Re: [Gnu-arch-users] [bug] import revisions should have a distinctive header in their patchlogs

[Aaron Bentley]

James Blackwell wrote:
> Aaron Bentley wrote:
>
> > Since currently, they are always at base-0, and *don't* have tag 
> > headers, it's pretty easy to determine whether a revision is an import 
> > revision from tree data.
> >
> > That will change if imports are permitted to occur anywhere (as Tom 
> > plans), and will make it impossible to get an accurate picture of 
> > ancestry from tree-local data.  Therefore, I suggest adding a 
> > distinctive header to the patchlogs of import revisions.
>
>
> What's the logic on allowing imports anyways? Offhand, I don't imagine
> it as a useful tool since in my mind it would break the chain of
> revisions, i.e.
>
> Base-0 -> Rev-1 -> Rev-2 -> Rev-3
>
> Base-0 -> Rev-1 -> Rev-2  Base-1 Rev-(3? 1?)

You mean Patch-1 etc?  Naming doesn't change.

So when this is done, you can have

Base-0  (import)
Patch-1 (simple revision)
Patch-2 (simple revision)
Patch-3 (import)
Patch-4 (simple revision)
Patch-5 (tag of patch-2)
Patch-6 (simple revision)
etc...

> Why wouldn't somebody just avoid this mess and just start a new branch?

It comes down to wanting to fix mistakes.  I got myself very confused 
with aba initially, because I wanted to use a particular branch name. 
So I kept trying to change history, doing it wrong, and getting bizarre 
changesets (since it was comparing against a previous timeline).

The two cases I can see:
1. Screwed up initial import (maybe they omitted stuff or put stuff in 
that didn't belong)
2. Change of tagging method (Import revisions should be smaller than the 
changeset that represents every file being deleted and re-added)

People are often very particular about the names they use.  Tom wants 
the namespace to be accomodating to users.

I'm not hugely in favour of this, but if it's going to happen, I'd like 
to be able to distinguish imports from simple revisions.

Aaron