Re: [Gnu-arch-users] Registered and official names

[Aaron Bentley]

James Blackwell wrote:
> In lists.arch.users, you wrote:

> > What about using registered names for the read side?  Currently, "tla 
> > tag foo-MIRROR/version foo/version" succeeds, but will do the wrong 
> > thing, writing the registered name to the CONTINUATION file and logs. 
> > Also, it erronously caches a revision.
>
>
> I think you're right; there's enough challenged people out there that
> this particular corner should be padded. 
>
> I suggest the following: 
>
> if (provided_archive !=  `=meta-info/name`)
>   {
>     require --im-smarter-than-you flag;
>   }

That's one way of handling it, and considering how rarely we want to 
access -SOURCE or -MIRROR archives, it might be worth putting the check 
in arch_archive_connect().

On the other hand, it's certainly feasible to make tagging (or getting 
from) a registered name work properly.

> > However, if we had read caching, we wouldn't 
> > need local mirrors.  That would reduce the need to write to registered 
> > names, and we could just mandate the use of official names everywhere 
> > (except archive-mirror).  
>
>
> With mirroring, I have a guarantee that, at the time of mirroring, I
> have everything that is in the archive that I'm mirroring. 
>
> Would read caching make the same guarantees?  I'd hate to mirror an
> archive, head off to the park, and find out I'm missing something from
> that archive.

That's not a use case for me, and I most certainly don't want my cache 
to be a mirror.  I want the cache to prevent me from downloading 
something twice, but I don't want to download anything I might not need.

That said, the implementation would probably be a mirror variant, so it 
probably wouldn't be too hard to make archive-mirror work with caches.

Aaron

--
Aaron Bentley
Director of Technology
Panometrics, Inc.