Table resize can't be acknowledged

[martinthomson]

Acknowledgments encode the number of insertions, which is nice because that's what you need to track for other reasons. However, this makes resizing the table invisible and not acknowledged. I think that's OK, but it means that an encoder can't rely on increased space being available until it receives acknowledgment for the next insertion after the resize.

I think that's worth calling out in the description of the Dynamic Table Size Update instruction.

(Marking this design in case someone disagrees with my conclusion.)

[afrind]

The inserts and the table size updates all happen on the encoder stream, which is in-order and reliable. Why can't the encoder assume the additional space is available immediately after encoding the table size update?

[MikeBishop]

Indeed, HPACK has no acknowledgements for this. HPACK has acknowledgements (via H2 SETTINGS) for changes in the decoder's maximum table size, but that's mostly immutable in HQ.

[afrind]

@martinthomson : are we good to close this with no-action or you think we could benefit from editorial text explaining the similarity to HPACK in this regard?

[martinthomson]

"I think that's worth calling out in the description of the Dynamic Table Size Update instruction." - just a few words noting that the change isn't important from the perspective of the encoder, so it doesn't get acknowledged, except transitively.