c.post(t.newMetadataExtensionMessage(c, pp.DataMetadataExtensionMsgType, piece, t.metadataBytes[start:start+t.metadataPieceSize(piece)]))
+		c.write(t.newMetadataExtensionMessage(c, pp.DataMetadataExtensionMsgType, piece, t.metadataBytes[start:start+t.metadataPieceSize(piece)]))
 		return nil
 	case pp.RejectMetadataExtensionMsgType:
 		return nil
@@ -1139,7 +1149,6 @@ func (cl *Client) newTorrent(ih metainfo.Hash, specStorage storage.ClientImpl) (
 		webSeeds: make(map[string]*Peer),
 	}
 	t._pendingPieces.NewSet = priorityBitmapStableNewSet
-	t.requestStrategy = cl.config.DefaultRequestStrategy(t.requestStrategyCallbacks(), &cl._mu)
 	t.logger = cl.logger.WithContextValue(t)
 	t.setChunkSize(defaultChunkSize)
 	return
@@ -1405,11 +1414,9 @@ func (cl *Client) newConnection(nc net.Conn, outgoing bool, remoteAddr PeerRemot
 		},
 		connString: connString,
 		conn:       nc,
-		writeBuffer: new(bytes.Buffer),
 	}
 	c.peerImpl = c
 	c.logger = cl.logger.WithDefaultLevel(log.Warning).WithContextValue(c)
-	c.writerCond.L = cl.locker()
 	c.setRW(connStatsReadWriter{nc, c})
 	c.r = &rateLimitedReader{
 		l: cl.config.DownloadRateLimiter,

client_test.go

@@ -322,7 +322,7 @@ func TestDhtInheritBlocklist(t *testing.T) {
 	numServers := 0
 	cl.eachDhtServer(func(s DhtServer) {
 		t.Log(s)
-		assert.Equal(t, ipl, s.(anacrolixDhtServerWrapper).Server.IPBlocklist())
+		assert.Equal(t, ipl, s.(AnacrolixDhtServerWrapper).Server.IPBlocklist())
 		numServers++
 	})
 	assert.EqualValues(t, 2, numServers)
@@ -554,6 +554,8 @@ func TestPeerInvalidHave(t *testing.T) {
 		t: tt,
 	}}
 	cn.peerImpl = cn
+	cl.lock()
+	defer cl.unlock()
 	assert.NoError(t, cn.peerSentHave(0))
 	assert.Error(t, cn.peerSentHave(1))
 }

cmd/torrent/main.go

@@ -173,6 +173,7 @@ type DownloadCmd struct {
 	TestPeer           []string       `help:"addresses of some starting peers"`
 	Seed               bool           `help:"seed after download is complete"`
 	Addr               string         `help:"network listen addr"`
+	MaxUnverifiedBytes tagflag.Bytes  `help:"maximum number bytes to have pending verification"`
 	UploadRate         *tagflag.Bytes `help:"max piece bytes to send per second"`
 	DownloadRate       *tagflag.Bytes `help:"max bytes per second down from peers"`
 	PackedBlocklist    string
@@ -311,6 +312,7 @@ func downloadErr() error {
 	if flags.Quiet {
 		clientConfig.Logger = log.Discard
 	}
+	clientConfig.MaxUnverifiedBytes = flags.MaxUnverifiedBytes.Int64()
 
 	var stop missinggo.SynchronizedEvent
 	defer func() {

config.go

@@ -59,6 +59,8 @@ type ClientConfig struct {
 	// (~4096), and the requested chunk size (~16KiB, see
 	// TorrentSpec.ChunkSize).
 	DownloadRateLimiter *rate.Limiter
+	// Maximum unverified bytes across all torrents. Not used if zero.
+	MaxUnverifiedBytes int64
 
 	// User-provided Client peer ID. If not present, one is generated automatically.
 	PeerID string
@@ -137,8 +139,6 @@ type ClientConfig struct {
 	// OnQuery hook func
 	DHTOnQuery func(query *krpc.Msg, source net.Addr) (propagate bool)
 
-	DefaultRequestStrategy requestStrategyMaker
-
 	Extensions PeerExtensionBits
 
 	DisableWebtorrent bool
@@ -185,9 +185,6 @@ func NewDefaultClientConfig() *ClientConfig {
 		CryptoSelector: mse.DefaultCryptoSelector,
 		CryptoProvides: mse.AllSupportedCrypto,
 		ListenPort:     42069,
-
-		DefaultRequestStrategy: RequestStrategyDuplicateRequestTimeout(5 * time.Second),
-
 		Extensions:     defaultPeerExtensionBytes(),
 	}
 	//cc.ConnTracker.SetNoMaxEntries()

conn_stats.go

@@ -23,6 +23,7 @@ type ConnStats struct {
 	BytesRead                   Count
 	BytesReadData               Count
 	BytesReadUsefulData         Count
+	BytesReadUsefulIntendedData Count
 
 	ChunksWritten Count
 
@@ -78,14 +79,9 @@ func (cs *ConnStats) wroteMsg(msg *pp.Message) {
 	}
 }
 
-func (cs *ConnStats) readMsg(msg *pp.Message) {
+func (cs *ConnStats) receivedChunk(size int64) {
-	// We want to also handle extended metadata pieces here, but we wouldn't
-	// have decoded the extended payload yet.
-	switch msg.Type {
-	case pp.Piece:
 	cs.ChunksRead.Add(1)
-		cs.BytesReadData.Add(int64(len(msg.Piece)))
+	cs.BytesReadData.Add(size)
-	}
 }
 
 func (cs *ConnStats) incrementPiecesDirtiedGood() {

dht.go

@@ -29,11 +29,11 @@ type DhtAnnounce interface {
 	Peers() <-chan dht.PeersValues
 }
 
-type anacrolixDhtServerWrapper struct {
+type AnacrolixDhtServerWrapper struct {
 	*dht.Server
 }
 
-func (me anacrolixDhtServerWrapper) Stats() interface{} {
+func (me AnacrolixDhtServerWrapper) Stats() interface{} {
 	return me.Server.Stats()
 }
 
@@ -45,13 +45,13 @@ func (me anacrolixDhtAnnounceWrapper) Peers() <-chan dht.PeersValues {
 	return me.Announce.Peers
 }
 
-func (me anacrolixDhtServerWrapper) Announce(hash [20]byte, port int, impliedPort bool) (DhtAnnounce, error) {
+func (me AnacrolixDhtServerWrapper) Announce(hash [20]byte, port int, impliedPort bool) (DhtAnnounce, error) {
 	ann, err := me.Server.Announce(hash, port, impliedPort)
 	return anacrolixDhtAnnounceWrapper{ann}, err
 }
 
-func (me anacrolixDhtServerWrapper) Ping(addr *net.UDPAddr) {
+func (me AnacrolixDhtServerWrapper) Ping(addr *net.UDPAddr) {
 	me.Server.Ping(addr)
 }
 
-var _ DhtServer = anacrolixDhtServerWrapper{}
+var _ DhtServer = AnacrolixDhtServerWrapper{}

file.go

@@ -66,20 +66,20 @@ func fileBytesLeft(
 	switch numPiecesSpanned {
 	case 0:
 	case 1:
-		if !torrentCompletedPieces.Get(fileFirstPieceIndex) {
+		if !torrentCompletedPieces.Get(bitmap.BitIndex(fileFirstPieceIndex)) {
 			left += fileLength
 		}
 	default:
-		if !torrentCompletedPieces.Get(fileFirstPieceIndex) {
+		if !torrentCompletedPieces.Get(bitmap.BitIndex(fileFirstPieceIndex)) {
 			left += torrentUsualPieceSize - (fileTorrentOffset % torrentUsualPieceSize)
 		}
-		if !torrentCompletedPieces.Get(fileEndPieceIndex - 1) {
+		if !torrentCompletedPieces.Get(bitmap.BitIndex(fileEndPieceIndex - 1)) {
 			left += fileTorrentOffset + fileLength - int64(fileEndPieceIndex-1)*torrentUsualPieceSize
 		}
 		completedMiddlePieces := torrentCompletedPieces.Copy()
-		completedMiddlePieces.RemoveRange(0, fileFirstPieceIndex+1)
+		completedMiddlePieces.RemoveRange(0, bitmap.BitRange(fileFirstPieceIndex+1))
-		completedMiddlePieces.RemoveRange(fileEndPieceIndex-1, bitmap.ToEnd)
+		completedMiddlePieces.RemoveRange(bitmap.BitRange(fileEndPieceIndex-1), bitmap.ToEnd)
-		left += int64(numPiecesSpanned-2-completedMiddlePieces.Len()) * torrentUsualPieceSize
+		left += int64(numPiecesSpanned-2-pieceIndex(completedMiddlePieces.Len())) * torrentUsualPieceSize
 	}
 	return
 }

global.go

@@ -9,7 +9,6 @@ import (
 
 const (
 	pieceHash        = crypto.SHA1
-	maxRequests      = 250    // Maximum pending requests we allow peers to send us.
 	defaultChunkSize = 0x4000 // 16KiB
 )
 

go.mod

@@ -5,6 +5,7 @@ require (
 	crawshaw.io/sqlite v0.3.3-0.20210127221821-98b1f83c5508
 	github.com/RoaringBitmap/roaring v0.6.0 // indirect
 	github.com/alexflint/go-arg v1.3.0
+	github.com/anacrolix/chansync v0.0.0-20210524073341-a336ebc2de92 // indirect
 	github.com/anacrolix/confluence v1.7.1-0.20210311004351-d642adb8546c // indirect
 	github.com/anacrolix/dht/v2 v2.9.1
 	github.com/anacrolix/envpprof v1.1.1
@@ -12,9 +13,9 @@ require (
 	github.com/anacrolix/log v0.9.0
 	github.com/anacrolix/missinggo v1.2.1
 	github.com/anacrolix/missinggo/perf v1.0.0
-	github.com/anacrolix/missinggo/v2 v2.5.0
+	github.com/anacrolix/missinggo/v2 v2.5.1-0.20210520011502-b3d95d6b1d02
-	github.com/anacrolix/multiless v0.1.0
+	github.com/anacrolix/multiless v0.1.1-0.20210520040635-10ee7b5f3cff
-	github.com/anacrolix/sync v0.2.0
+	github.com/anacrolix/sync v0.3.0
 	github.com/anacrolix/tagflag v1.3.0
 	github.com/anacrolix/upnp v0.1.2-0.20200416075019-5e9378ed1425
 	github.com/anacrolix/utp v0.1.0

go.sum

@@ -8,6 +8,7 @@ cloud.google.com/go v0.37.0 h1:69FNAINiZfsEuwH3fKq8QrAAnHz+2m4XL4kVYi5BX0Q=
 cloud.google.com/go v0.37.0/go.mod h1:TS1dMSSfndXH133OKGwekG838Om/cQT0BUHV3HcBgoo=
 crawshaw.io/iox v0.0.0-20181124134642-c51c3df30797 h1:yDf7ARQc637HoxDho7xjqdvO5ZA2Yb+xzv/fOnnvZzw=
 crawshaw.io/iox v0.0.0-20181124134642-c51c3df30797/go.mod h1:sXBiorCo8c46JlQV3oXPKINnZ8mcqnye1EkVkqsectk=
+crawshaw.io/sqlite v0.3.2/go.mod h1:igAO5JulrQ1DbdZdtVq48mnZUBAPOeFzer7VhDWNtW4=
 crawshaw.io/sqlite v0.3.3-0.20210127221821-98b1f83c5508 h1:fILCBBFnjnrQ0whVJlGhfv1E/QiaFDNtGFBObEVRnYg=
 crawshaw.io/sqlite v0.3.3-0.20210127221821-98b1f83c5508/go.mod h1:igAO5JulrQ1DbdZdtVq48mnZUBAPOeFzer7VhDWNtW4=
 dmitri.shuralyov.com/app/changes v0.0.0-20180602232624-0a106ad413e3 h1:hJiie5Bf3QucGRa4ymsAUOxyhYwGEz1xrsVk0P8erlw=
@@ -55,6 +56,8 @@ github.com/alexflint/go-arg v1.3.0 h1:UfldqSdFWeLtoOuVRosqofU4nmhI1pYEbT4ZFS34Bd
 github.com/alexflint/go-arg v1.3.0/go.mod h1:9iRbDxne7LcR/GSvEr7ma++GLpdIU1zrghf2y2768kM=
 github.com/alexflint/go-scalar v1.0.0 h1:NGupf1XV/Xb04wXskDFzS0KWOLH632W/EO4fAFi+A70=
 github.com/alexflint/go-scalar v1.0.0/go.mod h1:GpHzbCOZXEKMEcygYQ5n/aa4Aq84zbxjy3MxYW0gjYw=
+github.com/anacrolix/chansync v0.0.0-20210524073341-a336ebc2de92 h1:WGk37RyXPWcIALJxTkTNrXN3yLQp7hSFa3x5GkrK/Rs=
+github.com/anacrolix/chansync v0.0.0-20210524073341-a336ebc2de92/go.mod h1:DZsatdsdXxD0WiwcGl0nJVwyjCKMDv+knl1q2iBjA2k=
 github.com/anacrolix/confluence v1.7.1-0.20210221224747-9cb14aa2c53a/go.mod h1:T0JHvSaf9UfoiUdCtCOUuRroHm/tauUJTbLc6/vd5YA=
 github.com/anacrolix/confluence v1.7.1-0.20210221225853-90405640e928/go.mod h1:NoLcfoRet+kYttjLXJRmh4qBVrylJsfIItik5GGj21A=
 github.com/anacrolix/confluence v1.7.1-0.20210311004351-d642adb8546c h1:HfbeiZS/0hwdotwtQhllrd3PagmuLgCN9O8CHJgzPGQ=
@@ -108,14 +111,18 @@ github.com/anacrolix/missinggo/v2 v2.3.1/go.mod h1:3XNH0OEmyMUZuvXmYdl+FDfXd0vvS
 github.com/anacrolix/missinggo/v2 v2.4.1-0.20200227072623-f02f6484f997/go.mod h1:KY+ij+mWvwGuqSuecLjjPv5LFw5ICUc1UvRems3VAZE=
 github.com/anacrolix/missinggo/v2 v2.5.0 h1:75aciOVrzVV1bTH9rl8tYLbXO9A7HXFtHexTChawe/U=
 github.com/anacrolix/missinggo/v2 v2.5.0/go.mod h1:HYuCbwvJXY3XbcmcIcTgZXHleoDXawxPWx/YiPzFzV0=
+github.com/anacrolix/missinggo/v2 v2.5.1-0.20210520011502-b3d95d6b1d02 h1:wf3HKUunewks4FdGJqkViby+vr3n5/IFpPsyEMokxYE=
+github.com/anacrolix/missinggo/v2 v2.5.1-0.20210520011502-b3d95d6b1d02/go.mod h1:WEjqh2rmKECd0t1VhQkLGTdIWXO6f6NLjp5GlMZ+6FA=
 github.com/anacrolix/mmsg v0.0.0-20180515031531-a4a3ba1fc8bb/go.mod h1:x2/ErsYUmT77kezS63+wzZp8E3byYB0gzirM/WMBLfw=
 github.com/anacrolix/mmsg v1.0.0 h1:btC7YLjOn29aTUAExJiVUhQOuf/8rhm+/nWCMAnL3Hg=
 github.com/anacrolix/mmsg v1.0.0/go.mod h1:x8kRaJY/dCrY9Al0PEcj1mb/uFHwP6GCJ9fLl4thEPc=
 github.com/anacrolix/multiless v0.0.0-20191223025854-070b7994e841/go.mod h1:TrCLEZfIDbMVfLoQt5tOoiBS/uq4y8+ojuEVVvTNPX4=
 github.com/anacrolix/multiless v0.0.0-20200413040533-acfd16f65d5d/go.mod h1:TrCLEZfIDbMVfLoQt5tOoiBS/uq4y8+ojuEVVvTNPX4=
 github.com/anacrolix/multiless v0.0.0-20210222022749-ef43011a77ec/go.mod h1:TrCLEZfIDbMVfLoQt5tOoiBS/uq4y8+ojuEVVvTNPX4=
-github.com/anacrolix/multiless v0.1.0 h1:gjR3SdJ+E0avnmEoAV/7K7n2kILZhVu/M6aQEtz8H3s=
+github.com/anacrolix/multiless v0.1.1-0.20210510014912-3f17cb19bda9 h1:fmNDxh5ysBPinRq249xYYYOLQ/h95DoyGE9e9Gp9xqo=
-github.com/anacrolix/multiless v0.1.0/go.mod h1:TrCLEZfIDbMVfLoQt5tOoiBS/uq4y8+ojuEVVvTNPX4=
+github.com/anacrolix/multiless v0.1.1-0.20210510014912-3f17cb19bda9/go.mod h1:TrCLEZfIDbMVfLoQt5tOoiBS/uq4y8+ojuEVVvTNPX4=
+github.com/anacrolix/multiless v0.1.1-0.20210520040635-10ee7b5f3cff h1:ve99yq2FjiN3OANMjRz2rjCX4f2PSKMf3NeDFnmcs8s=
+github.com/anacrolix/multiless v0.1.1-0.20210520040635-10ee7b5f3cff/go.mod h1:TrCLEZfIDbMVfLoQt5tOoiBS/uq4y8+ojuEVVvTNPX4=
 github.com/anacrolix/stm v0.1.0/go.mod h1:ZKz7e7ERWvP0KgL7WXfRjBXHNRhlVRlbBQecqFtPq+A=
 github.com/anacrolix/stm v0.1.1-0.20191106051447-e749ba3531cf/go.mod h1:zoVQRvSiGjGoTmbM0vSLIiaKjWtNPeTvXUSdJQA4hsg=
 github.com/anacrolix/stm v0.2.0/go.mod h1:zoVQRvSiGjGoTmbM0vSLIiaKjWtNPeTvXUSdJQA4hsg=
@@ -128,6 +135,10 @@ github.com/anacrolix/sync v0.0.0-20180611022320-3c4cb11f5a01/go.mod h1:+u91KiUuf
 github.com/anacrolix/sync v0.0.0-20180808010631-44578de4e778/go.mod h1:s735Etp3joe/voe2sdaXLcqDdJSay1O0OPnM0ystjqk=
 github.com/anacrolix/sync v0.2.0 h1:oRe22/ZB+v7v/5Mbc4d2zE0AXEZy0trKyKLjqYOt6tY=
 github.com/anacrolix/sync v0.2.0/go.mod h1:BbecHL6jDSExojhNtgTFSBcdGerzNc64tz3DCOj/I0g=
+github.com/anacrolix/sync v0.2.1-0.20210520084835-26aa6614542f h1:7KqmZoEOIXa0UbR2WQ/YPF4H+MPV6rhWk4E4tcv5eDg=
+github.com/anacrolix/sync v0.2.1-0.20210520084835-26aa6614542f/go.mod h1:BbecHL6jDSExojhNtgTFSBcdGerzNc64tz3DCOj/I0g=
+github.com/anacrolix/sync v0.3.0 h1:ZPjTrkqQWEfnYVGTQHh5qNjokWaXnjsyXTJSMsKY0TA=
+github.com/anacrolix/sync v0.3.0/go.mod h1:BbecHL6jDSExojhNtgTFSBcdGerzNc64tz3DCOj/I0g=
 github.com/anacrolix/tagflag v0.0.0-20180109131632-2146c8d41bf0/go.mod h1:1m2U/K6ZT+JZG0+bdMK6qauP49QT4wE5pmhJXOKKCHw=
 github.com/anacrolix/tagflag v0.0.0-20180605133421-f477c8c2f14c/go.mod h1:1m2U/K6ZT+JZG0+bdMK6qauP49QT4wE5pmhJXOKKCHw=
 github.com/anacrolix/tagflag v0.0.0-20180803105420-3a8ff5428f76/go.mod h1:1m2U/K6ZT+JZG0+bdMK6qauP49QT4wE5pmhJXOKKCHw=
@@ -135,7 +146,6 @@ github.com/anacrolix/tagflag v1.0.0/go.mod h1:1m2U/K6ZT+JZG0+bdMK6qauP49QT4wE5pm
 github.com/anacrolix/tagflag v1.0.1/go.mod h1:gb0fiMQ02qU1djCSqaxGmruMvZGrMwSReidMB0zjdxo=
 github.com/anacrolix/tagflag v1.1.0/go.mod h1:Scxs9CV10NQatSmbyjqmqmeQNwGzlNe0CMUMIxqHIG8=
 github.com/anacrolix/tagflag v1.1.1-0.20200411025953-9bb5209d56c2/go.mod h1:Scxs9CV10NQatSmbyjqmqmeQNwGzlNe0CMUMIxqHIG8=
-github.com/anacrolix/tagflag v1.2.0 h1:WdSv10SpxOI97++f5FUKnKPFkVGMiPlpYm52XPaMkp4=
 github.com/anacrolix/tagflag v1.2.0/go.mod h1:Scxs9CV10NQatSmbyjqmqmeQNwGzlNe0CMUMIxqHIG8=
 github.com/anacrolix/tagflag v1.3.0 h1:5NI+9CniDnEH0BWA4UcQbERyFPjKJqZnVkItGVIDy/s=
 github.com/anacrolix/tagflag v1.3.0/go.mod h1:Scxs9CV10NQatSmbyjqmqmeQNwGzlNe0CMUMIxqHIG8=
@@ -268,8 +278,6 @@ github.com/frankban/quicktest v1.11.3/go.mod h1:wRf/ReqHper53s+kmmSZizM8NamnL3IM
 github.com/fsnotify/fsnotify v1.4.7/go.mod h1:jwhsz4b93w/PPRr/qN1Yymfu8t87LnFCMoQvtojpjFo=
 github.com/fsnotify/fsnotify v1.4.9 h1:hsms1Qyu0jgnwNXIxa+/V/PDsU6CfLf6CNO8H7IWoS4=
 github.com/fsnotify/fsnotify v1.4.9/go.mod h1:znqG4EE+3YCdAaPaxE2ZRY/06pZUdp0tY4IgpuI1SZQ=
-github.com/getlantern/sqlite v0.3.3-0.20210215090556-4f83cf7731f0 h1:zvFSvII5rTbMZ3idAqSUjUCDgZFbWMKzxQot3/Y7nzA=
-github.com/getlantern/sqlite v0.3.3-0.20210215090556-4f83cf7731f0/go.mod h1:igAO5JulrQ1DbdZdtVq48mnZUBAPOeFzer7VhDWNtW4=
 github.com/ghodss/yaml v1.0.0 h1:wQHKEahhL6wmXdzwWG11gIVCkOv05bNOh+Rxn0yngAk=
 github.com/ghodss/yaml v1.0.0/go.mod h1:4dBDuWmgqj2HViK6kFavaiC9ZROes6MMH2rRYeMEF04=
 github.com/gliderlabs/ssh v0.1.1 h1:j3L6gSLQalDETeEg/Jg0mGY0/y/N6zI2xX1978P0Uqw=
@@ -424,6 +432,7 @@ github.com/hashicorp/go-version v1.2.0/go.mod h1:fltr4n8CU8Ke44wwGCBoEymUuxUHl09
 github.com/hashicorp/go.net v0.0.1 h1:sNCoNyDEvN1xa+X0baata4RdcpKwcMS6DH+xwfqPgjw=
 github.com/hashicorp/go.net v0.0.1/go.mod h1:hjKkEWcCURg++eb33jQU7oqQcI9XDCnUzHA0oac0k90=
 github.com/hashicorp/golang-lru v0.5.0/go.mod h1:/m3WP610KZHVQ1SGc6re/UDhFvYD7pJ4Ao+sR/qLZy8=
+github.com/hashicorp/golang-lru v0.5.1 h1:0hERBMJE1eitiLkihrMvRVBYAkpHzc/J3QdDN+dAcgU=
 github.com/hashicorp/golang-lru v0.5.1/go.mod h1:/m3WP610KZHVQ1SGc6re/UDhFvYD7pJ4Ao+sR/qLZy8=
 github.com/hashicorp/golang-lru v0.5.4 h1:YDjusn29QI/Das2iO9M0BHnIbxPeyuCHsjMW+lJfyTc=
 github.com/hashicorp/golang-lru v0.5.4/go.mod h1:iADmTwqILo4mZ8BN3D2Q6+9jd8WM5uGBxy+E8yxSoD4=

misc.go

@@ -5,29 +5,27 @@ import (
 	"net"
 
 	"github.com/anacrolix/missinggo/v2"
+	"github.com/anacrolix/torrent/types"
 	"golang.org/x/time/rate"
 
 	"github.com/anacrolix/torrent/metainfo"
 	pp "github.com/anacrolix/torrent/peer_protocol"
 )
 
-type ChunkSpec struct {
+type (
-	Begin, Length pp.Integer
+	Request       = types.Request
-}
+	ChunkSpec     = types.ChunkSpec
+	piecePriority = types.PiecePriority
+)
 
-type Request struct {
+const (
-	Index pp.Integer
+	PiecePriorityNormal    = types.PiecePriorityNormal
-	ChunkSpec
+	PiecePriorityNone      = types.PiecePriorityNone
-}
+	PiecePriorityNow       = types.PiecePriorityNow
-
+	PiecePriorityReadahead = types.PiecePriorityReadahead
-func (r Request) ToMsg(mt pp.MessageType) pp.Message {
+	PiecePriorityNext      = types.PiecePriorityNext
-	return pp.Message{
+	PiecePriorityHigh      = types.PiecePriorityHigh
-		Type:   mt,
+)
-		Index:  r.Index,
-		Begin:  r.Begin,
-		Length: r.Length,
-	}
-}
 
 func newRequest(index, begin, length pp.Integer) Request {
 	return Request{index, ChunkSpec{begin, length}}
@@ -151,6 +149,16 @@ func min(as ...int64) int64 {
 	return ret
 }
 
+func minInt(as ...int) int {
+	ret := as[0]
+	for _, a := range as[1:] {
+		if a < ret {
+			ret = a
+		}
+	}
+	return ret
+}
+
 var unlimited = rate.NewLimiter(rate.Inf, 0)
 
 type (

misc_test.go

@@ -35,7 +35,7 @@ func BenchmarkIterBitmapsDistinct(t *testing.B) {
 		output := iter.ToSlice(iterBitmapsDistinct(&skipCopy, first, second))
 		t.StopTimer()
 		assert.Equal(t, []interface{}{0, 3, 2}, output)
-		assert.Equal(t, []int{1}, skip.ToSortedSlice())
+		assert.Equal(t, []bitmap.BitIndex{1}, skip.ToSortedSlice())
 	}
 }
 

peer-conn-msg-writer.go

@@ -0,0 +1,133 @@
+package torrent
+
+import (
+	"bytes"
+	"io"
+	"time"
+
+	"github.com/anacrolix/chansync"
+	"github.com/anacrolix/log"
+	"github.com/anacrolix/sync"
+
+	pp "github.com/anacrolix/torrent/peer_protocol"
+)
+
+func (pc *PeerConn) startWriter() {
+	w := &pc.messageWriter
+	*w = peerConnMsgWriter{
+		fillWriteBuffer: func() {
+			pc.locker().Lock()
+			defer pc.locker().Unlock()
+			if pc.closed.IsSet() {
+				return
+			}
+			pc.fillWriteBuffer()
+		},
+		closed: &pc.closed,
+		logger: pc.logger,
+		w:      pc.w,
+		keepAlive: func() bool {
+			pc.locker().Lock()
+			defer pc.locker().Unlock()
+			return pc.useful()
+		},
+		writeBuffer: new(bytes.Buffer),
+	}
+	go func() {
+		defer pc.locker().Unlock()
+		defer pc.close()
+		defer pc.locker().Lock()
+		pc.messageWriter.run(time.Minute)
+	}()
+}
+
+type peerConnMsgWriter struct {
+	// Must not be called with the local mutex held, as it will call back into the write method.
+	fillWriteBuffer func()
+	closed          *chansync.SetOnce
+	logger          log.Logger
+	w               io.Writer
+	keepAlive       func() bool
+
+	mu        sync.Mutex
+	writeCond chansync.BroadcastCond
+	// Pointer so we can swap with the "front buffer".
+	writeBuffer *bytes.Buffer
+}
+
+// Routine that writes to the peer. Some of what to write is buffered by
+// activity elsewhere in the Client, and some is determined locally when the
+// connection is writable.
+func (cn *peerConnMsgWriter) run(keepAliveTimeout time.Duration) {
+	var (
+		lastWrite      time.Time = time.Now()
+		keepAliveTimer *time.Timer
+	)
+	keepAliveTimer = time.AfterFunc(keepAliveTimeout, func() {
+		cn.mu.Lock()
+		defer cn.mu.Unlock()
+		if time.Since(lastWrite) >= keepAliveTimeout {
+			cn.writeCond.Broadcast()
+		}
+		keepAliveTimer.Reset(keepAliveTimeout)
+	})
+	cn.mu.Lock()
+	defer cn.mu.Unlock()
+	defer keepAliveTimer.Stop()
+	frontBuf := new(bytes.Buffer)
+	for {
+		if cn.closed.IsSet() {
+			return
+		}
+		if cn.writeBuffer.Len() == 0 {
+			func() {
+				cn.mu.Unlock()
+				defer cn.mu.Lock()
+				cn.fillWriteBuffer()
+			}()
+		}
+		if cn.writeBuffer.Len() == 0 && time.Since(lastWrite) >= keepAliveTimeout && cn.keepAlive() {
+			cn.writeBuffer.Write(pp.Message{Keepalive: true}.MustMarshalBinary())
+			torrent.Add("written keepalives", 1)
+		}
+		if cn.writeBuffer.Len() == 0 {
+			writeCond := cn.writeCond.Signaled()
+			cn.mu.Unlock()
+			select {
+			case <-cn.closed.Done():
+			case <-writeCond:
+			}
+			cn.mu.Lock()
+			continue
+		}
+		// Flip the buffers.
+		frontBuf, cn.writeBuffer = cn.writeBuffer, frontBuf
+		cn.mu.Unlock()
+		n, err := cn.w.Write(frontBuf.Bytes())
+		cn.mu.Lock()
+		if n != 0 {
+			lastWrite = time.Now()
+			keepAliveTimer.Reset(keepAliveTimeout)
+		}
+		if err != nil {
+			cn.logger.WithDefaultLevel(log.Debug).Printf("error writing: %v", err)
+			return
+		}
+		if n != frontBuf.Len() {
+			panic("short write")
+		}
+		frontBuf.Reset()
+	}
+}
+
+func (cn *peerConnMsgWriter) write(msg pp.Message) bool {
+	cn.mu.Lock()
+	defer cn.mu.Unlock()
+	cn.writeBuffer.Write(msg.MustMarshalBinary())
+	cn.writeCond.Broadcast()
+	return !cn.writeBufferFull()
+}
+
+func (cn *peerConnMsgWriter) writeBufferFull() bool {
+	return cn.writeBuffer.Len() >= writeBufferHighWaterLen
+}

peer-impl.go

@@ -8,16 +8,20 @@ import (
 // BitTorrent protocol connections. Some methods are underlined so as to avoid collisions with
 // legacy PeerConn methods.
 type peerImpl interface {
+	onNextRequestStateChanged()
 	updateRequests()
 	writeInterested(interested bool) bool
-	cancel(Request) bool
+
-	// Return true if there's room for more activity.
+	// Neither of these return buffer room anymore, because they're currently both posted. There's
-	request(Request) bool
+	// also PeerConn.writeBufferFull for when/where it matters.
+	_cancel(Request) bool
+	_request(Request) bool
+
 	connectionFlags() string
 	onClose()
-	_postCancel(Request)
 	onGotInfo(*metainfo.Info)
 	drop()
 	String() string
 	connStatusString() string
+	writeBufferFull() bool
 }

peerconn_test.go

@@ -5,7 +5,6 @@ import (
 	"net"
 	"sync"
 	"testing"
-	"time"
 
 	"github.com/anacrolix/missinggo/pubsub"
 	"github.com/bradfitz/iter"
@@ -32,7 +31,7 @@ func TestSendBitfieldThenHave(t *testing.T) {
 	r, w := io.Pipe()
 	//c.r = r
 	c.w = w
-	go c.writer(time.Minute)
+	c.startWriter()
 	c.locker().Lock()
 	c.t._completedPieces.Add(1)
 	c.postBitfield( /*[]bool{false, true, false}*/ )
@@ -98,7 +97,7 @@ func BenchmarkConnectionMainReadLoop(b *testing.B) {
 	ts := &torrentStorage{}
 	t := &Torrent{
 		cl:                cl,
-		storage:           &storage.Torrent{TorrentImpl: ts},
+		storage:           &storage.Torrent{TorrentImpl: storage.TorrentImpl{Piece: ts.Piece, Close: ts.Close}},
 		pieceStateChanges: pubsub.NewPubSub(),
 	}
 	require.NoError(b, t.setInfo(&metainfo.Info{

pexconn_test.go

@@ -21,7 +21,7 @@ func TestPexConnState(t *testing.T) {
 	c := cl.newConnection(nil, false, addr, addr.Network(), "")
 	c.PeerExtensionIDs = make(map[pp.ExtensionName]pp.ExtensionNumber)
 	c.PeerExtensionIDs[pp.ExtensionNamePex] = pexExtendedId
-	c.writerCond.L.Lock()
+	c.messageWriter.mu.Lock()
 	c.setTorrent(torrent)
 	torrent.addPeerConn(c)
 
@@ -36,7 +36,7 @@ func TestPexConnState(t *testing.T) {
 		out = m
 		return true
 	}
-	c.writerCond.Wait()
+	<-c.messageWriter.writeCond.Signaled()
 	c.pex.Share(testWriter)
 	require.True(t, writerCalled)
 	require.EqualValues(t, pp.Extended, out.Type)

piece.go

@@ -11,33 +11,6 @@ import (
 	"github.com/anacrolix/torrent/storage"
 )
 
-// Describes the importance of obtaining a particular piece.
-type piecePriority byte
-
-func (pp *piecePriority) Raise(maybe piecePriority) bool {
-	if maybe > *pp {
-		*pp = maybe
-		return true
-	}
-	return false
-}
-
-// Priority for use in PriorityBitmap
-func (me piecePriority) BitmapPriority() int {
-	return -int(me)
-}
-
-const (
-	PiecePriorityNone      piecePriority = iota // Not wanted. Must be the zero value.
-	PiecePriorityNormal                         // Wanted.
-	PiecePriorityHigh                           // Wanted a lot.
-	PiecePriorityReadahead                      // May be required soon.
-	// Succeeds a piece where a read occurred. Currently the same as Now,
-	// apparently due to issues with caching.
-	PiecePriorityNext
-	PiecePriorityNow // A Reader is reading in this piece. Highest urgency.
-)
-
 type Piece struct {
 	// The completed piece SHA1 hash, from the metainfo "pieces" field.
 	hash  *metainfo.Hash
@@ -55,6 +28,7 @@ type Piece struct {
 
 	publicPieceState PieceState
 	priority         piecePriority
+	availability     int64
 
 	// This can be locked when the Client lock is taken, but probably not vice versa.
 	pendingWritesMutex sync.Mutex
@@ -79,7 +53,7 @@ func (p *Piece) Storage() storage.Piece {
 }
 
 func (p *Piece) pendingChunkIndex(chunkIndex int) bool {
-	return !p._dirtyChunks.Contains(chunkIndex)
+	return !p._dirtyChunks.Contains(bitmap.BitIndex(chunkIndex))
 }
 
 func (p *Piece) pendingChunk(cs ChunkSpec, chunkSize pp.Integer) bool {
@@ -95,12 +69,12 @@ func (p *Piece) numDirtyChunks() pp.Integer {
 }
 
 func (p *Piece) unpendChunkIndex(i int) {
-	p._dirtyChunks.Add(i)
+	p._dirtyChunks.Add(bitmap.BitIndex(i))
 	p.t.tickleReaders()
 }
 
 func (p *Piece) pendChunkIndex(i int) {
-	p._dirtyChunks.Remove(i)
+	p._dirtyChunks.Remove(bitmap.BitIndex(i))
 }
 
 func (p *Piece) numChunks() pp.Integer {
@@ -144,7 +118,7 @@ func (p *Piece) chunkIndexSpec(chunk pp.Integer) ChunkSpec {
 func (p *Piece) chunkIndexRequest(chunkIndex pp.Integer) Request {
 	return Request{
 		pp.Integer(p.index),
-		chunkIndexSpec(chunkIndex, p.length(), p.chunkSize()),
+		p.chunkIndexSpec(chunkIndex),
 	}
 }
 
@@ -221,14 +195,11 @@ func (p *Piece) SetPriority(prio piecePriority) {
 	p.t.updatePiecePriority(p.index)
 }
 
-func (p *Piece) uncachedPriority() (ret piecePriority) {
+func (p *Piece) purePriority() (ret piecePriority) {
-	if p.t.pieceComplete(p.index) || p.t.pieceQueuedForHash(p.index) || p.t.hashingPiece(p.index) {
-		return PiecePriorityNone
-	}
 	for _, f := range p.files {
 		ret.Raise(f.prio)
 	}
-	if p.t.readerNowPieces().Contains(int(p.index)) {
+	if p.t.readerNowPieces().Contains(bitmap.BitIndex(p.index)) {
 		ret.Raise(PiecePriorityNow)
 	}
 	// if t._readerNowPieces.Contains(piece - 1) {
@@ -241,6 +212,13 @@ func (p *Piece) uncachedPriority() (ret piecePriority) {
 	return
 }
 
+func (p *Piece) uncachedPriority() (ret piecePriority) {
+	if p.t.pieceComplete(p.index) || p.t.pieceQueuedForHash(p.index) || p.t.hashingPiece(p.index) {
+		return PiecePriorityNone
+	}
+	return p.purePriority()
+}
+
 // Tells the Client to refetch the completion status from storage, updating priority etc. if
 // necessary. Might be useful if you know the state of the piece data has changed externally.
 func (p *Piece) UpdateCompletion() {
@@ -256,11 +234,7 @@ func (p *Piece) completion() (ret storage.Completion) {
 }
 
 func (p *Piece) allChunksDirty() bool {
-	return p._dirtyChunks.Len() == int(p.numChunks())
+	return p._dirtyChunks.Len() == bitmap.BitRange(p.numChunks())
-}
-
-func (p *Piece) requestStrategyPiece() requestStrategyPiece {
-	return p
 }
 
 func (p *Piece) dirtyChunks() bitmap.Bitmap {
@@ -270,3 +244,15 @@ func (p *Piece) dirtyChunks() bitmap.Bitmap {
 func (p *Piece) State() PieceState {
 	return p.t.PieceState(p.index)
 }
+
+func (p *Piece) iterUndirtiedChunks(f func(cs ChunkSpec) bool) bool {
+	for i := pp.Integer(0); i < p.numChunks(); i++ {
+		if p.chunkIndexDirty(i) {
+			continue
+		}
+		if !f(p.chunkIndexSpec(i)) {
+			return false
+		}
+	}
+	return true
+}

request-strategy-defaults.go

@@ -1,45 +0,0 @@
-package torrent
-
-import (
-	"github.com/anacrolix/missinggo/iter"
-	"github.com/anacrolix/missinggo/v2/bitmap"
-	pp "github.com/anacrolix/torrent/peer_protocol"
-)
-
-// Provides default implementations for requestStrategy methods. Could be embedded, or delegated to.
-type requestStrategyDefaults struct{}
-
-func (requestStrategyDefaults) hooks() requestStrategyHooks {
-	return requestStrategyHooks{
-		sentRequest:    func(Request) {},
-		deletedRequest: func(Request) {},
-	}
-}
-
-func (requestStrategyDefaults) iterUndirtiedChunks(p requestStrategyPiece, f func(ChunkSpec) bool) bool {
-	chunkIndices := p.dirtyChunks().Copy()
-	chunkIndices.FlipRange(0, bitmap.BitIndex(p.numChunks()))
-	return iter.ForPerm(chunkIndices.Len(), func(i int) bool {
-		ci, err := chunkIndices.RB.Select(uint32(i))
-		if err != nil {
-			panic(err)
-		}
-		return f(p.chunkIndexRequest(pp.Integer(ci)).ChunkSpec)
-	})
-}
-
-func (requestStrategyDefaults) nominalMaxRequests(cn requestStrategyConnection) int {
-	return int(
-		max(
-			64,
-			cn.stats().ChunksReadUseful.Int64()-
-				(cn.stats().ChunksRead.Int64()-cn.stats().ChunksReadUseful.Int64())))
-}
-
-func (requestStrategyDefaults) piecePriority(cn requestStrategyConnection, piece pieceIndex, tpp piecePriority, prio int) int {
-	return prio
-}
-
-func (requestStrategyDefaults) shouldRequestWithoutBias(cn requestStrategyConnection) bool {
-	return false
-}

request-strategy.go

@@ -1,223 +0,0 @@
-package torrent
-
-import (
-	"math"
-	"sync"
-	"time"
-
-	"github.com/anacrolix/missinggo/v2/bitmap"
-	"github.com/anacrolix/missinggo/v2/prioritybitmap"
-
-	pp "github.com/anacrolix/torrent/peer_protocol"
-)
-
-type requestStrategyPiece interface {
-	numChunks() pp.Integer
-	dirtyChunks() bitmap.Bitmap
-	chunkIndexRequest(i pp.Integer) Request
-}
-
-type requestStrategyTorrent interface {
-	numConns() int
-	numReaders() int
-	numPieces() int
-	readerPiecePriorities() (now, readahead bitmap.Bitmap)
-	ignorePieces() bitmap.Bitmap
-	pendingPieces() *prioritybitmap.PriorityBitmap
-}
-
-type requestStrategyConnection interface {
-	torrent() requestStrategyTorrent
-	peerPieces() bitmap.Bitmap
-	pieceRequestOrder() *prioritybitmap.PriorityBitmap
-	fastest() bool
-	stats() *ConnStats
-	totalExpectingTime() time.Duration
-	peerMaxRequests() int
-	chunksReceivedWhileExpecting() int64
-}
-
-type requestStrategy interface {
-	iterPendingPieces(requestStrategyConnection, func(pieceIndex) bool) bool
-	iterUndirtiedChunks(requestStrategyPiece, func(ChunkSpec) bool) bool
-	nominalMaxRequests(requestStrategyConnection) int
-	shouldRequestWithoutBias(requestStrategyConnection) bool
-	piecePriority(requestStrategyConnection, pieceIndex, piecePriority, int) int
-	hooks() requestStrategyHooks
-}
-
-type requestStrategyHooks struct {
-	sentRequest    func(Request)
-	deletedRequest func(Request)
-}
-
-type requestStrategyCallbacks interface {
-	requestTimedOut(Request)
-}
-
-type requestStrategyFuzzing struct {
-	requestStrategyDefaults
-}
-
-type requestStrategyFastest struct {
-	requestStrategyDefaults
-}
-
-func newRequestStrategyMaker(rs requestStrategy) requestStrategyMaker {
-	return func(requestStrategyCallbacks, sync.Locker) requestStrategy {
-		return rs
-	}
-}
-
-// The fastest connection downloads strictly in order of priority, while all others adhere to their
-// piece inclinations.
-func RequestStrategyFastest() requestStrategyMaker {
-	return newRequestStrategyMaker(requestStrategyFastest{})
-}
-
-// Favour higher priority pieces with some fuzzing to reduce overlaps and wastage across
-// connections.
-func RequestStrategyFuzzing() requestStrategyMaker {
-	return newRequestStrategyMaker(requestStrategyFuzzing{})
-}
-
-func (requestStrategyFastest) shouldRequestWithoutBias(cn requestStrategyConnection) bool {
-	if cn.torrent().numReaders() == 0 {
-		return false
-	}
-	if cn.torrent().numConns() == 1 {
-		return true
-	}
-	if cn.fastest() {
-		return true
-	}
-	return false
-}
-
-type requestStrategyDuplicateRequestTimeout struct {
-	requestStrategyDefaults
-	// How long to avoid duplicating a pending request.
-	duplicateRequestTimeout time.Duration
-
-	callbacks requestStrategyCallbacks
-
-	// The last time we requested a chunk. Deleting the request from any connection will clear this
-	// value.
-	lastRequested map[Request]*time.Timer
-	// The lock to take when running a request timeout handler.
-	timeoutLocker sync.Locker
-}
-
-// Generates a request strategy instance for a given torrent. callbacks are probably specific to the torrent.
-type requestStrategyMaker func(callbacks requestStrategyCallbacks, clientLocker sync.Locker) requestStrategy
-
-// Requests are strictly by piece priority, and not duplicated until duplicateRequestTimeout is
-// reached.
-func RequestStrategyDuplicateRequestTimeout(duplicateRequestTimeout time.Duration) requestStrategyMaker {
-	return func(callbacks requestStrategyCallbacks, clientLocker sync.Locker) requestStrategy {
-		return requestStrategyDuplicateRequestTimeout{
-			duplicateRequestTimeout: duplicateRequestTimeout,
-			callbacks:               callbacks,
-			lastRequested:           make(map[Request]*time.Timer),
-			timeoutLocker:           clientLocker,
-		}
-	}
-}
-
-func (rs requestStrategyDuplicateRequestTimeout) hooks() requestStrategyHooks {
-	return requestStrategyHooks{
-		deletedRequest: func(r Request) {
-			if t, ok := rs.lastRequested[r]; ok {
-				t.Stop()
-				delete(rs.lastRequested, r)
-			}
-		},
-		sentRequest: rs.onSentRequest,
-	}
-}
-
-func (rs requestStrategyDuplicateRequestTimeout) iterUndirtiedChunks(p requestStrategyPiece, f func(ChunkSpec) bool) bool {
-	for i := pp.Integer(0); i < pp.Integer(p.numChunks()); i++ {
-		if p.dirtyChunks().Get(bitmap.BitIndex(i)) {
-			continue
-		}
-		r := p.chunkIndexRequest(i)
-		if rs.wouldDuplicateRecent(r) {
-			continue
-		}
-		if !f(r.ChunkSpec) {
-			return false
-		}
-	}
-	return true
-}
-
-func (requestStrategyFuzzing) piecePriority(cn requestStrategyConnection, piece pieceIndex, tpp piecePriority, prio int) int {
-	switch tpp {
-	case PiecePriorityNormal:
-	case PiecePriorityReadahead:
-		prio -= int(cn.torrent().numPieces())
-	case PiecePriorityNext, PiecePriorityNow:
-		prio -= 2 * int(cn.torrent().numPieces())
-	default:
-		panic(tpp)
-	}
-	prio += int(piece / 3)
-	return prio
-}
-
-func (requestStrategyDuplicateRequestTimeout) iterPendingPieces(cn requestStrategyConnection, f func(pieceIndex) bool) bool {
-	return iterUnbiasedPieceRequestOrder(cn, f)
-}
-func defaultIterPendingPieces(rs requestStrategy, cn requestStrategyConnection, f func(pieceIndex) bool) bool {
-	if rs.shouldRequestWithoutBias(cn) {
-		return iterUnbiasedPieceRequestOrder(cn, f)
-	} else {
-		return cn.pieceRequestOrder().IterTyped(func(i int) bool {
-			return f(pieceIndex(i))
-		})
-	}
-}
-func (rs requestStrategyFuzzing) iterPendingPieces(cn requestStrategyConnection, cb func(pieceIndex) bool) bool {
-	return defaultIterPendingPieces(rs, cn, cb)
-}
-func (rs requestStrategyFastest) iterPendingPieces(cn requestStrategyConnection, cb func(pieceIndex) bool) bool {
-	return defaultIterPendingPieces(rs, cn, cb)
-}
-
-func (rs requestStrategyDuplicateRequestTimeout) onSentRequest(r Request) {
-	rs.lastRequested[r] = time.AfterFunc(rs.duplicateRequestTimeout, func() {
-		rs.timeoutLocker.Lock()
-		delete(rs.lastRequested, r)
-		rs.timeoutLocker.Unlock()
-		rs.callbacks.requestTimedOut(r)
-	})
-}
-
-// The actual value to use as the maximum outbound requests.
-func (rs requestStrategyDuplicateRequestTimeout) nominalMaxRequests(cn requestStrategyConnection) (ret int) {
-	expectingTime := int64(cn.totalExpectingTime())
-	if expectingTime == 0 {
-		expectingTime = math.MaxInt64
-	} else {
-		expectingTime *= 2
-	}
-	return int(clamp(
-		1,
-		int64(cn.peerMaxRequests()),
-		max(
-			// It makes sense to always pipeline at least one connection, since latency must be
-			// non-zero.
-			2,
-			// Request only as many as we expect to receive in the duplicateRequestTimeout
-			// window. We are trying to avoid having to duplicate requests.
-			cn.chunksReceivedWhileExpecting()*int64(rs.duplicateRequestTimeout)/expectingTime,
-		),
-	))
-}
-func (rs requestStrategyDuplicateRequestTimeout) wouldDuplicateRecent(r Request) bool {
-	// This piece has been requested on another connection, and the duplicate request timer is still
-	// running.
-	_, ok := rs.lastRequested[r]
-	return ok
-}

request-strategy/order.go

@@ -0,0 +1,349 @@
+package request_strategy
+
+import (
+	"fmt"
+	"sort"
+
+	"github.com/anacrolix/multiless"
+
+	pp "github.com/anacrolix/torrent/peer_protocol"
+	"github.com/anacrolix/torrent/types"
+)
+
+type (
+	Request       = types.Request
+	pieceIndex    = types.PieceIndex
+	piecePriority = types.PiecePriority
+	// This can be made into a type-param later, will be great for testing.
+	ChunkSpec = types.ChunkSpec
+)
+
+type ClientPieceOrder struct{}
+
+type filterTorrent struct {
+	Torrent
+	unverifiedBytes int64
+	// Potentially shared with other torrents.
+	storageLeft *int64
+}
+
+func sortFilterPieces(pieces []filterPiece) {
+	sort.Slice(pieces, func(_i, _j int) bool {
+		i := pieces[_i]
+		j := pieces[_j]
+		return multiless.New().Int(
+			int(j.Priority), int(i.Priority),
+		).Bool(
+			j.Partial, i.Partial,
+		).Int64(
+			i.Availability, j.Availability,
+		).Int(
+			i.index, j.index,
+		).Uintptr(
+			i.t.StableId, j.t.StableId,
+		).MustLess()
+	})
+}
+
+type requestsPeer struct {
+	Peer
+	nextState                  PeerNextRequestState
+	requestablePiecesRemaining int
+}
+
+func (rp *requestsPeer) canFitRequest() bool {
+	return len(rp.nextState.Requests) < rp.MaxRequests
+}
+
+func (rp *requestsPeer) addNextRequest(r Request) {
+	_, ok := rp.nextState.Requests[r]
+	if ok {
+		panic("should only add once")
+	}
+	rp.nextState.Requests[r] = struct{}{}
+}
+
+type peersForPieceRequests struct {
+	requestsInPiece int
+	*requestsPeer
+}
+
+func (me *peersForPieceRequests) addNextRequest(r Request) {
+	me.requestsPeer.addNextRequest(r)
+	me.requestsInPiece++
+}
+
+type requestablePiece struct {
+	index             pieceIndex
+	t                 Torrent
+	alwaysReallocate  bool
+	NumPendingChunks  int
+	IterPendingChunks ChunksIter
+}
+
+type filterPiece struct {
+	t     *filterTorrent
+	index pieceIndex
+	Piece
+}
+
+func getRequestablePieces(input Input) (ret []requestablePiece) {
+	// Storage capacity left for this run, keyed by the storage capacity pointer on the storage
+	// TorrentImpl.
+	storageLeft := make(map[*func() *int64]*int64)
+	var pieces []filterPiece
+	for _, _t := range input.Torrents {
+		// TODO: We could do metainfo requests here.
+		t := &filterTorrent{
+			Torrent:         _t,
+			unverifiedBytes: 0,
+		}
+		key := t.Capacity
+		if key != nil {
+			if _, ok := storageLeft[key]; !ok {
+				storageLeft[key] = (*key)()
+			}
+			t.storageLeft = storageLeft[key]
+		}
+		for i, tp := range t.Pieces {
+			pieces = append(pieces, filterPiece{
+				t:     t,
+				index: i,
+				Piece: tp,
+			})
+		}
+	}
+	sortFilterPieces(pieces)
+	var allTorrentsUnverifiedBytes int64
+	for _, piece := range pieces {
+		if left := piece.t.storageLeft; left != nil {
+			if *left < int64(piece.Length) {
+				continue
+			}
+			*left -= int64(piece.Length)
+		}
+		if !piece.Request || piece.NumPendingChunks == 0 {
+			// TODO: Clarify exactly what is verified. Stuff that's being hashed should be
+			// considered unverified and hold up further requests.
+			continue
+		}
+		if piece.t.MaxUnverifiedBytes != 0 && piece.t.unverifiedBytes+piece.Length > piece.t.MaxUnverifiedBytes {
+			continue
+		}
+		if input.MaxUnverifiedBytes != 0 && allTorrentsUnverifiedBytes+piece.Length > input.MaxUnverifiedBytes {
+			continue
+		}
+		piece.t.unverifiedBytes += piece.Length
+		allTorrentsUnverifiedBytes += piece.Length
+		ret = append(ret, requestablePiece{
+			index:             piece.index,
+			t:                 piece.t.Torrent,
+			NumPendingChunks:  piece.NumPendingChunks,
+			IterPendingChunks: piece.iterPendingChunksWrapper,
+			alwaysReallocate:  piece.Priority >= types.PiecePriorityNext,
+		})
+	}
+	return
+}
+
+type Input struct {
+	Torrents           []Torrent
+	MaxUnverifiedBytes int64
+}
+
+// TODO: We could do metainfo requests here.
+func Run(input Input) map[PeerId]PeerNextRequestState {
+	requestPieces := getRequestablePieces(input)
+	torrents := input.Torrents
+	allPeers := make(map[uintptr][]*requestsPeer, len(torrents))
+	for _, t := range torrents {
+		peers := make([]*requestsPeer, 0, len(t.Peers))
+		for _, p := range t.Peers {
+			peers = append(peers, &requestsPeer{
+				Peer: p,
+				nextState: PeerNextRequestState{
+					Requests: make(map[Request]struct{}, p.MaxRequests),
+				},
+			})
+		}
+		allPeers[t.StableId] = peers
+	}
+	for _, piece := range requestPieces {
+		for _, peer := range allPeers[piece.t.StableId] {
+			if peer.canRequestPiece(piece.index) {
+				peer.requestablePiecesRemaining++
+			}
+		}
+	}
+	for _, piece := range requestPieces {
+		allocatePendingChunks(piece, allPeers[piece.t.StableId])
+	}
+	ret := make(map[PeerId]PeerNextRequestState)
+	for _, peers := range allPeers {
+		for _, rp := range peers {
+			if rp.requestablePiecesRemaining != 0 {
+				panic(rp.requestablePiecesRemaining)
+			}
+			if _, ok := ret[rp.Id]; ok {
+				panic(fmt.Sprintf("duplicate peer id: %v", rp.Id))
+			}
+			ret[rp.Id] = rp.nextState
+		}
+	}
+	return ret
+}
+
+// Checks that a sorted peersForPiece slice makes sense.
+func ensureValidSortedPeersForPieceRequests(peers []*peersForPieceRequests, sortLess func(_, _ int) bool) {
+	if !sort.SliceIsSorted(peers, sortLess) {
+		panic("not sorted")
+	}
+	peerMap := make(map[*peersForPieceRequests]struct{}, len(peers))
+	for _, p := range peers {
+		if _, ok := peerMap[p]; ok {
+			panic(p)
+		}
+		peerMap[p] = struct{}{}
+	}
+}
+
+func allocatePendingChunks(p requestablePiece, peers []*requestsPeer) {
+	peersForPiece := make([]*peersForPieceRequests, 0, len(peers))
+	for _, peer := range peers {
+		peersForPiece = append(peersForPiece, &peersForPieceRequests{
+			requestsInPiece: 0,
+			requestsPeer:    peer,
+		})
+	}
+	defer func() {
+		for _, peer := range peersForPiece {
+			if peer.canRequestPiece(p.index) {
+				peer.requestablePiecesRemaining--
+			}
+		}
+	}()
+	sortPeersForPiece := func(req *Request) {
+		less := func(i, j int) bool {
+			byHasRequest := func() multiless.Computation {
+				ml := multiless.New()
+				if req != nil {
+					_, iHas := peersForPiece[i].nextState.Requests[*req]
+					_, jHas := peersForPiece[j].nextState.Requests[*req]
+					ml = ml.Bool(jHas, iHas)
+				}
+				return ml
+			}()
+			ml := multiless.New()
+			// We always "reallocate", that is force even striping amongst peers that are either on
+			// the last piece they can contribute too, or for pieces marked for this behaviour.
+			// Striping prevents starving peers of requests, and will always re-balance to the
+			// fastest known peers.
+			if !p.alwaysReallocate {
+				ml = ml.Bool(
+					peersForPiece[j].requestablePiecesRemaining == 1,
+					peersForPiece[i].requestablePiecesRemaining == 1)
+			}
+			if p.alwaysReallocate || peersForPiece[j].requestablePiecesRemaining == 1 {
+				ml = ml.Int(
+					peersForPiece[i].requestsInPiece,
+					peersForPiece[j].requestsInPiece)
+			} else {
+				ml = ml.AndThen(byHasRequest)
+			}
+			ml = ml.Int(
+				peersForPiece[i].requestablePiecesRemaining,
+				peersForPiece[j].requestablePiecesRemaining,
+			).Float64(
+				peersForPiece[j].DownloadRate,
+				peersForPiece[i].DownloadRate,
+			)
+			ml = ml.AndThen(byHasRequest)
+			return ml.Int64(
+				int64(peersForPiece[j].Age), int64(peersForPiece[i].Age),
+				// TODO: Probably peer priority can come next
+			).Uintptr(
+				peersForPiece[i].Id.Uintptr(),
+				peersForPiece[j].Id.Uintptr(),
+			).MustLess()
+		}
+		sort.Slice(peersForPiece, less)
+		ensureValidSortedPeersForPieceRequests(peersForPiece, less)
+	}
+	// Chunks can be preassigned several times, if peers haven't been able to update their "actual"
+	// with "next" request state before another request strategy run occurs.
+	preallocated := make(map[ChunkSpec][]*peersForPieceRequests, p.NumPendingChunks)
+	p.IterPendingChunks(func(spec ChunkSpec) {
+		req := Request{pp.Integer(p.index), spec}
+		for _, peer := range peersForPiece {
+			if h := peer.HasExistingRequest; h == nil || !h(req) {
+				continue
+			}
+			if !peer.canFitRequest() {
+				continue
+			}
+			if !peer.canRequestPiece(p.index) {
+				continue
+			}
+			preallocated[spec] = append(preallocated[spec], peer)
+			peer.addNextRequest(req)
+		}
+	})
+	pendingChunksRemaining := int(p.NumPendingChunks)
+	p.IterPendingChunks(func(chunk types.ChunkSpec) {
+		if _, ok := preallocated[chunk]; ok {
+			return
+		}
+		req := Request{pp.Integer(p.index), chunk}
+		defer func() { pendingChunksRemaining-- }()
+		sortPeersForPiece(nil)
+		for _, peer := range peersForPiece {
+			if !peer.canFitRequest() {
+				continue
+			}
+			if !peer.HasPiece(p.index) {
+				continue
+			}
+			if !peer.pieceAllowedFastOrDefault(p.index) {
+				// TODO: Verify that's okay to stay uninterested if we request allowed fast pieces.
+				peer.nextState.Interested = true
+				if peer.Choking {
+					continue
+				}
+			}
+			peer.addNextRequest(req)
+			break
+		}
+	})
+chunk:
+	for chunk, prePeers := range preallocated {
+		pendingChunksRemaining--
+		req := Request{pp.Integer(p.index), chunk}
+		for _, pp := range prePeers {
+			pp.requestsInPiece--
+		}
+		sortPeersForPiece(&req)
+		for _, pp := range prePeers {
+			delete(pp.nextState.Requests, req)
+		}
+		for _, peer := range peersForPiece {
+			if !peer.canFitRequest() {
+				continue
+			}
+			if !peer.HasPiece(p.index) {
+				continue
+			}
+			if !peer.pieceAllowedFastOrDefault(p.index) {
+				// TODO: Verify that's okay to stay uninterested if we request allowed fast pieces.
+				peer.nextState.Interested = true
+				if peer.Choking {
+					continue
+				}
+			}
+			peer.addNextRequest(req)
+			continue chunk
+		}
+	}
+	if pendingChunksRemaining != 0 {
+		panic(pendingChunksRemaining)
+	}
+}

request-strategy/order_test.go

@@ -0,0 +1,297 @@
+package request_strategy
+
+import (
+	"math"
+	"testing"
+
+	"github.com/bradfitz/iter"
+	qt "github.com/frankban/quicktest"
+
+	pp "github.com/anacrolix/torrent/peer_protocol"
+)
+
+func r(i pieceIndex, begin int) Request {
+	return Request{pp.Integer(i), ChunkSpec{pp.Integer(begin), 1}}
+}
+
+func chunkIterRange(end int) func(func(ChunkSpec)) {
+	return func(f func(ChunkSpec)) {
+		for offset := range iter.N(end) {
+			f(ChunkSpec{pp.Integer(offset), 1})
+		}
+	}
+}
+
+func chunkIter(offsets ...int) func(func(ChunkSpec)) {
+	return func(f func(ChunkSpec)) {
+		for _, offset := range offsets {
+			f(ChunkSpec{pp.Integer(offset), 1})
+		}
+	}
+}
+
+func requestSetFromSlice(rs ...Request) (ret map[Request]struct{}) {
+	ret = make(map[Request]struct{}, len(rs))
+	for _, r := range rs {
+		ret[r] = struct{}{}
+	}
+	return
+}
+
+type intPeerId int
+
+func (i intPeerId) Uintptr() uintptr {
+	return uintptr(i)
+}
+
+func TestStealingFromSlowerPeer(t *testing.T) {
+	c := qt.New(t)
+	basePeer := Peer{
+		HasPiece: func(i pieceIndex) bool {
+			return true
+		},
+		MaxRequests:  math.MaxInt16,
+		DownloadRate: 2,
+	}
+	// Slower than the stealers, but has all requests already.
+	stealee := basePeer
+	stealee.DownloadRate = 1
+	stealee.HasExistingRequest = func(r Request) bool {
+		return true
+	}
+	stealee.Id = intPeerId(1)
+	firstStealer := basePeer
+	firstStealer.Id = intPeerId(2)
+	secondStealer := basePeer
+	secondStealer.Id = intPeerId(3)
+	results := Run(Input{Torrents: []Torrent{{
+		Pieces: []Piece{{
+			Request:           true,
+			NumPendingChunks:  5,
+			IterPendingChunks: chunkIterRange(5),
+		}},
+		Peers: []Peer{
+			stealee,
+			firstStealer,
+			secondStealer,
+		},
+	}}})
+
+	c.Assert(results, qt.HasLen, 3)
+	check := func(p PeerId, l int) {
+		c.Check(results[p].Requests, qt.HasLen, l)
+		c.Check(results[p].Interested, qt.Equals, l > 0)
+	}
+	check(stealee.Id, 1)
+	check(firstStealer.Id, 2)
+	check(secondStealer.Id, 2)
+}
+
+func checkNumRequestsAndInterest(c *qt.C, next PeerNextRequestState, num int, interest bool) {
+	c.Check(next.Requests, qt.HasLen, num)
+	c.Check(next.Interested, qt.Equals, interest)
+}
+
+func TestStealingFromSlowerPeersBasic(t *testing.T) {
+	c := qt.New(t)
+	basePeer := Peer{
+		HasPiece: func(i pieceIndex) bool {
+			return true
+		},
+		MaxRequests:  math.MaxInt16,
+		DownloadRate: 2,
+	}
+	stealee := basePeer
+	stealee.DownloadRate = 1
+	stealee.HasExistingRequest = func(r Request) bool {
+		return true
+	}
+	stealee.Id = intPeerId(1)
+	firstStealer := basePeer
+	firstStealer.Id = intPeerId(2)
+	secondStealer := basePeer
+	secondStealer.Id = intPeerId(3)
+	results := Run(Input{Torrents: []Torrent{{
+		Pieces: []Piece{{
+			Request:           true,
+			NumPendingChunks:  2,
+			IterPendingChunks: chunkIter(0, 1),
+		}},
+		Peers: []Peer{
+			stealee,
+			firstStealer,
+			secondStealer,
+		},
+	}}})
+
+	checkNumRequestsAndInterest(c, results[firstStealer.Id], 1, true)
+	checkNumRequestsAndInterest(c, results[secondStealer.Id], 1, true)
+	checkNumRequestsAndInterest(c, results[stealee.Id], 0, false)
+}
+
+func TestPeerKeepsExistingIfReasonable(t *testing.T) {
+	c := qt.New(t)
+	basePeer := Peer{
+		HasPiece: func(i pieceIndex) bool {
+			return true
+		},
+		MaxRequests:  math.MaxInt16,
+		DownloadRate: 2,
+	}
+	// Slower than the stealers, but has all requests already.
+	stealee := basePeer
+	stealee.DownloadRate = 1
+	keepReq := r(0, 0)
+	stealee.HasExistingRequest = func(r Request) bool {
+		return r == keepReq
+	}
+	stealee.Id = intPeerId(1)
+	firstStealer := basePeer
+	firstStealer.Id = intPeerId(2)
+	secondStealer := basePeer
+	secondStealer.Id = intPeerId(3)
+	results := Run(Input{Torrents: []Torrent{{
+		Pieces: []Piece{{
+			Request:           true,
+			NumPendingChunks:  4,
+			IterPendingChunks: chunkIter(0, 1, 3, 4),
+		}},
+		Peers: []Peer{
+			stealee,
+			firstStealer,
+			secondStealer,
+		},
+	}}})
+
+	c.Assert(results, qt.HasLen, 3)
+	check := func(p PeerId, l int) {
+		c.Check(results[p].Requests, qt.HasLen, l)
+		c.Check(results[p].Interested, qt.Equals, l > 0)
+	}
+	check(firstStealer.Id, 2)
+	check(secondStealer.Id, 1)
+	c.Check(results[stealee.Id], qt.ContentEquals, PeerNextRequestState{
+		Interested: true,
+		Requests:   requestSetFromSlice(keepReq),
+	})
+}
+
+func TestDontStealUnnecessarily(t *testing.T) {
+	c := qt.New(t)
+	basePeer := Peer{
+		HasPiece: func(i pieceIndex) bool {
+			return true
+		},
+		MaxRequests:  math.MaxInt16,
+		DownloadRate: 2,
+	}
+	// Slower than the stealers, but has all requests already.
+	stealee := basePeer
+	stealee.DownloadRate = 1
+	keepReqs := requestSetFromSlice(
+		r(3, 2), r(3, 4), r(3, 6), r(3, 8),
+		r(4, 0), r(4, 1), r(4, 7), r(4, 8))
+	stealee.HasExistingRequest = func(r Request) bool {
+		_, ok := keepReqs[r]
+		return ok
+	}
+	stealee.Id = intPeerId(1)
+	firstStealer := basePeer
+	firstStealer.Id = intPeerId(2)
+	secondStealer := basePeer
+	secondStealer.Id = intPeerId(3)
+	secondStealer.HasPiece = func(i pieceIndex) bool {
+		switch i {
+		case 1, 3:
+			return true
+		default:
+			return false
+		}
+	}
+	results := Run(Input{Torrents: []Torrent{{
+		Pieces: []Piece{
+			{
+				Request:           true,
+				NumPendingChunks:  0,
+				IterPendingChunks: chunkIterRange(9),
+			},
+			{
+				Request:           true,
+				NumPendingChunks:  7,
+				IterPendingChunks: chunkIterRange(7),
+			},
+			{
+				Request:           true,
+				NumPendingChunks:  0,
+				IterPendingChunks: chunkIterRange(0),
+			},
+			{
+				Request:           true,
+				NumPendingChunks:  9,
+				IterPendingChunks: chunkIterRange(9),
+			},
+			{
+				Request:           true,
+				NumPendingChunks:  9,
+				IterPendingChunks: chunkIterRange(9),
+			}},
+		Peers: []Peer{
+			firstStealer,
+			stealee,
+			secondStealer,
+		},
+	}}})
+
+	c.Assert(results, qt.HasLen, 3)
+	check := func(p PeerId, l int) {
+		c.Check(results[p].Requests, qt.HasLen, l)
+		c.Check(results[p].Interested, qt.Equals, l > 0)
+	}
+	check(firstStealer.Id, 5)
+	check(secondStealer.Id, 7+9)
+	c.Check(results[stealee.Id], qt.ContentEquals, PeerNextRequestState{
+		Interested: true,
+		Requests:   requestSetFromSlice(r(4, 0), r(4, 1), r(4, 7), r(4, 8)),
+	})
+}
+
+// This tests a situation where multiple peers had the same existing request, due to "actual" and
+// "next" request states being out of sync. This reasonable occurs when a peer hasn't fully updated
+// its actual request state since the last request strategy run.
+func TestDuplicatePreallocations(t *testing.T) {
+	peer := func(id int, downloadRate float64) Peer {
+		return Peer{
+			HasExistingRequest: func(r Request) bool {
+				return true
+			},
+			MaxRequests: 2,
+			HasPiece: func(i pieceIndex) bool {
+				return true
+			},
+			Id:           intPeerId(id),
+			DownloadRate: downloadRate,
+		}
+	}
+	results := Run(Input{
+		Torrents: []Torrent{{
+			Pieces: []Piece{{
+				Request:           true,
+				NumPendingChunks:  1,
+				IterPendingChunks: chunkIterRange(1),
+			}, {
+				Request:           true,
+				NumPendingChunks:  1,
+				IterPendingChunks: chunkIterRange(1),
+			}},
+			Peers: []Peer{
+				// The second peer was be marked as the preallocation, clobbering the first. The
+				// first peer is preferred, and the piece isn't striped, so it gets preallocated a
+				// request, and then gets reallocated from the peer the same request.
+				peer(1, 2),
+				peer(2, 1),
+			},
+		}},
+	})
+	c := qt.New(t)
+	c.Assert(2, qt.Equals, len(results[intPeerId(1)].Requests)+len(results[intPeerId(2)].Requests))
+}

request-strategy/peer.go

@@ -0,0 +1,38 @@
+package request_strategy
+
+import (
+	"time"
+)
+
+type PeerNextRequestState struct {
+	Interested bool
+	Requests   map[Request]struct{}
+}
+
+type PeerId interface {
+	Uintptr() uintptr
+}
+
+type Peer struct {
+	HasPiece           func(i pieceIndex) bool
+	MaxRequests        int
+	HasExistingRequest func(r Request) bool
+	Choking            bool
+	PieceAllowedFast   func(pieceIndex) bool
+	DownloadRate       float64
+	Age                time.Duration
+	// This is passed back out at the end, so must support equality. Could be a type-param later.
+	Id PeerId
+}
+
+func (p *Peer) pieceAllowedFastOrDefault(i pieceIndex) bool {
+	if f := p.PieceAllowedFast; f != nil {
+		return f(i)
+	}
+	return false
+}
+
+// TODO: This might be used in more places I think.
+func (p *Peer) canRequestPiece(i pieceIndex) bool {
+	return p.HasPiece(i) && (!p.Choking || (p.PieceAllowedFast != nil && p.PieceAllowedFast(i)))
+}

request-strategy/piece.go

@@ -0,0 +1,24 @@
+package request_strategy
+
+import (
+	"github.com/anacrolix/torrent/types"
+)
+
+type ChunksIter func(func(types.ChunkSpec))
+
+type Piece struct {
+	Request           bool
+	Priority          piecePriority
+	Partial           bool
+	Availability      int64
+	Length            int64
+	NumPendingChunks  int
+	IterPendingChunks ChunksIter
+}
+
+func (p Piece) iterPendingChunksWrapper(f func(ChunkSpec)) {
+	i := p.IterPendingChunks
+	if i != nil {
+		i(f)
+	}
+}

request-strategy/torrent.go

@@ -0,0 +1,11 @@
+package request_strategy
+
+type Torrent struct {
+	Pieces   []Piece
+	Capacity *func() *int64
+	Peers    []Peer // not closed.
+	// Some value that's unique and stable between runs. Could even use the infohash?
+	StableId uintptr
+
+	MaxUnverifiedBytes int64
+}

requesting.go

@@ -0,0 +1,136 @@
+package torrent
+
+import (
+	"time"
+	"unsafe"
+
+	"github.com/anacrolix/missinggo/v2/bitmap"
+
+	"github.com/anacrolix/chansync"
+	request_strategy "github.com/anacrolix/torrent/request-strategy"
+	"github.com/anacrolix/torrent/types"
+)
+
+func (cl *Client) requester() {
+	for {
+		update := func() chansync.Signaled {
+			cl.lock()
+			defer cl.unlock()
+			cl.doRequests()
+			return cl.updateRequests.Signaled()
+		}()
+		// We can probably tune how often to heed this signal. TODO: Currently disabled to retain
+		// existing behaviour, while the signalling is worked out.
+		update = nil
+		select {
+		case <-cl.closed.LockedChan(cl.locker()):
+			return
+		case <-update:
+		case <-time.After(100 * time.Millisecond):
+		}
+	}
+}
+
+func (cl *Client) tickleRequester() {
+	cl.updateRequests.Broadcast()
+}
+
+func (cl *Client) doRequests() {
+	ts := make([]request_strategy.Torrent, 0, len(cl.torrents))
+	for _, t := range cl.torrents {
+		rst := request_strategy.Torrent{
+			StableId: uintptr(unsafe.Pointer(t)),
+		}
+		if t.storage != nil {
+			rst.Capacity = t.storage.Capacity
+		}
+		for i := range t.pieces {
+			p := &t.pieces[i]
+			rst.Pieces = append(rst.Pieces, request_strategy.Piece{
+				Request:          !t.ignorePieceForRequests(i),
+				Priority:         p.purePriority(),
+				Partial:          t.piecePartiallyDownloaded(i),
+				Availability:     p.availability,
+				Length:           int64(p.length()),
+				NumPendingChunks: int(t.pieceNumPendingChunks(i)),
+				IterPendingChunks: func(f func(types.ChunkSpec)) {
+					p.iterUndirtiedChunks(func(cs ChunkSpec) bool {
+						f(cs)
+						return true
+					})
+				},
+			})
+		}
+		t.iterPeers(func(p *Peer) {
+			if p.closed.IsSet() {
+				return
+			}
+			if p.piecesReceivedSinceLastRequestUpdate > p.maxPiecesReceivedBetweenRequestUpdates {
+				p.maxPiecesReceivedBetweenRequestUpdates = p.piecesReceivedSinceLastRequestUpdate
+			}
+			p.piecesReceivedSinceLastRequestUpdate = 0
+			rst.Peers = append(rst.Peers, request_strategy.Peer{
+				HasPiece:    p.peerHasPiece,
+				MaxRequests: p.nominalMaxRequests(),
+				HasExistingRequest: func(r request_strategy.Request) bool {
+					_, ok := p.actualRequestState.Requests[r]
+					return ok
+				},
+				Choking: p.peerChoking,
+				PieceAllowedFast: func(i pieceIndex) bool {
+					return p.peerAllowedFast.Contains(bitmap.BitIndex(i))
+				},
+				DownloadRate: p.downloadRate(),
+				Age:          time.Since(p.completedHandshake),
+				Id:           (*peerId)(p),
+			})
+		})
+		ts = append(ts, rst)
+	}
+	nextPeerStates := request_strategy.Run(request_strategy.Input{
+		Torrents:           ts,
+		MaxUnverifiedBytes: cl.config.MaxUnverifiedBytes,
+	})
+	for p, state := range nextPeerStates {
+		setPeerNextRequestState(p, state)
+	}
+}
+
+type peerId Peer
+
+func (p *peerId) Uintptr() uintptr {
+	return uintptr(unsafe.Pointer(p))
+}
+
+func setPeerNextRequestState(_p request_strategy.PeerId, rp request_strategy.PeerNextRequestState) {
+	p := (*Peer)(_p.(*peerId))
+	p.nextRequestState = rp
+	p.onNextRequestStateChanged()
+}
+
+func (p *Peer) applyNextRequestState() bool {
+	next := p.nextRequestState
+	current := p.actualRequestState
+	if !p.setInterested(next.Interested) {
+		return false
+	}
+	for req := range current.Requests {
+		if _, ok := next.Requests[req]; !ok {
+			if !p.cancel(req) {
+				return false
+			}
+		}
+	}
+	for req := range next.Requests {
+		more, err := p.request(req)
+		if err != nil {
+			panic(err)
+		} else {
+			//log.Print(req)
+		}
+		if !more {
+			return false
+		}
+	}
+	return true
+}

storage/bolt.go

@@ -43,7 +43,11 @@ func (me *boltClient) Close() error {
 }
 
 func (me *boltClient) OpenTorrent(info *metainfo.Info, infoHash metainfo.Hash) (TorrentImpl, error) {
-	return &boltTorrent{me, infoHash}, nil
+	t := &boltTorrent{me, infoHash}
+	return TorrentImpl{
+		Piece: t.Piece,
+		Close: t.Close,
+	}, nil
 }
 
 func (me *boltTorrent) Piece(p metainfo.Piece) PieceImpl {

storage/disabled/disabled.go

@@ -0,0 +1,58 @@
+package disabled
+
+import (
+	"errors"
+
+	"github.com/anacrolix/torrent/metainfo"
+	"github.com/anacrolix/torrent/storage"
+)
+
+type Client struct{}
+
+var capacity int64
+
+func (c Client) OpenTorrent(info *metainfo.Info, infoHash metainfo.Hash) (storage.TorrentImpl, error) {
+	capFunc := func() *int64 {
+		return &capacity
+	}
+	return storage.TorrentImpl{
+		Piece: func(piece metainfo.Piece) storage.PieceImpl {
+			return Piece{}
+		},
+		Close: func() error {
+			return nil
+		},
+		Capacity: &capFunc,
+	}, nil
+}
+
+func (c Client) capacity() *int64 {
+	return &capacity
+}
+
+type Piece struct{}
+
+func (Piece) ReadAt(p []byte, off int64) (n int, err error) {
+	err = errors.New("disabled")
+	return
+}
+
+func (Piece) WriteAt(p []byte, off int64) (n int, err error) {
+	err = errors.New("disabled")
+	return
+}
+
+func (Piece) MarkComplete() error {
+	return errors.New("disabled")
+}
+
+func (Piece) MarkNotComplete() error {
+	return errors.New("disabled")
+}
+
+func (Piece) Completion() storage.Completion {
+	return storage.Completion{
+		Complete: false,
+		Ok:       true,
+	}
+}

storage/file.go

@@ -67,14 +67,16 @@ func (me *fileClientImpl) Close() error {
 	return me.pc.Close()
 }
 
-func (fs *fileClientImpl) OpenTorrent(info *metainfo.Info, infoHash metainfo.Hash) (TorrentImpl, error) {
+func (fs *fileClientImpl) OpenTorrent(info *metainfo.Info, infoHash metainfo.Hash) (_ TorrentImpl, err error) {
 	dir := fs.pathMaker(fs.baseDir, info, infoHash)
 	upvertedFiles := info.UpvertedFiles()
 	files := make([]file, 0, len(upvertedFiles))
 	for i, fileInfo := range upvertedFiles {
-		s, err := ToSafeFilePath(append([]string{info.Name}, fileInfo.Path...)...)
+		var s string
+		s, err = ToSafeFilePath(append([]string{info.Name}, fileInfo.Path...)...)
 		if err != nil {
-			return nil, fmt.Errorf("file %v has unsafe path %q: %w", i, fileInfo.Path, err)
+			err = fmt.Errorf("file %v has unsafe path %q: %w", i, fileInfo.Path, err)
+			return
 		}
 		f := file{
 			path:   filepath.Join(dir, s),
@@ -83,16 +85,21 @@ func (fs *fileClientImpl) OpenTorrent(info *metainfo.Info, infoHash metainfo.Has
 		if f.length == 0 {
 			err = CreateNativeZeroLengthFile(f.path)
 			if err != nil {
-				return nil, fmt.Errorf("creating zero length file: %w", err)
+				err = fmt.Errorf("creating zero length file: %w", err)
+				return
 			}
 		}
 		files = append(files, f)
 	}
-	return &fileTorrentImpl{
+	t := &fileTorrentImpl{
 		files,
 		segments.NewIndex(common.LengthIterFromUpvertedFiles(upvertedFiles)),
 		infoHash,
 		fs.pc,
+	}
+	return TorrentImpl{
+		Piece: t.Piece,
+		Close: t.Close,
 	}, nil
 }
 

storage/interface.go

@@ -17,9 +17,11 @@ type ClientImpl interface {
 }
 
 // Data storage bound to a torrent.
-type TorrentImpl interface {
+type TorrentImpl struct {
-	Piece(metainfo.Piece) PieceImpl
+	Piece func(p metainfo.Piece) PieceImpl
-	Close() error
+	Close func() error
+	// Storages that share the same value, will provide a pointer to the same function.
+	Capacity *func() *int64
 }
 
 // Interacts with torrent piece data. Optional interfaces to implement include:

storage/mmap.go

@@ -30,14 +30,14 @@ func NewMMapWithCompletion(baseDir string, completion PieceCompletion) *mmapClie
 	}
 }
 
-func (s *mmapClientImpl) OpenTorrent(info *metainfo.Info, infoHash metainfo.Hash) (t TorrentImpl, err error) {
+func (s *mmapClientImpl) OpenTorrent(info *metainfo.Info, infoHash metainfo.Hash) (_ TorrentImpl, err error) {
 	span, err := mMapTorrent(info, s.baseDir)
-	t = &mmapTorrentStorage{
+	t := &mmapTorrentStorage{
 		infoHash: infoHash,
 		span:     span,
 		pc:       s.pc,
 	}
-	return
+	return TorrentImpl{Piece: t.Piece, Close: t.Close}, err
 }
 
 func (s *mmapClientImpl) Close() error {

storage/piece-resource.go

@@ -26,6 +26,7 @@ type ResourcePiecesOpts struct {
 	// Sized puts require being able to stream from a statement executed on another connection.
 	// Without them, we buffer the entire read and then put that.
 	NoSizedPuts bool
+	Capacity    *int64
 }
 
 func NewResourcePieces(p PieceProvider) ClientImpl {
@@ -49,10 +50,11 @@ func (piecePerResourceTorrentImpl) Close() error {
 }
 
 func (s piecePerResource) OpenTorrent(info *metainfo.Info, infoHash metainfo.Hash) (TorrentImpl, error) {
-	return piecePerResourceTorrentImpl{
+	t := piecePerResourceTorrentImpl{
 		s,
 		make([]sync.RWMutex, info.NumPieces()),
-	}, nil
+	}
+	return TorrentImpl{Piece: t.Piece, Close: t.Close}, nil
 }
 
 func (s piecePerResourceTorrentImpl) Piece(p metainfo.Piece) PieceImpl {

storage/sqlite/direct.go

@@ -61,9 +61,24 @@ func NewDirectStorage(opts NewDirectStorageOpts) (_ storage.ClientImplCloser, er
 	if opts.BlobFlushInterval != 0 {
 		cl.blobFlusher = time.AfterFunc(opts.BlobFlushInterval, cl.blobFlusherFunc)
 	}
+	cl.capacity = cl.getCapacity
 	return cl, nil
 }
 
+func (cl *client) getCapacity() (ret *int64) {
+	cl.l.Lock()
+	defer cl.l.Unlock()
+	err := sqlitex.Exec(cl.conn, "select value from setting where name='capacity'", func(stmt *sqlite.Stmt) error {
+		ret = new(int64)
+		*ret = stmt.ColumnInt64(0)
+		return nil
+	})
+	if err != nil {
+		panic(err)
+	}
+	return
+}
+
 type client struct {
 	l           sync.Mutex
 	conn        conn
@@ -71,6 +86,7 @@ type client struct {
 	blobFlusher *time.Timer
 	opts        NewDirectStorageOpts
 	closed      bool
+	capacity    func() *int64
 }
 
 func (c *client) blobFlusherFunc() {
@@ -91,7 +107,8 @@ func (c *client) flushBlobs() {
 }
 
 func (c *client) OpenTorrent(info *metainfo.Info, infoHash metainfo.Hash) (storage.TorrentImpl, error) {
-	return torrent{c}, nil
+	t := torrent{c}
+	return storage.TorrentImpl{Piece: t.Piece, Close: t.Close, Capacity: &c.capacity}, nil
 }
 
 func (c *client) Close() error {

test/issue377_test.go

@@ -124,7 +124,7 @@ func (me *diskFullStorage) Close() error {
 }
 
 func (d *diskFullStorage) OpenTorrent(info *metainfo.Info, infoHash metainfo.Hash) (storage.TorrentImpl, error) {
-	return d, nil
+	return storage.TorrentImpl{Piece: d.Piece, Close: d.Close}, nil
 }
 
 type pieceImpl struct {

test/transfer_test.go

@@ -12,6 +12,7 @@ import (
 	"testing/iotest"
 	"time"
 
+	"github.com/anacrolix/missinggo/v2/bitmap"
 	"github.com/anacrolix/missinggo/v2/filecache"
 	"github.com/anacrolix/torrent"
 	"github.com/anacrolix/torrent/internal/testutil"
@@ -168,7 +169,7 @@ func testClientTransfer(t *testing.T, ps testClientTransferParams) {
 	for _, pc := range leecherPeerConns {
 		completed := pc.PeerPieces().Len()
 		t.Logf("peer conn %v has %v completed pieces", pc, completed)
-		if completed == leecherTorrent.Info().NumPieces() {
+		if completed == bitmap.BitRange(leecherTorrent.Info().NumPieces()) {
 			foundSeeder = true
 		}
 	}
@@ -192,7 +193,6 @@ func testClientTransfer(t *testing.T, ps testClientTransferParams) {
 type fileCacheClientStorageFactoryParams struct {
 	Capacity    int64
 	SetCapacity bool
-	Wrapper     func(*filecache.Cache) storage.ClientImplCloser
 }
 
 func newFileCacheClientStorageFactory(ps fileCacheClientStorageFactoryParams) storageFactory {
@@ -201,10 +201,22 @@ func newFileCacheClientStorageFactory(ps fileCacheClientStorageFactoryParams) st
 		if err != nil {
 			panic(err)
 		}
+		var sharedCapacity *int64
 		if ps.SetCapacity {
+			sharedCapacity = &ps.Capacity
 			fc.SetCapacity(ps.Capacity)
 		}
-		return ps.Wrapper(fc)
+		return struct {
+			storage.ClientImpl
+			io.Closer
+		}{
+			storage.NewResourcePiecesOpts(
+				fc.AsResourceProvider(),
+				storage.ResourcePiecesOpts{
+					Capacity: sharedCapacity,
+				}),
+			ioutil.NopCloser(nil),
+		}
 	}
 }
 
@@ -212,17 +224,13 @@ type storageFactory func(string) storage.ClientImplCloser
 
 func TestClientTransferDefault(t *testing.T) {
 	testClientTransfer(t, testClientTransferParams{
-		LeecherStorage: newFileCacheClientStorageFactory(fileCacheClientStorageFactoryParams{
+		LeecherStorage: newFileCacheClientStorageFactory(fileCacheClientStorageFactoryParams{}),
-			Wrapper: fileCachePieceResourceStorage,
-		}),
 	})
 }
 
 func TestClientTransferDefaultNoMetadata(t *testing.T) {
 	testClientTransfer(t, testClientTransferParams{
-		LeecherStorage: newFileCacheClientStorageFactory(fileCacheClientStorageFactoryParams{
+		LeecherStorage:               newFileCacheClientStorageFactory(fileCacheClientStorageFactoryParams{}),
-			Wrapper: fileCachePieceResourceStorage,
-		}),
 		LeecherStartsWithoutMetadata: true,
 	})
 }
@@ -244,16 +252,6 @@ func TestClientTransferRateLimitedDownload(t *testing.T) {
 	})
 }
 
-func fileCachePieceResourceStorage(fc *filecache.Cache) storage.ClientImplCloser {
-	return struct {
-		storage.ClientImpl
-		io.Closer
-	}{
-		storage.NewResourcePieces(fc.AsResourceProvider()),
-		ioutil.NopCloser(nil),
-	}
-}
-
 func testClientTransferSmallCache(t *testing.T, setReadahead bool, readahead int64) {
 	testClientTransfer(t, testClientTransferParams{
 		LeecherStorage: newFileCacheClientStorageFactory(fileCacheClientStorageFactoryParams{
@@ -261,7 +259,6 @@ func testClientTransferSmallCache(t *testing.T, setReadahead bool, readahead int
 			// Going below the piece length means it can't complete a piece so
 			// that it can be hashed.
 			Capacity: 5,
-			Wrapper:  fileCachePieceResourceStorage,
 		}),
 		SetReadahead: setReadahead,
 		// Can't readahead too far or the cache will thrash and drop data we
@@ -324,9 +321,7 @@ func sqliteLeecherStorageTestCase(numConns int) leecherStorageTestCase {
 func TestClientTransferVarious(t *testing.T) {
 	// Leecher storage
 	for _, ls := range []leecherStorageTestCase{
-		{"Filecache", newFileCacheClientStorageFactory(fileCacheClientStorageFactoryParams{
+		{"Filecache", newFileCacheClientStorageFactory(fileCacheClientStorageFactoryParams{}), 0},
-			Wrapper: fileCachePieceResourceStorage,
-		}), 0},
 		{"Boltdb", storage.NewBoltDB, 0},
 		{"SqliteDirect", func(s string) storage.ClientImplCloser {
 			path := filepath.Join(s, "sqlite3.db")

torrent.go

@@ -12,17 +12,12 @@ import (
 	"net/http"
 	"net/url"
 	"sort"
+	"strings"
 	"sync"
 	"text/tabwriter"
 	"time"
 	"unsafe"
 
-	"github.com/anacrolix/torrent/common"
-	"github.com/anacrolix/torrent/segments"
-	"github.com/anacrolix/torrent/webseed"
-	"github.com/davecgh/go-spew/spew"
-	"github.com/pion/datachannel"
-
 	"github.com/anacrolix/dht/v2"
 	"github.com/anacrolix/log"
 	"github.com/anacrolix/missinggo"
@@ -32,12 +27,18 @@ import (
 	"github.com/anacrolix/missinggo/slices"
 	"github.com/anacrolix/missinggo/v2/bitmap"
 	"github.com/anacrolix/missinggo/v2/prioritybitmap"
+	"github.com/anacrolix/multiless"
+	"github.com/davecgh/go-spew/spew"
+	"github.com/pion/datachannel"
 
 	"github.com/anacrolix/torrent/bencode"
+	"github.com/anacrolix/torrent/common"
 	"github.com/anacrolix/torrent/metainfo"
 	pp "github.com/anacrolix/torrent/peer_protocol"
+	"github.com/anacrolix/torrent/segments"
 	"github.com/anacrolix/torrent/storage"
 	"github.com/anacrolix/torrent/tracker"
+	"github.com/anacrolix/torrent/webseed"
 	"github.com/anacrolix/torrent/webtorrent"
 )
 
@@ -55,9 +56,6 @@ type Torrent struct {
 	dataUploadDisallowed   bool
 	userOnWriteChunkErr    func(error)
 
-	// Determines what chunks to request from peers.
-	requestStrategy requestStrategy
-
 	closed   missinggo.Event
 	infoHash metainfo.Hash
 	pieces   []Piece
@@ -95,7 +93,6 @@ type Torrent struct {
 	// Set of addrs to which we're attempting to connect. Connections are
 	// half-open until all handshakes are completed.
 	halfOpen map[string]PeerInfo
-	fastestPeer *Peer
 
 	// Reserve of peers to connect to. A peer can be both here and in the
 	// active connections if were told about the peer after connecting with
@@ -150,6 +147,34 @@ type Torrent struct {
 	pex pexState
 }
 
+func (t *Torrent) pieceAvailabilityFromPeers(i pieceIndex) (count int) {
+	t.iterPeers(func(peer *Peer) {
+		if peer.peerHasPiece(i) {
+			count++
+		}
+	})
+	return
+}
+
+func (t *Torrent) decPieceAvailability(i pieceIndex) {
+	if !t.haveInfo() {
+		return
+	}
+	p := t.piece(i)
+	if p.availability <= 0 {
+		panic(p.availability)
+	}
+	p.availability--
+}
+
+func (t *Torrent) incPieceAvailability(i pieceIndex) {
+	// If we don't the info, this should be reconciled when we do.
+	if t.haveInfo() {
+		p := t.piece(i)
+		p.availability++
+	}
+}
+
 func (t *Torrent) numConns() int {
 	return len(t.conns)
 }
@@ -166,15 +191,8 @@ func (t *Torrent) readerReadaheadPieces() bitmap.Bitmap {
 	return t._readerReadaheadPieces
 }
 
-func (t *Torrent) ignorePieces() bitmap.Bitmap {
+func (t *Torrent) ignorePieceForRequests(i pieceIndex) bool {
-	ret := t._completedPieces.Copy()
+	return !t.wantPieceIndex(i)
-	ret.Union(t.piecesQueuedForHash)
-	for i := 0; i < t.numPieces(); i++ {
-		if t.piece(i).hashing {
-			ret.Set(i, true)
-		}
-	}
-	return ret
 }
 
 func (t *Torrent) pendingPieces() *prioritybitmap.PriorityBitmap {
@@ -417,8 +435,14 @@ func (t *Torrent) onSetInfo() {
 		p.onGotInfo(t.info)
 	})
 	for i := range t.pieces {
-		t.updatePieceCompletion(pieceIndex(i))
 		p := &t.pieces[i]
+		// Need to add availability before updating piece completion, as that may result in conns
+		// being dropped.
+		if p.availability != 0 {
+			panic(p.availability)
+		}
+		p.availability = int64(t.pieceAvailabilityFromPeers(i))
+		t.updatePieceCompletion(pieceIndex(i))
 		if !p.storageCompletionOk {
 			// t.logger.Printf("piece %s completion unknown, queueing check", p)
 			t.queuePieceCheck(pieceIndex(i))
@@ -536,6 +560,26 @@ func (t *Torrent) newMetadataExtensionMessage(c *PeerConn, msgType int, piece in
 	}
 }
 
+type pieceAvailabilityRun struct {
+	count        pieceIndex
+	availability int64
+}
+
+func (me pieceAvailabilityRun) String() string {
+	return fmt.Sprintf("%v(%v)", me.count, me.availability)
+}
+
+func (t *Torrent) pieceAvailabilityRuns() (ret []pieceAvailabilityRun) {
+	rle := missinggo.NewRunLengthEncoder(func(el interface{}, count uint64) {
+		ret = append(ret, pieceAvailabilityRun{availability: el.(int64), count: int(count)})
+	})
+	for i := range t.pieces {
+		rle.Append(t.pieces[i].availability, 1)
+	}
+	rle.Flush()
+	return
+}
+
 func (t *Torrent) pieceStateRuns() (ret PieceStateRuns) {
 	rle := missinggo.NewRunLengthEncoder(func(el interface{}, count uint64) {
 		ret = append(ret, PieceStateRun{
@@ -606,17 +650,26 @@ func (t *Torrent) writeStatus(w io.Writer) {
 		}
 		fmt.Fprintln(w)
 	}
-	fmt.Fprintf(w, "Piece length: %s\n", func() string {
+	fmt.Fprintf(w, "Piece length: %s\n",
+		func() string {
 			if t.haveInfo() {
-			return fmt.Sprint(t.usualPieceSize())
+				return fmt.Sprintf("%v (%v chunks)",
+					t.usualPieceSize(),
+					float64(t.usualPieceSize())/float64(t.chunkSize))
 			} else {
-			return "?"
+				return "no info"
 			}
-	}())
+		}(),
+	)
 	if t.info != nil {
 		fmt.Fprintf(w, "Num Pieces: %d (%d completed)\n", t.numPieces(), t.numPiecesCompleted())
-		fmt.Fprintf(w, "Piece States: %s", t.pieceStateRuns())
+		fmt.Fprintf(w, "Piece States: %s\n", t.pieceStateRuns())
-		fmt.Fprintln(w)
+		fmt.Fprintf(w, "Piece availability: %v\n", strings.Join(func() (ret []string) {
+			for _, run := range t.pieceAvailabilityRuns() {
+				ret = append(ret, run.String())
+			}
+			return
+		}(), " "))
 	}
 	fmt.Fprintf(w, "Reader Pieces:")
 	t.forReaderOffsetPieces(func(begin, end pieceIndex) (again bool) {
@@ -651,8 +704,15 @@ func (t *Torrent) writeStatus(w io.Writer) {
 	spew.Fdump(w, t.statsLocked())
 
 	peers := t.peersAsSlice()
-	sort.Slice(peers, func(i, j int) bool {
+	sort.Slice(peers, func(_i, _j int) bool {
-		return worseConn(peers[i], peers[j])
+		i := peers[_i]
+		j := peers[_j]
+		if less, ok := multiless.New().EagerSameLess(
+			i.downloadRate() == j.downloadRate(), i.downloadRate() < j.downloadRate(),
+		).LessOk(); ok {
+			return less
+		}
+		return worseConn(i, j)
 	})
 	for i, c := range peers {
 		fmt.Fprintf(w, "%2d. ", i+1)
@@ -700,7 +760,7 @@ func (t *Torrent) bytesMissingLocked() int64 {
 }
 
 func (t *Torrent) bytesLeft() (left int64) {
-	bitmap.Flip(t._completedPieces, 0, bitmap.BitIndex(t.numPieces())).IterTyped(func(piece int) bool {
+	bitmap.Flip(t._completedPieces, 0, bitmap.BitRange(t.numPieces())).IterTyped(func(piece int) bool {
 		p := &t.pieces[piece]
 		left += int64(p.length() - p.numDirtyBytes())
 		return true
@@ -735,8 +795,8 @@ func (t *Torrent) numPieces() pieceIndex {
 	return pieceIndex(t.info.NumPieces())
 }
 
-func (t *Torrent) numPiecesCompleted() (num int) {
+func (t *Torrent) numPiecesCompleted() (num pieceIndex) {
-	return t._completedPieces.Len()
+	return pieceIndex(t._completedPieces.Len())
 }
 
 func (t *Torrent) close() (err error) {
@@ -746,7 +806,9 @@ func (t *Torrent) close() (err error) {
 		func() {
 			t.storageLock.Lock()
 			defer t.storageLock.Unlock()
-			t.storage.Close()
+			if f := t.storage.Close; f != nil {
+				f()
+			}
 		}()
 	}
 	t.iterPeers(func(p *Peer) {
@@ -844,7 +906,7 @@ func (t *Torrent) haveAllPieces() bool {
 	if !t.haveInfo() {
 		return false
 	}
-	return t._completedPieces.Len() == bitmap.BitIndex(t.numPieces())
+	return t._completedPieces.Len() == bitmap.BitRange(t.numPieces())
 }
 
 func (t *Torrent) havePiece(index pieceIndex) bool {
@@ -890,12 +952,15 @@ func chunkIndex(cs ChunkSpec, chunkSize pp.Integer) int {
 }
 
 func (t *Torrent) wantPieceIndex(index pieceIndex) bool {
+	// TODO: Are these overly conservative, should we be guarding this here?
+	{
 		if !t.haveInfo() {
 			return false
 		}
 		if index < 0 || index >= t.numPieces() {
 			return false
 		}
+	}
 	p := &t.pieces[index]
 	if p.queuedForHash() {
 		return false
@@ -906,7 +971,7 @@ func (t *Torrent) wantPieceIndex(index pieceIndex) bool {
 	if t.pieceComplete(index) {
 		return false
 	}
-	if t._pendingPieces.Contains(bitmap.BitIndex(index)) {
+	if t._pendingPieces.Contains(int(index)) {
 		return true
 	}
 	// t.logger.Printf("piece %d not pending", index)
@@ -965,7 +1030,7 @@ func (t *Torrent) pieceNumPendingChunks(piece pieceIndex) pp.Integer {
 }
 
 func (t *Torrent) pieceAllDirty(piece pieceIndex) bool {
-	return t.pieces[piece]._dirtyChunks.Len() == int(t.pieceNumChunks(piece))
+	return t.pieces[piece]._dirtyChunks.Len() == bitmap.BitRange(t.pieceNumChunks(piece))
 }
 
 func (t *Torrent) readersChanged() {
@@ -1024,11 +1089,11 @@ func (t *Torrent) updatePiecePriority(piece pieceIndex) {
 	newPrio := p.uncachedPriority()
 	// t.logger.Printf("torrent %p: piece %d: uncached priority: %v", t, piece, newPrio)
 	if newPrio == PiecePriorityNone {
-		if !t._pendingPieces.Remove(bitmap.BitIndex(piece)) {
+		if !t._pendingPieces.Remove(int(piece)) {
 			return
 		}
 	} else {
-		if !t._pendingPieces.Set(bitmap.BitIndex(piece), newPrio.BitmapPriority()) {
+		if !t._pendingPieces.Set(int(piece), newPrio.BitmapPriority()) {
 			return
 		}
 	}
@@ -1084,7 +1149,7 @@ func (t *Torrent) forReaderOffsetPieces(f func(begin, end pieceIndex) (more bool
 }
 
 func (t *Torrent) piecePriority(piece pieceIndex) piecePriority {
-	prio, ok := t._pendingPieces.GetPriority(bitmap.BitIndex(piece))
+	prio, ok := t._pendingPieces.GetPriority(piece)
 	if !ok {
 		return PiecePriorityNone
 	}
@@ -1232,7 +1297,7 @@ func (t *Torrent) readerPiecePriorities() (now, readahead bitmap.Bitmap) {
 	t.forReaderOffsetPieces(func(begin, end pieceIndex) bool {
 		if end > begin {
 			now.Add(bitmap.BitIndex(begin))
-			readahead.AddRange(bitmap.BitIndex(begin)+1, bitmap.BitIndex(end))
+			readahead.AddRange(bitmap.BitRange(begin)+1, bitmap.BitRange(end))
 		}
 		return true
 	})
@@ -1319,6 +1384,16 @@ func (t *Torrent) deletePeerConn(c *PeerConn) (ret bool) {
 	return
 }
 
+func (t *Torrent) decPeerPieceAvailability(p *Peer) {
+	if !t.haveInfo() {
+		return
+	}
+	p.newPeerPieces().IterTyped(func(i int) bool {
+		p.t.decPieceAvailability(i)
+		return true
+	})
+}
+
 func (t *Torrent) numActivePeers() (num int) {
 	t.iterPeers(func(*Peer) {
 		num++
@@ -1905,7 +1980,7 @@ func (t *Torrent) tryCreatePieceHasher() bool {
 		return false
 	}
 	p := t.piece(pi)
-	t.piecesQueuedForHash.Remove(pi)
+	t.piecesQueuedForHash.Remove(bitmap.BitIndex(pi))
 	p.hashing = true
 	t.publishPieceChange(pi)
 	t.updatePiecePriority(pi)
@@ -2036,30 +2111,6 @@ func (t *Torrent) piece(i int) *Piece {
 	return &t.pieces[i]
 }
 
-func (t *Torrent) requestStrategyTorrent() requestStrategyTorrent {
-	return t
-}
-
-type torrentRequestStrategyCallbacks struct {
-	t *Torrent
-}
-
-func (cb torrentRequestStrategyCallbacks) requestTimedOut(r Request) {
-	torrent.Add("Request timeouts", 1)
-	cb.t.cl.lock()
-	defer cb.t.cl.unlock()
-	cb.t.iterPeers(func(cn *Peer) {
-		if cn.peerHasPiece(pieceIndex(r.Index)) {
-			cn.updateRequests()
-		}
-	})
-
-}
-
-func (t *Torrent) requestStrategyCallbacks() requestStrategyCallbacks {
-	return torrentRequestStrategyCallbacks{t}
-}
-
 func (t *Torrent) onWriteChunkErr(err error) {
 	if t.userOnWriteChunkErr != nil {
 		go t.userOnWriteChunkErr(err)
@@ -2121,7 +2172,7 @@ func (t *Torrent) SetOnWriteChunkError(f func(error)) {
 	t.userOnWriteChunkErr = f
 }
 
-func (t *Torrent) iterPeers(f func(*Peer)) {
+func (t *Torrent) iterPeers(f func(p *Peer)) {
 	for pc := range t.conns {
 		f(&pc.Peer)
 	}
@@ -2154,7 +2205,6 @@ func (t *Torrent) addWebSeed(url string) {
 			outgoing:                 true,
 			Network:                  "http",
 			reconciledHandshakeStats: true,
-			peerSentHaveAll:          true,
 			// TODO: Raise this limit, and instead limit concurrent fetches.
 			PeerMaxRequests: 32,
 			RemoteAddr:      remoteAddrFromUrl(url),
@@ -2180,6 +2230,7 @@ func (t *Torrent) addWebSeed(url string) {
 		ws.onGotInfo(t.info)
 	}
 	t.webSeeds[url] = &ws.peer
+	ws.peer.onPeerHasAllPieces()
 }
 
 func (t *Torrent) peerIsActive(p *Peer) (active bool) {

torrent_test.go

@@ -9,6 +9,7 @@ import (
 	"testing"
 
 	"github.com/anacrolix/missinggo"
+	"github.com/anacrolix/missinggo/v2/bitmap"
 	"github.com/bradfitz/iter"
 	"github.com/stretchr/testify/assert"
 	"github.com/stretchr/testify/require"
@@ -96,8 +97,8 @@ func BenchmarkUpdatePiecePriorities(b *testing.B) {
 		r.Seek(3500000, io.SeekStart)
 	}
 	assert.Len(b, t.readers, 7)
-	for i := 0; i < int(t.numPieces()); i += 3 {
+	for i := 0; i < t.numPieces(); i += 3 {
-		t._completedPieces.Set(i, true)
+		t._completedPieces.Set(bitmap.BitIndex(i), true)
 	}
 	t.DownloadPieces(0, t.numPieces())
 	for range iter.N(b.N) {

types/types.go

@@ -0,0 +1,52 @@
+package types
+
+import (
+	pp "github.com/anacrolix/torrent/peer_protocol"
+)
+
+type PieceIndex = int
+
+type ChunkSpec struct {
+	Begin, Length pp.Integer
+}
+
+type Request struct {
+	Index pp.Integer
+	ChunkSpec
+}
+
+func (r Request) ToMsg(mt pp.MessageType) pp.Message {
+	return pp.Message{
+		Type:   mt,
+		Index:  r.Index,
+		Begin:  r.Begin,
+		Length: r.Length,
+	}
+}
+
+// Describes the importance of obtaining a particular piece.
+type PiecePriority byte
+
+func (pp *PiecePriority) Raise(maybe PiecePriority) bool {
+	if maybe > *pp {
+		*pp = maybe
+		return true
+	}
+	return false
+}
+
+// Priority for use in PriorityBitmap
+func (me PiecePriority) BitmapPriority() int {
+	return -int(me)
+}
+
+const (
+	PiecePriorityNone      PiecePriority = iota // Not wanted. Must be the zero value.
+	PiecePriorityNormal                         // Wanted.
+	PiecePriorityHigh                           // Wanted a lot.
+	PiecePriorityReadahead                      // May be required soon.
+	// Succeeds a piece where a read occurred. Currently the same as Now,
+	// apparently due to issues with caching.
+	PiecePriorityNext
+	PiecePriorityNow // A Reader is reading in this piece. Highest urgency.
+)

webseed-peer.go

@@ -24,6 +24,10 @@ type webseedPeer struct {
 
 var _ peerImpl = (*webseedPeer)(nil)
 
+func (me *webseedPeer) writeBufferFull() bool {
+	return false
+}
+
 func (me *webseedPeer) connStatusString() string {
 	return me.client.Url
 }
@@ -37,20 +41,15 @@ func (ws *webseedPeer) onGotInfo(info *metainfo.Info) {
 	ws.client.Info = info
 }
 
-func (ws *webseedPeer) _postCancel(r Request) {
-	ws.cancel(r)
-}
-
 func (ws *webseedPeer) writeInterested(interested bool) bool {
 	return true
 }
 
-func (ws *webseedPeer) cancel(r Request) bool {
+func (ws *webseedPeer) _cancel(r Request) bool {
 	active, ok := ws.activeRequests[r]
-	if !ok {
+	if ok {
-		return false
-	}
 		active.Cancel()
+	}
 	return true
 }
 
@@ -58,7 +57,7 @@ func (ws *webseedPeer) intoSpec(r Request) webseed.RequestSpec {
 	return webseed.RequestSpec{ws.peer.t.requestOffset(r), int64(r.Length)}
 }
 
-func (ws *webseedPeer) request(r Request) bool {
+func (ws *webseedPeer) _request(r Request) bool {
 	ws.requesterCond.Signal()
 	return true
 }
@@ -79,7 +78,7 @@ func (ws *webseedPeer) requester() {
 	defer ws.requesterCond.L.Unlock()
 start:
 	for !ws.peer.closed.IsSet() {
-		for r := range ws.peer.requests {
+		for r := range ws.peer.actualRequestState.Requests {
 			if _, ok := ws.activeRequests[r]; ok {
 				continue
 			}
@@ -99,7 +98,6 @@ func (ws *webseedPeer) connectionFlags() string {
 func (ws *webseedPeer) drop() {}
 
 func (ws *webseedPeer) updateRequests() {
-	ws.peer.doRequestState()
 }
 
 func (ws *webseedPeer) onClose() {
@@ -112,6 +110,10 @@ func (ws *webseedPeer) onClose() {
 
 func (ws *webseedPeer) requestResultHandler(r Request, webseedRequest webseed.Request) {
 	result := <-webseedRequest.Result
+	// We do this here rather than inside receiveChunk, since we want to count errors too. I'm not
+	// sure if we can divine which errors indicate cancellation on our end without hitting the
+	// network though.
+	ws.peer.doChunkReadStats(int64(len(result.Bytes)))
 	ws.peer.t.cl.lock()
 	defer ws.peer.t.cl.unlock()
 	if result.Err != nil {
@@ -146,3 +148,7 @@ func (ws *webseedPeer) requestResultHandler(r Request, webseedRequest webseed.Re
 		}
 	}
 }
+
+func (me *webseedPeer) onNextRequestStateChanged() {
+	me.peer.applyNextRequestState()
+}