On Thu, Mar 11, 2021 at 5:39 PM Mark Thomas <ma...@apache.org> wrote:
> On 11/03/2021 16:11, Rémy Maucherat wrote: > > On Thu, Mar 11, 2021 at 3:17 PM Mark Thomas <ma...@apache.org> wrote: > > > >> A question mainly for Rémy I guess. > >> > >> Any reason not to back-port the asyncIO HTTP/2 implementation to 8.5.x? > >> > > > > Today's my lucky day it seems :) > > > > Ok, so there's a balance between the usefulness of the feature, the > > complexity of the changes, the risk introduced and the stability of more > > ancient branches. The older the branch, the higher the expected stability > > IMO. There's often the scenario of backporting too early: the feature > seems > > fine, but actually it's just not been tested enough yet. > > > > Here, I was thinking the feature is not a must have, and 8.5 is supposed > to > > be really stable now, so I limited the backport to 9 and never considered > > 8.5. I understand the increased usefulness it has if the idea is to > really > > harmonize the h2 code. > > Yes, it is a balance. > > Given the recent flurry of HTTP/2 issues that have affected 8.5.x, 9.0.x > and 10.0.x, I think there is more benefit in harmonization than there is > risk in introducing the async code. > > I think I have most of the alignment complete apart from the async > stuff. I'll try and keep that in a separate commit in case it causes > problems. > After checking, +1 for fully harmonizing the HTTP/2 code, the bug reports are now quite limited and less serious, while we can assume there's significant use of HTTP/2. I looked at the connector code and the backports seem to be there. Rémy