|
|
@@ -11,7 +11,6 @@ import (
|
|
|
"io"
|
|
|
"sync"
|
|
|
"time"
|
|
|
-
|
|
|
"github.com/calmh/syncthing/xdr"
|
|
|
)
|
|
|
|
|
|
@@ -96,6 +95,15 @@ type rawConnection struct {
|
|
|
outbox chan []encodable
|
|
|
closed chan struct{}
|
|
|
once sync.Once
|
|
|
+
|
|
|
+ incomingIndexes chan incomingIndex
|
|
|
+}
|
|
|
+
|
|
|
+type incomingIndex struct {
|
|
|
+ update bool
|
|
|
+ id NodeID
|
|
|
+ repo string
|
|
|
+ files []FileInfo
|
|
|
}
|
|
|
|
|
|
type asyncResult struct {
|
|
|
@@ -116,19 +124,20 @@ func NewConnection(nodeID NodeID, reader io.Reader, writer io.Writer, receiver M
|
|
|
wb := bufio.NewWriterSize(cw, 65536)
|
|
|
|
|
|
c := rawConnection{
|
|
|
- id: nodeID,
|
|
|
- name: name,
|
|
|
- receiver: nativeModel{receiver},
|
|
|
- state: stateInitial,
|
|
|
- cr: cr,
|
|
|
- xr: xdr.NewReader(rb),
|
|
|
- cw: cw,
|
|
|
- wb: wb,
|
|
|
- xw: xdr.NewWriter(wb),
|
|
|
- awaiting: make([]chan asyncResult, 0x1000),
|
|
|
- outbox: make(chan []encodable),
|
|
|
- nextID: make(chan int),
|
|
|
- closed: make(chan struct{}),
|
|
|
+ id: nodeID,
|
|
|
+ name: name,
|
|
|
+ receiver: nativeModel{receiver},
|
|
|
+ state: stateInitial,
|
|
|
+ cr: cr,
|
|
|
+ xr: xdr.NewReader(rb),
|
|
|
+ cw: cw,
|
|
|
+ wb: wb,
|
|
|
+ xw: xdr.NewWriter(wb),
|
|
|
+ awaiting: make([]chan asyncResult, 0x1000),
|
|
|
+ outbox: make(chan []encodable),
|
|
|
+ nextID: make(chan int),
|
|
|
+ closed: make(chan struct{}),
|
|
|
+ incomingIndexes: make(chan incomingIndex, 100), // should be enough for anyone, right?
|
|
|
}
|
|
|
|
|
|
go c.indexSerializerLoop()
|
|
|
@@ -307,15 +316,6 @@ func (c *rawConnection) readerLoop() (err error) {
|
|
|
}
|
|
|
}
|
|
|
|
|
|
-type incomingIndex struct {
|
|
|
- update bool
|
|
|
- id NodeID
|
|
|
- repo string
|
|
|
- files []FileInfo
|
|
|
-}
|
|
|
-
|
|
|
-var incomingIndexes = make(chan incomingIndex, 100) // should be enough for anyone, right?
|
|
|
-
|
|
|
func (c *rawConnection) indexSerializerLoop() {
|
|
|
// We must avoid blocking the reader loop when processing large indexes.
|
|
|
// There is otherwise a potential deadlock where both sides has the model
|
|
|
@@ -325,7 +325,7 @@ func (c *rawConnection) indexSerializerLoop() {
|
|
|
// routine and buffered channel.
|
|
|
for {
|
|
|
select {
|
|
|
- case ii := <-incomingIndexes:
|
|
|
+ case ii := <-c.incomingIndexes:
|
|
|
if ii.update {
|
|
|
if debug {
|
|
|
l.Debugf("calling IndexUpdate(%v, %v, %d files)", ii.id, ii.repo, len(ii.files))
|
|
|
@@ -360,7 +360,7 @@ func (c *rawConnection) handleIndex() error {
|
|
|
if debug {
|
|
|
l.Debugf("queueing Index(%v, %v, %d files)", c.id, im.Repository, len(im.Files))
|
|
|
}
|
|
|
- incomingIndexes <- incomingIndex{false, c.id, im.Repository, im.Files}
|
|
|
+ c.incomingIndexes <- incomingIndex{false, c.id, im.Repository, im.Files}
|
|
|
}
|
|
|
return nil
|
|
|
}
|
|
|
@@ -374,7 +374,7 @@ func (c *rawConnection) handleIndexUpdate() error {
|
|
|
if debug {
|
|
|
l.Debugf("queueing IndexUpdate(%v, %v, %d files)", c.id, im.Repository, len(im.Files))
|
|
|
}
|
|
|
- incomingIndexes <- incomingIndex{true, c.id, im.Repository, im.Files}
|
|
|
+ c.incomingIndexes <- incomingIndex{true, c.id, im.Repository, im.Files}
|
|
|
}
|
|
|
return nil
|
|
|
}
|