Item8971: Working copy versions are pure trouble
Having beaten my head off various version history problems for the last week, I've come to some conclusions:
- In most cases, the version number in the working copy has to be ignored.
- The store has to be able to understand the possibility of non-sequential versioning.
- RCS version numbers are just confusing the issue.
There are still corner cases that will fail, mostly to do with manual manipulations of the version history, so a new testcase LoadedRevTests
covers the important cases.
Note that it is possible now to have a meta object with undefined text, when said object does not exist. This was a deliberate change, and is an attempt to try and clean up the semantics of topic objects further.
There are no functionality changes
- this is all bugfixing.
Note that the work doesn't make non-sequential versioning entirely functional, as the RDiff code assumes sequential version numbers when it diffs over version ranges.
- 04 May 2010
No complaints (yet), so marking as ready for release.
- 06 May 2010
A test implementation of a subversion store - which is non-linear - has shown up some issues with the API, so making some mods.
- 13 May 2010
Subversion store is working, so guess this can be closed now.
- 23 May 2010