Jakob Borg
a65cf5791e
Add a magic word in front of the header
...
This identifies the connection as a Syncthing connection, makes it look
less like uTP for interested viewers. Passes the test suite here, and
works when copied into our vendoring directory.
2016-05-26 15:18:31 +02:00
Matt Joiner
d529dcb23b
Terminate extension chains when marshalling header
...
This is important if the buffer is being reused, and isn't necessarily zeroed.
2016-05-26 22:13:24 +10:00
Matt Joiner
0a3e0d100a
Don't do resends in a separate goroutine
...
~10% performance improvement
2016-05-26 15:13:26 +10:00
Matt Joiner
0950f6188a
Bump to 20MB echo test
2016-05-26 15:12:50 +10:00
Matt Joiner
fc1b0ce1d6
TestMain: Wait a second for sockets to be closed before writing status
2016-05-26 15:10:44 +10:00
Matt Joiner
3c03170741
Avoid a small allocation
2016-05-26 15:09:46 +10:00
Matt Joiner
cc6c46a38d
Add Echo benchmark
2016-05-26 15:09:25 +10:00
Matt Joiner
e64a46c6c6
Slow down the raw UTP test
2016-05-23 09:32:51 +10:00
Matt Joiner
f203455078
Clean up accept code
2016-05-14 13:59:06 +10:00
Matt Joiner
2f283dacc4
Observations while fixing the last
2016-05-14 13:15:41 +10:00
Matt Joiner
fb5d273ddc
Socket didn't have a destroyed Event, Accept would never return without explicit Close despite error
2016-05-14 13:15:19 +10:00
Matt Joiner
ea7646aa4a
Add TestAcceptOnDestroyedSocket
...
Noticed by @whyrusleeping
2016-05-14 13:04:48 +10:00
Matt Joiner
7a25702542
Improve consistency of TestWriteUnderlyingPacketConnClosed
2016-05-14 13:03:58 +10:00
Matt Joiner
2ae532de0f
missinggo.Event changes
2016-05-07 18:59:21 +10:00
Matt Joiner
e509da3316
Don't do send buffer allocation in header.Marshal, and don't allocate minMTU for outgoing non-data packets, throw away sendBufferPool
...
sendBufferPool will be easier to work with in the future with allocation no longer occuring in header.Marshal. I'm not even sure it's necessary.
2016-04-30 04:06:29 +10:00
Matt Joiner
40801b2499
Track how many state packets are saved by batching them
2016-04-29 21:21:13 +10:00
Matt Joiner
f40a4e70a4
Missed hunk from last commit
...
Fixes #11 .
2016-04-29 20:52:37 +10:00
Matt Joiner
96252a858d
Fix handling of net.Addr.Network throughout, and create a utp addr type
...
Fixes #11 .
2016-04-29 20:50:24 +10:00
Matt Joiner
e5dca62cbd
Cap the read buffer to 1MiB, test Conn write deadline, and large transfer
2016-04-26 23:43:03 +10:00
Matt Joiner
5eef1671c7
Tidy up write
2016-04-26 23:16:42 +10:00
Matt Joiner
68ec4a72c8
Trial a dedicated read sync.Cond for improved performance
...
I tried channels and they're just too slow. Read events occur too quickly.
2016-04-26 12:48:06 +10:00
Matt Joiner
a0d40bd787
Remove dedicated Conn.deliveryProcessor goroutine, and fix Conn.deliver deadlock
2016-04-26 01:08:19 +10:00
Matt Joiner
e57d077423
Clean up a test
2016-04-24 21:18:02 +10:00
Matt Joiner
a3167f3d18
Clean up deadline handling
...
https://github.com/anacrolix/utp/issues/10
2016-04-24 21:17:41 +10:00
Matt Joiner
276b2e01aa
Conn.Write should fail if Socket is destroyed
...
https://github.com/ipfs/go-ipfs/issues/2566
2016-04-24 20:19:48 +10:00
Matt Joiner
336395439c
Rework connPair to use the same Socket for both Conns
2016-04-24 20:02:45 +10:00
Matt Joiner
4ff5ca0f7d
Count dropped and unused packets
2016-04-24 20:01:19 +10:00
Matt Joiner
bb8da3e04a
Doc comments and tidy
2016-04-24 20:00:52 +10:00
Matt Joiner
0163ffe960
Do dispatching inline with the socket reader and destroy socket on read error
2016-04-24 19:58:43 +10:00
Matt Joiner
d2fb2b0385
On dial timeout, don't send a FIN, as they'll treat it like EOF
2016-04-22 02:39:16 +10:00
Matt Joiner
78425cef94
TestAcceptGone: Check that SetReadDeadline worked
2016-04-21 22:39:05 +10:00
Matt Joiner
c5226b7368
Add some new tests
2016-04-21 22:38:50 +10:00
Matt Joiner
c452c69132
Make it easier to set and unset testing and default timeouts
2016-04-21 22:38:16 +10:00
Matt Joiner
719fbbd05e
updateStates is just lazyDestroy, and use it in Conn.Write
2016-04-21 22:34:24 +10:00
Matt Joiner
8e5eae6555
Also hook addr resolving for tests
2016-04-21 22:33:32 +10:00
Matt Joiner
6d38c865d7
Use in-memory implementation of PacketConn for tests
2016-04-20 00:01:08 +10:00
Matt Joiner
5486f285e7
Treat closed as gotFin for purposes of destroying Conn after sends are acked
2016-04-19 18:33:01 +10:00
Matt Joiner
974bb7b58f
Close triggers EOF on Read
2016-04-19 18:32:09 +10:00
Matt Joiner
83c99a29af
registerConn add connKey to Conn only after it's added
2016-04-19 18:29:00 +10:00
Matt Joiner
c4b73fb4aa
Trigger Conn detach when destroyed
2016-04-19 18:03:24 +10:00
Matt Joiner
3b47958457
Fix naming of Conn.{closing,closed}
...
Closed means the user has called Close(). Destroyed is when the Conn is being forcefully removed, and no more activity should occur.
2016-04-19 17:45:11 +10:00
Matt Joiner
39b33ffeec
Fix nixxing of ackSkipped resend when already acked
2016-04-19 17:35:44 +10:00
Matt Joiner
6865c4baa2
Make resend a fixed method on send
2016-04-19 17:27:18 +10:00
Matt Joiner
d313cb3d3d
Single global mutex and cond
2016-04-19 11:27:05 +10:00
Matt Joiner
59c10150c5
Break up into smaller files
2016-04-19 10:26:44 +10:00
Matt Joiner
a0ed6ae2f3
Don't do TestConnectSelf in parallel
2016-04-18 21:31:00 +10:00
Matt Joiner
e597926497
Make the timedOut callback a method
2016-04-18 01:23:18 +10:00
Matt Joiner
2aa2a1db54
Don't resend packet when ack is skipped if it's already acked
2016-04-16 02:49:36 +10:00
Matt Joiner
0035676f02
Tidy up Conn.unackedSends trim logic
2016-04-15 15:49:29 +10:00
Matt Joiner
11118b11ca
Nil send function fields once acked
...
May allow GC to free payload buffers and other captured variables earlier.
2016-04-15 15:49:08 +10:00
Jakob Borg
Matt Joiner