Announce: rsync spec + conformance test harness

[danshearer]

Hello all,

I've pushed an early version of rsync-rfc-conformance, with thanks to Dan Mestas for assistance and encouragement. This specification is sufficient to write a program that can have a protocol v32 conversation with rsync as client or server, with all optional features. There's the odd wobbly bit (I'm not convinced the xattr/ACL protocol conformance test is quite right yet, for example) but it does seem to work pretty well. It also handles protocols 27-31 for various definitions of 'handle'. I've really only tested it against 32 and 31, mostly 32. There are many ways this might potentially be useful, for example, it could be a source of a new category of test case just for wire protocol.

Best,

--
Dan Shearer
dan@shearer.org / https://shearer.org

[steadytao]

Thanks for publishing this. A protocol conformance harness could be very useful for rsync, especially for catching wire-format regressions separately from the normal integration tests. It may be worth tracking possible rsync test-suite integration separately once the v32 cases are properly stable.

[danshearer]

Sounds good. I have derived v32 behaviour by wire snooping and the source code, and based the RFC and implemented test cases against these.

Do you see v32 behaviours and featureset changing much in the near future?

[steadytao]

Do you see v32 behaviours and featureset changing much in the near future?

Incompatible changes to the existing v32 wire behaviour should not occur but new compat-flagged features and conformance clarifications may still be added so I would record the exact rsync version or commit tested by each harness result.