Re: [Gnu-arch-users] Changing ID

[Miles Bader]

On Wed, Apr 21, 2004 at 04:07:00PM -0400, Stefan Monnier wrote:
> I wish there was a way for tla to say "hmmm, let's represent this
> remove+add on the same file name as an ID-change operation".  The ID-change
> operation would of course be accompagnied with a diff of the content of the
> file has changed (which is basically always the case if you're using
> taglines).

Yeah, this has always been something I've wanted (I was mostly thinking of
cases where the user _wants_ to change an explicit id to a tagline or
vice-versa, but it's always annoying to see tagline correction changesets
being so large, and prone to cause conflicts).

Since it probably requires changing the way changesets are handled a bit, it
has the new-archive-version-required problem that seems a good way to kill
any further discussion (I think this is one reason similar topics always
seemed to end up with people saying "just live with the add+delete").

As for the actual representation, it seems like the current changeset format
could actually represent this nicely:

A normal changeset represents a patch like:

   old-id-mapping:      ID_TAG  OLD_FILENAME
   new-id-mapping:      ID_TAG  NEW_FILENAME

   if both mappings exist for a given ID_TAG:
      patch-file:       patches/NEW_FILENAME.patch
   else if only the old-id-mapping exists:
      old-file:         old-files/OLD_FILENAME
   else
      new-file:         new-files/NEW_FILENAME

Currently, an id change in RETAGGED_FILE looks like:

   old-id-mapping:      OLD_ID_TAG  RETAGGED_FILE
   new-id-mapping:      NEW_ID_TAG  RETAGGED_FILE
   old-file:            old-files/RETAGGED_FILE
   new-file:            new-files/RETAGGED_FILE

So it seems like you could just change the `if' test above to something like
`if both mappings exist for a given ID_TAG _or_ both mappings exist for a
given FILENAME', and then use the obvious:

   old-id-mapping:      OLD_ID_TAG  RETAGGED_FILE
   new-id-mapping:      NEW_ID_TAG  RETAGGED_FILE
   patch-file:          patches/RETAGGED_FILE.patch

instead.

-Miles
-- 
.Numeric stability is probably not all that important when you're guessing.