[Monotone-devel] Re: Branch naming conventions

[Bruce Stephens]

Florian Weimer <address@hidden> writes:

[...]

> This is just so true!  What's worse, there's no way to map an
> archive name to a network location, so you actually need those
> archives to find a hint to the location of an archive.

The problem is that you don't see the archive name, so not being able
to map the name to a network location isn't an issue.

One can map from a network location to an archive name (indeed, Arch
can do that conveniently).  So why not just forget the archive name,
and use the location (a URI)?  The idea (as far as I understand it) is
to deliberately provide a level of indirection, so that the same
archive name can map to different network locations for different
people (that's how mirroring is done: you register the remote archive
as <archive-name>-SOURCE, and mirror it to a local archive which you
(locally) refer to as <archive-name>).

> Nowadays, DNS could support such a mapping, using NAPTR records.
> However, library support is still rather poor, so it's quite a
> challenge to implement it.  (If the archive name is just a domain,
> it's a bit easier because you could use SRV records which are much
> simpler.)

There have been a number of ideas for such things, but none of them
can work as things stand (the non-unique mapping from archive names to
URIs is a feature that's used).  There are obvious ways you could
reengineer things (make explicit the idea of a primary URI for the
archive, and various mirrors), but that's adding complexity for (I
suspect) limited gain.  Given an archive name, it's relatively
straightforward to find a URI, I find, simply by using normal web
searching methods.