Re: H2 vs HPACK header table size... again

On Fri, Dec 24, 2021, at 10:54, Kazuho Oku wrote:
> In case of H2O /, this is a FIXME that never has been fixed. 

Ahh, that's not great.  It makes <4096 pretty perilous for clients.  I wonder if that is the same reason for other implementations.

I didn't realize that Cory merged #1003; I'm not that far through my email yet...  My intent was to keep this open a little.  I'll revert that and reopen it so that we can keep discussing it.

Given where things stand, the change you suggest in #1004 is probably positive from a compatibility perspective.  It allows implementations to ignore increases in the maximum.  For your implementation, all you have to do is deal with a reduction in the maximum table size more elegantly (I'd start with a connection error of type INTERNAL_ERROR).


I thought for a while we needed to update RFC 7541 as a result, but a careful read reveals that we're OK.  The most interesting part is the caution about reducing the maximum size.  It was a little fiddly to write.

Received on Friday, 24 December 2021 00:51:00 UTC