There is also an analog to the renegotiate + certifiable request pattern in TLS 1.3 which is similarly HTTP/2-incompatible. I don't think TLS 1.3 changes anything about the situation. On Tue, May 23, 2023, 10:47 Bence Béky <bnc@chromium.org> wrote: > Hi Lucas, > > Chrome had also been seeing it in the context of NTLM (and there were > complaints that Chrome would not try HTTP/2 again afterwards), see > https://crbug.com/516237, https://crbug.com/685741 and > https://crbug.com/713851. Discussion was most active in 2017, and the > last comment is from 2021, not sure how often HTTP_1_1_REQUIRED is > still being used. Unfortunately Chrome does not record error codes of > incoming RST_STREAM frames in HTTP/2. > > Bence > > On Tue, May 23, 2023 at 10:37 AM Daniel Stenberg <daniel@haxx.se> wrote: > > > > On Tue, 23 May 2023, Lucas Pardue wrote: > > > > > I thought I'd ask here to crowdsource some answers about whether these > codes > > > are actually used in practice. > > > > In curl we have certainly seen HTTP_1_1_REQUIRED used in the past and we > have > > support for it in code (and nghttp2 knows about it can can return info > that it > > arrived). I can't say how often or even if it still actually is used in > the > > wild. > > > > The last report I find now of a server returning it was in March 2019 > (because > > curl misbehaved). Used for downgrading from HTTP/2 to be able to do > NTLM... > > > > -- > > > > / daniel.haxx.se > > > >
Received on Tuesday, 23 May 2023 14:50:30 UTC