D2593: state: add logic to parse the state file in old way if cbor fails

[martinvonz (Martin von Zweigbergk)]

martinvonz added a comment.


  In https://phab.mercurial-scm.org/D2593#44291, @indygreg wrote:
  
  > Not sure where to record this comment in this series. So I'll pick this commit.
  >
  > I think we want an explicit version header in the state files so clients know when they may be reading a file in an old format. For example, if I start a merge in one terminal window, I sometimes move to another terminal window to resolve parts of it. The multiple windows may be running different Mercurial versions. For example, sometimes one of the shells has a virtualenv activated and that virtualenv is running an older Mercurial. We don't want the older Mercurial trampling on state needed by the new Mercurial.
  
  
  Also, it doesn't seem like CBOR defines any magic bytes to start the top-level object with, so it's not obvious to me that an old state file (e.g. on containing just a nodeid) could not be parsed as a valid CBOR file. Perhaps cmdstate should help with that? We could make it always add a first item that's just "CBOR" or something (it seem very unlikely that we'd have that in an old state file), and we could fail when reading a state file that doesn't have that. Or would could require any new state files to have a new name than the old ones?

REPOSITORY
  rHG Mercurial

REVISION DETAIL
  https://phab.mercurial-scm.org/D2593

To: pulkit, #hg-reviewers
Cc: martinvonz, indygreg, durin42, mercurial-devel