Re: [quicwg/base-drafts] DATA frame encoding is inefficient for long dynamically generated bodies (#1885)

[Kazuho Oku <[e-mail]>]

@dtikhonov 
> Isn't this as simple as foregoing the initial size check and closing the connection if you do hit the internal limit?

You also need to handle EOS differently. In the bounded frame case, seeing EOS before consuming the entire frame is a protocol error. In the non-bounded case, seeing it is a non-error.

To generalize, allowing the use of EOS as a terminator for non-DATA frames introduces new complexity to *some* decoders. OTOH, allowing that does not simplify on the encoder side, because you need the code to emit such frames in the middle of the stream.

This is very different from DATA frames:
* For many HQ stacks, DATA would be the last frame to be emitted on a stream. Therefore, you can get rid of the code that emits bounded DATA frames if we introduce the use of EOS as a terminator.
* It does not make the decoder complex, because in practice no decoder will block until it receives an entire DATA frame.
* DATA will be the most frequently used frame (compared to nothing else). Therefore, it is beneficial to reduce the number of the frames (by allowing use of EOS as the terminator).

-- 
You are receiving this because you are subscribed to this thread.
Reply to this email directly or view it on GitHub:
https://github.com/quicwg/base-drafts/issues/1885#issuecomment-431570158