Don't retransmit old MAX_STREAM_DATA

[ekr]

The current text says that one ought to retransmit MAX_STREAM_DATA:

   o  All application data sent in STREAM frames MUST be retransmitted,
      unless the endpoint has sent a RST_STREAM for that stream.  When
      an endpoint sends a RST_STREAM frame, data outstanding on that
      stream SHOULD NOT be retransmitted, since subsequent data on this
      stream is expected to not be delivered by the receiver.

   o  ACK and PADDING frames MUST NOT be retransmitted.  ACK frames
      containing updated information will be sent as described in
      Section 8.13.

   o  STOP_SENDING frames MUST be retransmitted, unless the stream has
      become closed in the appropriate direction.  See Section 10.3.
   o  All other frames MUST be retransmitted.

But this is silly. Say I transmitted MAX_STREAM_DATA=100, MAX_STREAM_DATA=200, MAX_STREAM_DATA=300 and then none of them were acknowledged. Why am I retransmitting frames which must be ignored.

[ianswett]

Agreed, this is silly.

[janaiyengar]

This is silly. We could say something along the lines of "the newest MAX_STREAM_DATA and MAX_DATA MUST be retransmitted until acked; all other previous flow control advertisements SHOULD NOT be retransmitted."

[martinthomson]

Or as I have said on numerous occasions, we should rewrite this section so that it concentrates on the retransmission of the information contained in frames rather than the frames themselves. We have the same problem with all frame types that are used to carry data that needs to be reliably delivered to the other side, why not fix it systematically rather than piecemeal. The text above shows that we've done that for STREAM and ACK. The only frame that doesn't carry information that needs to be reliably delivered is PADDING.

[ianswett]

Agreed, misuse of the word retransmit is a widespread issue. On the positive side, it should be an uncontroversial editorial fix.

[ekr]

I think as a short-term fix we should do what Jana suggests. Minq presently does that. And the spec is so wrong now we should fix that ASAP

As a long-term fix, I think what we want is actually that when you discover it is time to retransmit MAX_STREAM_DATA (e.g., timer expired, you discover the data was lost) you send the then-current flow control window. That seems like it will be more accurate.