@normanmaurer I've added new logic that allows a number of buffers to be forwarded before a readComplete, and a test. I hope that works for you. The logic is getting complicated though. PTAL.

I've also changed the handler API to create the decompressor lazily, in handlerAdded. This looks a bit odd (you pass the decompressor builder, not the decompressor, to the handler) but I think it works as an API and it's safer.

I've added a HttpContentDecoder and -decompressor replacement, based on the BackpressureDecompressionHandler. I've also deprecated all the old decoders.

I still need to migrate DelegatingDecompressorFrameListener. It is a bit more difficult because it uses a different flow control mechanism. Probably it needs an independent Decompressor-based implementation, not based on BackpressureDecompressionHandler.

Finally, I changed the decompressor API a little bit so that the alloc is required by the .build call, not when creating the builder. This complements the change in 9279fc0. Now, the BackpressureDecompressionHandler will use the alloc from the context, and you don't need to manually pass in the alloc at all before that.

Okay, I rebuilt the decompression codecs again. HTTP/1.1 pipelining has a lot of edge cases to handle. I decided to build a central AbstractBackpressureDecompressionHandler which is designed for HTTP/1.1 pipelining, and then also extend it for BackpressureDecompressionHandler, since BackpressureDecompressionHandler is basically just the base case without pipelining. Easier than maintaining two separate backpressure implementations.

I've also built a Http2FrameListener based on the new decompressors. It appears to be a lot simpler than the ChannelHandler approach, unless I missed some edge cases.

I'm going on vacation for a few days so I might be slow to answer now.