[Gnu-arch-users] Inexact patching and directory renames

[Aaron Bentley]

Hi all,

As far as I can tell the changeset format doesn't provide a way of 
handling the case where a file is added to a renamed directory.

In changesets, file inventory ids are stored in mod-files-index, 
orig-dirs-index, and so on.  But it's just a partial list, containing 
only entries for things that have changed.

While it provides an inventory id for the new file, I don't see any way 
of determining the inventory id of the parent directory.  In theory, you 
could look it up in a revlib, but that's just ugly.

I think this could be fixed by providing the inventory IDs of all 
directories that have files added or deleted.  Since the directory 
itself hasn't been added or deleted, the ID should be stored in both 
orig-dirs-index and mod-dirs-index.

For compatibility with old archives, it would make sense to support 
falling back to pathnames when no parent directory is available.  This 
also makes sense because the tree root directory has no inventory ID.

The problem with having these two behavours is that old changesets 
occasionally store the parent id by coincidence.  In order to 
distinguish between them, it might be necessary to add a changeset 
version identifier.

Current versions of tla seem to ignore extra directory info, so there 
would be minimal breakage.

Comments?  Flames?

Aaron

--
Aaron Bentley
Director of Technology
Panometrics, Inc.