Datasets:

---

feedback-to-code

/

mozzarella

like 1

Follow

Feedback-2-Code 7

[ "timestamp(timedelta=83.0, similarity=0.9042778390261851)" ]

[ "@ngocdaothanh would you mind submitting a PR ?\n", "Sure! I wanted to confirm if my guess was right.\n", "I've sent pull requests for branch 4.0, 4.1, and master.\n" ]

[]

[]

[ "codec-http/src/main/java/io/netty/handler/codec/http/websocketx/WebSocketServerHandshaker.java", "codec-http/src/main/java/io/netty/handler/codec/http/websocketx/WebSocketServerHandshaker07.java" ]

[ "codec-http/src/main/java/io/netty/handler/codec/http/websocketx/WebSocketServerHandshaker.java", "codec-http/src/main/java/io/netty/handler/codec/http/websocketx/WebSocketServerHandshaker07.java" ]

[]

[ "timestamp(timedelta=81.0, similarity=0.9042778390261851)" ]

[ "@ngocdaothanh would you mind submitting a PR ?\n", "Sure! I wanted to confirm if my guess was right.\n", "I've sent pull requests for branch 4.0, 4.1, and master.\n" ]

[]

[]

[ "codec-http/src/main/java/io/netty/handler/codec/http/websocketx/WebSocketServerHandshaker.java", "codec-http/src/main/java/io/netty/handler/codec/http/websocketx/WebSocketServerHandshaker07.java" ]

[ "codec-http/src/main/java/io/netty/handler/codec/http/websocketx/WebSocketServerHandshaker.java", "codec-http/src/main/java/io/netty/handler/codec/http/websocketx/WebSocketServerHandshaker07.java" ]

[]

[ "timestamp(timedelta=44.0, similarity=0.9042778390261851)" ]

[ "@ngocdaothanh would you mind submitting a PR ?\n", "Sure! I wanted to confirm if my guess was right.\n", "I've sent pull requests for branch 4.0, 4.1, and master.\n" ]

[]

[]

[ "codec-http/src/main/java/io/netty/handler/codec/http/websocketx/WebSocketServerHandshaker.java", "codec-http/src/main/java/io/netty/handler/codec/http/websocketx/WebSocketServerHandshaker07.java" ]

[ "codec-http/src/main/java/io/netty/handler/codec/http/websocketx/WebSocketServerHandshaker.java", "codec-http/src/main/java/io/netty/handler/codec/http/websocketx/WebSocketServerHandshaker07.java" ]

[]

[ "timestamp(timedelta=51.0, similarity=0.8715944770161461)" ]

[ "I was interested in the performance of this, so I ran a little JMH benchmark (disclaimer: I don't know the first thing about benchmarking) on my laptop (too lazy to connect to Azure :)). Well, this is what I got https://gist.github.com/bucjac/9401713 and here is the code https://gist.github.com/bucjac/9401700.\n\nI am not sure if this would bring any performance gains?\n", "That's interesting. I just ran it on a Ubuntu VM on Windows Azure (AMD Opteron) and the bitwise optimization resulted in a 2.13 speedup, while for non-poweroftwos it was only 4% slower. https://gist.github.com/bucjac/9450506\n", "It only will work with power of two\n\n> Am 09.03.2014 um 17:48 schrieb Jakob Buchgraber [email protected]:\n> \n> That's interesting. I just ran it on a Ubuntu VM on Windows Azure (AMD Opteron) and the bitwise optimization resulted in a 2.13 speedup, while for non-poweroftwos it was only 4%. https://gist.github.com/bucjac/9450506\n> \n> —\n> Reply to this email directly or view it on GitHub.\n", "Yeah, but on my laptop it didn't run any faster for power of twos, while on azure it was twice as fast?\n", "Fixed by #2299\n" ]

[]

[ "improvement" ]

[ "common/src/main/java/io/netty/util/concurrent/MultithreadEventExecutorGroup.java" ]

[ "common/src/main/java/io/netty/util/concurrent/MultithreadEventExecutorGroup.java" ]

[]

[ "timestamp(timedelta=166.0, similarity=0.8448387943046178)" ]

[ "Let me check... maybe I can do something smart with an `AtomicReference`\n", "@trustin I wonder why we even lazy start the Thread and add the task to the priority queue. Why not just do this in the constructor right away ?\n", "A user could create relatively large executor group and then only register smaller number of channels, then we will be wasting threads.\n", "@trustin true... not sure how this would happen in reality though ... Let me just use some AtomicInteger with compareAndSet and spinning. This should work.\n", "@valodzka should be fixed now by #2310 . Please re-test and let me know.\n", "Tested with 4.0.18.Final-20140314.085205-40, issue has been solved\n", "@valodzka awesome... thanks for the feedback and like I said. Keep them coming ;)\n" ]

[ "`case ST_NOT_STARTED:` could also just fall-through.\n", "Maybe like this?\n\n``` java\nfor (;;) {\n int oldState = state.get();\n if (oldState >= ST_SHUTTING_DOWN || state.compareAndSet(oldState, ST_SHUTTING_DOWN)) {\n break;\n }\n}\n```\n", "Could we also use `PlatformDependent.newAtomicIntegerFieldUpdater()` if possible?\n" ]

[ "improvement" ]

[ "common/src/main/java/io/netty/util/concurrent/SingleThreadEventExecutor.java" ]

[ "common/src/main/java/io/netty/util/concurrent/SingleThreadEventExecutor.java" ]

[]

[ "timestamp(timedelta=25226.0, similarity=0.9999999999999998)" ]

[ "I think that it may be closed too\n" ]

[]

[ "feature" ]

[ "transport/src/main/java/io/netty/channel/socket/nio/NioDatagramChannel.java", "transport/src/main/java/io/netty/channel/socket/nio/NioServerSocketChannel.java", "transport/src/main/java/io/netty/channel/socket/nio/NioSocketChannel.java" ]

[ "transport/src/main/java/io/netty/channel/socket/nio/NioDatagramChannel.java", "transport/src/main/java/io/netty/channel/socket/nio/NioServerSocketChannel.java", "transport/src/main/java/io/netty/channel/socket/nio/NioSocketChannel.java" ]

[]

[ "timestamp(timedelta=27.0, similarity=0.9606678489503532)" ]

[ "We love pull-requests ;)\n\n> Am 24.03.2014 um 13:26 schrieb Alexey [email protected]:\n> \n> Hi\n> \n> https://github.com/netty/netty/blob/4.0/codec/src/main/java/io/netty/handler/codec/protobuf/ProtobufVarint32LengthFieldPrepender.java\n> \n> Now we calculate headerLen for start, and some later create CodedOutputStream with code\n> \n> CodedOutputStream headerOut =\n> CodedOutputStream.newInstance(new ByteBufOutputStream(out));\n> default constructor allocate 4096 buffer, but headerLen always less than 5.\n> More correct way it's set size of buffer for new instance with code:\n> \n> CodedOutputStream headerOut =\n> CodedOutputStream.newInstance(new ByteBufOutputStream(out), headerLen);\n> this decrease allocation for header prepender from 4096 bytes to [1-5] bytes for every protobuf message\n> \n> —\n> Reply to this email directly or view it on GitHub.\n", "how i can remove first \"referenced this issue from a commit in\"\nbecause he have incorrect committer and i did't pull request for he =\\\n\npull request: [#2340]\n", "Close it an open a new one? Also please update tge commit message to match tge template we use.\n\n> Am 24.03.2014 um 16:44 schrieb Alexey [email protected]:\n> \n> how i can remove first pull request,\n> because he have incorrect committer and i did't pull request for he =\\\n> \n> —\n> Reply to this email directly or view it on GitHub.\n" ]

[]

[ "improvement" ]

[ "codec/src/main/java/io/netty/handler/codec/protobuf/ProtobufVarint32LengthFieldPrepender.java" ]

[ "codec/src/main/java/io/netty/handler/codec/protobuf/ProtobufVarint32LengthFieldPrepender.java" ]

[]

[ "timestamp(timedelta=54.0, similarity=0.8885609524883314)" ]

[]

[ "Maybe we should not throw an exception but delegate to not break existing apps?\n", "Good point, let me fix that. Thanks\n\nsöndagen den 30:e mars 2014 skrev Norman Maurer [email protected]:\n\n> In\n> codec-http/src/main/java/io/netty/handler/codec/http/cors/CorsConfig.java:\n> \n> > @@ -228,15 +250,28 @@ public static Builder anyOrigin() {\n> > \\* @return {@link Builder} to support method chaining.\n> > */\n> > public static Builder withOrigin(final String origin) {\n> > - if (origin.equals(\"*\")) {\n> \n> Maybe we should not throw an exception but delegate to not break existing\n> apps?\n> \n> ## \n> \n> Reply to this email directly or view it on GitHubhttps://github.com/netty/netty/pull/2347/files#r11097182\n> .\n" ]

[]

[ "codec-http/src/main/java/io/netty/handler/codec/http/cors/CorsConfig.java", "codec-http/src/main/java/io/netty/handler/codec/http/cors/CorsHandler.java", "example/src/main/java/io/netty/example/http/cors/HttpServerInitializer.java" ]

[ "codec-http/src/main/java/io/netty/handler/codec/http/cors/CorsConfig.java", "codec-http/src/main/java/io/netty/handler/codec/http/cors/CorsHandler.java", "example/src/main/java/io/netty/example/http/cors/HttpServerInitializer.java" ]

[ "codec-http/src/test/java/io/netty/handler/codec/http/cors/CorsConfigTest.java", "codec-http/src/test/java/io/netty/handler/codec/http/cors/CorsHandlerTest.java" ]

[ "timestamp(timedelta=96.0, similarity=0.882593498387067)" ]

[ "Assigned to me... all the JNI fun for me :)\n" ]

[ "Use `{@code ...}` to signify parameter names, everywhere possible.\n", "@trustin I think I could even remove the javadocs as these are on the interfaces anyway ?\n", "`s/ / /` - Please fix anything like this everywhere.\n" ]

[ "feature" ]

[ "transport-native-epoll/src/main/c/io_netty_channel_epoll_Native.c", "transport-native-epoll/src/main/c/io_netty_channel_epoll_Native.h", "transport-native-epoll/src/main/java/io/netty/channel/epoll/AbstractEpollChannel.java", "transport-native-epoll/src/main/java/io/netty/channel/epoll/EpollServerSocketChannel.java", "transport-native-epoll/src/main/java/io/netty/channel/epoll/EpollSocketChannel.java", "transport-native-epoll/src/main/java/io/netty/channel/epoll/Native.java" ]

[ "transport-native-epoll/src/main/c/io_netty_channel_epoll_Native.c", "transport-native-epoll/src/main/c/io_netty_channel_epoll_Native.h", "transport-native-epoll/src/main/java/io/netty/channel/epoll/AbstractEpollChannel.java", "transport-native-epoll/src/main/java/io/netty/channel/epoll/EpollDatagramChannel.java", "transport-native-epoll/src/main/java/io/netty/channel/epoll/EpollDatagramChannelConfig.java", "transport-native-epoll/src/main/java/io/netty/channel/epoll/EpollDatagramChannelOutboundBuffer.java", "transport-native-epoll/src/main/java/io/netty/channel/epoll/EpollServerSocketChannel.java", "transport-native-epoll/src/main/java/io/netty/channel/epoll/EpollSocketChannel.java", "transport-native-epoll/src/main/java/io/netty/channel/epoll/Native.java" ]

[ "transport-native-epoll/src/test/java/io/netty/channel/epoll/EpollDatagramUnicastTest.java", "transport-native-epoll/src/test/java/io/netty/channel/epoll/EpollSocketTestPermutation.java" ]

[ "timestamp(timedelta=27.0, similarity=0.9437644332585227)" ]

[ "There are a couple places where client/server context is important, so it might end up propagating itself anyway.\n\nFor example, push_promise setting should only be sent by the client, push promises should only be sent by the server. client vs server stream ids (already handled I think).\n\nThat said, I don't mind however this is implemented, as long as it becomes implemented!\n", "I went with option 2 to see what that would look like, and I kind of like the result.\n", "Fixed by #2403\n" ]

[ "This is to prevent control from passing on if we're still reading the preface. Could probably change this to checking if any bytes are left in the incoming ByteBuf that haven't been consumed yet.\n", "Note! Advice wanted.\n", "nit accidental wildcard\n", "IntelliJ keeps doing this automatically no matter what I do, but only for the Netty project. I assume this is actually what the Netty style settings want?\n", "NM: not accidental!! lots of this in the netty codebase... ignore me.\n", "Maybe both ?\n", "You could just use `preface.isReadable()` :)\n", "Both of which of the 3?\n- Close channel\n- Throw exception synchronously up the I/O stack\n- ctx.fire?\n", "Nice, will fix.\n", "So I debugged through and it turns out simply throwing a synchronous exception does exactly what you'd want.\n\n1) It's caught by the catch in ChannelHandlerInvokerUtil.invokeChannelReadNow which fires the exception down the handler chain.\n\n2) Http2ConnectionHandler.exceptionCaught sees the connection error, sends a GoAway frame, and closes the channel.\n\nSo I think anything other than synchronous throw is redundant here, assuming the handler chain is already setup.\n", "You need to call ((ByteBuf) msg).release() before you break to not introduce a buffer leak here.\n", "Oh, I see. Because I'm not sending it on down the pipeline, right?\n", "Exactly :)\n\n-- \nNorman Maurer\n\nAm 17. April 2014 bei 21:28:19, Scott Blum ([email protected]) schrieb:\n\nIn codec-http2/src/main/java/io/netty/handler/codec/http2/draft10/frame/decoder/Http2ServerPrefaceReader.java:\n\n> +\n> +/**\n> - \\* Reads the initial client preface, then removes itself from the pipeline.\n> - \\* Only the server pipeline should do this.\n> - *\n> - \\* https://tools.ietf.org/html/draft-ietf-httpbis-http2-10#section-3.5\n> - */\n> +public class Http2ServerPrefaceReader extends ChannelHandlerAdapter {\n> +\n> - private final ByteBuf preface = connectionPrefaceBuf();\n> +\n> - @Override\n> - public void channelRead(ChannelHandlerContext ctx, Object msg) throws Exception {\n> - if (preface.isReadable() && msg instanceof ByteBuf) {\n> - processHttp2Preface(ctx, (ByteBuf) msg);\n> - if (preface.isReadable()) {\n> Oh, I see. Because I'm not sending it on down the pipeline, right?\n\n—\nReply to this email directly or view it on GitHub.\n" ]

[]

[ "codec-http2/src/main/java/io/netty/handler/codec/http2/draft10/Http2OrHttpChooser.java", "codec-http2/src/main/java/io/netty/handler/codec/http2/draft10/connection/Http2ConnectionHandler.java", "codec-http2/src/main/java/io/netty/handler/codec/http2/draft10/frame/decoder/Http2FrameDecoder.java", "codec-http2/src/main/java/io/netty/handler/codec/http2/draft10/frame/encoder/Http2FrameEncoder.java", "example/src/main/java/io/netty/example/http2/client/Http2ClientInitializer.java" ]

[ "codec-http2/src/main/java/io/netty/handler/codec/http2/draft10/Http2OrHttpChooser.java", "codec-http2/src/main/java/io/netty/handler/codec/http2/draft10/connection/Http2ConnectionHandler.java", "codec-http2/src/main/java/io/netty/handler/codec/http2/draft10/frame/decoder/Http2FrameDecoder.java", "codec-http2/src/main/java/io/netty/handler/codec/http2/draft10/frame/decoder/Http2ServerPrefaceReader.java", "codec-http2/src/main/java/io/netty/handler/codec/http2/draft10/frame/encoder/Http2ClientPrefaceWriter.java", "codec-http2/src/main/java/io/netty/handler/codec/http2/draft10/frame/encoder/Http2FrameEncoder.java", "example/src/main/java/io/netty/example/http2/client/Http2ClientInitializer.java" ]

[]

[ "timestamp(timedelta=22.0, similarity=0.856703916914292)" ]

[ "Sure a fix would be awesome. We love contributions :)\n\n> Am 18.04.2014 um 20:30 schrieb mkrueger92 [email protected]:\n> \n> Netty version: 4.0.18.Final\n> \n> Context:\n> I am using snappy to send a compressed TCP data stream of arbitrary data. The peer application has spit out a warning that the stream started with the wrong chunk type. (0x80 instead of 0xff)\n> \n> In my case this is not such a big problem because 0x80 frames are within the skippable chunk range. But the warning points to a potential problem for other cases.\n> \n> Please see https://code.google.com/p/snappy/source/browse/trunk/framing_format.txt#68 for more information.\n> \n> If wanted could look into the Netty code and provide a patch. Just let me know.\n> \n> —\n> Reply to this email directly or view it on GitHub.\n", "@normanmaurer: Please have a look at the commit. If this one meets your quality requirements I can provide an additional patch for the 4.0 and 4.1 branch as well if needed.\n", "Looks correct :)\n\n> Am 18.04.2014 um 22:19 schrieb mkrueger92 [email protected]:\n> \n> @normanmaurer: Please have a look at the commit. If this one meets your quality requirements I can provide an additional patch for the 4.0 and 4.1 branch as well if needed.\n> \n> —\n> Reply to this email directly or view it on GitHub.\n", "@mkrueger92 create a PR and I will merge in\n", "Fixed via #2411 thanks!\n" ]

[]

[]

[ "codec/src/main/java/io/netty/handler/codec/compression/SnappyFramedDecoder.java", "codec/src/main/java/io/netty/handler/codec/compression/SnappyFramedEncoder.java" ]

[ "codec/src/main/java/io/netty/handler/codec/compression/SnappyFramedDecoder.java", "codec/src/main/java/io/netty/handler/codec/compression/SnappyFramedEncoder.java" ]

[ "codec/src/test/java/io/netty/handler/codec/compression/SnappyFramedDecoderTest.java", "codec/src/test/java/io/netty/handler/codec/compression/SnappyFramedEncoderTest.java" ]

[ "timestamp(timedelta=27.0, similarity=0.8576720900504698)" ]

[ "@Teots can I see your \"dirty reflection hack\" ;) ?\n", "Sure you can. Here we go:\n\n```\n// channel is passed into the function and is of type Channel\nif (channel instanceof LocalChannel) {\n Class<?> clazz = channel.getClass();\n Field stateField = clazz.getDeclaredField(\"state\");\n stateField.setAccessible(true);\n stateField.setInt(channel, 3);\n}\n\nchannel.deregister().addListener(new ChannelFutureListener() {\n @Override\n public void operationComplete(ChannelFuture future) throws Exception {\n try {\n Channel channel = future.channel();\n EventLoopGroup eventLoopGroup = \n if (channel instanceof LocalChannel) {\n eventLoopGroup = Helper.localInputEventLoopGroup;\n\n Class<?> clazz = channel.getClass();\n Field stateField = clazz.getDeclaredField(\"state\");\n stateField.setAccessible(true);\n stateField.setInt(channel, 0);\n\n Field peerField = clazz.getDeclaredField(\"peer\");\n peerField.setAccessible(true);\n LocalChannel peer = (LocalChannel) peerField.get(channel);\n\n Field connectPromiseField = clazz.getDeclaredField(\"connectPromise\");\n connectPromiseField.setAccessible(true);\n connectPromiseField.set(peer, peer.unsafe().voidPromise());\n\n // Change event loop group.\n eventLoopGroup.register(channel).sync();\n } else {\n eventLoopGroup = Helper.networkInputEventLoopGroup;\n\n // Change event loop group.\n eventLoopGroup.register(channel).sync();\n }\n\n // Enable auto read again.\n channel.config().setAutoRead(true);\n } catch (Throwable t) {\n LOG.error(t.getLocalizedMessage(), t);\n }\n }\n}).sync();\n```\n", "@Teots could you please check if the PR #2413 fix the issue for you ?\n" ]

[ "'Check if both peer and parent are non-null because this channel was created by a ..'\n\nI would always prefer 'because' to 'as' because it's less confusing to a reader.\n" ]

[ "defect" ]

[ "transport/src/main/java/io/netty/channel/local/LocalChannel.java" ]

[ "transport/src/main/java/io/netty/channel/local/LocalChannel.java" ]

[ "transport/src/test/java/io/netty/channel/local/LocalChannelTest.java" ]

[ "timestamp(timedelta=7073.0, similarity=0.8421856332718755)" ]

[ "Forgot to mention that I used nightly build from oss.sonatype.org: netty-example-5.0.0.Alpha2-20140508.064249-382.jar and netty-all-5.0.0.Alpha2-20140508.064500-374.jar\n", "@nmittler could you check ?\n", "Yup, will take a look tomorrow. Should be easy enough to fix. Thanks for catching this @tatsuhiro-t \n", "Created PR https://github.com/netty/netty/pull/2480\n", "I think this should be fixed now that #2480 is pulled in\n" ]

[]

[]

[ "codec-http2/src/main/java/io/netty/handler/codec/http2/AbstractHttp2ConnectionHandler.java", "codec-http2/src/main/java/io/netty/handler/codec/http2/DefaultHttp2Connection.java", "codec-http2/src/main/java/io/netty/handler/codec/http2/DefaultHttp2FrameReader.java", "codec-http2/src/main/java/io/netty/handler/codec/http2/DefaultHttp2FrameWriter.java", "codec-http2/src/main/java/io/netty/handler/codec/http2/DefaultHttp2PriorityTree.java", "codec-http2/src/main/java/io/netty/handler/codec/http2/Http2CodecUtil.java", "codec-http2/src/main/java/io/netty/handler/codec/http2/Http2Connection.java", "example/src/main/java/io/netty/example/http2/client/Http2ClientConnectionHandler.java" ]

[ "codec-http2/src/main/java/io/netty/handler/codec/http2/AbstractHttp2ConnectionHandler.java", "codec-http2/src/main/java/io/netty/handler/codec/http2/DefaultHttp2Connection.java", "codec-http2/src/main/java/io/netty/handler/codec/http2/DefaultHttp2FrameReader.java", "codec-http2/src/main/java/io/netty/handler/codec/http2/DefaultHttp2FrameWriter.java", "codec-http2/src/main/java/io/netty/handler/codec/http2/DefaultHttp2PriorityTree.java", "codec-http2/src/main/java/io/netty/handler/codec/http2/Http2CodecUtil.java", "codec-http2/src/main/java/io/netty/handler/codec/http2/Http2Connection.java", "example/src/main/java/io/netty/example/http2/client/Http2ClientConnectionHandler.java" ]

[ "codec-http2/src/test/java/io/netty/handler/codec/http2/DefaultHttp2ConnectionTest.java", "codec-http2/src/test/java/io/netty/handler/codec/http2/DefaultHttp2FrameIOTest.java", "codec-http2/src/test/java/io/netty/handler/codec/http2/DelegatingHttp2ConnectionHandlerTest.java", "codec-http2/src/test/java/io/netty/handler/codec/http2/Http2FrameRoundtripTest.java" ]

[ "timestamp(timedelta=42.0, similarity=0.9299720569086561)" ]

[ "The following test demonstrates the problem. This is run with the netty 4.0.15.Final version of Netty.\n\n``` java\n\nimport io.netty.bootstrap.Bootstrap;\nimport io.netty.bootstrap.ServerBootstrap;\nimport io.netty.buffer.ByteBuf;\nimport io.netty.buffer.Unpooled;\nimport io.netty.channel.Channel;\nimport io.netty.channel.ChannelFuture;\nimport io.netty.channel.ChannelFutureListener;\nimport io.netty.channel.ChannelHandlerContext;\nimport io.netty.channel.ChannelInboundHandlerAdapter;\nimport io.netty.channel.ChannelInitializer;\nimport io.netty.channel.nio.NioEventLoopGroup;\nimport io.netty.channel.socket.nio.NioServerSocketChannel;\nimport io.netty.channel.socket.nio.NioSocketChannel;\nimport io.netty.handler.traffic.ChannelTrafficShapingHandler;\n\nimport java.net.SocketAddress;\nimport java.util.zip.CRC32;\n\nimport org.junit.Assert;\nimport org.junit.Test;\n\n\npublic class TrafficTest {\n\n @Test\n public void testTrafficShaping() throws Exception {\n testTrafficShapingHandler(100, false);\n System.out.println(\"OK without ChannelTrafficShapingHandler\");\n\n for (int i = 1; i <= 10; i++) {\n long sleep = i * 100;\n System.out.print(\"Test with \" + sleep + \"ms sleep..\");\n testTrafficShapingHandler(sleep, true);\n System.out.println(\" OK\");\n }\n\n }\n\n public void testTrafficShapingHandler(long sleepWhenNotWritable, final boolean useTrafficShaping) throws Exception {\n\n final Crc32Sink sink = new Crc32Sink();\n\n ServerBootstrap serverBootstrap = new ServerBootstrap();\n serverBootstrap.channel(NioServerSocketChannel.class).group(new NioEventLoopGroup()).localAddress(\"localhost\", 0);\n\n serverBootstrap.childHandler(new ChannelInitializer<Channel>() {\n\n @Override\n protected void initChannel(Channel channel) throws Exception {\n if (useTrafficShaping) {\n channel.pipeline().addLast(new ChannelTrafficShapingHandler(0, 128 * 1024));\n }\n channel.pipeline().addLast(sink);\n }\n });\n\n Bootstrap clientBootstrap = new Bootstrap();\n clientBootstrap.channel(NioSocketChannel.class).group(new NioEventLoopGroup());\n clientBootstrap.handler(new ChannelInboundHandlerAdapter());\n\n try {\n\n ChannelFuture cf = serverBootstrap.bind().sync();\n SocketAddress serverAddress = cf.channel().localAddress();\n\n System.out.println(\"Server address = \" + serverAddress);\n\n Channel clientChannel = clientBootstrap.connect(serverAddress).sync().channel();\n\n byte[] data = new byte[512 * 1024];\n\n for (int i = 0; i < data.length; i++) {\n data[i] = (byte) (i % 31);\n }\n\n CRC32 crc32 = new CRC32();\n for (int i = 0; i < data.length; i++) {\n crc32.update(data[i]);\n }\n\n for (int i = 0; i < 256; i++) {\n ByteBuf slice = Unpooled.wrappedBuffer(data, 2048 * i, 2048);\n while (!clientChannel.isWritable()) {\n Thread.sleep(sleepWhenNotWritable);\n }\n clientChannel.writeAndFlush(slice);\n }\n\n while (sink.getByteCount() < data.length) {\n Thread.sleep(100);\n }\n\n clientChannel.writeAndFlush(Unpooled.EMPTY_BUFFER).addListener(ChannelFutureListener.CLOSE);\n sink.getChannel().closeFuture().sync();\n\n Assert.assertEquals(data.length, sink.getByteCount());\n Assert.assertEquals(crc32.getValue(), sink.getCrc32());\n\n } finally {\n clientBootstrap.group().shutdownGracefully();\n serverBootstrap.group().shutdownGracefully();\n }\n\n }\n\n private static class Crc32Sink extends ChannelInboundHandlerAdapter {\n\n private final CRC32 crc32 = new CRC32();\n\n private volatile long byteCount = 0;\n\n private Channel channel;\n\n @Override\n public void channelActive(ChannelHandlerContext ctx) throws Exception {\n channel = ctx.channel();\n }\n\n @Override\n public void channelRead(ChannelHandlerContext ctx, Object msg) throws Exception {\n ByteBuf buf = (ByteBuf) msg;\n byteCount += buf.readableBytes();\n for (int i = buf.readerIndex(); i < buf.writerIndex(); i++) {\n crc32.update(buf.readByte());\n }\n\n buf.release();\n }\n\n long getCrc32() {\n return crc32.getValue();\n }\n\n long getByteCount() {\n return byteCount;\n }\n\n Channel getChannel() {\n return channel;\n }\n\n }\n\n}\n\n```\n", "@fredericBregier Care to check?\n", "I am having the same problem!\n", "In AbstractTrafficShapingHandler.java, why are we scheduling fireChannelRead (inside channelRead method). We received a byte seq, we started processing it by passing it to pipeline and by scheduling we are just delaying the process of next handler. I don't think there is a need for delaying the next handler from processing the inbound data. When I remove that scheduling code it is working fine.\n", "@trustin I had lost this one (sorry) but proposal seems ok for me.\nHowever build seems in error: unstable for ByteBuf memory check since not released, but it seems totally unrelated to the patch, right ?\n", "Fix merged...\n" ]

[]

[ "defect" ]

[ "handler/src/main/java/io/netty/handler/traffic/AbstractTrafficShapingHandler.java" ]

[ "handler/src/main/java/io/netty/handler/traffic/AbstractTrafficShapingHandler.java" ]

[]

[ "timestamp(timedelta=87.0, similarity=0.8885360305547618)" ]

[ "Yeah, I was quite lazy when writing this. Another food for thought is: do we really need a swapped buffer? Why don't we just add a little endian getters and setters into `ByteBuf`? It will make it a little bit crowded but it actually should simplify things.\n", "@trustin lol :) What you mean with getter and setter here ? Just allow to change the endianess on the fly (like what ByteBuffer does) ?\n", "Nope. Just having additional methods:\n\n```\nByteBuf buf = ...;\nint bigEndianValue = buf.readInt();\nint littleEndialValue = buf.readIntL();\n```\n", "sounds legit... This would also reduce the GC pressure as we can eliminate the extra object creation for the SwappedByteBuf. But I think we can only do this for 4.1 and master. I would like to come up with some \"fix/workaround\" for 4.0 too. \n", "Go ahead.\n", "@trustin lol yeah... any preferences how to fix this for 4.0.x ? (I can also come up with some PR based on your idea for 4.1 and master).\n", "I prefer to leave 4.0 as is unless it's a critical problem (you can define 'critical' by yourself :-))\n", "All the power for me :) I will think about it buddy and come up with something for 4.1 and master in the meantime.\n", "@pavanka I just pushed a commit into 4.0, 4.1 and master branches. This should be part of the next 4.0.20.Final release. Thanks for reporting !\n" ]

[]

[ "improvement" ]

[ "buffer/src/main/java/io/netty/buffer/AbstractByteBuf.java", "buffer/src/main/java/io/netty/buffer/PooledUnsafeDirectByteBuf.java", "buffer/src/main/java/io/netty/buffer/SwappedByteBuf.java", "buffer/src/main/java/io/netty/buffer/UnpooledUnsafeDirectByteBuf.java", "microbench/pom.xml" ]

[ "buffer/src/main/java/io/netty/buffer/AbstractByteBuf.java", "buffer/src/main/java/io/netty/buffer/PooledUnsafeDirectByteBuf.java", "buffer/src/main/java/io/netty/buffer/SwappedByteBuf.java", "buffer/src/main/java/io/netty/buffer/UnpooledUnsafeDirectByteBuf.java", "buffer/src/main/java/io/netty/buffer/UnsafeDirectSwappedByteBuf.java", "microbench/pom.xml" ]

[ "microbench/src/test/java/io/netty/microbench/buffer/SwappedByteBufBenchmark.java" ]

[ "timestamp(timedelta=48801.0, similarity=0.8672136286651443)" ]

[ "@philbaxter thanks merged in :)\n" ]

[]

[]

[ "pom.xml" ]

[ "pom.xml" ]

[]

[ "timestamp(timedelta=120.0, similarity=0.9140302153138531)" ]

[ "@trustin so either we let ioBuffer() not use size of 0 as default or allow to configure it. I would prefer the first. WDYT ?\n", "It's due to a historical reason - long time ago in the early days of 4.0.AlphaX, we allocated a buffer per handler. Having a non-zero capacity buffer from the beginning of the connection was a huge waste.\n\nWe are not affected by that problem anymore, so I think we are fine to use the same default initial capacity for `ioBuffer()`.\n\nActually, an encoder implementation should know better about the initial capacity of the buffer. How about adding a protected method?\n\n``` java\n protected ByteBuf allocateBuffer(ChannelHandlerContext ctx, I msg) throws Exception {\n // default impl:\n if (preferDirect) {\n return ctx.alloc().ioBuffer();\n } else {\n return ctx.alloc().heapBuffer();\n }\n }\n```\n\nThe overriding methods need to know if the handler prefers a direct buffer or not, so maybe we should make `preferDirect` protected, add a new protected method `isDirectBufferPreferred()`, or just pass it as a parameter to `allocateBuffer()`.\n", "We could also remove the constructor that accepts `preferDirect` and let `allocateBuffer()` use `ioBuffer()` by default. If a user prefers a heap buffer, he or she can still override it.\n\nFor 4.0/4.1, we could simply deprecate that constructor and let the default `allocateBuffer()` impl respect `preferDirect`.\n", "What about adding another constructor that take preferDirect and initialCapacity as parameter and deprecate the old constructors that accept preferDirect? The initialCapacity will be used for direct and heap buffers.\n\nWDYT?\n\n> Am 21.06.2014 um 12:35 schrieb Trustin Lee [email protected]:\n> \n> We could also remove the constructor that accepts preferDirect and let allocateBuffer() use ioBuffer() by default. If a user prefers a heap buffer, he or she can still override it.\n> \n> For 4.0/4.1, we could simply deprecate that constructor and let the default allocateBuffer() impl respect preferDirect.\n> \n> —\n> Reply to this email directly or view it on GitHub.\n", "I think we should add a protected method that allows a handler to allocate a buffer in a contextual way, because the initial capacity of a buffer really depends on a message being encoded.\n", "Let me come up with a PR\n\n> Am 22.06.2014 um 05:04 schrieb Trustin Lee [email protected]:\n> \n> I think we should add a protected method that allows a handler to allocate a buffer in a contextual way, because the initial capacity of a buffer really depends on a message being encoded.\n> \n> —\n> Reply to this email directly or view it on GitHub.\n", "Fixed by #2591 \n" ]

[]

[ "improvement" ]

[ "buffer/src/main/java/io/netty/buffer/AbstractByteBufAllocator.java", "buffer/src/main/java/io/netty/buffer/ByteBufAllocator.java", "codec/src/main/java/io/netty/handler/codec/MessageToByteEncoder.java" ]

[ "buffer/src/main/java/io/netty/buffer/AbstractByteBufAllocator.java", "buffer/src/main/java/io/netty/buffer/ByteBufAllocator.java", "codec/src/main/java/io/netty/handler/codec/MessageToByteEncoder.java" ]

[]

[ "timestamp(timedelta=28887.0, similarity=0.8797419366114805)" ]

[ "Pull request accepted...issue closed.\n" ]

[]

[]

[ "example/src/main/java/io/netty/example/http2/client/Http2ClientConnectionHandler.java" ]

[ "example/src/main/java/io/netty/example/http2/client/Http2ClientConnectionHandler.java" ]

[]

[ "timestamp(timedelta=35.0, similarity=1.0000000000000002)" ]

[ "> roughly 1.79 GB or 1.40% of the GC pressure during performance testing\n\nOn a curious sidenote, what profiler do you use to get to these numbers?\n", "@jakobbuchgraber you can use Java Mission Control for this.\n", "@jakobbuchgraber I used Java Mission Control's Flight Record to get this data.\n" ]

[]

[ "improvement" ]

[ "buffer/src/main/java/io/netty/buffer/CompositeByteBuf.java" ]

[ "buffer/src/main/java/io/netty/buffer/CompositeByteBuf.java" ]

[]

[ "timestamp(timedelta=41.0, similarity=0.9051057703458023)" ]

[ "opinali@ thanks for the find! Do you think you could change the code to capture the state of the map when the assertion occurs?\n", "@nmittler shouldn't be hard; let me try this.\n", "well, it's being harder than expected... I have added code to dump the map state, but after redeploy, couldn't reproduce the error anymore. I hope it's not some of those obscure javac or JIT bugs where any perturbation to the application code makes the symptoms go away.\n\nWell, I will keep this debug code in my test deployments for some time, I usually have a test server running 24x7 anyway. Will post if I get lucky to capture the bug again.\n", "Ok, testing failed so I'm reviewing the code. This part of insertAt() looks fishy:\n\n```\n } else if (available == 0) {\n // Open addressing requires that we have at least 1 slot available. Need to refresh\n // the arrays to clear any removed elements.\n rehash(capacity());\n }\n```\n\nShouldn't that be rehash(capacity() + 1) at least? This will rehash the hashmap to the same size it already has, violating the rule of keeping at least one slot available. And if you violate that, the next call to put() won't find any AVAILABLE/REMOVED slot, and because the capacity check/rehash is only performed by insertAt(), put() doesn't call that unless there's some available slot, so this will produce exactly the assertion I saw.\n", "@opinali sounds fishy for sure. Shouldn't we also double it and not just use +1 ?\n\nSomething like:\n\n``` java\nrehash(capacity() <<1);\n```\n\n@nmittler @opinali wdyt ?\n", "Yes I see no reason to not double (or use whatever same growing factor used by the size>maxSize case). The code could be simplified to:\n\n```\n if (size > maxSize || available == 0) {\n rehash(capacity() * 2);\n }\n```\n", "@opinali yep sounds good! I would make the capacity odd, however (add 1). Would you mind making a pull request?\n", "Actually I take that back, the +1 will be done automatically, so rehash(capacity() \\* 2) should be sufficient.\n", "@nmittler could you submit a patch with a unit case?\n\n> Am 16.07.2014 um 21:35 schrieb Nathan Mittler [email protected]:\n> \n> Actually I take that back, the +1 will be done automatically, so rehash(capacity() \\* 2) should be sufficient.\n> \n> —\n> Reply to this email directly or view it on GitHub.\n", "size*1.5+1 is usually a better space/speed deal unless the algorithm\nrequires power-of-2 sizes which is not the case here. Wouldn't hurt also to\nguard against overflow when size approaches MAX_VALUE... not likely to ever\nhappen for the channel map but maybe you want to use this for something\nelse later.\nOn Jul 16, 2014 3:57 PM, \"Norman Maurer\" [email protected] wrote:\n\n> @nmittler could you submit a patch with a unit case?\n> \n> > Am 16.07.2014 um 21:35 schrieb Nathan Mittler [email protected]:\n> > \n> > Actually I take that back, the +1 will be done automatically, so\n> > rehash(capacity() \\* 2) should be sufficient.\n> > \n> > —\n> > Reply to this email directly or view it on GitHub.\n> \n> —\n> Reply to this email directly or view it on GitHub\n> https://github.com/netty/netty/issues/2659#issuecomment-49218191.\n", "@opinali I'm starting to think the original code was correct. maxSize is the size at which we have to grow (defaults to 0.5 \\* capacity). If we add a bunch of entries and then delete them, the number of AVAILABLE slots could eventually become 0 since they have been marked as REMOVED. In this case, we rehash with the current capacity to clear out the REMOVED nodes and make them AVAILABLE again. We shouldn't need to grow in this case.\n\nI suspect that the problem you're running into is related to the choice of capacity. This choice is important for reducing collisions in open addressing and to this point I've just been ensuring that it's odd. Using prime numbers for capacity, however, is the typical approach and I think we should migrate to this before making any other changes.\n\nSound reasonable?\n", "Hmm yeah\n\n> Am 17.07.2014 um 19:34 schrieb Nathan Mittler [email protected]:\n> \n> @opinali I'm starting to think the original code was correct. maxSize is the size at which we have to grow (defaults to 0.5 \\* capacity). If we add a bunch of entries and then delete them, the number of AVAILABLE slots could eventually become 0 since they have been marked as REMOVED. In this case, we rehash with the current capacity to clear out the REMOVED nodes and make them AVAILABLE again. We shouldn't need to grow in this case.\n> \n> I suspect that the problem you're running into is related to the choice of capacity. This choice is important for reducing collisions in open addressing and to this point I've just been ensuring that it's odd. Using prime numbers for capacity, however, is the typical approach and I think we should migrate to this before making any other changes.\n> \n> Sound reasonable?\n> \n> —\n> Reply to this email directly or view it on GitHub.\n", "@normanmaurer do you have any concerns?\n", "@nmittler nope... go ahead\n", "Here's a easy test that triggers the bug. (Forcing certain values including loadFactor=1.0 just to make it quicker to catch in step-by-step debugging; it's trivial to reproduce with any loadFactor.)\n\npublic class IntObjectHashMapTest {\n @Test\n public void test() {\n IntObjectHashMap<String> map = new IntObjectHashMap<>(5 \\* 2, 1.0f);\n for (int i = 0; i < 5; ++i) {\n map.put(i \\* 5 /\\* hash collision */, \"\");\n }\n }\n}\n\n@nmittler was right, the problem is elsewhere. The probing logic in put() is incorrect; see:\ncapacity=10, key=20 => increment=5\nindex will be: 0, 5, 10 wraps to 0 which stops the loop.\nOnly slots 0 and 5 are ever visited by the loop, then it fails.\nThis logic will never work in all cases, unless the capacity size is guaranteed to always be a prime number (there's a comment about that; but it's probably a bad idea because, for large N, finding the next prime >N is very inefficient --- best choice is BigInteger.nextProbablePrime() but this is overkill so you need to reimplement that algorithm for ints and that will still be outrageously slow for this purpose).\nI understand that the intention is to scatter spills across the whole array to avoid generate performance for specific buckets, but there's always some scenario where this will happen. You could just use increment=1 and let bucket distribution depend uniquely from element hashes.\n", "@opinali btw we love contributions ;) So feel free to submit a PR with fix + text-case.\n", "@opinali The way Trove (and likely other libraries) deal with this is to use a lookup table for the next prime. Basically they take advantage of the fact that they always expand by a factor of 2 to allow them to reduce the table size. When they expand, they do a binary search to get the first value in the table that's >= the requested capacity. This should be trivial to implement and the adjustCapacity() method was already separated out so that we could add the logic there.\n", "Good idea the lookup table. For my emergency needs I've implemented the ++index with wrapping to zero; tested it and works perfectly. It seems the table size stabilizes at 11264 with my test load. Maybe a good option if you think this issue must be fixed quickly in a maintenance release. \n\nThe rehash function could be optimized, right now it just loops invoking put() but it could have a much simpler loop without special cases like REMOVED entries and checks for capacity etc. Rehashing cost can be a problem for huge hashmaps, remarkably for high-scale & latency-sensitive code; once again it's not a priority for the current usage of this map, just another thing we might improve easily.\n\nI'm always suspicious of each new attempt to optimize open hashing, I immediately remember Peter Brass's book where he strongly discourages this technique and and the production of any new papers about the subject. :) But admittedly, in a system like the JVM where linked nodes are too expensive for primitive-value keys like ints, the open hashing (being the only way to resolve conflicts without boxing) should beat its disadvantages. Gonna check this later, I hope to have some time in the next few days to play with this and send a good patch.\n", "@opinali @nmittler I think we can do an easy fix fast and then think about how to improve it. WDYT ?\n", "Oops, I was too eager with the last comment (deleted), had a test bug. So I'm working on a patch, but also worried with indexOf(). This method works correctly (after applying the same probing fix), but here the problem is that it degenerates to linear scan of potentially the whole array if we have too many REMOVED entries. I guess this is already know, there's a TODO for implementing re-compaction which is necessary to avoid that problem. This is critical because the problem is not some edge case, e.g. doesn't affect only maps that have too much conflicts for a specific key: any map that's completely filled up, and then completely drained (but only with individual removes, never clear()), will have all states set to REMOVED, and once this happens, if you add any two keys that conflict and remove the first and try to find the second, this will always cost O(capacity). Anyway, probably punting on this for now. I could add a simple check to remove—if the map becomes empty, then fill states with AVAILABLE—but this would backfire if the map repeatedly transitions between few elements and empty.\n", "@opinali @normanmaurer I know we're concerned about external dependencies in Netty but I'm concerned that we're reinventing the wheel here. Other projects already have working implementations that have been through more rigorous testing. Should we start shopping?\n", "@nmittler @normanmaurer +1 to shop versus build it. Hacking fundamental data structures is fun, but it's an endless rabbit hole. For my emergency needs I need the patched IntObjectHashMap so I'm working on it even if next Netty will drop it, will send the patch anyway, in case the shopping option is not easy due to compatible licenses or other reason.\n", "@nmittler @opinali the problem is that Trove is LGPL so we can only add is as optional dependency and this makes it also impossible to just fork their impl :( \n", "@nmittler @opinali I did not check in detail but I found this impl to be ASL2:\n\nhttps://github.com/android/platform_frameworks_base/blob/master/core/java/android/util/SparseArray.java\nhttps://github.com/libgdx/libgdx/blob/master/gdx/src/com/badlogic/gdx/utils/IntMap.java\n\nSo maybe we could just fork SparseArray ?\n", "+1 for SparseArray\n", "@nmittler so maybe just fork this into internal package and deprecate the collection package with classes. @nmittler @opinali sounds good ?\n", "@normanmaurer That sounds good to me. I'll see if I have time this weekend to work on it.\n", "@opinali beside this we could also just use HashMap in EpollEventLoop for now as I think it should be good enough at the end. WDYT?\n\n> Am 18.07.2014 um 21:35 schrieb Nathan Mittler [email protected]:\n> \n> @normanmaurer That sounds good to me. I'll see if I have time this weekend to work on it.\n> \n> —\n> Reply to this email directly or view it on GitHub.\n", "@normanmaurer @nmittler SparseArray seems to be the better of these two options, but according to its docs only good for really small maps, up to \"hundreds\" of elements. My test goes up to 10K elements, and I could go higher than that with a big instance with 10Gbit connection. But I have no idea how critical is the performance of this map for Netty (I suppose it's at least significant enough to have motivated the recent introduction of IntObjectHashMap?). Another important factor in the shop-vs-build decision is that, if we have to fork and maybe do any adaptation in the \"shopped\" code, and tracking changes to the source project and merging new stuff etc., that's what was already done with IntObjectHashMap so we might just as well keep it (once fixed). If we don't have the option of black-box reuse (just depend on someone else's jar file and not mess with its impl), having to cherry-pick and fork some code, then we should fully own that code.\n\nI've got carried away and made significant improvements on this code already, including limited recompaction and new tests. Sending patch later today or tomorrow... but let's keep options open anyway.\n", "True enough... So maybe let us just fix ours :)\n\n> Am 19.07.2014 um 16:45 schrieb Osvaldo Pinali Doederlein [email protected]:\n> \n> @normanmaurer @nmittler SparseArray seems to be the better of these two options, but according to its docs only good for really small maps, up to \"hundreds\" of elements. My test goes up to 10K elements, and I could go higher than that with a big instance with 10Gbit connection. But I have no idea how critical is the performance of this map for Netty (I suppose it's at least significant enough to have motivated the recent introduction of IntObjectHashMap?). Another important factor in the shop-vs-build decision is that, if we have to fork and maybe do any adaptation in the \"shopped\" code, and tracking changes to the source project and merging new stuff etc., that's what was already done with IntObjectHashMap so we might just as well keep it (once fixed). If we don't have the option of black-box reuse (just depend on someone else's jar file and not mess w ith its impl), having to cherry-pick and fork some code, then we should fully own that code.\n> \n> I've got carried away and made significant improvements on this code already, including limited recompaction and new tests. Sending patch later today or tomorrow... but let's keep options open anyway.\n> \n> —\n> Reply to this email directly or view it on GitHub.\n", "@opinali agreed - I recall looking at SparseArray but I think I went down a different path for that reason. Ping me once you get the patch together ... thanks for the contrib!\n", "@nmittler @normanmaurer Pull request sent!\n", "Nice ! Will review today\n\n> Am 20.07.2014 um 00:24 schrieb Osvaldo Pinali Doederlein [email protected]:\n> \n> @nmittler @normanmaurer Pull request sent!\n> \n> —\n> Reply to this email directly or view it on GitHub.\n", "Fix was pushed\n" ]

[ "Please use for (;;) as we use for(;;) for this kind of stuff everywhere in netty. Just for the sake of consistency\n", "IllegalStateException ?\n", "I think if we do v.equals(values[i]) we can eliminate the null check\n", "return false if null is given as value\n", "same as above\n", "Please replace with for (;;)\n", "Store oldVals[i] in local variable so we not need to double access the array in case it is not null\n", "for (;;)\n", "Store oldKeys[i] local variable so we can use it again without the need to access the array again\n", "for (;;)\n", "Maybe use some better initialCapacit for the builder ?\n", "Store source.values[i] in local variable so we can re-use it if not null and not need again to lookup the array.\n", "Maybe keep the AVAILABLE just to make it easier to understand ?\n", "Store keys[i] in local variable so we not need to access the array again\n", "I prefer that too, done\n", "This would be incorrect here, as we now use values, not keys, to know which fields are available... and for reference types 'null' has a natural \"unused\" meaning by default, so I think a constant AVAILABLE=null would only obfuscate things.\n", "done\n", "The map still supports null :) there's a trick here, will add as documentation.\n", "This is a very low-level optimization: nulls are expected to be very common (at least 50% of all slots even in a \"full\" map with the default load factor 0.5), so I fast-path this case before making any invocation, will be faster even if the equals call is inlined (which I don't expect to happen, in the general case it's a megamorphic call since you can put any ind of object in the map).\n", "Ah right... Maybe add a comment to make it clear why the null check is done. Just to be sure we not refactor it later\n\n> Am 20.07.2014 um 17:21 schrieb Osvaldo Pinali Doederlein [email protected]:\n> \n> In common/src/main/java/io/netty/util/collection/IntObjectHashMap.java:\n> \n> > @@ -212,10 +183,9 @@ public boolean containsKey(int key) {\n> > \n> > ```\n> > @Override\n> > public boolean containsValue(V value) {\n> > ```\n> > - int i = -1;\n> > - while ((i = nextEntryIndex(i + 1)) >= 0) {\n> > - V next = values[i];\n> > - if (value == next || value != null && value.equals(next)) {\n> > - V v = setValue(value);\n> > - for (int i = 0; i < values.length; ++i) {\n> > - if (values[i] != null && values[i].equals(v)) {\n> > This is a very low-level optimization: nulls are expected to be very common (at least 50% of all slots even in a \"full\" map with the default load factor 0.5), so I fast-path this case before making any invocation, will be faster even if the equals call is inlined (which I don't expect to happen, in the general case it's a megamorphic call since you can put any ind of object in the map).\n> \n> —\n> Reply to this email directly or view it on GitHub.\n", "going for a faster loop... the standard formula of multiply-and-accummulate-with-prime is a good default, but here the array can be big and this needs to be as fast as possible including potential unrolling/SIMD optimization.\n", "seems different to me; both sides can either be null or non-null (occupied vs. non-occupied)\n", "done\n", "done\n", "done\n", "done\n", "done\n", "forgot that one... done\n", "True :)\n\n> Am 20.07.2014 um 17:34 schrieb Osvaldo Pinali Doederlein [email protected]:\n> \n> In common/src/main/java/io/netty/util/collection/IntObjectHashMap.java:\n> \n> > ```\n> > if (value == null) {\n> > ```\n> > - if (otherValue != null || !other.containsKey(entry.key())) {\n> > seems different to me; both sides can either be null or non-null (occupied vs. non-occupied)\n> \n> —\n> Reply to this email directly or view it on GitHub.\n", "Ok\n", "Thanks\n", "Seems like this is a left-over... Remove it\n", "Remove merge stuff\n", "Fix up\n", "Fix up\n", "Fix up\n", "Fix up\n", "argh, github rookie here. will fix these.\n", "If you have any issues let me know\n\n> Am 20.07.2014 um 17:55 schrieb Osvaldo Pinali Doederlein [email protected]:\n> \n> In common/src/main/java/io/netty/util/collection/IntObjectHashMap.java:\n> \n> > @@ -212,10 +197,13 @@ public boolean containsKey(int key) {\n> > \n> > ```\n> > @Override\n> > public boolean containsValue(V value) {\n> > ```\n> > - int i = -1;\n> > - while ((i = nextEntryIndex(i + 1)) >= 0) {\n> > - V next = values[i];\n> > - if (value == next || value != null && value.equals(next)) {\n> > - V v = setValue(value);\n> > - for (int i = 0; i < values.length; ++i) {\n> > +<<<<<<< HEAD\n> > - // The map supports null values; this will be matched as NULL_VALUE.equals(NULL_VALUE).\n> > +=======\n> > +>>>>>>> 5813e35a5182578fe7c7fa646012dd4b7d72a2e8\n> > argh, github rookie here. will fix these.\n> \n> —\n> Reply to this email directly or view it on GitHub.\n", "nit: the names of these methods are a little unclear. Maybe something like valueToReturn() and valueToStore()? Not crazy about these either ... maybe you can come up with something better?\n", "Maybe wrap() and unwrap()? Or mask() and unmask(), or toInternal() and toExternal()\n", "+1 for toInternal/toExternal\n", "@nmittler @opinali: Shouldn't this be:\n\n``` java\nif (otherValue != null || other.containsKey(key)) {\n```\n\nWith this code merged into master the http2 tests fails because of the failed equals tests.\n", "Aww, I was running only the netty-comon tests because the full test suite failed for me since first checkout. The bug here is more subtle, I'm forgetting to handle the null vs. NULL_VALUE stuff. Fixing now..\n", "Ah ok... let me know one you fixed it\n", "Fixed already, working on improved tests for this now... in fact the code was handling NULL_VALUE, but incorrectly. Also, the original code has a bug in hashCode(); if you examine it, the hashCode function depends on the internal order of entries: values=[1,2] and values=[2,1] will produce different hashes. This difference of ordering is possible when the maps had different hashing conflicts, e.g. the natural placement would be [1,2] but insertion history was something like [-,-] =(put 8)=> [8,-] =(put 1)=> [8,1] =(remove 8)=> [-,1] =(put 2)=> [2,1]. And even when you hit this scenario, which may be very hard in artificial unit tests not designed to build hashtables in a messy way, your test may still succeed if it uses only equals() but not hashCode() (rare real-world code will use hashtables as values inside other hashtable). The new hashCode() function already avoids this bug, but I'm adding a commente there and also tests to watch for this...\n", "@normanmaurer fixed and pushed. Also reported an issue for my full-build problem (this is on Goobuntu Trusty, and I don't normally compile native code so maybe something else missing in my env.) But I have ran most tests, everything before that failure works including http2.\n" ]

[ "defect" ]

[ "common/src/main/java/io/netty/util/collection/IntObjectHashMap.java" ]

[ "common/src/main/java/io/netty/util/collection/IntObjectHashMap.java" ]

[ "common/src/test/java/io/netty/util/collection/IntObjectHashMapTest.java" ]

[ "timestamp(timedelta=38.0, similarity=0.8639139103559901)" ]

[ "@nmittler can you take care ?\n", "Sure thing ... created PR #2701 \n", "@Scottmitch1 FYI, I think you only need to follow the netty commit message rules for pull requests, not issues. When I first looked at this I thought it was a PR :)\n", "@nmittler - That makes sense because I also confused @normanmaurer by doing this on another issue.\n\nSeparate question. Is there something that needs to be explicitly done to have the PR auto built by the build system and notifications sent by @trustin-notification?\n", "@Scottmitch1 nope, it's all automatically done when you create the PR.\n", "Nope... Every PR is picked up\n\n> Am 23.07.2014 um 18:52 schrieb Scottmitch1 [email protected]:\n> \n> @nmittler - That makes sense because I also confused @normanmaurer by doing this on another issue.\n> \n> Separate question. Is there something that needs to be explicitly done to have the PR auto built by the build system and notifications sent by @trustin-notification?\n> \n> —\n> Reply to this email directly or view it on GitHub.\n", "Fixed by #2701 \n" ]

[]

[]

[ "codec-http2/src/main/java/io/netty/handler/codec/http2/AbstractHttp2ConnectionHandler.java", "codec-http2/src/main/java/io/netty/handler/codec/http2/DefaultHttp2OutboundFlowController.java", "codec-http2/src/main/java/io/netty/handler/codec/http2/DelegatingHttp2ConnectionHandler.java", "codec-http2/src/main/java/io/netty/handler/codec/http2/Http2OutboundFlowController.java", "example/src/main/java/io/netty/example/http2/server/HelloWorldHttp2Handler.java" ]

[ "codec-http2/src/main/java/io/netty/handler/codec/http2/AbstractHttp2ConnectionHandler.java", "codec-http2/src/main/java/io/netty/handler/codec/http2/DefaultHttp2OutboundFlowController.java", "codec-http2/src/main/java/io/netty/handler/codec/http2/DelegatingHttp2ConnectionHandler.java", "codec-http2/src/main/java/io/netty/handler/codec/http2/Http2OutboundFlowController.java", "example/src/main/java/io/netty/example/http2/server/HelloWorldHttp2Handler.java" ]

[ "codec-http2/src/test/java/io/netty/handler/codec/http2/DefaultHttp2OutboundFlowControllerTest.java", "codec-http2/src/test/java/io/netty/handler/codec/http2/DelegatingHttp2ConnectionHandlerTest.java", "codec-http2/src/test/java/io/netty/handler/codec/http2/Http2ConnectionRoundtripTest.java" ]

[ "timestamp(timedelta=17.0, similarity=0.8508248784792203)" ]

[ "It's a bug. Key password must be respected. I would be more than happy if you are coming up with a patch!\n", "And, here's the ICLA form: https://docs.google.com/spreadsheet/viewform?formkey=dHBjc1YzdWhsZERUQnhlSklsbG1KT1E6MQ\n", "Alright, I filled in the ICLA form and will do my best to help.\n" ]

[ "Why not make this class final and add a private constructor? Seems like it only contains a static method anyway.\n", "Add empty line \n", "Also I think we can make it package private\n", "After thinking a bit more about it maybe we could also just merge it into `JdkSslServerContext`. Or is there any reasons to not do so ?\n", "It was part of `JSSC` but then I extracted a class in order to separate concerns since key spec generation doesn't require the class context of `JSSC`. However, `JSSC` is the only class using it, so I might as well merge it back into it, for the sake of simplicity.\n", "Yep just merge it for now.\n" ]

[ "defect" ]

[ "handler/src/main/java/io/netty/handler/ssl/JdkSslServerContext.java" ]

[ "handler/src/main/java/io/netty/handler/ssl/JdkSslServerContext.java" ]

[ "handler/src/test/java/io/netty/handler/ssl/JdkSslServerContextTest.java", "handler/src/test/resources/io/netty/handler/ssl/netty_test", "handler/src/test/resources/io/netty/handler/ssl/netty_test.crt", "handler/src/test/resources/io/netty/handler/ssl/netty_test_unencrypted" ]

[ "timestamp(timedelta=62530.0, similarity=0.8478815462737131)" ]

[ "[This build](https://secure.motd.kr/jenkins/job/netty-4.0/4803/) reduces the initial capacity of `ChannelOutboundBuffer` to 4. (commit: 7e362277b93bbf4d59fd80485f3fe534a3ed8bba)\n", "@trustin I think we can close this ?\n", "Nope. The second item was not resolved yet.\n", "Ah got it\n\n> Am 23.07.2014 um 22:44 schrieb Trustin Lee [email protected]:\n> \n> Nope. The second item was not resolved yet.\n> \n> —\n> Reply to this email directly or view it on GitHub.\n", "Another thing we could do is to make the DefaultChannelPromise an entry element by itself and just store it. The downside would be that we need to wrap any other ChsnnelPromise impls.\n\nJust thinking out loud...\n\n> Am 23.07.2014 um 22:44 schrieb Trustin Lee [email protected]:\n> \n> Nope. The second item was not resolved yet.\n> \n> —\n> Reply to this email directly or view it on GitHub.\n", "I had to switch to Netty 3 to avoid this memory usage problem. I'm not sure what the difference is, but with Netty 3 my program uses 8-16mb of RAM while the Netty 4 version will utilize all that I can afford to give it and then crash.\n", "There must be something else wrong. Coukd you share a reproducer?\n\n> Am 25.07.2014 um 00:58 schrieb Blake Beaupain [email protected]:\n> \n> I had to switch to Netty 3 to avoid this memory usage problem. I'm not sure what the difference is, but with Netty 3 my program uses 8-16mb of RAM while the Netty 4 version will utilize all that I can afford to give it and then crash.\n> \n> —\n> Reply to this email directly or view it on GitHub.\n", "@trustin for a fix see https://github.com/netty/netty/pull/2704\n", "I have meet this issue, I try to disable AUTO_READ, and manually trigger read() operation by my self, and this issue disappear, Java heap memory consumption is quite stable." ]

[ "This line should not be changed.\n", "If this is not public, a user who extends `AbstractChannel` (and consequently `AbstractUnsafe`) will not be able to access the return value of `AbstractUnsafe.outboundBuffer()`.\n", "Any reason this is not part of the interface?\n" ]

[ "improvement" ]

[ "transport/src/main/java/io/netty/channel/AbstractChannel.java", "transport/src/main/java/io/netty/channel/ChannelOutboundBuffer.java" ]

[ "transport/src/main/java/io/netty/channel/AbstractChannel.java", "transport/src/main/java/io/netty/channel/ChannelOutboundBuffer.java" ]

[ "transport/src/test/java/io/netty/channel/ChannelOutboundBufferTest.java" ]

[ "timestamp(timedelta=89.0, similarity=0.893472531598197)" ]

[ "Fixed in 3f3e66c31ae3da70c36cc125ca9bcac8215390e4\n" ]

[ "This line should not be changed.\n", "If this is not public, a user who extends `AbstractChannel` (and consequently `AbstractUnsafe`) will not be able to access the return value of `AbstractUnsafe.outboundBuffer()`.\n", "Any reason this is not part of the interface?\n" ]

[ "defect" ]

[ "transport/src/main/java/io/netty/channel/AbstractChannel.java", "transport/src/main/java/io/netty/channel/ChannelOutboundBuffer.java" ]

[ "transport/src/main/java/io/netty/channel/AbstractChannel.java", "transport/src/main/java/io/netty/channel/ChannelOutboundBuffer.java" ]

[ "transport/src/test/java/io/netty/channel/ChannelOutboundBufferTest.java" ]

[ "timestamp(timedelta=39985.0, similarity=0.9999999999999999)" ]

[ "Do we also need something like getIntAndRemove(...) ?\n", "@plucury thanks for all the work. This can be closed now. \n" ]

[ "maybe start with 2 ? Seems more likely then 4.\n", "why is this even needed ?\n", "Again I think we should use 2\n", "Same as above\n", "I don't think an exception is the right thing here. I would even say that we should better change the return type to Long everywhere and return null in the case that it is not found. Sounds more sane to me.\n\n@trustin wdyt ? \n", "I found that it is 4 in method `DefaultTextHeaders.getAll`. If we change the number here maybe should change that too.\n", "It seems that we always add new header before the origin head. If we want return the list ordered by its append time, we need reverse it. \n", "@normanmaurer Yes, I agree with you. Using Long or Integer as return type seems more approach in this situation.\n", "Sounds good to me. Let me handle it in another commit.\n", "Will take care in another commit.\n" ]

[ "feature" ]

[ "codec/src/main/java/io/netty/handler/codec/DefaultTextHeaders.java", "codec/src/main/java/io/netty/handler/codec/EmptyTextHeaders.java", "codec/src/main/java/io/netty/handler/codec/TextHeaders.java" ]

[ "codec/src/main/java/io/netty/handler/codec/DefaultTextHeaders.java", "codec/src/main/java/io/netty/handler/codec/EmptyTextHeaders.java", "codec/src/main/java/io/netty/handler/codec/TextHeaders.java" ]

[]

[ "timestamp(timedelta=96494.0, similarity=0.8982404311961517)" ]

[ "Do `LZF` and `FastLZ` employ different compression algorithms? Also, could you explain why 6PACK is not suitable for Netty? For example, we could use an arbitrary file name like `data`?\n", "I guess I know why:\n- too much overhead\n- 6pack is only provided as an example to show how to use FastLZ.\n\nHere's my feed back about your proposed format:\n- We don't need magic number for all chunks. Shouldn't we have it only at the beginning of the stream?\n- We should make checksum optional. For example, we can use one bit of the 'type' field to signify if it has checksum or not.\n\nWDYT?\n", "Another thing to consider is the community around the compression algorithm and format. Looking from the FastLZ's repository, there's no commit since 2007. If you don't mind, it would be nice if we could focus on more actively maintained formats like LZ4.\n\nAlso, according to [the benchmark from ning](https://github.com/ning/jvm-compressor-benchmark/wiki), LZ4, Snappy, and LZF are the winners across multiple test cases, although they did not include FastLZ in the list.\n\nWhat do you think?\n", "@trustin nign considered FastLZ algorithm but said that FastLZ doesn't have a Java version (see at the bottom of [Codecs included](https://github.com/ning/jvm-compressor-benchmark/wiki#codecs-included)). But I found jfastlz library.\n\nFastLZ (according to sources) has good results:\n- http://fastlz.org/index.html\n- http://fastlz.org/lzf.htm\n- https://cwiki.apache.org/confluence/display/TS/RamCache\n- http://encode.ru/threads/851-introducing-FastLZ\n\nand may be used for real-time compression/decompression.\n\nFastLZ is used by Apache Traffic Server and PHP-memcached. But they have their own header format, for example in Apache Traffic Server:\n\n``` c++\nstruct RamCacheCLFUSEntry {\n INK_MD5 key;\n uint32_t auxkey1;\n uint32_t auxkey2;\n uint64_t hits;\n uint32_t size; // memory used including paddding in buffer\n uint32_t len; // actual data length\n uint32_t compressed_len;\n union {\n struct {\n uint32_t compressed:3; // compression type\n uint32_t incompressible:1;\n uint32_t lru:1;\n uint32_t copy:1; // copy-in-copy-out\n } flag_bits;\n uint32_t flags;\n };\n LINK(RamCacheCLFUSEntry, lru_link);\n LINK(RamCacheCLFUSEntry, hash_link);\n Ptr<IOBufferData> data;\n};\n```\n\nThey use only original `fastlz.c` for compression/decompression. And I didn't found any other compression libraries which provide FastLZ compression/decompression.\nSo I think that we also may use only `JFastLZ.fastlzCompress()` and `JFastLZ.fastlzDecompress()` and create our custom header, because (you are right) 6pack's format has unnecessary overhead and it looks like an archiver format.\n\nWhat about your feed back:\n- I think that framed format is more convenient (when you don't have a stream and all chunks are independent). And 3 magic bytes is a very little overhead. It will look like UDP or LZF frame.\n- I'm agree with you about checksum.\n\nNew proposal:\n\n| Length (bytes) | What |\n| :-: | :-: |\n| 3 | Magic 'FLZ' (like extension of 6PACK's compressed files) |\n| 1 | Options (checksum and type) |\n| 4 | Checksum (optional) |\n| 2 | ChunkLength |\n| 2 | OriginalLength (optional) |\n\nOptions:\n 0x00: Non-compressed chunk without checksum\n 0x01: Compressed chunk without checksum\n 0x10: Non-compressed chunk with checksum\n 0x11: Compressed chunk with checksum\n", "OK. The proposed frame format looks reasonable. Please go ahead and make modification. Please ping me once it's ready for a review.\n", "Superceded by #2755 \n" ]

[]

[ "feature", "won't fix" ]

[ "NOTICE.txt" ]

[ "NOTICE.txt", "codec/src/main/java/io/netty/handler/codec/compression/FastLz.java", "codec/src/main/java/io/netty/handler/codec/compression/FastLzFramedDecoder.java", "codec/src/main/java/io/netty/handler/codec/compression/FastLzFramedEncoder.java", "license/LICENSE.jfastlz.txt" ]

[ "codec/src/test/java/io/netty/handler/codec/compression/FastLzIntegrationTest.java" ]

[ "timestamp(timedelta=14.0, similarity=0.8792972262052686)" ]

[]

[ "no need for this.\n", "just remove the @return and merge it with the previous line. Also please use {@code -1} to highlight it. \n", "When there are two cases for return values, I find `@return` useful.\n\n```\nReturns the length of the input.\n@return the length of the input if the length of the input is known.\n a negative value if the length of the input is unknown.\n```\n", "Shouldn't this be `endOffset - startOffset`?\n", "This also should be `endOffset - startOffset`.\n" ]

[ "feature" ]

[ "codec-http/src/main/java/io/netty/handler/codec/http/HttpChunkedInput.java", "codec-http/src/main/java/io/netty/handler/codec/http/multipart/HttpPostRequestEncoder.java", "handler/src/main/java/io/netty/handler/stream/ChunkedFile.java", "handler/src/main/java/io/netty/handler/stream/ChunkedInput.java", "handler/src/main/java/io/netty/handler/stream/ChunkedNioFile.java", "handler/src/main/java/io/netty/handler/stream/ChunkedNioStream.java", "handler/src/main/java/io/netty/handler/stream/ChunkedStream.java", "handler/src/main/java/io/netty/handler/stream/ChunkedWriteHandler.java" ]

[ "codec-http/src/main/java/io/netty/handler/codec/http/HttpChunkedInput.java", "codec-http/src/main/java/io/netty/handler/codec/http/multipart/HttpPostRequestEncoder.java", "handler/src/main/java/io/netty/handler/stream/ChunkedFile.java", "handler/src/main/java/io/netty/handler/stream/ChunkedInput.java", "handler/src/main/java/io/netty/handler/stream/ChunkedNioFile.java", "handler/src/main/java/io/netty/handler/stream/ChunkedNioStream.java", "handler/src/main/java/io/netty/handler/stream/ChunkedStream.java", "handler/src/main/java/io/netty/handler/stream/ChunkedWriteHandler.java" ]

[ "handler/src/test/java/io/netty/handler/stream/ChunkedWriteHandlerTest.java" ]

[ "timestamp(timedelta=16.0, similarity=0.9867336731736773)" ]

[ "@normanmaurer Maybe it is already done? It seems can be assigned by `ChannelConfig`. \n", "@plucury this is used for all. I may need this per ChannelHandlerContext.\n", "@normanmaurer Ah, I see. So we need keep a `ByteBufAllocator` instance in `AbstractChannelHandlerContext`, right?\n", "I don't think we need this anymore. Let us close this\n", "@normanmaurer That's OK\n" ]

[]

[ "feature" ]

[ "transport/src/main/java/io/netty/channel/AbstractChannelHandlerContext.java", "transport/src/main/java/io/netty/channel/ChannelHandlerContext.java" ]

[ "transport/src/main/java/io/netty/channel/AbstractChannelHandlerContext.java", "transport/src/main/java/io/netty/channel/ChannelHandlerContext.java" ]

[]

[ "timestamp(timedelta=2429.0, similarity=0.9154252787991187)" ]

[ "The HTTP/1.x translation layer needs to be notified when ever a node's parent changes. The interface for `Http2Connection.Listener.streamPriorityChanged` lends itself to this but is not invoked for all parent change events. Should this be invoked in a different spot(s)? For example is the `DefaultHttp2OutboundFlowController$streamPriorityChanged` method correctly updating state with the current invocation of `streamPriorityChanged` in `DefaultHttp2Connection`?\n", "Here is the initial discussion extracted from PR #2775\n\n@nmittler - I would like to make some changes to support tracking the priority information in the HTTP/1.x headers.\n**Change Summary**\n1. Add method: `Http2Connection.Listener.parentChange(Http2Stream stream, Http2Stream oldParent)` which is invoked in `DefaultHttp2Connection.addChild` (and possibly other spots).\n2. Re-order `stream.setPriority(...)` calls to be after the `onXXXRead(...)` calls in `AbstractHttp2ConnectionHandler`\n\n**Potential negatives**\n1. The new parentChange method is fine grained and will be invoked once for each listener for each parent change.\n2. `DefaultHttp2OutboundFlowController` consumes priority listener events and may be impacted (although I don't think it will be because I don't think I need to change the onXXXWrite flow).\n3. (Regardless of implementation) Depending on when the stream events come in it may be past the HTTP/1.x message flow and the event may be ignored.\n\n**Justification**\nThe HTTP/1.x headers currently list\n- Parent Stream Id\n- Is Exclusive\n- Weight\n\nWeight and IsExclusive can be obtained reliably from the onXXXRead methods. The parent stream id is currently not being kept up to date in all situations (if tree operations move a node unrelated to the onXXXRead method). To mitigate this I would like to add a method to the listener interface that is invoked when a parent link changes.\nAfter this change there will still be an ordering issue between when the priority tree events are dispatched (via `stream.setPriority(...)`) and the `onHeadersRead` or `onPriorityRead` event. The ordering issue is if I subscribe the the priority tree listener interface for events it is possible (in fact likely) that I will not know that the stream exists (because the onXXXRead method hasn't been called). So to solve this I would need to move the `stream.setPriority(...)` method to be after the onXXXRead method.\n", "@nmittler - Can you explain the conditional statements leading up to setting the stream priority when a header is read on [line 862 on master](https://github.com/netty/netty/blob/master/codec-http2/src/main/java/io/netty/handler/codec/http2/AbstractHttp2ConnectionHandler.java#L862)? For example what if the first header message contains prioritization information, will we be setting the priority on the stream? Also what if the stream is in the half-closed local state?\n", "@Scottmitch maybe we can generalize this change a bit. The current Listener interface has two event handlers:\n\nstreamPriorityChanged(Http2Stream stream, Http2Stream previousParent) for simple changes, possibly just changing the parent.\n\nstreamPrioritySubtreeChanged(Http2Stream stream, Http2Stream subtreeRoot) when a node was made to be dependent on one of its descendants. In this case, the subtree starts at the new parent (previously was the descendant).\n\nThese events were tailored for use by the outbound flow controller logic, but I think we can simplify this interface so it will work for all use cases.\n\nstreamPriorityChanging(Http2Stream stream, Http2Stream previousParent) Called for all affected nodes before they are moved or their subtree is restructured. Need to think about it more, but event order may matter (bottom-up?) to make sure that the outbound flow controller properly removes weight calculations up the tree.\n\nstreamPriorityChanged(Http2Stream stream, Http2Stream newParent);\nCalled after each affect node's priority has changed or it's final position in the tree has been restored. Again, event order may matter (top-down?) ... need to iron out the details.\n\nWe likely will want to do this for all affected nodes in order:\n1) streamPriorityChanging for all nodes about to be affected\n2) make all the changes to the tree/nodes.\n3) streamPriorityChanged for all nodes that were affected.\n\nThoughts?\n", "The generalization you suggest is exactly what I had in mind. I agree the order of events should be well coordinated to avoid notification at incorrect times when manipulation is still occurring and also to handle the case that you currently are handling when a node is moved into its child tree structure.\n", "@Scottmitch ok sounds great! Let's get the draft 14 framing PR cherrypicked and I can start working on this.\n", "@nmittler - We can handle the coordination between http2 core codec and the http translation layer in the same manner as the draft-14 updates (I will ultimately submit a PR based upon your cherry picked changes).\n", "@Scottmitch so your PR will be blocked on the priority changes? Would it be easier for you to just take over this bug and incorporate the changes into your PR?\n", "@nmittler - Do you mean handle this as part of PR https://github.com/netty/netty/pull/2778? If so I think it would be best to handle this as a separate PR to break up concerns. In terms of who does the work on the listener interface (and related) changes and I can take a stab at it if you want.\n", "@Scottmitch that was my thought, since I didn't want to further block #2778. However, if you can implement it without the priority changes for now, that would definitely make it easier to divide up the work ... and I'll just take care of this in parallel. Will that work?\n", "@nmittler - That sounds ideal. Feel free to work on this. I can apply the updates to the HTTP/1.x translation layer after #2778 is accepted (and likely work with you in parallel again to ensure compatibility/timely completion of my changes).\n", "Here is an example (which can be made more minimal if necessary) of the non-restructure case which demonstrates the issue and may make a good test case.\n\n**Initial State**\nEach of the following nodes represents a stream, and the letter is representative of their stream ID.\n\n> ```\n> A\n> / | \\\n> B C D\n> / \\ / \\ / \\\n> E F G H I J\n> ```\n1. Call `C.setPriority(D, 4, true)` from `DefaultHttp2Connection`\n\n**End Result**\n\n> ```\n> A\n> / \\\n> B D\n> / \\ |\n> E F C\n> / / \\ \\\n> G H I J\n> ```\n\n**Current Implementation Result**\n1. `notifyPriorityChanged(C, A)` is called but is the `priorityBytes` in `DefaultHttp2OutboundFlowController` for I and J ever added into C (and up the tree)?\n2. There is no notification for I and J even though their parents have changed (new functionality requested by this issue).\n", "@Scottmitch So I think the new logic would do the following:\n1. streamPriorityChanging for C, G, H, I, and J\n2. Update the tree\n3. streamPriorityChanged for G, H, I, J, and C\n\nDoes that sound right to you?\n", "@nmittler - I made need some clarification on the meaning of `streamPriorityChang[ed|ing]` but are G, H, I, and J changing priority? I am fixed on the mindset of notifying when parents change and so I may have tunnel vision.\n\nI think the nodes that are changing priority (if priority is the sum of all immediate decedent priorities) are (bottom up): C, D, A.\n\nThe nodes that are changing parents are (top down): C (from A to D), I and J (from D to C)\n", "Getting notified for each element in the affected subtree(s) is useful for the outbound flow controller, which maintains state (i.e. total number of pending bytes for the subtree) at each node. I was imagining that streamPriorityChanging would allow the flow controller to remove the pending bytes upward in the tree for that node. When streamPriorityChanged is invoked, it would then add the pending bytes up the tree for that node ... if it was assigned a new parent, that's fine.\n\nSo all of these events are useful for this case, but will they cause confusion for other cases? We can always throw a ton of documentation at the interface :)\n", "I think I see where you are coming from by referencing `incrementPriorityBytes`. I think my mental model of the interface was a bit different:\n\n**Notify in a top down manner**\n_parentChanging(Http2Stream stream, Http2Stream newParent)_\ntree changes happen here\nparentChanged(Http2Stream stream, Http2Stream oldParent)\n\nThis structures the interface more around general tree operations and I think decreases the chances of clients missing relevant events (or double counting some events). I would need to think about how this generalizes to the case when we are moving a node into its decedent sub-tree.\n\nI put the changing events in _italics_ because I'm not sure if they are needed?\n\nIn the previous example the order of events -> and resulting flow controller operations:\nparentChanged(C, A) -> subtract C.priority from A up to root, add C.priority to C.parent up to root\nparentChanged(I, D) -> subtract I.priority from D up to root, add I.priority to I.parent up to root\nparentChanged(J, D) -> subtract J.priority from D up to root, add J.priority to J.parent up to root\n", "Lets try an example where a node is moved into its own sub-tree. The spec conveniently has an example for this case http://http2.github.io/http2-spec/index.html#rfc.section.5.3.3.\n\n**Non-exclusive order of events**\nparentChanged(D, C) -> subtract D.priority from C up to root, add D.priority to D.parent up to root\nparentChanged(A, ?) -> subtract A.priority from ? up to root, add A.priority to A.parent up to root\n\n**Exclusive order of events**\nparentChanged(D, C) -> subtract D.priority from C up to root, add D.priority to D.parent up to root\nparentChanged(A, ?) -> subtract A.priority from ? up to root, add A.priority to A.parent up to root\nparentChanged(F, D) -> subtract F.priority from D up to root, add F.priority to F.parent up to root\n\nI think these both work, but does it generalize?\n\nThis interface also facilitates the information I am after, being able to generate messages when individual node parent change events occur. If this generalizes then it also reduces the required interface for priority tree events from 2 to 1.\n", "@nmittler - I opened PR https://github.com/netty/netty/pull/2802 to address this issue.\n", "@Scottmitch can we close this now?\n", "@nmittler - Closed.\n\n@nmittler & @normanmaurer reviewed PR #2802 and @nmittler cherry picked into master.\n\nClosing this bug/improvement.\n" ]

[ "@nmittler - I could use some feedback/advice from you here.\n", "I think we might need to notify the listener in 2 passes: once before we make the changes and once after the changes are complete.\n\nNotifying the listener before the changes allows the flow controller to properly decrement pendingBytes for a node from it's branch (up the tree) ... we need the previous tree structure in place to do this.\n", "Agreed, we should definitely have a unit test for this if we don't already :). \n\nIf we have the pre/post-change events as I suggested above, this method will be replaced by the new event handlers. On pre-change, we decrement the bytes up the old branch. On post-change we add them to the new branch. \n", "@nmittler - Verify this logic please.\n", "@nmittler - Verify this logic please.\n", "@Scottmitch this looks right.\n", "@Scottmitch this looks right\n", "Not sure out of my head if listeners is an ArrayList if so you should better use `for (int i = 0; i < 0; i++) { ... }` to reduce object creation as the foreachloop will create a new Iterator everytime.\n", "Coouldn't we use let `DefaultStream.parent()` use `DefaultStream` as return type and so remove the cast ?\n", "Same as above\n", "Listeners is a set (hashset), but the new functionality I added uses an ArrayList with a foreach. I will update to a traditional for loop.\n", "Done.\n", "This code has changed and the cast is no longer necessary, but all other cases in this class have been updated.\n", "@nmittler - Note that I changed the new interface from 'childRemoving' to 'parentChanging'. I feel it makes the interface more consistent to be notifying in terms of parent events. I updated the outbound flow controller & unit tests accordingly.\n", "parentChanging sounds good to me :)\n", "@Scottmitch Agreed, parentChanging sounds good.\n", "@nmittler - Can you confirm this is the correct state after removal? For example, is there another way I should be removing A to test if A's priority bytes is just A (because it is its own tree)?\n", "If you're opening this up for testing, maybe make all mutation methods private?\n", "just to simplify things, you probably only need to set the priority on one of them. You should also verify that no structure changing events were received.\n", "nit: maybe make the verification a method ... might make this a little more readable.\n", "maybe move this to below public methods?\n", "nit: maybe make a state(stream) method that does the cast?\n\nor maybe even just priorityBytes(stream) ... that might be cleaner since that's all you're looking for.\n", "Update javadoc ... You're \"removing\" a node, not \"adding\" one.\n", "I'm actually not sure the content of the removed state matters much since it won't be in the tree anymore anyway. Can we verify that the connection's priorityBytes is correct?\n", "Done.\n", "Done.\n", "Done.\n", "Done.\n", "Done.\n", "Good catch\n" ]

[]

[ "codec-http2/src/main/java/io/netty/handler/codec/http2/DefaultHttp2Connection.java", "codec-http2/src/main/java/io/netty/handler/codec/http2/DefaultHttp2OutboundFlowController.java", "codec-http2/src/main/java/io/netty/handler/codec/http2/Http2Connection.java", "codec-http2/src/main/java/io/netty/handler/codec/http2/Http2ConnectionAdapter.java" ]

[ "codec-http2/src/main/java/io/netty/handler/codec/http2/DefaultHttp2Connection.java", "codec-http2/src/main/java/io/netty/handler/codec/http2/DefaultHttp2OutboundFlowController.java", "codec-http2/src/main/java/io/netty/handler/codec/http2/Http2Connection.java", "codec-http2/src/main/java/io/netty/handler/codec/http2/Http2ConnectionAdapter.java" ]

[ "codec-http2/src/test/java/io/netty/handler/codec/http2/DefaultHttp2ConnectionTest.java", "codec-http2/src/test/java/io/netty/handler/codec/http2/DefaultHttp2OutboundFlowControllerTest.java" ]

[ "timestamp(timedelta=25.0, similarity=0.9598137742447445)" ]

[ "@nmittler is this still something we need ?\n", "@normanmaurer yeah we're still out of spec WRT padding. It's low priority though and can wait until after we get draft 14 support.\n", "Draft 14 is tracked in issue https://github.com/netty/netty/issues/2715. \n", "Fixed by #2805\n" ]

[ "Isn't this still deferring all the padding until the last frame on the split?\n", "I thought this was to address the TODO, but I see the PR description does not focus on addressing how to split padding. If this does not address the TODO then leave it in there until we cover it.\n", "Should we open another issue for this then or do you plan to address this before merge ?\n", "@Scottmitch ultimately it is addressing the TODO ... I just made the decision to have all the padding at the end. The main difference here is that the padding is now flow controlled as well.\n\n@jpinner do you have any thoughts on how to split data frames that contain padding? Are there any security concerns with keeping all of the padding at the end? Or is it preferred to split the padding as the frame is split?\n", "I plan on addressing in our catch-all bug for TODOs (#2782)\n", "@Scottmitch added the TODO back in.\n", "Can you move the static imports back where they were before ?\n", "@normanmaurer are you fine with keeping fully-qualified class names?\n\nAlso, is having static imports last a Netty convention? If so, perhaps it can be incorporated into the build checks?\n", "yeah this is fine...\n", "done.\n" ]

[]

[ "codec-http2/src/main/java/io/netty/handler/codec/http2/DefaultHttp2InboundFlowController.java", "codec-http2/src/main/java/io/netty/handler/codec/http2/DefaultHttp2OutboundFlowController.java", "codec-http2/src/main/java/io/netty/handler/codec/http2/Http2InboundFlowController.java" ]

[ "codec-http2/src/main/java/io/netty/handler/codec/http2/DefaultHttp2InboundFlowController.java", "codec-http2/src/main/java/io/netty/handler/codec/http2/DefaultHttp2OutboundFlowController.java", "codec-http2/src/main/java/io/netty/handler/codec/http2/Http2InboundFlowController.java" ]

[ "codec-http2/src/test/java/io/netty/handler/codec/http2/DefaultHttp2InboundFlowControllerTest.java", "codec-http2/src/test/java/io/netty/handler/codec/http2/DefaultHttp2OutboundFlowControllerTest.java" ]

[ "timestamp(timedelta=50.0, similarity=0.8558067741127687)" ]

[ "PR https://github.com/netty/netty/pull/2827 addresses this issue\n", "@Scottmitch cherry-picked your change. Thanks!\n" ]

[]

[ "cleanup" ]

[ ".gitignore" ]

[ ".gitignore" ]

[]

[ "timestamp(timedelta=25.0, similarity=0.905243869713583)" ]

[ "Pull request #2849 has a proposed fix.\n", "Was fixed \n" ]

[ "you should only check for `==`. `size` can never be greater than `maxCapacity`.\n", "`testMaxCapacity` and add a link to your bug report please.\n", "maybe wrap everything in a loop that tests this with a few random `maxCapacity` values between [1,1000).\n", "Well, that was actually part of the issue. The original commit (8d2ddae) checked with `==`, but also had `Math.max()` for proper capacity. That was later removed for some reason. If the original code had checked with `>=`, this bug wouldn't have been as bad as it is. The capacity would still be capped at the next power of two. It wouldn't have been unbounded.\n\nWith the new unit test, this kind of breakage during random change should be caught early. I personally prefer to still leave room for error. But if you want me to change to `==`, I will be glad to do it.\n", "I think this method should be package private and maybe `threadLocalCapacity()` would be a more accurate name?\n", "> I personally prefer to still leave room for error. But if you want me to change to ==, I will be glad to do it.\n\nI see what you mean, however I would argue that if the code is correct, `>` cannot happen and thus `==` makes much more sense when someone else reads the code a year from now :)\n", "Sure.\n", "Done.\n", "Added 50 random numbers between 256 and 1256.\n", "Done.\n", "you are missing an `a` after `Cap` in the name :-). Also we usually use javadoc comments like this\n\n```\n/**\n * Test for issue https://github.com/netty/netty/issues/2848\n */\n```\n" ]

[ "defect" ]

[ "common/src/main/java/io/netty/util/Recycler.java" ]

[ "common/src/main/java/io/netty/util/Recycler.java" ]

[ "common/src/test/java/io/netty/util/RecyclerTest.java" ]

[ "timestamp(timedelta=32.0, similarity=0.876096460955986)" ]

[ "Added to 4.0, 4.1 and master \n" ]

[]

[ "feature" ]

[ "transport-native-epoll/src/main/c/io_netty_channel_epoll_Native.c", "transport-native-epoll/src/main/c/io_netty_channel_epoll_Native.h", "transport-native-epoll/src/main/java/io/netty/channel/epoll/EpollDatagramChannel.java", "transport-native-epoll/src/main/java/io/netty/channel/epoll/EpollSocketChannel.java", "transport-native-epoll/src/main/java/io/netty/channel/epoll/IovArray.java", "transport-native-epoll/src/main/java/io/netty/channel/epoll/Native.java" ]

[ "transport-native-epoll/src/main/c/io_netty_channel_epoll_Native.c", "transport-native-epoll/src/main/c/io_netty_channel_epoll_Native.h", "transport-native-epoll/src/main/java/io/netty/channel/epoll/EpollDatagramChannel.java", "transport-native-epoll/src/main/java/io/netty/channel/epoll/EpollSocketChannel.java", "transport-native-epoll/src/main/java/io/netty/channel/epoll/IovArray.java", "transport-native-epoll/src/main/java/io/netty/channel/epoll/IovArrayThreadLocal.java", "transport-native-epoll/src/main/java/io/netty/channel/epoll/Native.java", "transport-native-epoll/src/main/java/io/netty/channel/epoll/NativeDatagramPacketArray.java" ]

[ "testsuite/src/test/java/io/netty/testsuite/transport/socket/DatagramUnicastTest.java" ]

[ "timestamp(timedelta=21.0, similarity=0.9999999999999999)" ]

[ "Addressed with aba28d4fef2e3bc7b5afe41662ad0672828d548d\n" ]

[ "replace `i * 2` with `i << 1`. Bit-shifting for the win ;)\n", "Good call! Done.\n", "This can be collapsed to else if (...)\n", "Done.\n", "Kill it or fill it ;)\n", "I was sneaky on this one. Look at the below lines for the `<ul>...</ul>`. Is OK?\n", "Ah sorry missed this.. Sounds good :)\n", "Does a range check... ?\n", "You got it. Fixed.\n", "@normanmaurer - Note that I introduced this method instead of using `Character.isDigit(c)`. Should I just be using `Character.isDigit(c)` or is this OK?\n", "No preference here.\n", "Do we need to have this package-private? Why not just make it private ?\n", "Missing '.' at end of line\n", "Missing '.' at end of line\n", "Missing '.' at end of line\n", "I think it would be nice to add some comments above the bit-shifting stuff to make it easier to understand what you are doing here when look at it again in 1 year or so ;)\n", "You could just do `i + 1` only one time and reuse the result.\n", "You could do `i + 2` only one time and reuse it\n", "Could you please add some comment why you do this ...\n", "Please add a comment explain this.\n", "We have multiple ip.length() calls here. Maybe we could just only call it one time and re-use the result as we never know what the CharSequence impl actually does and so it may be expensive.\n", "if it should never happen better to throw an Error or RuntimeException\n", "Missing '.' at end of line.\n", "Done.\n", "Done (both).\n", "Done\n", "Done.\n", "Done.\n", "Done.\n", "Good idea. Done.\n" ]

[ "feature" ]

[ "common/src/main/java/io/netty/util/NetUtil.java" ]

[ "common/src/main/java/io/netty/util/NetUtil.java" ]

[ "common/src/test/java/io/netty/util/NetUtilTest.java", "microbench/src/test/java/io/netty/microbenchmark/common/NetUtilBenchmark.java" ]

[ "timestamp(timedelta=249.0, similarity=0.902667567681132)" ]

[ "Here is my sendmmsg from bits/socket.h\n\n> /\\* Send a VLEN messages as described by VMESSAGES to socket FD.\n> Return the number of datagrams successfully written or -1 for errors.\n> This function is a cancellation point and therefore not marked with\n> __THROW. */\n> extern int sendmmsg (int __fd, struct mmsghdr *__vmessages,\n> unsigned int __vlen, int __flags);\n\nHere is the sendmmsg from io_netty_channel_epoll_Native.c\n\n> extern int sendmmsg(int sockfd, struct mmsghdr *msgvec, unsigned int vlen, unsigned int flags) **attribute**((weak));\n", "> gcc --version\n> gcc (Ubuntu/Linaro 4.6.3-1ubuntu5) 4.6.3\n", "Hold on a second I found http://netty.io/wiki/native-transports.html and am giving this a read ;)\n", "The above wiki did not help. Issue still exists.\n", "I opened up PR https://github.com/netty/netty/pull/2890 for this issue\n", "@normanmaurer Cherry picked fix as 94deea409ec6788cb3705aae319be1fdfcbe85f8\n" ]

[]

[ "defect" ]

[ "transport-native-epoll/pom.xml", "transport-native-epoll/src/main/c/io_netty_channel_epoll_Native.c" ]

[ "transport-native-epoll/pom.xml", "transport-native-epoll/src/main/c/io_netty_channel_epoll_Native.c" ]

[]

[ "timestamp(timedelta=15.0, similarity=0.9111227419072077)" ]

[ "@nmittler Do you see any issues with moving the `Http2InboundFlowController` from `AbstractHttp2ConnectionHandler` to `DefaultHttp2FrameReader`?\n\nIf this is OK then for consistency I may also doing the same thing for the `Http2OutboundFlowController` (moving it to `DefaultHttp2FrameWriter`)\n", "@nmittler - On second thought this may get ugly because the stream state hasn't been verified yet. As a result I may move the functionality from `DecompressorHttp2FrameReader` into `AbstractHttp2ConnectionHandler` after the flow control stuff. I think this will avoid complicating the flow control (and other) interactions.\n", "@Scottmitch I wonder if the right home for the decompression logic would be in `InboundHttp2ToHttpAdapter.onDataRead`. WDYT?\n", "@nmittler - I would like to have it be available for the HTTP/2 codec too. I think it can be but I just put it in the wrong place.\n", "@nmittler - I started working on this but realized this may create a messy merge situation with https://github.com/netty/netty/pull/2892. I think I will wait until after that is cherry-picked.\n", "@Scottmitch sgtm ... I'll try to get #2892 in soon.\n", "PR https://github.com/netty/netty/pull/2907 was opened to address this issue.\n", "Cherry picked as 3807f9fc8ee292eb35ea6c26f075f0185e2f4994\n" ]

[ "Maybe we should have a base class called Http2FrameListenerDecorator? WDYT?\n", "maybe just return here ... so no need for else?\n", "Should we do this in a finally block?\n", "May be just my personal preference, so feel free to ignore .... maybe consider if (stream == null ) return ... that way the rest of the method doesn't have to be indented and nested within an if block.\n", "done.\n", "Good idea. I have a connection listener to clean up just in-case but better to clean up when it is in my control.\n", "Missed this...No problem.\n", "Well it appears I blew away my changes from `Http2FrameAdapter` to `Http2FrameListener`. I will make this change now ;)\n" ]

[ "defect" ]

[ "codec-http2/src/main/java/io/netty/handler/codec/http2/DecompressorHttp2FrameReader.java", "codec-http2/src/main/java/io/netty/handler/codec/http2/DefaultHttp2FrameReader.java", "example/src/main/java/io/netty/example/http2/client/Http2ClientInitializer.java" ]

[ "codec-http2/src/main/java/io/netty/handler/codec/http2/DefaultHttp2FrameReader.java", "codec-http2/src/main/java/io/netty/handler/codec/http2/DelegatingDecompressorFrameListener.java", "codec-http2/src/main/java/io/netty/handler/codec/http2/Http2FrameListenerDecorator.java", "example/src/main/java/io/netty/example/http2/client/Http2ClientInitializer.java" ]

[ "codec-http2/src/test/java/io/netty/handler/codec/http2/DataCompressionHttp2Test.java" ]

[ "timestamp(timedelta=6595.0, similarity=0.8875179716498717)" ]

[ "Fix is provided in #2972 \n" ]

[ "Nit: Consider saving a reference to the `(FullHttpResponse) msg` here to avoid casting multiple times.\n", "done\n" ]

[ "defect" ]

[ "codec-http/src/main/java/io/netty/handler/codec/http/websocketx/WebSocketClientProtocolHandshakeHandler.java" ]

[ "codec-http/src/main/java/io/netty/handler/codec/http/websocketx/WebSocketClientProtocolHandshakeHandler.java" ]

[]

[ "timestamp(timedelta=24.0, similarity=0.8579470291274585)" ]

[ "Problems associated with this issue (if it is possible we are missing exceptions):\n1. Not setting promise object failure clauses (and thus not completing them)\n2. Leaking buffers (i.e. https://github.com/netty/netty/blob/master/codec-http2/src/main/java/io/netty/handler/codec/http2/AbstractHttp2ConnectionHandler.java#L380)\n", "@Scottmitch The read case should properly handle Throwables and convert them to Http2Exceptions. As for writing, we should probably do a safe release of the buffer on the line that you indicated.\n", "While looking through the code I stumbled upon the following kinda unrelated question:\n\n[Section 6.1](https://tools.ietf.org/html/draft-ietf-httpbis-http2-14#section-6.1) of the spec says that data frames which are not in the open or half closed local state must result in a STREAM_CLOSED error. We are also allowing STREAM_CLOSED [here](https://github.com/netty/netty/blob/master/codec-http2/src/main/java/io/netty/handler/codec/http2/AbstractHttp2ConnectionHandler.java#L849)...is this to handle the \"closed\" [section 5.1](https://tools.ietf.org/html/draft-ietf-httpbis-http2-14#section-5.1):\n\n> Flow controlled frames (i.e., DATA) received after sending\n> RST_STREAM are counted toward the connection flow control window.\n> Even though these frames might be ignored, because they are sent\n> before the sender receives the RST_STREAM, the sender will\n> consider the frames to count against the flow control window.\n\nIf so it seems like these sections in the specification give conflicting direction and you did a good job picking that up ;)\n\nHowever is it possible to be in the CLOSED state without receiving a reset? For example if we sent an END_STREAM and received an END_STREAM we will be in the CLOSED state? In this case will we still throw a STREAM_CLOSED error or will this skip through the current checks and call `AbstractHttp2ConnectionHandler.this.onDataRead`?\n", "@Scottmitch Yup ... looking at the code for onDataRead, I think if the stream was previously HALF_CLOSED_REMOTE, we would let it through ... which is a bug. Would you mind opening an issue for this? Good catch!\n", "@nmittler - Done. https://github.com/netty/netty/issues/2937\n", "@Scottmitch great - thanks! ... I'll put together a PR for this.\n", "Resolved by #2979\n" ]

[ "@normanmaurer - Can you weigh in on this. `isOpen()` seems to be lower in the state hierarchy...should we be using this method?\n", "remote -> local\n", "@nmittler - The `onException` method is calling `toHttp2Exception` which seems to do some duplicate work as the `getEmbeddedHttp2Exception` method called here. Would it make sense to save the reference that `getEmbeddedHttp2Exception` returns and create a new private final method to handle an Http2Exception?\n", "@nmittler What is the difference between the `closeListener` and creating a `new ClosingChannelFutureListener`? Is the `closeListener` just for streams? If so should we rename it to make that more clear?\n", "RST_STREAM -> `{@code RST_STREAM}`\n", "@nmittler - Remind me again why the EndPoint class `goAwayReceived` has to have the `Received` verbiage in the signature? I think I already suggested this but why wouldn't just `goAway(lastStreamId)` work?\n", "GO_AWAY -> `{@code GO_AWAY}`\n", "@nmittler - Is this class needed or used? Seems like it may be left over from development?\n", "nit: do we need a new line here?\n", "done.\n", "done.\n", "@Scottmitch the attempt was to help clarify what you're doing when you say <endpoint>.goAway(). Is the endpoint conceptually sending GO_AWAY or receiving it? I originally tried naming it goAway(), but the ambiguity of which direction we were talking about became confusing. Basically, I just picked a direction (sending/receiving) and renamed it :)\n", "done.\n", "doh ... no it's not :) ... removed!\n", "No ... autoformatter. Done.\n", "@Scottmitch I don't think the duplication here is all that bad. First off, it should be noted that I'm not entirely sure we need to override exceptionCaught() at all, since we don't have other upstream handlers that will be throwing Http2Exception. Originally we copied the SPDY implementation which had other upstream handlers that could throw and this was a way of communicating errors between handlers ... not sure it still applies.\n\nRegardless ... onException() is called from 2 other places, neither of which will need to conditionally call onException(). This is a special case where we only want to call onException if we're sure the exception came from the HTTP/2 layer. I think it's reasonable to just have this done here even if it is a slight duplication of what is done inside onException(). WDYT?\n", "@nmittler - Fair enough. Although I'm not sure if the situation improves by picking a direction and having the method used in the cotext of the opposite direciton ;).\n", "@nmittler - SGTM. Thanks for the clarification.\n", "Did this change get lost in the last commit?\n", "Did this change get lost in the last commit?\n", "Did this change get lost in the last commit?\n", "RST_STREAM -> `{@code RST_STREAM}`\n", "hmm .... not sure what happened. done.\n", "done.\n", "done.\n", "done.\n", "@Scottmitch I think the real solution would be to just move all GO_AWAY methods to the connection (e.g. goAwaySent()/goAwayReceived()) ... that way it would be more clear. This is a minor cleanup task that should probably be handled in another PR ... WDYT?\n", "I'm not sure why the diff isn't showing this ... the change is there - you can see it in the 3rd commit.\n", "@nmittler - I agree. Would you mind creating an issue?\n", "@Scottmitch done: #2986\n", "It looks like the 3rd commit only has changes in `Http2ConnectionHandler.java` and this is in `Http2LifecycleManager.java`.\n", "ah right ... I think I got lost in the diff :) .. done.\n", "closeListener is only set by the close() method (i.e. the connection was requested to shutdown by the local endpoint).\n", "@nmittler - Sounds good. I re-visted this after I made the comment and it seems fine.\n" ]

[ "defect" ]

[ "codec-http2/src/main/java/io/netty/handler/codec/http2/DefaultHttp2ConnectionDecoder.java", "codec-http2/src/main/java/io/netty/handler/codec/http2/DefaultHttp2ConnectionEncoder.java", "codec-http2/src/main/java/io/netty/handler/codec/http2/Http2CodecUtil.java", "codec-http2/src/main/java/io/netty/handler/codec/http2/Http2ConnectionDecoder.java", "codec-http2/src/main/java/io/netty/handler/codec/http2/Http2ConnectionEncoder.java", "codec-http2/src/main/java/io/netty/handler/codec/http2/Http2ConnectionHandler.java", "codec-http2/src/main/java/io/netty/handler/codec/http2/Http2LifecycleManager.java", "codec-http2/src/main/java/io/netty/handler/codec/http2/Http2StreamException.java" ]

[ "codec-http2/src/main/java/io/netty/handler/codec/http2/DefaultHttp2ConnectionDecoder.java", "codec-http2/src/main/java/io/netty/handler/codec/http2/DefaultHttp2ConnectionEncoder.java", "codec-http2/src/main/java/io/netty/handler/codec/http2/Http2CodecUtil.java", "codec-http2/src/main/java/io/netty/handler/codec/http2/Http2ConnectionDecoder.java", "codec-http2/src/main/java/io/netty/handler/codec/http2/Http2ConnectionEncoder.java", "codec-http2/src/main/java/io/netty/handler/codec/http2/Http2ConnectionHandler.java", "codec-http2/src/main/java/io/netty/handler/codec/http2/Http2LifecycleManager.java", "codec-http2/src/main/java/io/netty/handler/codec/http2/Http2StreamException.java" ]

[ "codec-http2/src/test/java/io/netty/handler/codec/http2/DefaultHttp2ConnectionDecoderTest.java", "codec-http2/src/test/java/io/netty/handler/codec/http2/DefaultHttp2ConnectionEncoderTest.java", "codec-http2/src/test/java/io/netty/handler/codec/http2/Http2ConnectionHandlerTest.java", "codec-http2/src/test/java/io/netty/handler/codec/http2/Http2ConnectionRoundtripTest.java" ]

[ "timestamp(timedelta=119735.0, similarity=0.9411294219327737)" ]

[ "@nmittler I think @normanmaurer cherry-picked and closed your PR. Can this be closed?\n", "I think so :)\n\n> Am 14.10.2014 um 22:39 schrieb Scott Mitchell [email protected]:\n> \n> @nmittler I think @normanmaurer cherry-picked and closed your PR. Can this be closed?\n> \n> —\n> Reply to this email directly or view it on GitHub.\n", "@normanmaurer @Scottmitch sorry for the delay, I was OOO. Thanks for cherry-picking!\n", "@nmittler NP. Just wanted to make sure there wasn't any loose ends you intended on tying up.\n", "@Scottmitch No I think it was good to go :)\n" ]

[]

[ "cleanup" ]

[ "codec-http2/src/main/java/io/netty/handler/codec/http2/DefaultHttp2Connection.java", "codec-http2/src/main/java/io/netty/handler/codec/http2/DefaultHttp2ConnectionDecoder.java", "codec-http2/src/main/java/io/netty/handler/codec/http2/Http2Connection.java", "codec-http2/src/main/java/io/netty/handler/codec/http2/Http2ConnectionHandler.java" ]

[ "codec-http2/src/main/java/io/netty/handler/codec/http2/DefaultHttp2Connection.java", "codec-http2/src/main/java/io/netty/handler/codec/http2/DefaultHttp2ConnectionDecoder.java", "codec-http2/src/main/java/io/netty/handler/codec/http2/Http2Connection.java", "codec-http2/src/main/java/io/netty/handler/codec/http2/Http2ConnectionHandler.java" ]

[ "codec-http2/src/test/java/io/netty/handler/codec/http2/DefaultHttp2ConnectionDecoderTest.java", "codec-http2/src/test/java/io/netty/handler/codec/http2/DefaultHttp2ConnectionTest.java" ]

[ "timestamp(timedelta=115825.0, similarity=0.9901557196938808)" ]

[ "Now I'm thinking this is an issue because our netty client example reliably times out in this case. The server (netty server example) prints that it sent everything, but the client hangs after processing the headers.\n\n**Server**\n\n``` bash\nSelected Protocol is HTTP_2\nOct 10, 2014 3:57:45 PM io.netty.handler.codec.http2.Http2FrameLogger log\nINFO: \n----------------OUTBOUND--------------------\nSETTINGS: ack=false, settings={HEADER_TABLE_SIZE=4096, MAX_CONCURRENT_STREAMS=2147483647, INITIAL_WINDOW_SIZE=65535, MAX_FRAME_SIZE=16384, MAX_HEADER_LIST_SIZE=2147483647}\n------------------------------------\nOct 10, 2014 3:57:45 PM io.netty.handler.codec.http2.Http2FrameLogger log\nINFO: \n----------------INBOUND--------------------\nSETTINGS: ack=false, settings={HEADER_TABLE_SIZE=4096, ENABLE_PUSH=1, MAX_CONCURRENT_STREAMS=2147483647, INITIAL_WINDOW_SIZE=65535, MAX_FRAME_SIZE=16384, MAX_HEADER_LIST_SIZE=2147483647}\n------------------------------------\nOct 10, 2014 3:57:45 PM io.netty.handler.codec.http2.Http2FrameLogger log\nINFO: \n----------------OUTBOUND--------------------\nSETTINGS ack=true\n------------------------------------\nOct 10, 2014 3:57:45 PM io.netty.handler.codec.http2.Http2FrameLogger log\nINFO: \n----------------INBOUND--------------------\nSETTINGS ack=true\n------------------------------------\nOct 10, 2014 3:57:46 PM io.netty.handler.codec.http2.Http2FrameLogger log\nINFO: \n----------------INBOUND--------------------\nHEADERS: streamId:3, headers=DefaultHttp2Headers[:authority: 127.0.0.1:13443, :method: GET, :path: /whatever, :scheme: https, Accept-Encoding: deflate, Accept-Encoding: gzip], streamDependency=0, weight=16, exclusive=false, padding=0, endStream=true\n------------------------------------\nOct 10, 2014 3:57:46 PM io.netty.handler.codec.http2.Http2FrameLogger log\nINFO: \n----------------OUTBOUND--------------------\nHEADERS: streamId:3, headers=DefaultHttp2Headers[:status: 200], padding=0, endStream=false\n------------------------------------\nOct 10, 2014 3:57:46 PM io.netty.handler.codec.http2.Http2FrameLogger log\nINFO: \n----------------OUTBOUND--------------------\nDATA: streamId=3, padding=0, endStream=true, length=16385, bytes=6161616....\n------------------------------------\n```\n\n**Client**\n\n``` bash\n./run-example.sh -Dssl -Dport=13443 http2-client\n[INFO] Running: http2-client (io.netty.example.http2.client.Http2Client -D_ -Dssl -Dport=13443)\n[GC (Allocation Failure) 8704K->2567K(31744K), 0.0048808 secs]\n[GC (Allocation Failure) 11271K->4654K(31744K), 0.0043717 secs]\n[GC (Allocation Failure) 13358K->6590K(31744K), 0.0086735 secs]\nConnected to [127.0.0.1:13443]\nOct 10, 2014 3:57:45 PM io.netty.handler.codec.http2.Http2FrameLogger log\nINFO: \n----------------OUTBOUND--------------------\nSETTINGS: ack=false, settings={HEADER_TABLE_SIZE=4096, ENABLE_PUSH=1, MAX_CONCURRENT_STREAMS=2147483647, INITIAL_WINDOW_SIZE=65535, MAX_FRAME_SIZE=16384, MAX_HEADER_LIST_SIZE=2147483647}\n------------------------------------\nOct 10, 2014 3:57:45 PM io.netty.handler.codec.http2.Http2FrameLogger log\nINFO: \n----------------INBOUND--------------------\nSETTINGS: ack=false, settings={HEADER_TABLE_SIZE=4096, MAX_CONCURRENT_STREAMS=2147483647, INITIAL_WINDOW_SIZE=65535, MAX_FRAME_SIZE=16384, MAX_HEADER_LIST_SIZE=2147483647}\n------------------------------------\nOct 10, 2014 3:57:45 PM io.netty.handler.codec.http2.Http2FrameLogger log\nINFO: \n----------------OUTBOUND--------------------\nSETTINGS ack=true\n------------------------------------\nSending request(s)...\n[Full GC (Ergonomics) 48062K->40805K(66560K), 0.0868109 secs]\nOct 10, 2014 3:57:46 PM io.netty.handler.codec.http2.Http2FrameLogger log\nINFO: \n----------------INBOUND--------------------\nSETTINGS ack=true\n------------------------------------\nOct 10, 2014 3:57:46 PM io.netty.handler.codec.http2.Http2FrameLogger log\nINFO: \n----------------OUTBOUND--------------------\nHEADERS: streamId:3, headers=DefaultHttp2Headers[:authority: 127.0.0.1:13443, :method: GET, :path: /whatever, :scheme: https, Accept-Encoding: deflate, Accept-Encoding: gzip], streamDependency=0, weight=16, exclusive=false, padding=0, endStream=true\n------------------------------------\nOct 10, 2014 3:57:46 PM io.netty.handler.codec.http2.Http2FrameLogger log\nINFO: \n----------------INBOUND--------------------\nHEADERS: streamId:3, headers=DefaultHttp2Headers[:status: 200], padding=0, endStream=false\n------------------------------------\nException in thread \"main\" java.lang.IllegalStateException: Timed out waiting for response on stream id 3\n at io.netty.example.http2.client.HttpResponseHandler.awaitResponses(HttpResponseHandler.java:67)\n at io.netty.example.http2.client.Http2Client.main(Http2Client.java:118)\n[ERROR] Failed to execute goal kr.motd.maven:exec-maven-plugin:1.0.0.Final:exec (default-cli) on project netty-example: Command execution failed. Process exited with an error: 1 (Exit value: 1) -> [Help 1]\n[ERROR] \n[ERROR] To see the full stack trace of the errors, re-run Maven with the -e switch.\n[ERROR] Re-run Maven using the -X switch to enable full debug logging.\n[ERROR] \n[ERROR] For more information about the errors and possible solutions, please read the following articles:\n[ERROR] [Help 1] http://cwiki.apache.org/confluence/display/MAVEN/MojoExecutionException\n```\n", "@nmittler - It looks like it may be a flush issue? I can confirm from the netty example client that it is not recieving the last partial frame partition (which happens to be non-full) when the timeout happens.\n\nI'm using wireshark and there is alot going on with my current setup but I'm not seeing an object of the appropriate size go accross in the failure case.\n", "@nmittler - Turns out the server example is no longer using the flow controllers properly for sending data. The `SimpleHttp2FrameListener` is taking a raw `Http2FrameWriter` and should be using a `Http2ConnectionEncoder`. I'll start by fixing that and see where that leaves us.\n", "Cherry picked\n", "@Scottmitch you've closed this issue so I'm assuming that adding flow control to the example was the fix here?\n", "@nmittler - Yes the Netty issue I was observing where the clients were waiting for stuff to happen was resolved by ensuring flow-control was used. PR #3000 addressed this issue.\n\nIt was a different issue for my personal server but this helped me also track down my own issues ;)\n", "@nmittler - As a result of this we can say that curl and netty are talking ;) see https://github.com/netty/netty/issues/2828\n", "@Scottmitch I saw that ... good stuff! :)\n" ]

[]

[ "defect" ]

[ "codec-http2/src/main/java/io/netty/handler/codec/http2/Http2ConnectionHandler.java", "example/src/main/java/io/netty/example/http2/server/HelloWorldHttp2Handler.java" ]

[ "codec-http2/src/main/java/io/netty/handler/codec/http2/Http2ConnectionHandler.java", "example/src/main/java/io/netty/example/http2/server/HelloWorldHttp2Handler.java" ]

[]

[ "timestamp(timedelta=13.0, similarity=0.864423459913442)" ]

[ "@nmittler fyi\n", "@Scottmitch this is addressed by #3016\n" ]

[]

[]

[ "codec-http2/pom.xml", "pom.xml" ]

[ "codec-http2/pom.xml", "pom.xml" ]

[]

[ "timestamp(timedelta=64.0, similarity=0.8550309335091857)" ]

[ "@nmittler - FYI. There is some info in section [5.1.1](https://tools.ietf.org/html/draft-ietf-httpbis-http2-14#section-5.1.1) related to this condition.\n", "For the writer portion of this it is also possible we put this responsibility on the users of the http2-codec? The http translation layer does allow users to specify streams, but may need some work in the case that stream ids are not provided by the user. WDYT?\n", "@Scottmitch The way it works now, an Http2Exception will be thrown when you try to create a stream after you've reached MAX_INT: https://github.com/netty/netty/blob/master/codec-http2/src/main/java/io/netty/handler/codec/http2/DefaultHttp2Connection.java#L725\n\nAn application with access to the Http2Connection can always do their own verification before they attempt to create a stream so that the Http2Exception isn't thrown.\n", "@nmittler - Good point...missed that. Should this be a protocol error or a GOAWAY as suggested in [5.1.1](https://tools.ietf.org/html/draft-ietf-httpbis-http2-14#section-5.1.1)?\n\nI agree on the second point.\n", "Throwing an Http2Exception (i.e. connection error) when reading an inbound frame will result in a GO_AWAY, since onConnectionError will be called: https://github.com/netty/netty/blob/master/codec-http2/src/main/java/io/netty/handler/codec/http2/Http2ConnectionHandler.java#L287\n", "@nmittler - Well then looks like you are a couple step ahead of me ;) Sorry for the noise.\n", "@Scottmitch no worries! Keep the bugs coming! :)\n", "@nmittler - Wait just a second... I think there is an issue here. The protocol error never gets triggered because of an IndexOutOfBoundsException which originates from here https://github.com/netty/netty/blob/master/codec-http2/src/main/java/io/netty/handler/codec/http2/DefaultHttp2Connection.java#L126 when the stream id wraps around and becomes negative. The root cause seems to be the `index` variable here is `-1` https://github.com/netty/netty/blob/d35ddf0648f06c726c7d6a2eb1ccd7e65ed5564c/common/src/main/java/io/netty/util/collection/IntObjectHashMap.java#L296 which seems to be because of the modulus operation here is not playing nice with negative values https://github.com/netty/netty/blob/d35ddf0648f06c726c7d6a2eb1ccd7e65ed5564c/common/src/main/java/io/netty/util/collection/IntObjectHashMap.java#L315.\n", "@Scottmitch good catch! We should add a unit test for IntObjectHashMap for this. Do you want to fix or should I?\n", "@nmittler - Unless you have objections I'll let you take this one ;)\n", "@Scottmitch created #3028 to address this.\n", "@Scottmitch anything else that needs to be done here?\n", "@nmittler - O yah I may have been a bit optimistic/premature on the closing of this issue. I had to rebase/rebuild my 3 PRs that I am using. I am about to try and see the behavior after the build completes.\n", "@nmittler - Behavior looks good. Protocol exception is now being hit. May be a good idea to add a unit test for this. WDYT?\n", "@Scottmitch good idea .... since you're already in there, would you mind taking a crack at it?\n", "@nmittler sure\n", "@nmittler - I think we can leave this closed. Thanks for your help!\n", "@nmittler - After doing some more testing (and paying attention a bit more) on the write side we are not sending the GOAWAY in this condition. Should we be? Thoughts?\n", "@nmittler - The promise gets correctly set here: https://github.com/netty/netty/blob/master/codec-http2/src/main/java/io/netty/handler/codec/http2/DefaultHttp2ConnectionEncoder.java#L242 but the exception never reaches `onConnectionError`.\n", "@Scottmitch right, for encoding it's a different situtation ... we just fail the promise. The exception is for decoding. Is that not what you would expect?\n", "@nmittler - Not sure..My current rational is the codec is already checking for this condition and is their any alternative outcome (from a http2 protocol point of view) other than sending the goaway? I can easily detect outside the codec but what do you think?\n", "@Scottmitch I think for this particular error, that it does make sense to send a GO_AWAY on the writing side, since no more streams can be created. Other protocol violations that occur during writing should probably just fail the promise, however. WDYT?\n", "I agree. This is the only exception I can think of at the moment that may result in special action on the write side.\n", "@nmittler - How do you want to handle this? One way would be to make a subclass of `Http2Exception` and check for it in `DefaultHttp2ConnectionEncoder.writeHeaders` catch clause.\n", "@Scottmitch that seems reasonable to me.\n", "@nmittler - PR is forthcoming...\n", "@nmittler PR https://github.com/netty/netty/issues/3034 was submitted to resolve this issue.\n", "Cherry picked #3034 \n" ]

[ "@nmittler - Can you verify this change...I think this was a bug.\n", "@nmittler - Note that I had an argument capture here to free the debug buffer but it is already freed.\n", "@Scottmitch oof yeah I think you're right. Good catch!\n", "I don't think we should catch it here. We should catch it in the encoder/decoder and call the lifecycle manager to write the GO_AWAY.\n", "I don't think this belongs here.\n", "I think we should revert this ... lifecycle manager (i.e. connection handler) is in charge of this.\n", "@Scottmitch I think that's fine.\n", "@nmittler - I agree...I was initially looking to use the `lifecycleManager` but for some reason thought it was only available at the `Http2ConnectionHandler` level.\n\nCan you clarify the following:\n\n> We should catch it in the encoder/decoder\n", "Agreed...miss on my part.\n", "Agreed.\n" ]

[ "defect" ]

[ "codec-http2/src/main/java/io/netty/handler/codec/http2/DefaultHttp2Connection.java", "codec-http2/src/main/java/io/netty/handler/codec/http2/DefaultHttp2ConnectionEncoder.java", "codec-http2/src/main/java/io/netty/handler/codec/http2/Http2ConnectionHandler.java", "codec-http2/src/main/java/io/netty/handler/codec/http2/Http2Error.java" ]

[ "codec-http2/src/main/java/io/netty/handler/codec/http2/DefaultHttp2Connection.java", "codec-http2/src/main/java/io/netty/handler/codec/http2/DefaultHttp2ConnectionEncoder.java", "codec-http2/src/main/java/io/netty/handler/codec/http2/Http2ConnectionHandler.java", "codec-http2/src/main/java/io/netty/handler/codec/http2/Http2Error.java", "codec-http2/src/main/java/io/netty/handler/codec/http2/Http2NoMoreStreamIdsException.java" ]

[ "codec-http2/src/test/java/io/netty/handler/codec/http2/Http2ConnectionRoundtripTest.java" ]

[ "timestamp(timedelta=13446.0, similarity=0.9305338663498393)" ]

[ "@trustin @nmittler @normanmaurer - This is more of a question at this point because I see that the existing JdkSslClientContext uses `null` for the `SSLContext.init(...)` calls for everything but the `TrustStore`. See [here](https://github.com/netty/netty/blob/master/handler/src/main/java/io/netty/handler/ssl/JdkSslClientContext.java#L123) and [here](https://github.com/netty/netty/blob/master/handler/src/main/java/io/netty/handler/ssl/JdkSslClientContext.java#L151). Are there any reasons why this has to be `null`, and if I tie in the `KeyManagerFactory` will this do the trick?\n", "To stay consistent with the server side I think may make sense to get rid of the `TrustManagerFactory` argument in this new constructor. This will also lend itself better to when we support OpenSSL on the client side. I'm pulling together a PR to try this out.\n", "On second thought I am going to wait until we resolve https://github.com/netty/netty/pull/2966. There will be a good amount of overlap between this and that PR.\n", "I would like to see mutual authentication support from both server and client. This is something I have run into a need for recently. Until Netty supports this, I'm creating my own JDK-based SSLContext. For the Netty-based SslContext API to be complete, all client and server SslContext implementations will require support for following parameters: (a) certFile, (b) privateKeyFile, (c) privateKeyPassword, (d) trustChainFile\n\nExplanation of each:\n- certFile: public X.509 certificate in PEM format\n- privateKeyFile: private key corresponding to public certFile\n- privateKeyPassword: password to decrypt private key file (if it's encrypted)\n- tustChainFile: PEM format for all accepted certificate authorities\n", "I am working on a PR for this (both sides...initially just implementing for JDK provider) but I am going to wait to submit it because I have another PR #2966 in the same area that would be messy for merging.\n", "I actually did not consider TLS mutual authentication so far. Please feel free to come up with a pull request.\n", "@trustin After #2966 is resolved I will submit one.\n", "PR https://github.com/netty/netty/issues/3041 was submitted to address this issue.\n", "Cherry-picked\n" ]

[]

[ "feature" ]

[ "handler/src/main/java/io/netty/handler/ssl/JdkSslClientContext.java", "handler/src/main/java/io/netty/handler/ssl/JdkSslContext.java", "handler/src/main/java/io/netty/handler/ssl/JdkSslServerContext.java", "handler/src/main/java/io/netty/handler/ssl/SslContext.java" ]

[ "handler/src/main/java/io/netty/handler/ssl/JdkSslClientContext.java", "handler/src/main/java/io/netty/handler/ssl/JdkSslContext.java", "handler/src/main/java/io/netty/handler/ssl/JdkSslServerContext.java", "handler/src/main/java/io/netty/handler/ssl/SslContext.java" ]

[ "handler/src/test/java/io/netty/handler/ssl/JdkSslEngineTest.java", "handler/src/test/java/io/netty/handler/ssl/JdkSslServerContextTest.java", "handler/src/test/resources/io/netty/handler/ssl/test2.crt", "handler/src/test/resources/io/netty/handler/ssl/test2_encrypted.pem", "handler/src/test/resources/io/netty/handler/ssl/test2_unencrypted.pem" ]

[ "timestamp(timedelta=3773.0, similarity=0.886181733809813)" ]

[ "Hi @Lekanich and thanks for reaching out. We will need a few more details to narrow in on the issue:\n- What version of netty are you using?\n- What is your system configuration (`mvn --version`)?\n- Are you using openssl or JDK providers for SSL?\n- Where is the location in the code that the buffer is not being released?\n- Have you tried to release this buffer and see if the behavior improves?\n", "1) netty 4.0.23 \n2) Maven home: /usr/lib/apache-maven/apache-maven-3.2.1\nJava version: 1.8.0, vendor: Oracle Corporation\nJava home: /usr/java/jdk1.8.0/jre\nDefault locale: en_US, platform encoding: UTF-8\nOS name: \"linux\", version: \"3.11.0-26-generic\", arch: \"amd64\", family: \"unix\"\n3) JDK provider\n4) line: 596 (method SslHandler.wrap(SslEngine, ByteBuf, ByteBuf)\nin line 599 ByteBuffer.allocateDirect(in0.remaining())\nand then it never dealloc\n5) It's Internall netty SslHandler call, so haven't try to release it\n\nPS I think it passed to SslEngine and then someone forget to release it (line 606)\n", "@Lekanich - Would you mind verifying your assumptions? If you could make the code change you suggested and test that your issues are resolved that would be very helpful. Also you could then submit a PR and get credit for your contributions ;)\n", "@Scottmitch I try...\n", "@Lekanich - I think your line numbers may be a bit off? Is [line 580](https://github.com/netty/netty/blob/netty-4.0.23.Final/handler/src/main/java/io/netty/handler/ssl/SslHandler.java#L580) where the allocation in question is done? If so can you describe why you feel this is an issue, and where/how you would suggest releasing the memory?\n", "@Scottmitch LIne is correct, because that DirectBuffer never clean?\nI think rewrite as\nFrom [line 587 ](https://github.com/netty/netty/blob/netty-4.0.23.Final/handler/src/main/java/io/netty/handler/ssl/SslHandler.java#L585)\n\n``` java\n for (;;) {\n ByteBuffer out0 = out.nioBuffer(out.writerIndex(), out.writableBytes());\n SSLEngineResult result = engine.wrap(in0, out0);\n cleanDirectBuffer(in0);\n cleanDirectBuffer(out0);\n\n```\n\nWhere <code>cleanDirectBuffer</code>\n\n``` java\n private void cleanDirectBuffer(ByteBuffer buffer) {\n if (!buffer.isDirect()) {\n return;\n }\n Cleaner cleaner = ((DirectBuffer) buffer).cleaner();\n if (cleaner == null) {\n return;\n }\n\n cleaner.clean();\n }\n```\n\nPS Maybe need clean only <code>in0</code>\n", "@Lekanich - I believe the allocations in the loop will be cleaned up by (and is the responsibility of) the `ByteBuf` after it has been released (for the internal buffers). In your situation is the memory ever released (despite the OOM exception)? Is the GC ever running when you observe the issue? Is it possible you are holding on to references to `ByteBuf` parameters or not calling `release()` on them?\n\nI'm investigating a potential for a leak on line 580 but I don't think the internal buffers to `ByteBuf` should be released here.\n", "@Scottmitch I think <code>ByteBuffer</code> <code>in0</code> is in no way connected with the sending <code>ByteBuf</code>, so when will be called <code>release()</code> it's not release <code>ByteBuffer</code> <code>in0</code>.I think it release only <code>ByteBuffer out0</code>\n", "@Lekanich - Can you confirm that you are not using a direct buffer in this method (i.e. `!in0.isDirect()` returns true). Also can you see if PR https://github.com/netty/netty/pull/3058 changes anything for you?\n", "@Lekanich - When you say:\n\n> I think ByteBuffer in0 is in no way connected with the sending ByteBuf\n\nDo you still think this is the case if `in.nioBuffer().isDirect()` is `true`?\n", "@Scottmitch During debug I was into this block\n", "@Lekanich - Great news...not that there is a bug but that #3058 may fix your issue. Please try this PR and let me know if this is the case.\n", "@Scottmitch Yes it seems to help.\n", "@Lekanich Thank you for testing this out and reporting this issue!\n\nJust to clarify when you say \"it seems to help\" can you be a bit more specific? In the issue description you provide details as to how much more memory is used while SSL is enabled.\n- Can you provide these metrics while using #3058? I'm more just looking to see that it is not 10x more and that allocation/releasing pattern follows a similar pattern in the ssl/plaintext cases.\n- Are you no longer seeing the OutOfMemory exception?\n- Are there other areas you suspect to be leaky?\n", "Fixed by 8db8aca1e7bac8a927bc82968291abbb5ade96be\n", "@Scottmitch Perf Test with fix uses 3GB of memory, and without over 13GB (memory + swap)\n", "@Lekanich - Sounds much better. Thanks again for reporting!\n", "@Scottmitch Thanks for the quick reply)\n" ]

[ "use `newDirectIn.internalNioBuffer(0, newDirectIn.readableBytes()` . This way you also not need to reset the reader index.\n", "Good call. Done.\n" ]

[ "defect" ]

[ "handler/src/main/java/io/netty/handler/ssl/SslHandler.java" ]

[ "handler/src/main/java/io/netty/handler/ssl/SslHandler.java" ]

[]

[ "timestamp(timedelta=18.0, similarity=0.9003473515919903)" ]

[ "@normanmaurer and @trustin - Thoughts?\n", "PR https://github.com/netty/netty/pull/3073 demonstrates using the `wantsDirectBuffer` variable to prevent requiring a direct byte buffer on every wrap. This could be a big savings if non-direct byte buffers are used and direct byte buffers are not required (i.e. JDK ssl provider?).\n", "@Scottmitch sounds legit to me... let us wait what @trustin things about it\n", "@normanmaurer - Ok sounds good.\n", "Cherry-picked changes into 4.0, 4.1 and master\n", "@normanmaurer Thanks!\n" ]

[ "This sounds legit... @trustin agree ? \n" ]

[ "improvement" ]

[ "handler/src/main/java/io/netty/handler/ssl/SslHandler.java" ]

[ "handler/src/main/java/io/netty/handler/ssl/SslHandler.java" ]

[]

[ "timestamp(timedelta=233.0, similarity=0.9226550345829467)" ]

[ "@nmittler - FYI. This may be more general problem to whenever the windows sizes are used. We should be handling the case where they can be negative.\n", "@Scottmitch yup, agreed. I'll take a look.\n", "Opps sorry about stealing the assignment. I have a fix I think.\n", "I'll submit a PR and we can go from there.\n", "Sgtm :)\n", "PR https://github.com/netty/netty/pull/3086 is to address this issue.\n", "Fix was cherry-picked\n" ]

[]

[ "defect" ]

[ "codec-http2/src/main/java/io/netty/handler/codec/http2/DefaultHttp2OutboundFlowController.java" ]

[ "codec-http2/src/main/java/io/netty/handler/codec/http2/DefaultHttp2OutboundFlowController.java" ]

[ "codec-http2/src/test/java/io/netty/handler/codec/http2/DefaultHttp2OutboundFlowControllerTest.java" ]

[ "timestamp(timedelta=72577.0, similarity=0.958106215902029)" ]

[ "Good catch\n", "Addressed by #3088\n", "Cherry-picked by @nmittler \n" ]

[ "Wording is a bit awkward. Consider clarifying..\n\nReturns the most {@link ChannelFuture} -> Returns the {@link ChannelFuture}\n", "What about continuation frames? Do we even send continuation frames or are we queuing up header writes until the endofstream happens somewhere?\n", "Is this needed if you are only adding 1 promise object to it? Why couldn't you just directly use `promise`?\n", "Does this change what we should be returning? Shouldn't we still return the `ChannelFuture` from the `frameWriter.writeHeaders(...)` call? I seem to remember returning the promise object during the restructure in a method by mistake and the behavior was undesirable.\n", "I like the consistency and cleanliness of code but what impacts will this have on the majority of time headers are written (the headers before data)? When will the `operationComplete` be called? Does this require going back to the event loop?\n", "nit: Consider renaming because of the context we are in. In the `OutboundFlowState` context this is just the last `newFrame` that was created.\n", "Also for consistency does it make sense to add/use an accessor for this member?\n", "Can you explain this change?\n", "Good catch! I guess we are being good users and not calling header/data methods out of order ;)\n", "Done.\n", "The interface doesn't let you write continuation frames directly. You write a `Headers` object and that may get broken up into multiple frames. I don't think this is an issue.\n", "Done. Split out the logic so we only do this when a previous DATA frame was written for the stream.\n", "I believe the code is correct. We want to return a future that completes when the write completes, which is what will be done in this case. Once `innerPromise` completes, the aggregator will automatically complete `promise`.\n", "Done ... this has been addressed since I'm not splitting the logic based on whether a previous DATA frame has been sent.\n", "Done.\n", "This should not have been changed ... good catch!\n", "This sounds familiar. So the Http2Headers must contain all headers for the current headers \"phase\" (because we don't support an interface to generate continuation frames).\n" ]

[ "defect" ]

[ "codec-http2/src/main/java/io/netty/handler/codec/http2/DefaultHttp2ConnectionEncoder.java", "codec-http2/src/main/java/io/netty/handler/codec/http2/DefaultHttp2OutboundFlowController.java", "codec-http2/src/main/java/io/netty/handler/codec/http2/Http2OutboundFlowController.java" ]

[ "codec-http2/src/main/java/io/netty/handler/codec/http2/DefaultHttp2ConnectionEncoder.java", "codec-http2/src/main/java/io/netty/handler/codec/http2/DefaultHttp2OutboundFlowController.java", "codec-http2/src/main/java/io/netty/handler/codec/http2/Http2OutboundFlowController.java" ]

[ "codec-http2/src/test/java/io/netty/handler/codec/http2/DefaultHttp2ConnectionEncoderTest.java", "codec-http2/src/test/java/io/netty/handler/codec/http2/DefaultHttp2OutboundFlowControllerTest.java", "codec-http2/src/test/java/io/netty/handler/codec/http2/Http2ConnectionRoundtripTest.java", "codec-http2/src/test/java/io/netty/handler/codec/http2/Http2TestUtil.java" ]

[ "timestamp(timedelta=45.0, similarity=0.8662422963744304)" ]

[ "`DefaultHttpHeaders` already implements the ascii checks (if the `validate` mode is turned on...which is exposed in the HTTP2 translation layer). I think all that needs to be done here is catch the exception that is thrown by this validation (`IllegalArgumentException`) when creating the HTTP headers object and re-throw as an `Http2Exception`.\n", "@Scottmitch sounds good!\n", "@nmittler - I think with the life cycle manager changes this exception should bubble up to `decode` and be caught here https://github.com/netty/netty/blob/master/codec-http2/src/main/java/io/netty/handler/codec/http2/Http2ConnectionHandler.java#L372. I think the first order of business is to add a unit test to verify this but functionally I think we may be covered.\n", "@Scottmitch sounds good!\n", "@nmittler - Turns out the exception didn't make it all the way up after all. See https://github.com/netty/netty/pull/3090.\n", "Cherry picked\n" ]

[ "Maybe you could use this utility method to help out here: https://github.com/netty/netty/blob/master/codec-http2/src/main/java/io/netty/handler/codec/http2/Http2CodecUtil.java#L134\n", "Good idea...done.\n" ]

[ "defect" ]

[ "codec-http2/src/main/java/io/netty/handler/codec/http2/HttpUtil.java" ]

[ "codec-http2/src/main/java/io/netty/handler/codec/http2/HttpUtil.java" ]

[ "codec-http2/src/test/java/io/netty/handler/codec/http2/InboundHttp2ToHttpAdapterTest.java" ]

[ "timestamp(timedelta=12.0, similarity=0.9442379339087925)" ]

[ "Note that https://github.com/netty/netty/pull/2907 was introduced to fix bugs in the original decompression code.\n", "@nmittler - So I have looked at this briefly and have come up with the following options:\n1. Create a class which inherits from `DefaultHttp2OutboundFlowController` that overrides `writeData()`, compresses the data, then calls `super.writeData()`\n2. Create a new interface which is passed into `AbstractHttp2ConnectionHandler` and transforms data in `writeData()` before calling `outboundFlow.writeData()`.\n\n| Option | Pros | Cons |\n| --- | --- | --- |\n| 1 | Existing interfaces don't change. Compression (and associated code) only used if runtime time is instantiated (no if checks) | Maybe not the most obvious choice to have a `Http2OutboundFlowController` also provide compression |\n| 2 | Clear resposibility of classes | Interfaces would change (new set of constructors). This will impact code even if no compression is used (require a call to a default `transformation` class or `if != null` check every `writeData` call). |\n\nWhat are your thoughts on this? Do you have any alternative approaches, or do you need any clarification on the above options?\n", "@Scottmitch I think we want to keep compression out of the flow controller ... it already has a lot of responsibilities and is complex enough as it is :).\n\nCompression can take a lot of forms and is not formally defined at the HTTP/2 framing layer, which is where flow control is applied. Instead, I think that compression is a concern of the HTTP/1->HTTP/2 translation layer, which would compress the data contained within a FullHttpMessage before writing it to the HTTP/2 layer. So I think this isn't quite equivalent to either of your options ... I guess this is number 3? WDYT?\n", "@nmittler - I agree that even though extending `DefaultHttp2OutboundFlowController` is convenient it is likely not the best solution.\n\nAlthough it is not formally defined in the http/2 spec (outside of the [warnings](https://tools.ietf.org/html/draft-ietf-httpbis-http2-14#section-10.6)) http/2 is just a proper framing layer around http and should retain all core functionality of http (see [abstract](https://tools.ietf.org/html/draft-ietf-httpbis-http2-14) and [section 2](https://tools.ietf.org/html/draft-ietf-httpbis-http2-14#section-2)). In http you can have chunked responses where each chunk is encoded according to the content-encoding header. The \"chunked\" response use case is inherently supported by http/2 with the DATA FRAME. So just as the recently added decompression code operates on the data portion of the DATA FRAME the compression would operate on the same section. The difference would be the decompression code operates post flow control and the compression operates pre flow control.\n\nI think it makes sense for the HTTP/1->HTTP/2 layer to be able to do compression/decompression however I think the responsibility to provide the functionality is in the HTTP/2 proper code. This would allow all use cases involving the HTTP2-codec to take advantage of this functionality.\n\nDo you agree with this?\n", "@Scottmitch I think we're basically on the same page here. Just to clarify where I'm coming from ... \n\nHTTP/1 is just one protocol that can be layered atop HTTP/2 framing. Other (e.g. new) protocols may have their own compression schemes that may \"chunk\" in different ways from HTTP/1 (i..e. defining where the compression boundaries are). For example, it may be the case that a protocol may not map compression boundaries to a single call to connectionHandler.writeData(). Maybe the protocol doesn't have all of the data yet and is waiting for more data before it can write the next compression block. Whatever the protocol decides to do, the details of compression are really a function of the upper-level protocol and not the framing layer.\n\nWith that said, I don't feel that AbstractHttp2ConnectionHandler is the right place for compression. Some HTTP/1 translation object will be responsible for calling connectionHandler.writeData() and I think that is the object that should be responsible for applying compression. I definitely think we should abstract this utility behind an interface and provide a default implementation - that makes a lot of sense.\n", "@nmittler - Hum I can't quite tell if we are on the same page...seems like we are close though ;)\n\nIn general I agree with your first paragraph. There can be many different compression schemes desired in the application layer. However I do think we can generalize a common use case which is the application calling some form of `writeData()` and having this data compressed before `flowControl.writeData()` is invoked. This compression is based upon some encoding (i.e. gzip, deflate, my-app-encoding, etc...) which was previously specified with a `connectionHandler.writeHeader()` call. This interface can be made generic so that they can implement queuing if they need to wait for more data before actually doing the compression and calling `connectionHandler.writeData()`. I think the default implementation (just like the decompression) would be to implement streaming compression based upon gzip and deflate with room for extension.\n\nIs the HTTP/1 translation layer the desired place because it takes a FullHttpMessage and has all the content when making a call to `connectionHandler.writeData()`? Either way isn't this just a subset of the use-case from my description in the first paragraph?\n", "@nmittler - I opened PR https://github.com/netty/netty/pull/2935 to help demo an approach.\n", "The re-factoring allowed for easy access in PR https://github.com/netty/netty/pull/3106.\n", "Cherry-picked as 685075f1a058c0d5d8fc59072278bc822dce26d8\n" ]

[ "@trustin - I think these got moved during the header back port (and then forward port). I think it is better to keep a common definition in the HttpHeaderValue class rather than an independent definition in http/2 and http codec (and potentially spdy). If you agree I will back port the portion related to this change after this PR is resolved.\n", "I think you could combine the two if's: if (compressor == null && !endOfStream)...\n", "Maybe add a comment here indicating why compressor might be null for a particular stream.\n", "Doesn't need to be an else.\n", "Shouldn't this be writing \"buf\", not \"data\"?\n", "maybe make a utility method newCompressionChannel(ZlibWrapper)?\n", "May be worth adding a comment that this does compression on the entire stream as a whole. Message-based applications will have to roll their own compression to demarcate compressed \"chunks\" within a single stream.\n", "Done.\n", "Done.\n", "Good catch.\n", "Yes! Great catch..knew I was going to do that...too much cpy/paste.\n", "Maybe combine the two writes into one to reduce the duplication...\n\n```\nboolean endOfStreamForBuffer = nextBuf == null ? endOfStream : false;\nsuper.writeData(ctx, streamId, buf, padding, endOfStreamForBuffer, newPromise);\nif (endOfStreamForBuffer) {\n break;\n}\n```\n", "Done.\n", "Any suggestions? I think `getTargetContentEncoding` should allow an extension point if they have another encoding on top of the compression encoding.\n", "Humm I'm not sure this will work. endOfStreamForBuffer can still be false (if endOfStream is false) even if there is no more next buffer.\n", "Actually, I don't see this class being extended to support message-based compression. I was just suggesting that we comment this class to indicate that when we're talking about compression, we mean for the entire stream. \n", "Ah ok, so the if would change:\n\n```\nboolean endOfStreamForBuffer = nextBuf == null ? endOfStream : false;\nsuper.writeData(ctx, streamId, buf, padding, endOfStreamForBuffer, newPromise);\nif (nextBuf == null) {\n break;\n}\n```\n", "I think that will work. Thanks.\n", "Oops I don't think I can do this. This may lead to `compressor` null pointer exception down at the else condition when `cleanup` is called.\n", "Check the comments I added (most recent commit) and let me know if this is what you had in mind.\n", "ah good point.\n" ]

[ "feature" ]

[ "codec-http/src/main/java/io/netty/handler/codec/http/HttpContentDecompressor.java", "codec-http/src/main/java/io/netty/handler/codec/http/HttpHeaderValues.java", "codec-http2/src/main/java/io/netty/handler/codec/http2/DefaultHttp2Connection.java", "codec-http2/src/main/java/io/netty/handler/codec/http2/DelegatingDecompressorFrameListener.java", "codec-http2/src/main/java/io/netty/handler/codec/http2/Http2Stream.java" ]

[ "codec-http/src/main/java/io/netty/handler/codec/http/HttpContentDecompressor.java", "codec-http/src/main/java/io/netty/handler/codec/http/HttpHeaderValues.java", "codec-http2/src/main/java/io/netty/handler/codec/http2/CompressorHttp2ConnectionEncoder.java", "codec-http2/src/main/java/io/netty/handler/codec/http2/DefaultHttp2Connection.java", "codec-http2/src/main/java/io/netty/handler/codec/http2/DelegatingDecompressorFrameListener.java", "codec-http2/src/main/java/io/netty/handler/codec/http2/Http2Stream.java" ]

[ "codec-http2/src/test/java/io/netty/handler/codec/http2/DataCompressionHttp2Test.java" ]

[ "timestamp(timedelta=104.0, similarity=0.8532341873259105)" ]

[ "@trustin Not trying to be picky or anything, but wouldn't it be better to call the enum `HttpStatus` or `HttpStatusType` or `HttpStatusCategory` or `HttpStatusGroup` instead of `HttpStatusClass`? Ending it with `Class` just feels weird to me. [Some other synonyms of classification/type](http://www.thesaurus.com/browse/category)\n", "@blucas - I'm guessing this terminology was used because it is used in the RFC http://www.w3.org/Protocols/rfc2616/rfc2616-sec10.html?\n", "@Scottmitch - I completely agree that is probably why trustin used `Class`. All I'm trying to say is that I don't believe `class` fits as well as some of my suggestions given the context. This is purely an opinion/preference.\n\n> 10.2 Successful 2xx\n> \n> This class of status code indicates that the client's request was successfully received, understood, and accepted.\n\nThe above quote is in the context of an RFC document. The use of the word `class` here is just shorthand for `classification`. This could easily be replaced by `category` `group(ing)` `type` or other synonyms. In the context of a document such as the RFC, `class` fits in quite well. My opinion is that `class` does not fit in as well in the context of a programming language.\n\nAgain, this is only my opinion/preference and by all means the name of the enum can remain as it is. I am merely suggesting some alternative names which may work better in the context of this programming language.\n", "@blucas - All good points. I agree that `class` in type names can be confusing in programming language constructs. However I also think it is important to use consistent terminology. I could go either way @trustin WDYT?\n", "Let me also chime in..... I think the current naming is good as it maps perfect to the rfc.\n", "@normanmaurer +1\n" ]

[]

[ "feature" ]

[ "codec-http/src/main/java/io/netty/handler/codec/http/HttpResponseStatus.java" ]

[ "codec-http/src/main/java/io/netty/handler/codec/http/HttpResponseStatus.java", "codec-http/src/main/java/io/netty/handler/codec/http/HttpStatusClass.java" ]

[]

[ "timestamp(timedelta=128033.0, similarity=0.8549278665067631)" ]

[ "@Scottmitch I created this based on your suggestion in review of #3098. I figured it made sense to assign to you since you've done the work on the HTTP layer.\n", "@nmittler - Thanks. I don't think this is tied to the HTTP layer but it is related to other code I have done. I'll put something together after your PR is resolved.\n", "See PR https://github.com/netty/netty/pull/3144.\n", "Cherry picked as b83f3850179ecb55dac382b7ade9d5d0ffbff375.\n" ]

[]

[ "improvement" ]

[ "codec-http2/src/main/java/io/netty/handler/codec/http2/CompressorHttp2ConnectionEncoder.java" ]

[ "codec-http2/src/main/java/io/netty/handler/codec/http2/CompressorHttp2ConnectionEncoder.java" ]

[ "codec-http2/src/test/java/io/netty/handler/codec/http2/DataCompressionHttp2Test.java" ]

[ "timestamp(timedelta=108306.0, similarity=0.8638679492524849)" ]

[ "@nmittler - FYI\n", "@nmittler - I'm going to take a first pass at this...there are no framing changes.\n", "See PR https://github.com/netty/netty/pull/3167\n", "Cherry-picked updates.\n" ]

[]

[ "feature" ]

[ "codec-http2/src/main/java/io/netty/handler/codec/http2/Http2SecurityUtil.java" ]

[ "codec-http2/src/main/java/io/netty/handler/codec/http2/Http2SecurityUtil.java" ]

[]

[ "timestamp(timedelta=14.0, similarity=0.8448786730231798)" ]

[ "please review #3224\n", "Fixed by #3224\n" ]

[ "`file = null;` in the `finally` block is evaluated even when `file` is `null`. I would move the null-check out of the try block and return early when `file` is `null`.\n" ]

[ "defect", "improvement" ]

[ "transport-native-epoll/src/main/c/io_netty_channel_epoll_Native.c", "transport-native-epoll/src/main/c/io_netty_channel_epoll_Native.h", "transport-native-epoll/src/main/java/io/netty/channel/epoll/Native.java", "transport/src/main/java/io/netty/channel/DefaultFileRegion.java" ]

[ "transport-native-epoll/src/main/c/io_netty_channel_epoll_Native.c", "transport-native-epoll/src/main/c/io_netty_channel_epoll_Native.h", "transport-native-epoll/src/main/java/io/netty/channel/epoll/Native.java", "transport/src/main/java/io/netty/channel/DefaultFileRegion.java" ]

[]

[ "timestamp(timedelta=61056.0, similarity=0.9268064321116396)" ]

[ "I have looked at the diffs....here are some notes from my first pass.\n\nWe need some changes to be able to send/receive priority frames in any state. I have a question on the spec to clarify the state a bit https://github.com/http2/http2-spec/issues/667. I am looking into adding another `EndPoint.createStream(...)` method with an additional `boolean` argument to create a stream in the idle state (as opposed to always in the OPEN state). This will have fallout in other areas that needs investigation.\n\n[section 5.3.4](https://tools.ietf.org/html/draft-ietf-httpbis-http2-16#section-5.3.4):\n\n> Similarly, streams that are in the \"idle\" state can be assigned\n> priority or become a parent of other streams. This allows for the\n> creation of a grouping node in the dependency tree, which enables\n> more flexible expressions of priority. Idle streams that are made a\n> parent of another stream are assigned a default priority\n> (<a href=\"https://tools.ietf.org/html/draft-ietf-httpbis-http2-16#section-5.3.5\">Section 5.3.5</a>).\n\nI have to review this a bit more (just a reminder to myself mostly).\n\n[section 5.5](https://tools.ietf.org/html/draft-ietf-httpbis-http2-16#section-5.5)\n\n> Note that treating any frame other\n> than DATA frames as flow controlled is such a change in semantics,\n> and can only be done through negotiation.\n\nI know we discussed having the flow controller a bit more flexible in some cases (lastFrameSent). I'm not sure if we want to make this change and investigate others to be more suitable for extension frames? If so we can open up a follow up issue to address this.\n\n[section 8.1.1.2](https://tools.ietf.org/html/draft-ietf-httpbis-http2-16#section-8.1.2.2)\n\n> The only exception to this is the TE header field, which MAY be\n> present in an HTTP/2 request; when it is, it MUST NOT contain any\n> value other than \"trailers\".\n\nI have to review this a bit more (just a reminder to myself mostly). This is not a change, but was moved in the spec so it showed up in my diff.\n\n[appendix a](https://tools.ietf.org/html/draft-ietf-httpbis-http2-16#appendix-A)\nI have removed the prohibited ciphers but I think it is worth another pass through [oracle's cipher suite list](http://docs.oracle.com/javase/8/docs/technotes/guides/security/SunProviders.html) to see if we can pull in more ciphers.\n", "I opened another issue https://github.com/http2/http2-spec/issues/668 asking for clarification on some terminology in the spec.\n", "I need to investigate if priority frames are allowed to be sent/received after go_away or RST frames have been sent/received.\n", "Discussion moved to PR https://github.com/netty/netty/pull/3217.\n", "resolved with #3217 \n" ]

[]

[ "feature" ]

[ "codec-http2/src/main/java/io/netty/handler/codec/http2/Http2CodecUtil.java" ]

[ "codec-http2/src/main/java/io/netty/handler/codec/http2/Http2CodecUtil.java" ]

[]

[ "timestamp(timedelta=61083.0, similarity=0.855996269826361)" ]

[ "Addressed as part of https://github.com/netty/netty/pull/3217\n" ]

[]

[ "feature" ]

[ "codec-http2/src/main/java/io/netty/handler/codec/http2/Http2CodecUtil.java" ]

[ "codec-http2/src/main/java/io/netty/handler/codec/http2/Http2CodecUtil.java" ]

[]

[ "timestamp(timedelta=17.0, similarity=0.9700502923967033)" ]

[ "@MichaelScofield sounds right... could you come up with PR ?\n", "I've created a PR to branch 3.9: \n\nhttps://github.com/netty/netty/pull/3253\n", "Fixed by #3253\n" ]

[]

[ "defect" ]

[ "src/main/java/org/jboss/netty/channel/socket/nio/AbstractNioBossPool.java", "src/main/java/org/jboss/netty/channel/socket/nio/AbstractNioWorkerPool.java" ]

[ "src/main/java/org/jboss/netty/channel/socket/nio/AbstractNioBossPool.java", "src/main/java/org/jboss/netty/channel/socket/nio/AbstractNioWorkerPool.java" ]

[]

[ "timestamp(timedelta=372.0, similarity=0.8748363603579907)" ]

[ "@Scottmitch thanks for the heads-up buddy.\n", "@normanmaurer - Sure I can take this, but I have some outstanding questions for the alpn-boot folks. They are not planning on back porting the alpn-boot bug fix because there was some updates in OpenJDK SSL related code (not related to this bug). Seems a bit limiting to only have the \"newest\" version \"functional\" (or at least no known bugs) because the alpn-boot versions are jdk dependent....\n", "@Scottmitch yeah this sound not good :(\n", "Two PRs were submitted upstream to the alpn-boot project to backport fixes:\n\nJDK7: https://github.com/jetty-project/jetty-alpn/pull/7 \nJDK8: https://github.com/jetty-project/jetty-alpn/pull/6\n", "Addressed by https://github.com/netty/netty/pull/3254.\n\nalpn-boot folks have decided to leave previous versions vulnerable to this and any future bugs. To me this sounds like versions of alpn-boot that do not correspond to the most current version (for their respective JDK 7 or 8) are deprecated. I have asked for them to clarify this on their website (or somewhere) but have not reached clarity on this issue yet.\n" ]

[]

[ "feature" ]

[ "pom.xml" ]

[ "pom.xml" ]

[]

[ "timestamp(timedelta=10.0, similarity=0.8847084238518407)" ]

[ "See http://linux.die.net/man/7/tcp\n", "Sounds awesome.\n", "@normanmaurer would you log them or so?\n", "@trustin working on it...\n\n@daschl nope just expose it on the EpollSocketChannel. You can do whatever you want to do with it\n" ]

[ "Because all the fields of TCP_INFO are unsigned, we should use wider types - int for u8, int for u16, long for u32.\n", "good idea.\n", "@trustin do we want to use byte for all u8 methods ? or better stick with int ?\n", "Should be short or int.\n", "Consistency: `JNIEnv* env`\n" ]

[ "feature" ]

[ "transport-native-epoll/src/main/c/io_netty_channel_epoll_Native.c", "transport-native-epoll/src/main/c/io_netty_channel_epoll_Native.h", "transport-native-epoll/src/main/java/io/netty/channel/epoll/EpollSocketChannel.java", "transport-native-epoll/src/main/java/io/netty/channel/epoll/Native.java" ]

[ "transport-native-epoll/src/main/c/io_netty_channel_epoll_Native.c", "transport-native-epoll/src/main/c/io_netty_channel_epoll_Native.h", "transport-native-epoll/src/main/java/io/netty/channel/epoll/EpollSocketChannel.java", "transport-native-epoll/src/main/java/io/netty/channel/epoll/EpollTcpInfo.java", "transport-native-epoll/src/main/java/io/netty/channel/epoll/Native.java" ]

[ "transport-native-epoll/src/test/java/io/netty/channel/epoll/EpollSocketChannelTest.java" ]

[ "timestamp(timedelta=75.0, similarity=0.8422273553057141)" ]

[ "@dittos I confirm.\nThe issue is in splitMultipartHeader where, once Content-Disposition is found, we split the data according to ';'. In this case, this is wrong.\nNothing prevents a filename to include a \";\" (even if ';' is really not compatible with most of OS like Linux).\nI do not have time right now to make a fix, but I feel like the splitMultipartHeader should be changed such that we split the additionnal arguments using ';' but including a check on ' \" ' pairs in order to not check within this pair.\nSomething like:\n- find first \"\n- find ; (possible multiple) between current position and first \" and split (keeping only right side), updating current position\n- loop on previous step as long as we found ;\n- find next \", set new position to after it\n- loop on first item until position is not the end\n", "@dittos could you propose a fix and a test case associated as you did already ?\n" ]

[]

[ "defect" ]

[ "codec-http/src/main/java/io/netty/handler/codec/http/multipart/HttpPostMultipartRequestDecoder.java" ]

[ "codec-http/src/main/java/io/netty/handler/codec/http/multipart/HttpPostMultipartRequestDecoder.java" ]

[ "codec-http/src/test/java/io/netty/handler/codec/http/multipart/HttpPostRequestDecoderTest.java" ]

[ "timestamp(timedelta=44275.0, similarity=0.9544150316866737)" ]

[ "@sammychen105 sounds legit... would you mind open a PR with a fix ?\n", "@normanmaurer ok, a PR(#3371) has been submit.\n", "Fixed by #3371\n" ]

[]

[ "defect" ]

[ "transport/src/main/java/io/netty/channel/nio/AbstractNioByteChannel.java" ]

[ "transport/src/main/java/io/netty/channel/nio/AbstractNioByteChannel.java" ]

[]

[ "timestamp(timedelta=67.0, similarity=0.9396625537231835)" ]

[ "That does sound a bit fishy...I can't give history for why this was done but I'm guessing the intention was to send something to the other end (as opposed to relying on timeouts or other failure detection mechanisms)? No matter what the intention was I agree based upon your description it seems like we are not meeting that now. If the content can be well formed w.r.t HTTP (either we have not sent any headers, we aren't providing a conflicting content length or any other headers, or we can still gracefully fail within the HTTP protocol) then I think it would make sense to provide the response immediately. If there is a chance the message would violate the protocol then either see if we can mitigate this or potentially throw an exception in the pipeline.\n\nOn a related note...\n- HTTP/2 provides feedback to the peer for a similar situation via a RST_STREAM frame [here](https://github.com/netty/netty/blob/master/codec-http2/src/main/java/io/netty/handler/codec/http2/InboundHttp2ToHttpAdapter.java#L273).\n- SPDY behaves in a similar manner [here](https://github.com/netty/netty/blob/master/codec-http/src/main/java/io/netty/handler/codec/spdy/SpdyHttpDecoder.java#L327) and throws an exception into the pipeline. I'm not as familiar with SPDY so I'm not sure if there is a mechanism setup to catch the exception and provide feedback to the peer.\n", "> I can't give history for why this was done but I'm guessing the intention was to send something to the other end \n\nYeah I think so too, although the current approach creates an awkward usage pattern of having to look at `decoderResult()` in `HttpObject`. This is easy to overlook and get into receiving incomplete HTTP messages.\n\nIMO, we should raise an exception and have the user decide how to handle this. \n\nI am not so sure if it is better to ignore all future content on the connection (today's behavior) or ignore it till the end of response (if exists) and then reuse the connection for further requests. The reason for reuse being that it's the receiver end's configuration to limit the header/initline/content size and not a part of the protocol, so theoretically, it is not a bad HTTP message. However, I can understand the scenario of a DOS attack if we keep reading content on the server till eternity.\nOn the contrary, if we choose to ignore all future content on the connection (as it is today), isn't it better to just close the connection as opposed to keep reading from the connection to ignore?\n\nThanks for providing HTTP/2, SPDY context. HTTP/2 then is equivalent to closing connection on HTTP/1\n\n@normanmaurer @trustin thoughts?\n", "> On the contrary, if we choose to ignore all future content on the connection (as it is today), isn't it better to just close the connection as opposed to keep reading from the connection to ignore?\n\nClosing the connection will indeed provide some notification to the peer but one thing to be mindful of is pipelining. Is it possible the user may want to consider the current request bad, but future messages on the same connection may be OK? This may be a bit of an esoteric boundary case as it depends upon the ability to detect when the current message ends and the next message begins (after the max length has been exceeded).\n\n> HTTP/2 then is equivalent to closing connection on HTTP/1\n\nI think there are a few subtle differences:\n1. There is rational as to what happened at the application layer for HTTP/2 as opposed to shutdown from the transport layer.\n2. Currently we are closing the connection in HTTP/2 but potentially we could allow other existing streams being processed to complete.\n", "Hi @NiteshKant thanks for filing an issue.\n\nSome history:\n\nThe HTTP spec requires us to introspect the message to determine the appropriate response code to send back to the peer. For example, there are two different codes depending on whether the initial line was too long or the headers block was too large (414 URI Too Long vs 431 Request Header Fields Too Large).\n\nFurther, when introducing SPDY we wanted to be able to share channel handlers between HTTP and SPDY (such as those responsible for retuning a 400 vs 414 vs 431 response). This meant that we couldn't signal these by sending different types of exceptions up the pipeline, since we would need to the Stream-Id to send the response on and we didn't want to invalidate other requests on the connection.\n\nOn to FullHttpMessage:\n\nThis change was put into place before the message objects were refactored (back when the 4.x branches still used HttpMessage/HttpChunk) and thus before the Full\\* classes were introduced. I think modifying the decoder to emit these objects as \"Full\" messages sounds feasible, but I'd like to see how this interacts with Channel Handlers designed to be shared between both HTTP and SPDY / HTTP/2 connections.\n", "Thanks @jpinner for the input!\n\n> For example, there are two different codes depending on whether the initial line was too long or the headers block was too large (414 URI Too Long vs 431 Request Header Fields Too Large).\n\nGlad you mentioned it, as there isn't a way today to do that (unless one wants to inspect the exception message string). `HttpObjectDecoder` should provide that information while calling `createInvalidMessage()` IMO.\n\n> Further, when introducing SPDY we wanted to be able to share channel handlers between HTTP and SPDY (such as those responsible for retuning a 400 vs 414 vs 431 response). This meant that we couldn't signal these by sending different types of exceptions up the pipeline, since we would need to the Stream-Id to send the response on and we didn't want to invalidate other requests on the connection.\n\nI am a bit confused about this. Is this still true in terms of reusing channel handlers between SPDY and HTTP? I did not see that in the examples & SPDY code. \n\n> I think modifying the decoder to emit these objects as \"Full\" messages sounds feasible, but I'd like to see how this interacts with Channel Handlers designed to be shared between both HTTP and SPDY / HTTP/2 connections.\n\nGreat, can you point me to a test/example that I can use to validate if it is working fine?\n\nBased on the inputs, I will categorize the issue in two behaviors:\n\n##### Incomplete `HttpRequest`/`HttpResponse`\n\nThis is the fact that [`HttpResponseDecoder`](https://github.com/netty/netty/blob/master/codec-http/src/main/java/io/netty/handler/codec/http/HttpResponseDecoder.java#L118) and [`HttpRequestDecoder`](https://github.com/netty/netty/blob/master/codec-http/src/main/java/io/netty/handler/codec/http/HttpRequestDecoder.java#L87) are not returning a Full\\* message. \nIt looks like, returning a Full\\* message should be a viable solution, provided we can validate SPDY/HTTP/2 usecases.\n\n##### All data ignored by `HttpObjectDecoder` after these errors\n\nThis is because [`HttpObjectDecoder`](https://github.com/netty/netty/blob/master/codec-http/src/main/java/io/netty/handler/codec/http/HttpObjectDecoder.java#L369) gets into `BAD_MESSAGE` state and ignores any other data received on this connection.\n\nIt looks like this is an undesirable behavior specifically for HTTP pipelining and keep-alive connections in general. Is there any consensus on a need for fixing this in some way?\n", "@NiteshKant \n\n> Glad you mentioned it, as there isn't a way today to do that (unless one wants to inspect the exception message string). HttpObjectDecoder should provide that information while calling createInvalidMessage() IMO.\n\nOne way to do this is to check and see if the URL is \"/bad-request\" -- if it is then createMessage(initialLine) was never called. But yes, this is hacky -- I have a better object decoder sitting on the back-burner that I should open source but have never gotten around to it.\n\n> I am a bit confused about this. Is this still true in terms of reusing channel handlers between SPDY and HTTP? I did not see that in the examples & SPDY code.\n\nA common way to set up the pipeline is to add a handler that responds to the NPN to select between HTTP and SPDY decoders, and then to add common handlers after that. If you do that, then exceptions on the pipeline need to correspond to connection-level errors not stream/request-level errors.\n\nTo your issue categorizations, yes the fact that we aren't returning \"Full\" messages could be fixed. As for HttpObjectDecoder itself ignoring future data, it has to because it hasn't read enough of the message to determine proper framing. In all common deployments, these sorts of parsing errors all cause connection closure over HTTP/1.1 connections.\n", "> A common way to set up the pipeline is to add a handler that responds to the NPN to select between HTTP and SPDY decoders, and then to add common handlers after that. If you do that, then exceptions on the pipeline need to correspond to connection-level errors not stream/request-level errors.\n\nGot it, thanks for the explanation.\n\n> In all common deployments, these sorts of parsing errors all cause connection closure over HTTP/1.1 connections.\n\nAgreed that connection closure is a better behavior. If I look at it purely from HTTP/1.1 point of view, I see the following behavior in client and server:\n\n##### Server\n\nServer receives such a request and emits a special `HttpRequest` message in the pipeline. As part of handling this request, typically a server would send a response with a header \"Connection: Close\" to indicate to the sender that the connection should not be reused. Upon close by the sender, the connection closes.\n\n##### Client\n\nClient receives such a response and emits a special `HttpResponse` message in the pipeline. As part of handling this response, the client is expected to close the connection.\n\nIn both the above cases, failure to follow the expected action, would result in an active connection, data on which is silently swallowed by `HttpObjectDecoder`. Should we fix this behavior such that the channel handler does the correct thing for HTTP/1.1 and allows override of this behavior per protocol?\n\nWhat I am referring to is having an abstract method in `HttpObjectDecoder` like:\n\n``` java\nprotected abstract void onInvalidMessage(Reason invalidationReason);\n```\n\nwhich is implemented in `HttpRequestDecoder` and `HttpResponseDecoder` to do the correct thing. Any other implementation or an extension of these classes can then override this behavior as needed.\n\nIf this makes sense, I can work on a PR to implement it.\n", "> I'd like to see how this interacts with Channel Handlers designed to be shared between both HTTP and SPDY / HTTP/2 connections.\n\nI'm not sure we have to be too concerned about HTTP/2. HTTP/2 provides translation from HTTP/2 frame events to HTTP/1.x objects [InboundHttp2ToHttpAdapter](https://github.com/netty/netty/blob/master/codec-http2/src/main/java/io/netty/handler/codec/http2/InboundHttp2ToHttpAdapter.java#L39) and translation from HTTP/1.x objects to HTTP/2 frames [HttpToHttp2ConnectionHandler](https://github.com/netty/netty/blob/master/codec-http2/src/main/java/io/netty/handler/codec/http2/HttpToHttp2ConnectionHandler.java#L29). Other than these classes (and some utility support classes) I don't think there is much dependency between HTTP/2 and HTTP/1.x.\n\n> I have a better object decoder sitting on the back-burner that I should open source but have never gotten around to it.\n\n@jpinner - Does your object decoder address this issue in a similar manner to @NiteshKant's proposal? What other improvements have you made? What is your anticipated time frame for making your improvements available as a PR?\n\n@NiteshKant - I would be interested in seeing a PR. Unless @jpinner's solution already accounts for this and it can be made available within a reasonable time frame.\n", "@Scottmitch we have an internal HTTP/2 implementation that we use for 3.x and 4.0. As for our object decoder, we've made slight changes to the HTTP objects themselves so that \"DecoderResult\" is a proper future that can have listeners attached to it. We've also changed how bodies are handled -- we use something similar to the Stream proposal for Netty 5. Let me see how quickly I can open source these to at least provide them as a reference in this discussion.\n\n@NiteshKant the problem with having the decoder have an \"onInvalidMessage\" that closes the connection is that when acting as a server, it would need to first write a response to the socket. To do this it would have to sit after the object encoder on the pipeline which places an ordering dependency and breaks the HttpServerCodec handler.\n\nCan we start with a simple change that emits \"Full\" invalid messages?\n", "PR #3379 does the change to emit Full\\* invalid messages from the decoders in question.\n\nI will try and put together a PR, attempting to fix the `HttpObjectDecoder` on similar lines as I proposed so that we can discuss the options, even if we do not go that route.\n", "The simple change that emits Full\\* on an invalid message has been merged. Thanks for great discussion, folks.\n\nIf I understood correctly, the problem is that there's no easy way to determine if the message is bad because of protocol violation or because of too large headers. In case the header is too large, we do not need to enter the `BAD_MESSAGE` state but just skip the traffic until the next request/response.\n\nAs some folks have mentioned, `DecoderResult` has some limitation and it stems from the lack of proper streaming construct. I hope to introduce/prototype a new construct that could solve this problem altogether, so please feel free to add more input to #859 \n" ]

[]

[]

[ "codec-http/src/main/java/io/netty/handler/codec/http/HttpRequestDecoder.java", "codec-http/src/main/java/io/netty/handler/codec/http/HttpResponseDecoder.java", "codec-http/src/main/java/io/netty/handler/codec/rtsp/RtspRequestDecoder.java", "codec-http/src/main/java/io/netty/handler/codec/rtsp/RtspResponseDecoder.java" ]

[ "codec-http/src/main/java/io/netty/handler/codec/http/HttpRequestDecoder.java", "codec-http/src/main/java/io/netty/handler/codec/http/HttpResponseDecoder.java", "codec-http/src/main/java/io/netty/handler/codec/rtsp/RtspRequestDecoder.java", "codec-http/src/main/java/io/netty/handler/codec/rtsp/RtspResponseDecoder.java" ]

[]

[ "timestamp(timedelta=103.0, similarity=0.9495401364451412)" ]

[ "@nmittler - FYI.\n", "Great initiative.\n", "@Scottmitch good idea! I'll start looking at this today.\n", "@nmittler - Thanks for taking care of this!\n", "@Scottmitch After playing with this a bit locally, I believe we can get rid of all of these with the exception of `isResetSent`. We need to know if we sent RST_STREAM so that we ignore (and don't fail) frames coming in on that stream. I should have a PR ready with a first cut of the changes shortly ... stay tuned.\n" ]

[ "nit: Auto format?\n", "I wanted to leave a comment on [line 233](https://github.com/netty/netty/pull/3386/files#diff-179f52a9de82166ed577701a1412b5a0R233)...but too far from modifications. If the stream state is HALF_CLOSED_REMOTE does it matter if we sent reset or not? The remote side shouldn't be sending data if it has sent END_STREAM flag right?\n\n[section 5.1](https://tools.ietf.org/html/draft-ietf-httpbis-http2-16#section-5.1):\n\n> half closed (remote):\n> ...\n> If an endpoint receives additional frames for a stream that is in\n> this state, other than WINDOW_UPDATE, PRIORITY or RST_STREAM, it\n> MUST respond with a stream error (Section 5.4.2) of type\n> STREAM_CLOSED.\n", "@Scottmitch that's a good point. We should always reset if receiving on HALF_CLOSED_REMOTE. I'll make that change.\n\nRelated to this section of code ... should we always be applying flow control? I suspect we should.\n", "Done.\n", "Yes we still want to count toward flow control for most instances. Here are some relevant snippets from the spec:\n\n[section 5.1](https://tools.ietf.org/html/draft-ietf-httpbis-http2-16#section-5.1):\n\n> closed:\n> ...\n> Flow controlled frames (i.e., DATA) received after sending\n> RST_STREAM are counted toward the connection flow control window.\n> Even though these frames might be ignored, because they are sent\n> before the sender receives the RST_STREAM, the sender will\n> consider the frames to count against the flow control window.\n\n[sectoin 5.4.2](https://tools.ietf.org/html/draft-ietf-httpbis-http2-16#section-5.4.2)\n\n> A RST_STREAM is the last frame that an endpoint can send on a stream.\n> The peer that sends the RST_STREAM frame MUST be prepared to receive\n> any frames that were sent or enqueued for sending by the remote peer.\n> These frames can be ignored, except where they modify connection\n> state (such as the state maintained for header compression\n> (Section 4.3), or flow control).\n\n[section 6.8](https://tools.ietf.org/html/draft-ietf-httpbis-http2-16#section-6.8)\n\n> After sending a GOAWAY frame, the sender can discard frames for\n> streams with identifiers higher than the identified last stream.\n> However, any frames that alter connection state cannot be completely\n> ignored. For instance, HEADERS, PUSH_PROMISE and CONTINUATION frames\n> MUST be minimally processed to ensure the state maintained for header\n> compression is consistent (see Section 4.3); similarly DATA frames\n> MUST be counted toward the connection flow control window. Failure\n> to process these frames can cause flow control or header compression\n> state to become unsynchronized.\n", "Great, thanks for the references, @Scottmitch! ... that actually makes it simpler. \n\nDone.\n", "The semantic of shouldIgnoreFrame is that we don't want to propagate the frame to the application layer. Consider clarifying it's the variable name to something like 'notifyListeners'\n", "Do we need to notify the listener here when they can already listen to the stream for closure? Separate issue?\n", "@louiscryan - This is a good point. However we should also consider that `ignore` is the terminology used in the HTTP/2 specification related to this context.\n", "@Scottmitch it looks like the only place we're using this is [InboundHttp2ToHttpAdapter](https://github.com/netty/netty/blob/master/codec-http2/src/main/java/io/netty/handler/codec/http2/InboundHttp2ToHttpAdapter.java#L306). Should this code be handling general stream closure rather than specifically listening for RST_STREAM?\n", "I think I'm leaning toward leaving the terminology as-is. The logic is not just around the listener call itself - there often is some minor state change involved, such as actually creating the stream upon receipt of HEADERS. So it does seem to be slightly more than `notifyListeners`. \n", "@nmittler - IIRC the `InboundHttp2ToHttpAdapter` is capturing this to see if there is an existing message, and if so just fire up the pipeline what we have received (because it is now \"complete\"). If by `handling general stream closure` you mean the `Http2Stream.Listener` interface then: yes `InboundHttp2ToHttpAdapter` is already a listener but I'm not sure we can achieve the same result (because there is no `ctx`). So it looks like we may end up dropping the messages (from the clients point of view) if we don't have this call.\n", "Ok, let's just keep things as-is for now then.\n", "Fair enough\n" ]

[ "improvement" ]

[ "codec-http2/src/main/java/io/netty/handler/codec/http2/DefaultHttp2Connection.java", "codec-http2/src/main/java/io/netty/handler/codec/http2/DefaultHttp2ConnectionDecoder.java", "codec-http2/src/main/java/io/netty/handler/codec/http2/DefaultHttp2ConnectionEncoder.java", "codec-http2/src/main/java/io/netty/handler/codec/http2/DefaultHttp2RemoteFlowController.java", "codec-http2/src/main/java/io/netty/handler/codec/http2/Http2Connection.java", "codec-http2/src/main/java/io/netty/handler/codec/http2/Http2ConnectionHandler.java", "codec-http2/src/main/java/io/netty/handler/codec/http2/Http2LifecycleManager.java", "codec-http2/src/main/java/io/netty/handler/codec/http2/Http2Stream.java" ]

[ "codec-http2/src/main/java/io/netty/handler/codec/http2/DefaultHttp2Connection.java", "codec-http2/src/main/java/io/netty/handler/codec/http2/DefaultHttp2ConnectionDecoder.java", "codec-http2/src/main/java/io/netty/handler/codec/http2/DefaultHttp2ConnectionEncoder.java", "codec-http2/src/main/java/io/netty/handler/codec/http2/DefaultHttp2RemoteFlowController.java", "codec-http2/src/main/java/io/netty/handler/codec/http2/Http2Connection.java", "codec-http2/src/main/java/io/netty/handler/codec/http2/Http2ConnectionHandler.java", "codec-http2/src/main/java/io/netty/handler/codec/http2/Http2LifecycleManager.java", "codec-http2/src/main/java/io/netty/handler/codec/http2/Http2Stream.java" ]

[ "codec-http2/src/test/java/io/netty/handler/codec/http2/DefaultHttp2ConnectionDecoderTest.java", "codec-http2/src/test/java/io/netty/handler/codec/http2/DefaultHttp2ConnectionEncoderTest.java", "codec-http2/src/test/java/io/netty/handler/codec/http2/DefaultHttp2LocalFlowControllerTest.java" ]

[ "timestamp(timedelta=26.0, similarity=0.9581606155234135)" ]

[ "@ngocdaothanh we love PR's ;)\n" ]

[]

[]

[ "pom.xml" ]

[ "pom.xml" ]

[]

[ "timestamp(timedelta=17.0, similarity=0.9305151939322108)" ]

[ "I think it may also not be unregistering closed datagram sockets, but I haven't traced that.\n", "Sounds like a bug... will investigate\n", "Fixed... sorry for the delay.\n" ]

[]

[ "defect" ]

[ "transport-native-epoll/src/main/java/io/netty/channel/epoll/EpollDatagramChannel.java" ]

[ "transport-native-epoll/src/main/java/io/netty/channel/epoll/EpollDatagramChannel.java" ]

[]

[ "timestamp(timedelta=66.0, similarity=0.9479850702428534)" ]

[ "+1. Thanks @buchgr for taking this one on!\n" ]

[ "add `final`?\n", "I removed it because, what's the point of a private final class? Nobody can extend it anyway.\n", "I think the package private access _maybe_ for 2 reasons:\n1. Used in tests?\n2. Avoid synthetic method generation?\n\nSo I'm not sure we need the package private access but if we could make more restrictive that is good too.\n", "A class within the `DefaultHttp2RemoteFlowController` scope could :) Not that is it too hard to derive but it may provide clarity into the hierarchy intentions/assumptions and also a hint to JIT.\n", "@Scottmitch hmm well it's not used in tests and IMHO we should not care about synthetic methods at all, unless I am missing something. WDYT?\n" ]

[ "improvement" ]

[ "codec-http2/src/main/java/io/netty/handler/codec/http2/DefaultHttp2RemoteFlowController.java" ]

[ "codec-http2/src/main/java/io/netty/handler/codec/http2/DefaultHttp2RemoteFlowController.java" ]

[]

[ "timestamp(timedelta=84.0, similarity=0.8501745656004406)" ]

[ "+1\n", "Closing this since we decided to keep the current naming for the time being (in #3491)\n" ]

[]

[]

[ "codec-http2/src/main/java/io/netty/handler/codec/http2/DefaultHttp2Connection.java", "codec-http2/src/main/java/io/netty/handler/codec/http2/DefaultHttp2ConnectionDecoder.java", "codec-http2/src/main/java/io/netty/handler/codec/http2/DefaultHttp2ConnectionEncoder.java", "codec-http2/src/main/java/io/netty/handler/codec/http2/Http2ConnectionHandler.java", "codec-http2/src/main/java/io/netty/handler/codec/http2/Http2LifecycleManager.java", "codec-http2/src/main/java/io/netty/handler/codec/http2/Http2Stream.java" ]

[ "codec-http2/src/main/java/io/netty/handler/codec/http2/DefaultHttp2Connection.java", "codec-http2/src/main/java/io/netty/handler/codec/http2/DefaultHttp2ConnectionDecoder.java", "codec-http2/src/main/java/io/netty/handler/codec/http2/DefaultHttp2ConnectionEncoder.java", "codec-http2/src/main/java/io/netty/handler/codec/http2/Http2ConnectionHandler.java", "codec-http2/src/main/java/io/netty/handler/codec/http2/Http2LifecycleManager.java", "codec-http2/src/main/java/io/netty/handler/codec/http2/Http2Stream.java" ]

[ "codec-http2/src/test/java/io/netty/handler/codec/http2/DefaultHttp2ConnectionDecoderTest.java", "codec-http2/src/test/java/io/netty/handler/codec/http2/DefaultHttp2ConnectionEncoderTest.java", "codec-http2/src/test/java/io/netty/handler/codec/http2/DefaultHttp2ConnectionTest.java", "codec-http2/src/test/java/io/netty/handler/codec/http2/DefaultHttp2RemoteFlowControllerTest.java" ]

[ "timestamp(timedelta=75.0, similarity=0.91124787193252)" ]

[ "@nmittler - Do you have any suggestions to improve clarity? By `direction` do you mean local/remote? The only thing that may be tough is the Endpoint interface is agnostic to direction due to it serving as both remote/local.\n", "@Scottmitch I haven't put much thought into it yet :). For push, we do clarify direction with the method name `endpoint.allowPushTo()`. Perhaps something like `endpoint.maxConcurrentStreamsFromPeer()` or `endpoint.maxRemotelyInitiatedConcurrentStreams`?\n\nThis would more clearly map to the `SETTINGS_MAX_CONCURRENT_STREAM` parameter from the spec. It would, however, invert our current logic which maintains the value as the `maxStreams` that can be created by the endpoint.\n\nWDYT?\n", "@nmittler - Seems a bit overloaded/confusing if the `remote` (or `local`) endpoint has a `remote` method. In this context it already takes a second to get oriented when the \"local\" settings uses the \"remote\" endpoint's `maxStreams` [here](https://github.com/netty/netty/blob/master/codec-http2/src/main/java/io/netty/handler/codec/http2/DefaultHttp2ConnectionDecoder.java#L146)...and if we add more directionality...does that conceptually invert the relationship :) A bit of devils advocate but I just want to be sure we can clarify the situation by adding more directionality context.\n", "@Scottmitch yes, the idea would be to invert the relationship so that the \"local\" settings use the local endpoint for everything. I think this would help to make things more clear for everyone wading through that code.\n\nNaming is always tricky and I agree with all of your comments. Suggestions are definitely welcome! :)\n", "@nmittler - Ok thanks for clarifying. I think I understand the scope of your suggestion...thought we were just changing names and keeping existing structure. Sounds like we are in agreement...I will think on it.\n", "@Scottmitch @buchgr here are my proposed changes to the `Endpoint` interface:\n\n`createStream` (no change)\n\n`acceptingNewStreams` becomes `canCreateStream`, and is documented to say that it compares against `maxConcurrentStreams`.\n\n`numActiveStreams` becomes `concurrentStreams` and is documented to say the number of concurrent streams created by this endpoint.\n\n`maxStreams` becomes `maxConcurrentStreams` and is documented to say that it's the number of concurrently active streams allowed to be created by this endpoint (i.e. the SETTINGS_MAX_CONCURRENT_STREAMS value provided by the opposite endpoint). This essentially keeps the logic as-is, but I think the documentation along with the other name changes help to clarify things a bit. I considered having it represent the max concurrent streams allowed by the opposite endpoint, but all of the other methods refer to creating streams from this endpoint so I think it makes sense the way it is.\n\nThoughts?\n", "@Scottmitch any thoughts on this? I think this change may be less controversial than the #3491 :)\n", "@nmittler - I'm sorry for the delay :( +1 to your most recent suggestions.\n" ]

[ "I think we have a naming disconnect between the `activate`/`deactivate` on the stream and `concurrent` on the endpoint. For example the stream still has `numActiveStreams()` and `activeStreams()` which is logically the same thing as the endpoint's `numConcurrentStreams()` but with a different name. Can make this consistent?\n", "I just double checked with the HTTP/2 spec and it sounds like our notion of \"active\" is exactly what should count towards `SETTINGS_MAX_CONCURRENT_STREAMS`. Since the spec doesn't really provide a term for streams in this state (i.e. open or half-open), I think \"active\" is reasonable. So I've changed everything over to use the \"active\" terminology. I also cleaned up a few other things related to the Http2Connection API as well.\n", "why we need the Math.min(...) at all ? \n", "@normanmaurer I'm assuming your suggestion is to directly cast the long to an int here? Wouldn't that allow us to get a negative value?\n", "Shouldn't we have a `try { .. }` around this? If 1 listener has an issue that shouldn't prevent us from notifying other listeners.\n", "This may be a general comment when ever we notify listeners...another issue perhaps?\n", "I thought the explicit \"old\" style loop was to avoid garbage being created from iterators and other objects?\n", "@Scottmitch +1... We should only use for each if we either use a LinkedList or an array\n", "`open` -> `{@code OPEN}`\n`half-closed` -> `{@code HALF CLOSED}`\n", "`open` -> `{@code OPEN}`\n`half-closed` -> `{@code HALF CLOSED}`\n", "+1. I like this change...no need to duplicate the interface at this level.\n", "Was there a reason why the deactivation and state change was decoupled? Was this so priority events would still be distributed to streams in this state?\n", "But in this case since the `try { ...}` was added to the outer scope we may just want to remove it from this PR and address all instances of this issue in another PR...or just address them all in this PR?\n", "Agreed, I think we should address this as a separate issue ... created #3501. \n\nWRT the finally block, I think I'd like to keep it in this PR since not doing this could lead to a memory leak if a listener throws.\n", "Ah right ... I was trying to make intellij green :) Reverted this change.\n", "Done ... scrubbed these terms throughout this file.\n", "Done.\n", "@Scottmitch I'm not sure what the original intent was here, but the initial call to `stream.close()` will deactivate the stream anyway so AFAICT the call to `connection().deactivate(stream)` is a no-op.\n", "+1. leave as is.\n", "Yes good catch...`deactivateInternal` was being called in `close()`.\n" ]

[ "improvement" ]

[ "codec-http2/src/main/java/io/netty/handler/codec/http2/DefaultHttp2Connection.java", "codec-http2/src/main/java/io/netty/handler/codec/http2/DefaultHttp2ConnectionDecoder.java", "codec-http2/src/main/java/io/netty/handler/codec/http2/DefaultHttp2ConnectionEncoder.java", "codec-http2/src/main/java/io/netty/handler/codec/http2/DefaultHttp2RemoteFlowController.java", "codec-http2/src/main/java/io/netty/handler/codec/http2/Http2Connection.java", "codec-http2/src/main/java/io/netty/handler/codec/http2/Http2ConnectionAdapter.java", "codec-http2/src/main/java/io/netty/handler/codec/http2/Http2ConnectionHandler.java", "codec-http2/src/main/java/io/netty/handler/codec/http2/Http2EventAdapter.java" ]

[ "codec-http2/src/main/java/io/netty/handler/codec/http2/DefaultHttp2Connection.java", "codec-http2/src/main/java/io/netty/handler/codec/http2/DefaultHttp2ConnectionDecoder.java", "codec-http2/src/main/java/io/netty/handler/codec/http2/DefaultHttp2ConnectionEncoder.java", "codec-http2/src/main/java/io/netty/handler/codec/http2/DefaultHttp2RemoteFlowController.java", "codec-http2/src/main/java/io/netty/handler/codec/http2/Http2Connection.java", "codec-http2/src/main/java/io/netty/handler/codec/http2/Http2ConnectionAdapter.java", "codec-http2/src/main/java/io/netty/handler/codec/http2/Http2ConnectionHandler.java", "codec-http2/src/main/java/io/netty/handler/codec/http2/Http2EventAdapter.java" ]

[ "codec-http2/src/test/java/io/netty/handler/codec/http2/DefaultHttp2ConnectionDecoderTest.java", "codec-http2/src/test/java/io/netty/handler/codec/http2/DefaultHttp2ConnectionEncoderTest.java", "codec-http2/src/test/java/io/netty/handler/codec/http2/DefaultHttp2ConnectionTest.java", "codec-http2/src/test/java/io/netty/handler/codec/http2/Http2ConnectionHandlerTest.java" ]

[ "timestamp(timedelta=730.0, similarity=0.8429262120825501)" ]

[ "I see a similar problem also in the spdy example.\n", "@trustin for this we will need to make the toInternLogger() method public... wdyt ?\n", "@trustin @normanmaurer is there any action that I can take on this currently? And if so, will it be backward compatible with 4.1?\n", "@nmittler The thing is `InternalLogger` and `InternalLogLevel` are internal API and thus a user or an example should not use and should not be exposed to them. My suggestion was to use `LogLevel` in the `io.netty.handler.logging` package instead and let user pass the name of the logger instead of `InternalLogger`, but I'm fine if there's a better idea.\n" ]

[ "Could we maybe add a javadoc comment telling this is for internal usage only ?\n", "Done.\n" ]

[ "defect" ]

[ "codec-http2/src/main/java/io/netty/handler/codec/http2/Http2FrameLogger.java", "example/src/main/java/io/netty/example/http2/client/Http2ClientInitializer.java", "example/src/main/java/io/netty/example/http2/server/HelloWorldHttp2Handler.java", "handler/src/main/java/io/netty/handler/logging/LogLevel.java" ]

[ "codec-http2/src/main/java/io/netty/handler/codec/http2/Http2FrameLogger.java", "example/src/main/java/io/netty/example/http2/client/Http2ClientInitializer.java", "example/src/main/java/io/netty/example/http2/server/HelloWorldHttp2Handler.java", "handler/src/main/java/io/netty/handler/logging/LogLevel.java" ]

[]

[ "timestamp(timedelta=126.0, similarity=0.9553155096245479)" ]

[ "@nmittler @jpinner FYI.\n", "HPACK-11 didn't last very long :) We now have draft 12.\n\nNote that the issue description and links have been updated.\n", "@Scottmitch ha! ... thanks for keeping on top of this!\n", "Will release hpack-1.0.0 once the spec gets published as an RFC. There will likely be minor API breaking since \"Header Table\" has been renamed in the spec to \"Dynamic Table\" (i.e. maxHeaderTableSize -> maxDynamicTableSize).\n", "@jpinner - That shouldn't be a problem. Thanks for the heads up.\n\nI have seen a lot of chatter on the email thread about fixing static tables...do you think there is a draft 13 on the horizon?\n", "I am reviewing the draft 17 specification. There are a few changes to the spec which appear to be made for clerical purposes but may have unintended implications. I have opened issues on the http2-spec to get more clarity.\n~~https://github.com/http2/http2-spec/issues/726~~ - No change. Spec clarified to indicate perspective of the stream states being called out.\n~~https://github.com/http2/http2-spec/issues/727~~ - No change. Spec clarified to expand stream states which SETTINGS can impact for window update.\n~~https://github.com/http2/http2-spec/issues/730~~ - No change. Redundancy in spec removed in draft 17, but there was still text from previous drafts which keeps 0 padding restriction.\n", "@Scottmitch great work ... thanks for doing this!\n", "@nmittler - What are your thoughts on the following addition in [section 8.2](https://tools.ietf.org/html/draft-ietf-httpbis-http2-17#section-8.2):\n\n> The server MUST include a value in the \":authority\" header field for\n> which the server is authoritative (see Section 10.1). A client MUST\n> treat a PUSH_PROMISE for which the server is not authoritative as a\n> stream error (Section 5.4.2) of type PROTOCOL_ERROR.\n\nShould we be providing an additional interface used to verify \"authority\" that the user must provide? Or should we follow the approach we used with the \"cacheability\" which is just to add [comments](https://github.com/netty/netty/blob/master/codec-http2/src/main/java/io/netty/handler/codec/http2/Http2FrameListener.java#L160)? Verifying authority can be relatively involved and context dependent, but we could have a simple interface which forces the issue (which we decided against doing for the \"cacheability\" case).\n", "+1 for a simple interface. Something like:\n\n``` java\ninterface AuthorityVerifier {\n void verifyAuthority(String authority) throws Http2Exception;\n}\n```\n\nAnd they provide a default that always allows the authority.\n\nI suppose we could do something similar for cacheability, but let's leave it as-is for now.\n\nWDYT?\n", "@nmittler - Yes I like making it explicit...comments are too easy to miss.\n", "Draft changes have been cherry-picked. I'm going to open a separate issue for updating HPACK.\n", "See https://github.com/netty/netty/issues/3535 for HPACK updates.\n" ]

[]

[ "feature" ]

[ "codec-http2/src/main/java/io/netty/handler/codec/http2/DefaultHttp2ConnectionDecoder.java", "codec-http2/src/main/java/io/netty/handler/codec/http2/DefaultHttp2HeadersDecoder.java", "codec-http2/src/main/java/io/netty/handler/codec/http2/Http2CodecUtil.java", "codec-http2/src/main/java/io/netty/handler/codec/http2/Http2ConnectionDecoder.java", "codec-http2/src/main/java/io/netty/handler/codec/http2/Http2FrameListener.java" ]

[ "codec-http2/src/main/java/io/netty/handler/codec/http2/DefaultHttp2ConnectionDecoder.java", "codec-http2/src/main/java/io/netty/handler/codec/http2/DefaultHttp2HeadersDecoder.java", "codec-http2/src/main/java/io/netty/handler/codec/http2/Http2CodecUtil.java", "codec-http2/src/main/java/io/netty/handler/codec/http2/Http2ConnectionDecoder.java", "codec-http2/src/main/java/io/netty/handler/codec/http2/Http2FrameListener.java", "codec-http2/src/main/java/io/netty/handler/codec/http2/Http2PromisedRequestVerifier.java" ]

[]

[ "timestamp(timedelta=45.0, similarity=0.9056461210732123)" ]

[ "@nmittler - FYI.\n", "We associate priority with the flow control window. Therefore we rely on both peers sharing a common view of the priority tree. However there seems to be some confusion (on my part or in the spec) with how to provide feedback when we don't accept a priority frame. We can't rely upon sending a RST_STREAM to the other end because in [section 5.1](https://tools.ietf.org/html/draft-ietf-httpbis-http2-17#section-5.1):\n\n> Endpoints MUST ignore\n> WINDOW_UPDATE or RST_STREAM frames received in this state, though\n> endpoints MAY choose to treat frames that arrive a significant\n> time after sending END_STREAM as a connection error\n> (Section 5.4.1) of type PROTOCOL_ERROR.\n\nSome thought is needed (perhaps when I'm not so tired) as to how we can maintain the priority & flow control association and keep the peers in sync.\n\nFor example say stream X is closed, and it has no children.\n1. We remove it from the priority tree.\n2. Later our peer sends a priority frame for stream X.\nWhat do we do:\n- Should we have been keeping track of stream X's window? (it can't be controlled via WINDOW_UPDATE frames at this point)\n- Should we ignore it? - What does that say about our windows throughout the tree?\n", "I'm working on a PR for this. I think I was confused in my late night thought process :) The window of the individual node doesn't particularly matter because we are considering how much the whole subtree _wants_ to send, and any individual node which is not eligible to send will not contribute to this number. So ignore my previous comment.\n", "ignored :)\n", "@nmittler - Humm this didn't get auto closed like we had hoped...closing manually.\n", "@Scottmitch hmm ... maybe updates don't count and the original description has to have the \"Fixes: \" text?\n", "@nmittler I wonder if a link has to be in the git commit too?\n", "@Scottmitch ah yeah that would make sense.\n" ]

[ "missing `.` on EOL\n", "missing `.` on EOL\n", "Done.\n", "Done.\n", "Boy, \"writable\" sure would be a good name .... (Nate giggles while sipping his tea) :)\n", "maybe call this `isViableBranch()` which first checks `isViable()`?\n", "Do we need 2 methods? Could `isViable` be recursive?\n", "Hmm ... I don't like `viable` :/\n\nI don't see it in the spec anywhere, so I suppose we're free to name this whatever we like. What about just using `isNonClosed` and `isBranchNonClosed` instead? Alternatively maybe something like `prioritizable`?\n", "> Hmm ... I don't like viable :/\n\n+1. My goal in life is to introduce a \"clear\" concept and then let the word smiths figure out the rest :)\n", "Seems like a viable strategy :)\n", "Lets handle the `Do we need 2 methods? Could isViable be recursive?` comment here too.\n\nWe may be able to do this but I'm of the mindset that it is makes sense to have them separated. For example there are places (takeChild, removeChild) where we don't want to recursively check the subtree and only want to check the immediate node. WDYT?\n", "When I choose this nebulous name I didn't use `closed` because I wasn't sure if other states would need to be included here. I like `isPrioritizable()` because we are really trying to define a concept that says \"if this node by itself belongs in the priority tree\". We also need something that says \"do any of this node's descendants belong in the priority tree\"...`hasPrioritizableDescendant()`?\n", "Yeah that's fine ... just a thought.\n", "sgtm\n", "Hah...\"writable\" in the flow controller context....maybe we can add a chapter in @normanmaurer's book explaining what we mean :)\n", "@nmittler - Please verify. We were created a circular dependency in the priority tree without this check. I found it in one of the new unit tests with a StackOverFlow excpetion.\n", "@Scottmitch should this check be `!newParent.isExclusiveChild(this)`, where `isExclusiveChild` would return `true` if `numChildren == 1` and the child is the one provided?\n", "@nmittler - I like it from a clarity point of view. However I think it will be the same result in this case and it sounds like we would have to re-check some of the stuff we already know in this block (either parents are equal, or check the parent's child list, and size == 1). If we were to add this new method do you have a preference for adding to `Http2Stream` or just private to `DefaultStream`? The question is will this method have utility beyond this block.\n", "Could we just make this the one and only initialization of `events` (and get rid of the `else`) to simplify the code?\n", "@Scottmitch Oh I see ... by the time we're checking exclusive we know that the parent isn't changing, we're handling the exclusive case. Yeah this change looks right to me.\n", "We could but we already have to do the if check...and the if check also tells us what the size of the collection needs to be.\n", "The arrays only differ in size by 1 ... just seemed like overkill to branch on creating the list. But I'm fine leaving it as-is if you have a strong preference here.\n", "Not a strong preference but I do think it clarifies expectations and we already have to evaluate the branch condition.\n\nI have no statistics to back it up but I have a gut feeling the `else` condition will be true most of the time...so most of the time we won't need the extra space in the array.\n", "@nmittler - Note that I change the notification for `onStreamRemoved` to be after it is removed from the priority tree and stream map. I think this lines up with how the other listeners are used (notify after event takes place) and also the comments on the `onStreamRemoved` method.\n", "@Scottmitch Yeah that makes sense. Thanks for pointing it out!\n" ]

[ "defect" ]

[ "codec-http2/src/main/java/io/netty/handler/codec/http2/DefaultHttp2Connection.java", "codec-http2/src/main/java/io/netty/handler/codec/http2/Http2Stream.java" ]

[ "codec-http2/src/main/java/io/netty/handler/codec/http2/DefaultHttp2Connection.java", "codec-http2/src/main/java/io/netty/handler/codec/http2/Http2Stream.java" ]

[ "codec-http2/src/test/java/io/netty/handler/codec/http2/DefaultHttp2ConnectionTest.java", "codec-http2/src/test/java/io/netty/handler/codec/http2/DefaultHttp2RemoteFlowControllerTest.java" ]

[ "timestamp(timedelta=73.0, similarity=0.9107620806503808)" ]

[]

[ "This is the main ugly bit. A chicken and the egg problem ... handler needs the encoder/decoder and they need the handler. If you have a better solution, I'm all ears.\n", "@nmittler could we just get rid of the builder altogether? It's only 5 arguments and we are not doing anything special in the builder. Why not pass those arguments to the constructor directly?\n", "@nmittler instead of calling `init()` here, could we do this as the last statement in the constructor? I know it's generally a anti-pattern to let `this` escape from within the constructor, but I think it's fine here? WDYT?\n", "Done.\n", "Done.\n", "Do we need these null checks here? Won't `DefaultHttp2ConnectionEncoder` and `DefaultHttp2ConnectionDecoder` do these?\n", "I think we should avoid the escaping constructor issue. We may be OK now but changes in the future, or extensions of our classes may experience half initialized object issues. Static factory method is the first thing that comes to mind...can anyone think of anything more clever?\n", "Done.\n", "Done ... doing everything when the handler is added now.\n", "Were we at one point checking if this was null before setting it?\n", "We check that the `encoder` and `decoder` are non-null in the constructor.\n", "Ok wasn't sure if we wanted to do the same here...but I guess we want to at least ensure they have a consistent lifecycle manager.\n", "@Scottmitch what exactly is the action you're asking for here? It seems that this code guarantees they have a consistent lifecycle managers. Or are you suggesting that we ensure the lifecycle managers for both are null in the constructor?\n", "@nmittler sry for being unclear. I'm wondering if we should check if the encoder and decoder already have a lifecycle manager set before we set them (also check to make sure the are the same after). Haven't thought about it too much and I would have to check if the http2connectionhandler has any limitations where it has to be the lifecyclemanager for the encoder/decoder that it uses. Not essential but wdyt?\n", "@Scottmitch I didn't provide a get method for the `lifecycleListener` to this point, since it's not needed for anything. I'm not convinced it's worth it just to handle this ... WDYT?\n", "@nmittler - sgtm. A user could extend the encoder/decoder and override the setter methods if they really wanted I guess.\n" ]

[ "improvement" ]

[ "codec-http2/src/main/java/io/netty/handler/codec/http2/CompressorHttp2ConnectionEncoder.java", "codec-http2/src/main/java/io/netty/handler/codec/http2/DefaultHttp2ConnectionDecoder.java", "codec-http2/src/main/java/io/netty/handler/codec/http2/DefaultHttp2ConnectionEncoder.java", "codec-http2/src/main/java/io/netty/handler/codec/http2/Http2ConnectionDecoder.java", "codec-http2/src/main/java/io/netty/handler/codec/http2/Http2ConnectionEncoder.java", "codec-http2/src/main/java/io/netty/handler/codec/http2/Http2ConnectionHandler.java", "codec-http2/src/main/java/io/netty/handler/codec/http2/HttpToHttp2ConnectionHandler.java" ]

[ "codec-http2/src/main/java/io/netty/handler/codec/http2/CompressorHttp2ConnectionEncoder.java", "codec-http2/src/main/java/io/netty/handler/codec/http2/DecoratingHttp2ConnectionDecoder.java", "codec-http2/src/main/java/io/netty/handler/codec/http2/DecoratingHttp2ConnectionEncoder.java", "codec-http2/src/main/java/io/netty/handler/codec/http2/DecoratingHttp2FrameWriter.java", "codec-http2/src/main/java/io/netty/handler/codec/http2/DefaultHttp2ConnectionDecoder.java", "codec-http2/src/main/java/io/netty/handler/codec/http2/DefaultHttp2ConnectionEncoder.java", "codec-http2/src/main/java/io/netty/handler/codec/http2/Http2ConnectionDecoder.java", "codec-http2/src/main/java/io/netty/handler/codec/http2/Http2ConnectionEncoder.java", "codec-http2/src/main/java/io/netty/handler/codec/http2/Http2ConnectionHandler.java", "codec-http2/src/main/java/io/netty/handler/codec/http2/HttpToHttp2ConnectionHandler.java" ]

[ "codec-http2/src/test/java/io/netty/handler/codec/http2/DataCompressionHttp2Test.java", "codec-http2/src/test/java/io/netty/handler/codec/http2/DefaultHttp2ConnectionDecoderTest.java", "codec-http2/src/test/java/io/netty/handler/codec/http2/DefaultHttp2ConnectionEncoderTest.java", "codec-http2/src/test/java/io/netty/handler/codec/http2/Http2ConnectionHandlerTest.java", "microbench/src/test/java/io/netty/microbench/http2/Http2FrameWriterBenchmark.java" ]

[ "timestamp(timedelta=57.0, similarity=0.9322689013154061)" ]

[]

[ "@nmittler why you changed this to public ? Was this by mistake ?\n", "@nmittler - Correct me if I'm wrong but I'll take a shot...\n\n@normanmaurer - I think because before this change we only allowed access to the \"standard\" set of settings (only 6 of them) via the other public methods in this class (i.e. `maxHeaderListSize`).\n\n@nmittler - We should consider the duplication between this and `Long get(int)`. The [spec](https://tools.ietf.org/html/draft-ietf-httpbis-http2-17#section-6.5.1) calls for `value` being `an unsigned 32-bit value.`....we should call this out in the comments if we decide to expose this method publicly. We may not want to expose this and just leave `Long get(int)` as the public method. WDYT?\n", "@Scottmitch I think this method would be useful, so let me just clarify in the javadocs.\n", "`on the value` -> `on the return of {@link #get(int)}`.\n", "Done.\n" ]

[ "defect" ]

[ "codec-http2/src/main/java/io/netty/handler/codec/http2/Http2CodecUtil.java", "codec-http2/src/main/java/io/netty/handler/codec/http2/Http2Settings.java", "common/src/main/java/io/netty/util/collection/IntObjectHashMap.java" ]

[ "codec-http2/src/main/java/io/netty/handler/codec/http2/Http2CodecUtil.java", "codec-http2/src/main/java/io/netty/handler/codec/http2/Http2Settings.java", "common/src/main/java/io/netty/util/collection/IntObjectHashMap.java" ]

[ "codec-http2/src/test/java/io/netty/handler/codec/http2/Http2SettingsTest.java" ]

[ "timestamp(timedelta=25.0, similarity=0.943774633710981)" ]

[ "PR pending for this....\n" ]

[]

[ "defect" ]

[ "codec-http2/src/main/java/io/netty/handler/codec/http2/DefaultHttp2ConnectionDecoder.java" ]

[ "codec-http2/src/main/java/io/netty/handler/codec/http2/DefaultHttp2ConnectionDecoder.java" ]

[ "codec-http2/src/test/java/io/netty/handler/codec/http2/DefaultHttp2ConnectionDecoderTest.java" ]

[ "timestamp(timedelta=20.0, similarity=0.9411251524949779)" ]

[ "The behavior should be the same as the PRIORITY frame. We should create the stream. So we need to move the set priority above the notification.\n" ]

[]

[ "defect" ]

[ "codec-http2/src/main/java/io/netty/handler/codec/http2/DefaultHttp2ConnectionDecoder.java" ]

[ "codec-http2/src/main/java/io/netty/handler/codec/http2/DefaultHttp2ConnectionDecoder.java" ]

[ "codec-http2/src/test/java/io/netty/handler/codec/http2/DefaultHttp2ConnectionDecoderTest.java" ]

[ "timestamp(timedelta=154.0, similarity=0.953447460037735)" ]

[ "@Scottmitch WDYT?\n", "We ignore these frames for non-existent streams if they are for stream Ids below the lastGoodStreamId\n", "That seems like a reasonable default policy. \n\nI guess the question here is whether we need to continue to allow for a pluggable policy or just bake this behavior in. \n", "@jpinner - What is `lastGoodStreamId`? Does this relate to the stream id from a [GOAWAY](https://tools.ietf.org/html/draft-ietf-httpbis-http2-17#section-6.8)?\n\n@nmittler - I would like to understand the implications of this statement:\n\n> This way we can always remove the streams as soon as they're closed to avoid potential memory leaks.\n\nDoes `DefaultHttp2StreamRemovalPolicy` itself have a problem, or the way the interfaces are designed make memory leaks likely?\n\nMy first thought are at a minimum we should do some re-naming because the [immediateRemovalPolicy](https://github.com/netty/netty/blob/master/codec-http2/src/main/java/io/netty/handler/codec/http2/Http2CodecUtil.java#L119) is really the [default](https://github.com/netty/netty/blob/master/codec-http2/src/main/java/io/netty/handler/codec/http2/DefaultHttp2Connection.java#L69). My next thoughts are I have not yet had a need for `DefaultHttp2StreamRemovalPolicy` or to use anything but the [immediateRemovalPolicy](https://github.com/netty/netty/blob/master/codec-http2/src/main/java/io/netty/handler/codec/http2/Http2CodecUtil.java#L119). I don't necessarily have a problem with the `Http2StreamRemovalPolicy` interface but if we are trying to simplify perhaps a better extension point would be to make [DefaultHttp2Connection.deactivateInternal](https://github.com/netty/netty/blob/master/codec-http2/src/main/java/io/netty/handler/codec/http2/DefaultHttp2Connection.java#L226)'s call to `removalPolicy.markForRemoval(stream);` a bit more accessible? Something like adding a method to DefaultHttp2Connection that is called instead of using the `removalPolicy`:\n\n``` java\nprotected void markForRemoval(Http2Stream stream) {\n removeStream(stream);\n}\n```\n\nWe could then get rid of `Http2StreamRemovalPolicy`. Less object creation/garbage, conditional checking, method call redirection, and supporting interfaces. WDYT?\n", "@Scottmitch sorry, should have been more explicit -- yes, it's the stream id that would go in a goaway frame if you were to start shutting down the connection\n", "@nmittler - I re-read your issue description and looked at the interfaces a bit more...I think I came to the same conclusion as you did :)\n\nJust call [DefaultHttp2Connection.removeStream](https://github.com/netty/netty/blob/master/codec-http2/src/main/java/io/netty/handler/codec/http2/DefaultHttp2Connection.java#L192) from [DefaultHttp2Connection.deactivateInternal](https://github.com/netty/netty/blob/master/codec-http2/src/main/java/io/netty/handler/codec/http2/DefaultHttp2Connection.java#L226). Consider making [DefaultHttp2Connection.removeStream](https://github.com/netty/netty/blob/master/codec-http2/src/main/java/io/netty/handler/codec/http2/DefaultHttp2Connection.java#L192) `protected` if we want to preserve the \"removal\" hooks. Is this what you were originally suggesting?\n", "@jpinner - Thanks for clarifying, and this makes sense.\n\n@nmittler - Could we just use [connection.remote().lastStreamCreated()](https://github.com/netty/netty/blob/master/codec-http2/src/main/java/io/netty/handler/codec/http2/Http2ConnectionHandler.java#L377) for this purpose? I'm hesitant because the application has the ability to send a go away with what ever stream id they want. We may need to consider providing an extension point which returns `connection.remote().lastStreamCreated()` by default. This would give the application the opportunity to override our somewhat speculative question \"if we were to send a go away now, what stream id would the application want to use\".\n", "@Scottmitch I think we should just get rid of everything related to stream \"removal\" and just have \"closed\" be the end state.\n\nI think for now we can just use `connection.remove().lastStreamCreated()` ... this would maintain the current behavior. We can look for a way of allowing the application to override this in the future, but I think that's beyond the scope here. WDYT?\n", "@nmittler - I think we are mostly in agreement and we can work out the details in the PR.\n\n> I think for now we can just use connection.remove().lastStreamCreated() ... this would maintain the current behavior. \n\nThis is not obvious to me yet. [onWindowUpdateRead](https://github.com/netty/netty/blob/master/codec-http2/src/main/java/io/netty/handler/codec/http2/DefaultHttp2ConnectionDecoder.java#L531) and [onRstStreamRead](https://github.com/netty/netty/blob/master/codec-http2/src/main/java/io/netty/handler/codec/http2/DefaultHttp2ConnectionDecoder.java#L388) both use `connection.requireStream(...)`. This will change to support the ignore \"frames for streams that don't exist but are lower than the stream id that would be sent in a go away frame\". The user currently has some control over when errors get returned in these situations via the \"removal\" mechanism. If we force `connection.remote().lastStreamCreated()` we may be taking this control away. I think this is a good default but the question is should we be providing an extension point as we introduce these changes.\n", "@Scottmitch agreed, that's a good point. Do you have any thoughts on what the API might look like?\n", "Hmm I can't really dig too deep yet but here is an option:\n\nAdd something like `int lastGoodStream()` to `Http2Connection.EndPoint` or `Http2Connection`. Then either:\n1. just return `connection.remote().lastKnownStream()` from the default classes (extension point is extend the class and override method). OR;\n2. create an interface which can be extended (see below) and call this interface to do the work for you.\n\n``` java\npublic interface Http2StreamStatus {\n public static final Http2RemoteStreamStatus DEFAULT = new Http2RemoteStreamStatus() {\n @Override\n public int lastGoodStreamId(Http2Connection connection) {\n return connection.remote().lastKnownStream();\n }\n };\n\n int lastGoodStreamId(Http2Connection connection);\n}\n```\n\nThoughts? Can you come up with anything better?\n", "We should also consider how to reconcile any potentially new interfaces with the stream id parameter on `Http2LifecycleManager.writeGoAway`, `Http2ConnectionHandler.writeGoAway`, `DefaultHttp2ConnectionEncoder.writeGoAway`, etc... Should we not allow a stream ID here, and just use the stream ID from this new interface? Is there a more straightforward way to present this?\n", "That seems reasonable. I might suggest some name changes (this can probably be improved as well):\n\n``` java\npublic interface Http2GoAwayProvider {\n public static final Http2GoAwayProvider DEFAULT = new Http2GoAwayProvider() {\n @Override\n public int goAwayStream(Http2Connection connection) {\n return connection.remote().lastKnownStream();\n }\n };\n\n int goAwayStream(Http2Connection connection);\n}\n```\n", "@nmittler - SGTM. Should have added a caveat: names are for demo purposes only...use at your own risk ;)\n", "Ha! We should have that disclaimer on all of our .java files :)\n", "@Scottmitch I'm just getting back to this. I think in general, we can get away with removing streams as we close them. In all the inbound frame handler methods, we could more-or-less ignore frames for streams that have been closed (i.e. existing-and-closed or the streamId <= `lastKnownStream` created by either `Endpoint`).\n\nWe could be more picky and maintain some history of streamIds for which we've sent `RST_STREAM` in either direction and trigger errors if an `Endpoint` sends a (disallowed) frame after it had already sent `RST_STREAM`. However, I'm not sure it would be worth it - it might be simpler and with less overhead to just to ignore all frames (to the extent possible) for any historical stream (existing-and-closed or possibly-existed).\n\nThe only caveat to this (that I can think of ATM) would be the priority tree for outbound flow control. I believe we need to keep any closed streams around so long as they still have children. Once all of their children are removed, the closed stream can then be removed from the connection entirely. Actually now that I'm thinking about it, this might be a separate task unto itself since this would currently be a problem if using the default (immediate) stream removal policy. We should fix this to delay stream removal until after all dependencies are removed.\n\nWDYT?\n", "> In all the inbound frame handler methods, we could more-or-less ignore frames for streams that have been closed (i.e. existing-and-closed or the streamId <= lastKnownStream created by either Endpoint).\n\n+1. Yes seems reasonable to me.\n\n> We could be more picky and maintain some history of streamIds for which we've sent RST_STREAM in either direction and trigger errors if an Endpoint sends a (disallowed) frame after it had already sent RST_STREAM\n\nPerhaps we don't maintain history by default....but it may be worth it to provide an interface which supports this? Kinda like we have today....humm\n\n> The only caveat to this (that I can think of ATM) would be the priority tree for outbound flow control.\n\nYes this may be problematic. Lets handle this as a separate issue. I think we need to think about this because `WINDOW_UPDATES` are only accepted for a `short period after a DATA or HEADERS frame containing an END_STREAM flag is sent` while PRIORITY frames can be received indefinitely. I'm wondering if some kind of deadlock may occur if we keep something in the priority tree and the window cannot be adjusted....\n", "> I'm wondering if some kind of deadlock may occur if we keep something in the priority tree and the window cannot be adjusted....\n\nIt seems reasonable to me that we always except WINDOW_UPDATES for streams that are still in the priority tree. Assume stream A is closed and has children B and C. If we receive a `WINDOW_UPDATE` for A, I would expect it to still impact the data written for B and C. After B and C are closed (and removed), we can then remove A. From that point forward we ignore WINDOW_UPDATES for stream A.\n", "> It seems reasonable to me that we always except WINDOW_UPDATES for streams that are still in the priority tree.\n\nI agree. However the specification may not allow sending of WINDOW_UPDATES in these states. Just need to think about this I guess....\n", "Created new issue for the priority tree management https://github.com/netty/netty/issues/3518.\n", "ahh there is already an issue for this, I was thinking about opening one myself this afternoon :-). +1\n", "@nmittler - Is this still on your plate?\n", "@Scottmitch I'm was waiting for the dust to settle on some of the other changes to flow control. I think everything is in now so I can take a look ... or were you looking to take this on?\n", "@nmittler - Nope I haven't touched this.\n", "@nmittler - Do you want https://github.com/netty/netty/pull/3543 to be pulled in before this?\n", "@Scottmitch yeah I think that makes sense.\n", "@nmittler - #3543 has been cherry-picked...you should now be unblocked.\n", "Nice to see this being fixed! We have seen some very large object allocations (several megabytes per object) due to the removal policy i.e. when opening and closing ~20,000 streams per second.\n", "@buchgr - Cool to see numbers and that you are pushing the codec hard...not cool that they appear to be less then optimal :) I think we are in a better place to tackle this and I hope it simplifies things and helps with the number too. Have you tried using the ImmediateRemovalPolicy instead? Do you see any other areas for improvement that we are not currently working on?\n", "@nmittler - A few PRs have been cherry-picked since we last visited this. Let me know if this is still blocked, or you are too busy to handle this. I would like to get this in before https://github.com/netty/netty/issues/3501.\n", "@Scottmitch I talked to @nmittler and I ll submit a PR for this.\n" ]

[ "Kind of strange to say \"Default implementation\" as this is not an implementation of an interface.\n", "@nmittler I think this comment is not right. Streams should be removed even if they have children, but the children should be reprioritised and get a new parent?\n", "@buchgr I believe the comment accurately reflects the current behavior\n\n@Scottmitch any thoughts?\n", "@nmittler hmm okay, would this then not allow for leaks i.e. \"closed parent streams\" sticking around after all their children have been removed?\n", "+1. I removed the interface and just kept the class as part of review. A comment similar to what you added for `interface Event` may make more sense?\n", "> @buchgr I believe the comment accurately reflects the current behavior\n\n+1.\n\nThe current behavior is such that streams remain in the priority tree until `prioritizableForTree() == 0`. Events which can change this count (parent change, stream removal from tree, state change, etc...) re-evaluate whether the stream still belongs in the tree, and calls `removeStream` if not.\n\n@nmittler - I'm not sure we need this comment here though...`removeStream` has a pretty detailed comment on it...maybe just clean that comment up if it is not adequite?\n", "> The current behavior is such that streams remain in the priority tree until prioritizableForTree() == 0. Events which can change this count (parent change, stream removal from tree, state change, etc...) re-evaluate whether the stream still belongs in the tree, and calls removeStream if not.\n\nnice. thanks for clarifying!\n", "Done.\n", "Comment removed.\n" ]

[ "improvement" ]

[ "codec-http2/src/main/java/io/netty/handler/codec/http2/DefaultHttp2Connection.java", "codec-http2/src/main/java/io/netty/handler/codec/http2/DefaultHttp2StreamRemovalPolicy.java", "codec-http2/src/main/java/io/netty/handler/codec/http2/Http2CodecUtil.java", "codec-http2/src/main/java/io/netty/handler/codec/http2/Http2StreamRemovalPolicy.java" ]

[ "codec-http2/src/main/java/io/netty/handler/codec/http2/DefaultHttp2Connection.java", "codec-http2/src/main/java/io/netty/handler/codec/http2/Http2CodecUtil.java" ]

[]

[ "timestamp(timedelta=18.0, similarity=0.8650718977788289)" ]

[ "+1. Thanks for reporting!\n", "@nmittler - I'll take this one. As part of the resolution to https://github.com/netty/netty/issues/3550 I needed to do additional iteration over methods which use listeners and so I had to do something about this issue. My approach was to catch `RuntimeExceptions` and only throw the last one:\n\n``` java\nRuntimeException lastException = null;\nfor (Listener l : listeneres) {\n try {\n l.notify(...);\n } catch (RuntimeException e) {\n lastException = e;\n }\n}\nif (lastException != null) {\n throw lastException;\n}\n```\n\nDoes this make sense? The goal being 1 listener's exception shouldn't prevent other listener's from being notified. However I am throwing 1 exception at the end of notification...just to ensure the exception is not swallowed...WDYT?\n", "@Scottmitch yep that sgtm ... that's sort of what I was thinking as well. Make it so!\n", "@nmittler - Fair warning...it is ugly to ensure we process all events and complete all tree operations and also handle the `RuntimeExceptions` :(\n", "@nmittler - An alternative approach could be that we document the `Http2Connection.Listener` interface such that thrown exceptions will result in the connection being closed (`GO_AWAY internal error`). This way we don't have to worry about tracking exceptions in recursive calls, and making sure state (stream state, priority tree, etc...) is maintained in the presence of exceptions...WDYT?\n", "@Scottmitch yeah, that actually sounds better to me. Let's keep it simple and just update the docs.\n", "@nmittler - I have to investigate but there may be code changes to ensure this always results in a GO_AWAY. For example the removal policy stuff makes this a bit more complicated. Perhaps we should handle this issue after https://github.com/netty/netty/issues/3448?\n", "@Scottmitch yeah that's fine.\n", "Reminder to me...also fix https://github.com/netty/netty/pull/3577/files#r27757163.\n", "@nmittler - I have looked at this a bit more. When events occur as a result of read operations I think we ultimately catch the `RuntimeExceptions` and translate these to GO_AWAY frames. However on the send side we don't do this, but instead we sometimes (depending on if try/catch blocks are placed appropriately) fail the promise that was provided in the `writeXXX` call. My concern is that we are in a situation where we use listeners to update stream state, and it is possible that 1 listener (potentially user listener) could interrupt this process, and then we just continue processing events while in a bad state. Should we translate runtime exceptions on the write side into `lifecycleManager` failure events? Any other ideas?\n", "@Scottmitch yeah, I think that makes sense. So fail the promise and send RST_STREAM/GO_AWAY. Are you thinking of just adding this as try/catch logic to each encoder.writeXXX method?\n", "@nmittler - I will investigate more but I think so....it is not ideal to handle these events up the call stack because then the classes which invoke the listeners are implicitly depending on those higher level classes to be present to clean up....I think the other options are to track exceptions at the call sites and either log them or throw the last one...but as I said this gets ugly...especially if we want to propagate 1 of the exceptions in the recursive calls we have.\n", "@nmittler - A \"cleaner\" alternative (from us tracking the exceptions perspective) is just to \"log and swallow\" these types of exceptions at the call site. We won't have to worry about tracking/propagating the exceptions...but swallowing exceptions is terrible too....\n", "@Scottmich actually, I think I like logging (WARN?) better. Do you want to throw a PR together?\n", "@nmittler - I'm glad you said that because I just finished a PR for this..and just added a unit test for it ;)\n", "@nmittler - I updated the javadocs to describe this behavior (and logged it as an error).\n", "#3615 cherry-picked\n" ]

[ "Just noticed this dependency. @nmittler - Is there no other way? :)\n", "Nooooo! Let's just use `AtomicReference` :-)\n", "+1 for `AtomicReference` :)\n", "Done :)\n", "Fixed in this PR for simplicity\n" ]

[ "defect" ]

[ "codec-http2/src/main/java/io/netty/handler/codec/http2/DefaultHttp2Connection.java", "codec-http2/src/main/java/io/netty/handler/codec/http2/Http2Connection.java", "codec-http2/src/main/java/io/netty/handler/codec/http2/Http2ConnectionHandler.java" ]

[ "codec-http2/src/main/java/io/netty/handler/codec/http2/DefaultHttp2Connection.java", "codec-http2/src/main/java/io/netty/handler/codec/http2/Http2Connection.java", "codec-http2/src/main/java/io/netty/handler/codec/http2/Http2ConnectionHandler.java" ]

[ "codec-http2/src/test/java/io/netty/handler/codec/http2/DefaultHttp2ConnectionTest.java" ]

[ "timestamp(timedelta=17.0, similarity=0.9976609011293613)" ]

[ "@nmittler @buchgr - FYI...I have a fix for this. Standby.\n", "https://github.com/netty/netty/issues/3618 fixed this\n" ]

[ "maybe just call this `retain`?\n", "Doesn't `removeStream` notify the listeners that the stream was removed?\n", "if `oldParent != null`, isn't it an error condition if removal of the child from `oldParent` is null?\n", "Isn't it an error condition if the `put` != null? Maybe we should throw if that's the case?\n", "I wonder if we really need a second hash map for this. To reduce object allocation, could we just do something like:\n\n``` java\nDefaultStream[] childArray = children.values();\nfor(int index=0; index < childArray.length; ++index) {\n DefaultStream grandChild = childArray[index];\n if (child.id() != grandChild.id()) {\n child.takeChild(grandChild, false, events);\n }\n}\n```\n", "Done.\n", "Yes if it is able to remove the stream. Is there a specific issue you are worried about here?\n", "Ah I think I just read this too fast ... looks like that is the intent here. Makes sense. Do you have a test for this?\n", "I don't think so. When a stream is created the connection stream does a `takeChild` on that stream....and the new stream's parent will be null. After this it would be an error condition though...were you thinking of having the connection stream override this method and have the mainstream case not have to worry about the null check?\n", "Yes I think with the new conditional statement guarding this we should no longer have to worry about a stream existing with the same ID. This condition seems like a programming error, should we just assert?\n", "I think this would lead to concurrent modification scenarios. Unless you are implying a copy operation via the `children.values();`.\n", "Ok. Yes `priorityChangeWithNoPrioritizableDependentsShouldRestructureTree` hits this.\n", "@nmittler - Your comment was collapsed:\n\n> > I wonder if we really need a second hash map for this. To reduce object allocation, could we just do something like:\n> > \n> > DefaultStream[] childArray = children.values();\n> > for(int index=0; index < childArray.length; ++index) {\n> > DefaultStream grandChild = childArray[index];\n> > if (child.id() != grandChild.id()) {\n> > child.takeChild(grandChild, false, events);\n> > }\n> > }\n\nmy response:\n\n> I think this would lead to concurrent modification scenarios. Unless you are implying a copy operation via the `children.values();`.\n", "@Scottmitch not quite sure I follow. So in the case of a new stream, I think you're saying that `oldParent == null`? I was actually referring to the case where `oldParent != null`.\n", "sgtm, thanks!\n", "Are we concerned with any side-effects with calling back the listener here while we're in the middle of rearranging the priority tree? I wonder if it would be better to store up these removal events and execute them at the end?\n", "sgtm ... I think what you have is fine.\n", "@nmittler - I think I'm confused now :)\n\n> if oldParent != null, isn't it an error condition if removal of the child from oldParent is null?\n\nSeems like you are saying \"if `oldParent == null` isn't this an error\". This is what my `I don't think so` applied to.\n\n> So in the case of a new stream, I think you're saying that oldParent == null\n\nYes. New streams start with `parent = null`. Then `connection.takeChild(newStream,...)` is called (for everything but the connection stream....connection stream parent remains `null`).\n", "> > if oldParent != null, isn't it an error condition if removal of the child from oldParent is null?\n> \n> Seems like you are saying \"if `oldParent == null` isn't this an error\". This is what my `I don't think so` applied to.\n\nAh ok, I think I understand your confusion now :). So looking at this line:\n\n``` java\nif (oldParent != null && oldParent.children.remove(child.id()) != null)\n```\n\nMy original question was: how can `oldParent.children.remove(child.id())` ever be `null`? In fact, it seems like it should be an error condition as this would imply an inconsistency in the tree ... or am I missing something?\n", "Yes there is concern not so much about notifying the listener but more about corrupting the tree links. Priority tree operations are happening and we could be removing elements from the tree...I want to avoid keeping extra state and queueing these events if possible. I'm not sure if this is entirely legitimate but I have been trying to justify it as \"these nodes don't belong in the tree and are only their because of their dependents require them to be there. If at any point they are no longer required through association they should immediately be removed.\".\n\nI went through a few scenarios (particularly when the new parent is a descendent of the current node) and didn't see any immediate issues with this approach...but would appreciate if you could come up with a case that would be problematic.\n", "Ahh this makes sense now....this is the `remove` portion to your https://github.com/netty/netty/pull/3618#discussion_r28195572. Yes, for the same rational as the other discussion.\n", "no scenarios at the moment :) ... let's leave it as-is. Maybe add a comment indicating the potential risk?\n", "Yeah assert seems reasonable.\n", "Yeah ok ... so add an assert here as well?\n", "@nmittler - Logically this makes sense...but I'm getting unit test failures...trying to understand what is going on....\n", "@nmittler - The `retain(..)` call when an exclusive dependency happens is messing things up. We remove from the children collection in a special way such that we don't update the child's parent link until the next `takeChild` recursive call. This means that the `oldParent` won't have the child in its `children` collection...\n", "Done. \n", "@nmittler - Turns out it is possible (and legit with the current code structure) for `oldParent.children.remove(child.id()) == null`.\n\n> @nmittler - The `retain(..)` call when an exclusive dependency happens is messing things up. We remove from the children collection in a special way such that we don't update the child's parent link until the next `takeChild` recursive call. This means that the `oldParent` won't have the child in its `children` collection...\n", "@nmittler - Please review this comment.\n" ]

[ "defect" ]

[ "codec-http2/src/main/java/io/netty/handler/codec/http2/DefaultHttp2Connection.java" ]

[ "codec-http2/src/main/java/io/netty/handler/codec/http2/DefaultHttp2Connection.java" ]

[ "codec-http2/src/test/java/io/netty/handler/codec/http2/DefaultHttp2ConnectionTest.java" ]