Hi, On 06/06/2020 14:07, Frank Crawford wrote: > On Thu, 2020-06-04 at 09:44 -0400, Derek Atkins wrote: >> So long as there isn't a client-server wire-format incompatiblity >> with >> older 2.x versions, I don't care! (I want to ensure we never make >> that >> same mistake again of a wire incompatibility due to a change in >> underlying tooling).
Just to be clear: we will break interfaces also in the future but we will do our best (but also not more) to break them across 3 versions instead of only 2, giving you one middle version to do your migration. Telling you anything else would be a promise to obsolete rdiff-backup. > And the only way to do that is to move to your own protocol over the > wire instead of using something that is tied to low-level python primitives. I slightly disagree with this statement: with a concept of client/server protocol version mapped to a pickle protocol version [1] we could probably avoid the issue. Actually, while writing this e-mail, I checked the code and noticed that it's enforcing pickle version 1 so the issue isn't the pickle protocol, the issue is solely the bytes vs. str vs. unicode change between python 2 and 3. There is not much we could have done about it IMHO. > Saying that, doubt that pickles will cause issues in the near future, > and so it is more likely to break due to change in function names > within rdiff-backup. We could enforce protocol 5 to infuriate our users but we won't do this ;-) Seriously, we should move to protocol 3 or 4 (known since Python 3.4 so no issue with incompatibility), it should improve speed and resilience. KR, Eric
